Datasets:

Hendrik-a

/

arxiv

like 0

this article is a contribution to the understanding of fluctuations in the out of equilibrium dynamics of glassy systems . by extending theoretical ideas based on the assumption that time - reparametrization invariance develops asymptotically we deduce the scaling properties of diverse high - order correlation functions . we examine these predictions with numerical tests in a standard glassy model , the edwards - anderson spin - glass , and in a system where time - reparametrization invariance is not expected to hold , the ferromagnetic ising model , both at low temperatures . our results enlighten a qualitative difference between the fluctuation properties of the two models and show that scaling properties conform to the time - reparametrization invariance scenario in the former but not in the latter . pacs : 05.70.ln , 75.40.gb , 05.40.-a

1105.2949

in a series of papers the idea that time - reparametrization invariance ( tri ) should be the reason for large non - equilibrium fluctuations in glassy systems was introduced . this claim was substantiated with the analysis of the martin - siggia - rose - janssen - dedominicis ( msrjd ) action for langevin stochastic processes @xcite and by numerical simulations of various glassy models @xcite . the state of the art of these ideas is summarized in @xcite . the non - equilibrium situation we are concerned with is a dynamic process where a system is quenched below a transition or dynamic crossover with some protocol and it is let relax subsequently . the most interesting time regime after the quench is the one in which the size of the sample is the largest ( though still finite ) scale and times are long but not as much as to allow for equilibration . the system is then in an asymptotic non - equilibrium regime in which a kind of dynamic scaling is very often realized . in this stage the behavior of time - dependent observables is rather well understood @xcite . the sources of fluctuations in the dynamics of these systems are disorder in the initial conditions , thermal noise , quenched disorder and external forces , if present . in @xcite and references therein the proposal was to average over any sort of quenched randomness with the purpose of bringing together disordered systems such as spin - glasses , superconductors , or other and conventional glasses of atomic , molecular or other nature . external forces were set to zero and had no effect on the dynamics . one is then left with initial condition and thermally induced fluctuations only . in order to study their effects , beside the fully averaged value @xmath2 of physical quantities , one has to consider the properties of their fluctuating parts @xmath3 , to be defined properly , which may be sample dependent . the most complete characterization of fluctuations is then obtained through the probability distribution @xmath4 . in particular , since the relevance of fully - averaged two - time quantities such as the self - correlation @xmath5 function , the linear autoresponse @xmath6 and their fluctuating parts , to the issue of tri was underlined in @xcite , one is interested in the probability distributions of @xmath7 and @xmath8 studied as a function of @xmath7 , @xmath9 , @xmath10 , the system size and the other parameters in the model . this is , however , a formidable task . less ambitious but still very instructive is to analyze the first moments of the distributions , such as the generalized variances that we shall define in the body of this paper . this is , indeed , the route that we shall follow in this article . resorting to a langevin description , in sec . [ formalism ] we set the formalism for continuous variables systems and then present analytic expressions for these variances in terms of the msrjd fields , deriving scaling laws based upon the tri scenario . next , recalling previous results on linear and beyond - linear response theory @xcite , we relate these variances to quantities that can be measured in ising spin systems in sec . [ 3dea ] . in sec . [ numerical ] we put the tri predictions to the test in the edwards - anderson ( ea ) model in @xmath11 by means of extensive numerical simulations . the results , which refine and extend the analysis performed in @xcite , are consistent with the predictions of tri . in the same section we present another set of numerical simulations performed on a coarsening system , the ferromagnetic ising model in @xmath12 , for which tri is not expected to hold @xcite , showing a different pattern of fluctuations , already at a qualitative level . a summary and the conclusions are contained in sec . [ concl ] . the stochastic dynamics of continuous variables in a macroscopic system are usually described with a markovian langevin equation with additive white noise . the ensemble of such coupled stochastic equations can be recast into a path - integral called martin - siggia - rose - jenssen - dedominicis ( msrjd ) generating functional that is better adapted for a general type of analysis . this formalism has been reviewed in several publications @xcite so we shall not repeat the construction of the path - integral here . it reads @xmath13= \int { \cal d}\phi { \cal d}i\hat \phi \ e^{-s[\phi;\eta,\hat \eta ] } \ p_{ic}[\phi ( x , t_0)]\end{aligned}\ ] ] with @xmath14 = \int_{t_0}^\infty dt dx \left\ { i\hat \phi(x , t ) [ \dot\phi(x , t ) + \frac{\delta v[\phi]}{\delta \phi(x , t ) } - h(x , t ) ] + t ( i\hat \phi(x , t))^2 + \eta(x , t ) \phi(x , t ) + \hat \eta(x , t ) i\hat \phi(x , t ) \right\ } \label{eq : action}\ ] ] where , to simplify the notation , we focused on a one - dimensional theory described by a scalar field @xmath15 . @xmath16 is the initial time , when the system is set in contact with a thermal bath at temperature @xmath17 . we set @xmath18 and we absorb the friction coefficient into a redefinition of time . @xmath19 $ ] is the probability distribution of the field initial conditions . henceforth we shall not write the limits of the time - intervals explicitly . @xmath20 is the potential energy from which the deterministic force in the langevin equation derives . @xmath21 is an external perturbation that couples linearly to the field @xmath22 in the potential . @xmath23 and @xmath24 are two space and time dependent sources . the path integral runs over all possible configurations of the field @xmath22 with initial condition @xmath25 . the measure is defined as @xmath26 and we jump over subtle discretization issues that are discussed in many publications by recalling that we do not need to worry about any determinant contribution to the action . the integral over the auxiliary field @xmath27 is similarly discretized , namely @xmath28 . without loss of generality for our purpose , we take @xmath29 to be flat . we use a compact notation for the two fields , @xmath30 , in the action and hereafter we label @xmath31 and @xmath32 its components . due to the normalization of the thermal noise distribution the path - integral is identical to one , @xmath33=1 $ ] , in the absence of sources ; this feature will be useful later . noise averages translate into averages computed within the path - integral formulation with weight @xmath34}$ ] . for definiteness , and to parallel the discussion in @xcite and references therein ( to which we refer for further details ) , let us separate a quadratic term from the non - linear one , @xmath35 , in the potential @xmath36 = \frac{1}{2 } \int dx dy \ j(x , y ) \phi(x , t ) \phi(y , t ) + v_{nl}[\phi]\ ] ] and assume that the symmetric time - independent exchange matrix @xmath37 is made of independent elements each of them gaussian - distributed @xmath38 \propto \prod_{xy } \exp[-(1/2 ) \int dx dy \ j^2(x , y ) k^{-1}(x , y)]$ ] with @xmath39 the connectivity matrix . for a lattice model with first nearest - neighbours and no self - interactions , as the @xmath0 edwards - anderson spin - glass that we studied numerically , @xmath40 for nearest - neighbours and zero otherwise , with @xmath37 the exchange constant and @xmath41 the connectivity of the lattice . quenched disorder is treated on average by exploiting the fact that @xmath33=1 $ ] . concretely , one averages the dynamic generating functional over the @xmath37-distribution with no need to use the replica trick . this gives rise to a quartic term in the action of the form @xmath42&=&-\frac{1}{8 } \int dx dy \int dt_1 dt_2 \ \left [ i\hat\phi(x , t_1)\,\phi(y , t_1 ) + \phi(x , t_1)\,i\hat\phi(y , t_1 ) \right ] k(x , y ) \left [ i\hat\phi(x , t_2)\,\phi(y , t_2 ) + \phi(x , t_2)\,i\hat\phi(y , t_2 ) \right ] \nonumber\\ & = & -\frac{1}{8 } \int dx dy \int dt_1 dt_2 \ \phi^a(x , t_1)\phi^{\bar a}(y , t_1)\ , k(x , y)\ , \phi^b(x , t_2)\phi^{\bar b}(y , t_2 ) , \end{aligned}\ ] ] where @xmath43 and @xmath44 and @xmath45 ( here repeated indices are summed ) . this term can be decoupled by introducing four spatially local two - time fields @xmath46 , @xmath43 , with a hubbard - stratonovich transformation . one then has @xmath42 & = & \int { \cal d}q \exp \left [ \frac{1}{2 } \int dx dy \ k^{-1}(x , y ) \int dt_1 dt_2 \ q^{ab}(x;t_1,t_2 ) q^{\overline b\overline a}(y , t_2,t_1 ) \right . \nonumber\\ & & \qquad\qquad\qquad\qquad \left . - \int dx \int dt_1 dt_2 \ \phi^a(x , t_1 ) q^{ab}(x;t_1,t_2 ) \phi^b(x , t_2 ) \right ] \;. \label{eq : actionq}\end{aligned}\ ] ] the zero - source generating functional can now be written as @xmath47 = \int { \cal d}q { \cal d}\phi \ e^{-s[q,\phi]}$ ] where the action @xmath48 $ ] includes all terms , those in eq . ( [ eq : actionq ] ) that depend on @xmath49 and those in eq . ( [ eq : action ] ) with @xmath20 replaced by @xmath35 that depend on @xmath50 only . the tri arguments @xcite , which are briefly recalled in sec . [ sec : tri ] , rely on the transformation properties of the fluctuating @xmath49 fields that , in turn , translate into relations between the average values of products of the @xmath49 fields . these objects , though , are auxiliary in the calculation and are not directly accessible in a simulation or experiment . in order to proceed towards a test of tri the next step , to which this paragraph is devoted , is to relate the statistical properties of the hubbard - stratonovich fields @xmath49 to those of the field @xmath50 . in particular , our program is to find relations between averages of products of @xmath49s and moments of the @xmath50s from which directly measurable correlation and response functions will be constructed . to this end we perform an infinitesimal translation of the field @xmath51 , which induces a change in the action @xmath48\to s[q,\phi]+\delta s[q,\phi]$ ] , with @xmath52 & = & - \int dx \int dt_1 dt_2 \ \phi^a(x , t_1 ) \phi^b(x , t_2 ) \delta q^{ab}(x;t_1,t_2 ) + \int dx dy \ k^{-1}(x , y ) \int dt_1 dt_2 \ \delta q^{ab}(x;t_1,t_2 ) q^{\overline b\overline a}(y , t_2,t_1 ) \nonumber\\ & & \qquad \qquad + \frac{1}{2 } \int dx dy \ k^{-1}(x , y ) \int dt_1 dt_2 \ \delta q^{ab}(x;t_1,t_2 ) \delta q^{\overline b\overline a}(y , t_2,t_1 ) \ ; . \end{aligned}\ ] ] the normalization @xmath33=1 $ ] implies @xmath53 where @xmath54}$ ] . to first and second order in @xmath55 one has @xmath56 and @xmath57 respectively . the first equation tells us that the average of the hubbard - stratonovich @xmath49 field , coarse - grained over the region in which @xmath58 is different from zero , is equal to the correlations of two @xmath50 fields . these in turn are related to the usual correlation and linear response functions , as will be recalled in sec . [ sec : observables ] . note that the rhs of eq . ) [ eq : average - dq ] ) is symmetric under the exchange of @xmath59 and @xmath60 and this implies that the average in the lhs should also be . since we are not only interested in the average of @xmath49 , but also in their generalized variances ( which are given by averages of four fields , as discussed in sec . [ sec : observables ] ) , the content of the second order equation ( [ eq : average - dqdq ] ) is also relevant . indeed , the expansion of the product in the right - hand - side of the second equation yields three kinds of terms , one of which is precisely the average of the product of four @xmath50 fields . another one is the average of the product of two @xmath49s . the last cross term is the average of the product of two @xmath50 and one @xmath49 fields . using standard manipulations of the path - integral , this term can also be recast into averages over products of @xmath49 fields : @xmath61 putting everything together , eq . ( [ eq : average - dqdq ] ) allows one to write the average of four @xmath50 fields as sums of terms with averages of products of two spatially coarse grained @xmath49 fields . as it will be discussed in sec . [ sec : tri ] , tri provides predictions on the transformation properties of composite @xmath49-fields operators under a certain class of transformations . because we have just shown above that averages of @xmath49 operators are equal to moments of @xmath50 fields , we arrive at the following important relation : _ all physical observables which can be written in terms of moments of @xmath50 fields should inherit the symmetry properties of the @xmath49 fields . _ an approximate treatment of the msrjd action @xcite , inspired by the solution to mean - field glassy models and mode - coupling approximations @xcite , is based on the assumption that it can be separated into two kinds of terms , involving fast and slow components of the fields . in the slow contribution time - derivatives are neglected , integrals involving products of fast and slow contributions are approximated , and the action is fully written in terms of the @xmath49 fields only . the @xmath62 , @xmath63 indices are linked in such a way that the ensemble of slow terms are invariant under the global transformation @xmath64 , where @xmath65 is any positive and monotonic function of time . this transformation is a time - reparametrization . in terms of the @xmath49 fields it implies @xmath66 and , for the averages of @xmath49-field products @xmath67 equation ( [ eq : transf - prod - q ] ) describes how a generic composite @xmath49-field operator transforms under time - reparametrization . in general , quantities characterized by the same set of indexes @xmath68 transform in the same way under the time - reparametrization change . a particular case is that of quantities that are left unchanged by the transformation , that is to say , tri - quantities . these can be obtained by integrating a composite operator such as the ones entering eq . ( [ eq : transf - prod - q ] ) over each time associated to an @xmath69 or @xmath70 subscript . indeed , the prefactors are eliminated by a change of integration variable @xmath71 . for the cases @xmath72 and @xmath73 , recalling eqs . ( [ eq : average - dq ] ) and ( [ eq : cross - average ] ) , one has that averages of products of @xmath74 @xmath50-fields are tri after integration over the times pertaining to the @xmath75 fields . the autocorrelation function @xmath76 and the associated linear susceptibility @xmath77 , as well as the second moments of their fluctuating parts , belong to this class , as it will be further discussed in sec . [ sec : ave - and - fluc ] . these quantities are the subject of the numerical study carried out in sec . [ 3dea ] . in this subsection we define the quantities we study in this paper , both analytically and numerically . the connected two - point two - time correlation function is @xmath78 where @xmath79 . the averaged linear response of any observable is defined as its variation with respect to an applied perturbation , in the limit in which the latter vanishes . the most studied linear response is the one of the averaged field itself with respect to a perturbation @xmath21 which couples linearly to @xmath15 in the hamiltonian ( a magnetic field , regarding @xmath15 as a spin ) . a simple calculation shows that , within the msrjd formalism , it is simply related to an average of the product of the fields @xmath22 and @xmath80 : @xmath81 note that a factor @xmath17 has been added in the definition of the linear response to ease the notation here and in the following . the integral of this quantity is the dynamic susceptibility @xmath82 which in the following will be more simply denoted as @xmath77 , omitting the superscript @xmath83 . we define a particular second order response , @xmath84 , as the variation of the averaged composite field @xmath85 with respect to two instantaneous perturbations applied at @xmath9 and @xmath10 on the same spatial points on which the fields are evaluated . again , a simple calculation allows one to show that within the msrjd formalism this quantity equals an average of , in this case , four fields in the absence of the perturbation : r^(2,2)(x , y;t , t_1,t_2 ) t^2 . |_h=0 = t^2 ( x , t ) i(x , t_1 ) ( y , t ) i(y , t_2 ) . [ r2var ] in general we define generalized responses @xmath86 as the response of a composite operator containing the product of @xmath87-fields with respect to @xmath88 magnetic fields . these are functions of @xmath89 spatial positions and @xmath89 times , but to keep the notation simple we will omit writing coinciding arguments ( for instance we will write @xmath90 instead of @xmath91 and @xmath92 instead of @xmath93 ) . quite generally , a variation with respect to the field @xmath21 can be replaced by a response field @xmath80 to compute higher - order response functions and , according to the convention introduced above , we include @xmath88 factors of @xmath17 in the definition of the generalized response functions . the next step is to introduce a couple of local fluctuating quantities , @xmath94 and @xmath95 , with the property @xmath96 and @xmath97 , to be interpreted as the _ fluctuating parts _ of @xmath76 and @xmath77 . since different fluctuating quantities can share the same average , this can be done in different ways . a detailed discussion of this issue is exposed in @xcite . for continuous variables described by a langevin equation , recalling the formalism of sec . [ formalism ] , one is naturally led to consider @xmath98 as being the variables whose msrjd average yield @xmath76 and @xmath77 . still , @xmath8 involves the auxiliary field @xmath80 and , in consequence , it is not directly accessible in an experiment or a simulation . this problem can be bypassed by studying the probability distribution of @xmath99 through its moments , since , as we show below , these can be expressed in terms of the field @xmath22 alone relating them to generalized response functions . starting from the lowest non - trivial moments , with the definitions ( [ eq : fluc_parts ] ) one can build the _ variances _ @xmath100 where for a generic quantity @xmath101 we have defined the fluctuation @xmath102 . as anticipated above , the variances containing @xmath103 can be expressed in terms of the _ physical _ field @xmath22 alone , by expressing them in terms of generalized responses . for instance , the variance ( [ eq : vchichi ] ) can be related to a second - order response , as given by eq . ( [ r2var ] ) , @xmath104 % [ \phi(y , t)i\hat \phi(y , t_2)-\langle \phi(y , t)i\hat \phi(y , t_2)\rangle ] % \rangle % \nonumber\\ % & = & \int_s^t dt_1 \int_s^t dt_2 \ [ r^{(2,2)}(\vec x,\vec y;t , t_1,t_2 ) - r^{(1,1)}(\vec x;t , t_1)r^{(1,1)}(\vec y;t , t_2 ) ] \ ; . \end{aligned}\ ] ] analogously , the co - variance @xmath105 can be expressed as @xmath106\ ; [ \phi(\vec y , t ) i\hat\phi(\vec y , t_1)-\langle\phi(\vec y , t ) i\hat\phi(\vec y , t_1)\rangle ] \;\rangle \nonumber\\ & = & \int_s^t dt_1 \ [ r^{(3,1)}(\vec x,\vec y;t , s , t_1 ) % = \phi(x , t ) \phi(x , s)\phi(y , t ) i\phi(y , t_1)\rangle - - c(\vec x;t , s ) % = \langle\phi(x , t ) i\phi(x , s)\rangle r^{(1,1)}(\vec y;t , t_1 ) % = \langle\phi(y , t ) i\phi(y , t_1)\rangle ] , \end{aligned}\ ] ] where @xmath107 is the autocorrelation function and @xmath108 . in this paper , we enforce these relations to compute the moments ( [ eq : vcc])-([eq : vchichi ] ) numerically . higher order moments are exceptionally demanding to compute , and we leave their study to further works . in order to improve the statistics of the concrete numerical measurements it will be convenient to compute the double spatial integral : @xmath109 where @xmath110 is a compact notation to denote the variances @xmath111 , @xmath105 and @xmath112 , simultaneously , @xmath113 is the spatial dimension and @xmath114 is the volume of the system . a comment is now in order . in spite of the name we gave to it , @xmath112 is not the variance of a physical quantity ; indeed , being linked to response functions , it can even take negative values . still , in the following we shall broadly use the term _ co - variances _ ( or simply variances ) when referring to @xmath111 , @xmath105 and @xmath112 . let us now come back to what tri implies for the quantities introduced above , following the prescriptions of sec . [ sec : tri ] . we start from the usual correlation @xmath115 and linear response @xmath116 . the former is written in terms of two @xmath22 fields , in terms of @xmath49 s it is then a function of @xmath117 only and it is invariant under eq . ( [ eq : transformation - tri ] ) . @xmath118 involves , besides one field @xmath22 , also one @xmath119 field , but there is an integral over the time argument of the latter ; thus @xmath77 does not transform either . proceeding analogously for the variances , @xmath111 involves only @xmath22 fields , @xmath105 and @xmath112 involve one and two fields @xmath119 respectively , together with one and two integrations over the corresponding time variables . then , also the variances are tri , similarly to @xmath76 and @xmath77 . a sum rule that generalizes fdt to higher powers is easily proven using the invariance of the action and the measure under the transformation @xmath120 and @xmath121 in equilibrium @xcite . indeed , @xmath122^n \right\rangle = \left\langle \left [ \int d^d x \left ( \widehat { \delta \chi}(\vec x ; t , s ) - \widehat{\delta c}(\vec x ; t , t ) + \widehat{\delta c}(\vec x ; t , s ) \right ) \right]^n \right\rangle , \ ] ] for @xmath123 . the first identity is due to causality and the second one is derived by using the symmetry . as a special case one has @xmath124 valid for @xmath73 and for variables such that @xmath125 , e.g. ising spins . for a generic fluctuating observable @xmath3 , besides the standard full average @xmath126 we also introduce the restricted average @xmath127 , namely an average taken only over the instances ( realizations ) with a given value @xmath7 of the fluctuating autocorrelation , as suggested in @xcite . clearly one has x=_cx_c p(c ) , where @xmath128 is the marginal probability that a chosen dynamical trajectory has autocorrelation @xmath7 . for a co - variance between two quantities @xmath129 and @xmath130 one has v^ab=_cv^ab_c p(c)+v^a b [ compos_var ] where @xmath131 is the restricted covariance of @xmath129 and @xmath130 ( for @xmath7 fixed ) , and @xmath132 is the covariance of the restricted averages @xmath133 and @xmath134 as @xmath7 is varied . in the late stage of the evolution of slowly relaxing systems , some sort of dynamical scaling is usually obeyed . indeed this is what we will find in our simulations . in this section , we discuss the implications of tri in such systems . let us start considering a couple of generic local observables @xmath135 , @xmath136 , such as the autocorrelation @xmath76 or the autoresponse @xmath77 , with scaling expressions @xmath137,\\ b(t , s)&=&\xi(s)^{b_b}f_b\left [ \frac{\xi ( t)}{\xi ( s)}\right ] , \end{aligned}\ ] ] where @xmath138 is a certain function of time . we emphasize that we always refer to the slow degrees of the dynamical process and that any explicit fast contribution must be subtracted out if present . in some cases , as in coarsening systems , @xmath139 can be readily interpreted as a growing correlation length , although this is not crucial in what follows . clearly , if the two quantities transform in the same way under tri it must be @xmath140 and @xmath141 in the double limit @xmath142 and @xmath143 , i.e. @xmath144 . if one time , say @xmath145 , can be eliminated in favor of @xmath146 , the quantity @xmath147 can also be expressed as @xmath148 $ ] , where @xmath149 , and one can introduce the _ slope _ x_ab(b , s)= = . ( s)^-b_b+b_a _ z=(s)^-b_bb . [ eq : x ] under time - reparametrization @xmath64 one has @xmath150 , \\ b(t , s)&\to & b_{tr}(t , s)=\xi(h(s))^{b_b } \ f_b\left[\frac{\xi ( h(t))}{\xi ( h(s))}\right ] \label{eq : b}.\end{aligned}\ ] ] defining @xmath151 similarly to what we did for @xmath152 , one has @xmath153 $ ] , and x_a_trb_tr(b_tr , s)== . ( h(s))^-b_b+b_a _ z=(h(s))^-b_bb_tr . [ eq : x_tr ] comparing eqs . ( [ eq : x ] ) and ( [ eq : x_tr ] ) we conclude that , if the two quantities transform in the same way , namely @xmath140 , the slope of the original parametric curve , @xmath154 , evaluated at a value @xmath146 is equal to the slope of the transformed parametric curve , @xmath155 , evaluated at @xmath156^{-b_b}b$ ] . in other words , under time - reparametrization @xmath146 and @xmath147 are transformed in such a way as to preserve the original curve with slope @xmath157 , although different parts of the curve are related by the transformation . considering the self - correlation and the self - response , @xmath158 and @xmath159 , @xmath160 turns out to be the so - called fluctuation - dissipation ratio @xmath161 , @xmath162 being usually denoted as the _ effective temperature _ ( here we simply quote the term , without entering into the delicate problem of its interpretation as a physical temperature ) @xcite . in this case , since according to the predictions of tri @xmath77 and @xmath76 are both invariant under time - reparametrization , eq . ( [ eq : x_tr ] ) implies a finite @xmath162 . on the other hand , in systems where tri is not obeyed @xmath76 and @xmath77 may transform differently and this prevents @xmath162 from being finite . this leads to the conjecture that tri may not be obeyed in coarsening systems with an infinite @xmath162 , as it was analytically shown in a solvable case in @xcite . for the ising model this issue will be discussed in sec . [ comparison ] . let us consider now non - local quantities such as , for instance @xmath163 . since spatial translational invariance as well as isotropy are respected in the mean , they are actually functions of @xmath164 only . assuming scaling , one has v^xy(x , y;t , s)=(s)^b_xy f_v^xy . [ eq : scald ] the @xmath165 component of such a quantity is then expected to scale as @xmath166 . \label{eq : scaldk}\ ] ] in sec . [ numerical ] we shall study the behavior of the variances @xmath167 , hereafter simply referred to as @xmath110 , numerically . since tri predicts that all the @xmath110 should transform in the same way , we expect them to scale as @xmath168 , \label{eq : scalvk}\ ] ] with a _ unique _ exponent @xmath169 . in this section we adapt the definitions of the variances and restricted averages presented in the previous section to discrete variables systems . we consider a spin system described by the hamiltonian h()=-_ijj_ij_i _ j , [ ham ] where @xmath170 are spin variables located on the @xmath171 sites @xmath172 of a cubic lattice of linear size @xmath173 . @xmath174 denotes a spin configuration and @xmath175 are the couplings between nearest neighbor sites @xmath176 . a constant @xmath177 defines the ferromagnetic ising model while in the ea model we choose random bimodal couplings , @xmath178 , with zero mean . in the following we shall assume @xmath179 . the evolution of the system is governed by the transition rates @xmath180 for going from a configuration @xmath174 to another @xmath181 . in the following we consider the dynamics whereby single spins are updated , namely @xmath182 . @xmath183 is the transition rate for flipping @xmath184 , and @xmath185 and @xmath181 may differ in the @xmath172-th spin only . we use @xmath183 of the glauber form @xmath186 $ ] , where @xmath187 is the energy variation due to the proposed spin flip and @xmath17 is the temperature of the heat bath . the autocorrelation function is defined as c(t , s)=_i(t)_i(s ) . here we have implicitly assumed @xmath188 @xmath189 , and @xmath190 denotes , for the spin glass model , the double average over thermal histories and initial conditions ( denoted by @xmath191 ) and over the quenched realization of the random couplings @xmath175 ( denoted by @xmath192 ) ; for the ferromagnetic ising model only the first average is present . the associated susceptibility is ( t , s)=_s^t dt_1 r^(1,1)(t , t_1 ) , where r ^(1,1)(t , s ) = t ._h=0 is the impulsive auto - response function and @xmath193 is an instantaneous magnetic field switched on and off at time @xmath194 . these averaged quantities do not depend on the position @xmath172 and we thus omit the @xmath172 subscript . in order to study fluctuating two - time quantities in spin models and to put the tri to the test , we have to provide expressions for the variances ( [ eq : vcc])-([eq : vchichi ] ) in systems with discrete variables , where the path integral representation discussed in sec . [ sec : generating - functional ] is not available . following @xcite , this can be done by using an extension of the fluctuation - dissipation theorem @xcite which allows one to relate response functions of any order to correlation functions in systems out of equilibrium . to linear order one has @xmath195 , \label{1.5}\ ] ] where @xmath196 w_i(\sigma ' \vert \sigma)$ ] . it easy to show that this operator rules the evolution of the local magnetizations , namely @xmath197 . it can also be shown @xcite that the expression ( [ 1.5 ] ) holds also in a langevin description if the analogue of @xmath198 , namely the deterministic force @xmath199 , is used . this strengthens the relation between the discrete variable formalism of this section and the one introduced in sec . [ formalism ] . with this fluctuating susceptibility and @xmath200 one readily builds @xmath111 and @xmath105 . the procedure to determine the expression of @xmath112 is more subtle as we now discuss . first , this quantity is related to the second order response in eq . ( [ r2var ] ) , which for discrete spins on a lattice is more properly written as r^(2,2)_ij(t , t_1,t_2 ) t^2 . |_h=0 , [ app1.2 ] by v^_ij(t , s)_s^t dt_1_s^t dt_2 . [ chi2r2 ] proceeding as before , we express @xmath84 in terms of correlation functions through the extension of the fluctuation - dissipation theorem mentioned above . using the expression for the second order response ( [ chi2r2 ] ) derived in @xcite one arrives at @xcite @xmath201[\sigma_j(t)-\sigma_j(s)]\rangle - \frac{1}{4 } \int_{s}^t dt_2 \ \langle\sigma_i(t)\sigma_j(t)[\sigma_i(t)-\sigma_i(s)]b_j(t_2)\rangle \nonumber \\ & & -\frac{1}{4 } \int_{s}^t dt_1 \ \langle\sigma_i(t)\sigma_j(t)b_i(t_1)[\sigma_j(t)-\sigma_j(s)]\rangle + \frac{1}{4 } \int_{s}^t dt_1 \int_{s}^t dt_2 \ \langle\sigma_i(t)\sigma_j(t)b_i(t_1)b_j(t_2)\rangle \nonumber \\ & & + \frac{1}{4}\delta_{ij}\int_{s}^t dt_1 \ \langle\sigma_i(t)^2 \tilde{b}_i(t_1)\rangle -\chi(t , s)\chi(t , s ) , \label{app1.4}\end{aligned}\ ] ] where @xmath202 ^ 2w(\sigma'|\sigma)$ ] . in this way all the variances are written in terms of the spin configurations and they can be computed in a numerical experiment . this method to compute response functions in the unperturbed evolution [ i.e. eqs . ( [ 1.5 ] ) and ( [ app1.4 ] ) ] , besides the advantage of having the vanishing perturbation limit built in , is by far more efficient @xcite as compared to the standard method where the perturbation is actually switched on , particularly for higher order responses , as discussed in @xcite . the accuracy of the results presented in sec . [ numerical ] would not have been possible without resorting to this technique . in the following we focus on the variances @xmath165 component , defined analogously to eq . ( [ eq : k=0 ] ) , v^xy(t , s)= _ ijv^xy_ij(t , s ) , [ kappazero ] where to ease the notation we omitted the under - script @xmath165 . notice that @xmath111 is also ( apart from an overall factor @xmath203 ) the variance of the space average of the autocorrelations , @xmath204 , and it is hence bound to be positive . this quantity is related to the one measured in @xcite with the only difference that here we take the coarse - graining linear size , @xmath205 , to be equal to the size of the system , @xmath206 . we recall that in equilibrium the variances are related by eq . ( [ eq : sumrule ] ) @xcite . the definitions in sec . [ subsubsec : restricted ] can be straightforwardly applied to problems with discrete variables . in this section we present the results obtained with monte carlo simulations of the @xmath0 edwards - anderson model and the @xmath1 ferromagnetic ising model . we choose to present , first , the bare data for the @xmath0 ea model and their scaling analysis . only later we interpret these forms in terms of tri . while doing so we confront to the @xmath1 ising model data . in this section we study the @xmath110s when the spin glass is in equilibrium at @xmath207 . the knowledge of their behavior will be useful in the interpretation of the non - equilibrium results , described in sec . [ secneq ] . with this perspective in mind , we selected very small sample sizes to allow for a rapid equilibration . we simultaneously monitored finite - size effects as explained below . these conditions led us to work with @xmath208 . in the following , time is measured in monte carlo steps ( mcs ) . the qualitative behavior of the @xmath110s and the role of the system size can be understood by inspection of fig . [ fig_eq_1 ] . equilibrium dynamics is stationary so the three @xmath110 depend on the time - difference only . @xmath111 is positive , for the reasons discussed in sec . [ secbasic ] . initially ( soon after a microscopic time ) it grows as a power - law , @xmath209 , and after a characteristic time difference @xmath210 it flattens and attains a constant value @xmath211 . this shows that the @xmath173-dependence of @xmath212 is a finite size effect due to @xmath213 . because of the rapid increase of the characteristic time @xmath214 with @xmath173 , the asymptotic value is fully reached only for @xmath215 in our simulations . for @xmath216 the flattening is completely out of reach . from the analysis of this dynamic curve we extract the value @xmath217 for the power - law exponent . the covariance @xmath105 is negative . this can be understood since the equilibrium parametric plot @xmath218 is a curve with local negative slope . apart from the sign , this quantity behaves initially as @xmath111 , exhibiting a power - law growth with an exponent that is compatible with the one found for @xmath111 . around @xmath214 the quantity @xmath105 reaches a minimum and then grows towards an asymptotic value @xmath219 which can be shown to be @xmath220 in cases in which the system is not magnetized , @xmath221 @xcite . indeed , in the case @xmath215 , where @xmath214 is sufficiently short to let the system reach the asymptotic value , one has a very small final numerical value of @xmath219 . finally , @xmath112 is initially positive and grows proportionally to the other two variances . around @xmath214 it reaches a maximum and then decreases to a negative asymptotic value given by @xmath222 @xcite . in short , the initial time - difference dependence is given by @xmath223 let us stress that curves with different @xmath173 collapse in this power - law regime . hence , for instance , the curves obtained for @xmath216 are also representative of a system with @xmath224 at least up to times @xmath225 mcs . this fact will be used in the next section , where a sample with @xmath226 will be studied up to @xmath227 mcs . the sum rule ( [ eq : sumrule ] ) is satisfied at all times within our numerical accuracy . we now turn to the study of @xmath110 in out of equilibrium conditions . we performed a quench from @xmath228 to the working temperature @xmath207 at time @xmath229 and we followed the relaxation of a system of size @xmath226 up to @xmath230 mcs . we checked that in the range of times accessed in our simulations the system is free from finite size effects and remains out of equilibrium . this is probably due to the fact that , although some kind of growing length @xmath231 is present in the system @xcite , in the range of times considered here it is still of the order of a few lattice spacings ( say 3 or 4 at most ) . the choice of the system size and range of times is dictated by the necessity to tame the colossal noise afflicting @xmath112 in order to have sufficiently clean data . indeed , increasing the system size or running for longer time enhances the noise . concerning size , this is due to the fact that only the couples @xmath232 within a distance of order @xmath231 in eq . ( [ kappazero ] ) contribute significantly to the sum , as expressed by eq . ( [ eq : scald ] ) , while the others introduce noisy contributions ( in spite of having zero average ) . regarding time , as @xmath231 increases the number @xmath233 of independent regions decreases and this also enhances the noise in the course of time furthermore , being two - time correlations , the @xmath110 decay to zero as @xmath234 increases , making the signal / noise ratio worse . the data presented below are obtained by averaging over more than @xmath235x@xmath236 realizations of thermal histories , initial conditions and coupling constants . the behavior of the variances is shown in fig . [ fig_neq_0 ] . one finds @xmath237 and @xmath238 , basically for the same reasons discussed in sec . [ seceq ] , while @xmath112 is positive . for small time differences @xmath239 one finds a quasi - equilibrium behavior : the variances depend on the time difference @xmath234 only , they superimpose on the curves found in equilibrium ( for the largest size @xmath216 , see the discussion at the end of sec . [ seceq ] ) and the relation ( [ eq : sumrule ] ) is verified ( see the inset ) . in short , one has |v^xy(t , s)|(t - s)^ 0.4 [ eq - scaling ] for larger values of @xmath234 the system enters the aging regime in which non - equilibrium effects become important . one observes an explicit dependence on the two times , and a departure from the equilibrium behavior . in this regime , the @xmath110 start growing faster than the power - law characteristic of the equilibrium behavior . for still larger values of @xmath234 , @xmath111 and @xmath105 continue their growth steadily but in a slower logarithmic way ( see fig . [ fig_neq_1 ] ) , @xmath240 . the exponents @xmath241 were found to be @xmath242 and @xmath243 in the region @xmath244 . on the other hand , @xmath112 seem to saturate to a constant value ( or there could be a maximum after which it would decay , we can not exclude this possibility ) . the difference of the scaling function with the ones of the other variables remains , however , only logarithmic and , in some sense , marginal . notice that the equilibrium part of the variances is not negligible in the aging regime . instead , increasing as a power - law , it is expected to be asymptotically dominant ( although we can not access such a long time behavior in the simulations ) , since @xmath110 either converge to a constant value ( @xmath112 ) or grow logarithmically ( @xmath111 and @xmath105 ) . interestingly , as shown in fig . [ fig_neq_1 ] , all variances obey the scaling form v^xy(t , s ) s^a f_xy(t / s ) [ scaling ] with the _ same _ exponent @xmath245 , the same dependence on the two times @xmath145 and @xmath246 through their ratio in the remaining factor , but different scaling functions . this fact has important consequences towards the tests of the tri scenario as we shall discuss in sect . [ numvsrep ] . the restricted average of the susceptibility , @xmath247 , is shown in fig . [ fig_neq_3 ] for three values of the two - field product , @xmath248 , @xmath249 , and @xmath250 . interestingly enough , for all @xmath7 the sign of the restricted variance is negative , namely the opposite of the sign of the whole @xmath112 . from the parametric plot of @xmath118 versus @xmath76 ( included in the upper inset ) one concludes that the value @xmath251 belongs to the quasi - equilibrium regime , where @xmath252 , whereas @xmath253 is far into the aging regime . the value @xmath249 is somehow in between . for the choice @xmath248 the curves for different values of @xmath246 collapse , showing that @xmath254 depend only on @xmath234 and , moreover , the data are well described by the power - law @xmath255 with an exponent that is consistent with @xmath256 as found for the full variance @xmath112 in equilibrium and in the quasi - stationary regime . therefore for values of @xmath7 in the quasi - equilibrium regime of the parametric plot @xmath77 vs. @xmath76 , the restricted averages have an equilibrium behavior ( even for times @xmath257 where the whole @xmath258 falls out of equilibrium ) . moving towards the aging regime , for @xmath249 and @xmath253 the data collapse is progressively lost , starting from the largest values of @xmath259 . interestingly , while time translation invariance is lost , the scaling of the data transform into the non - equilibrium one , eq . ( [ scaling ] ) , found for the totally averaged variances , with the same exponent @xmath260 . this is shown in the lower inset of fig . [ fig_neq_3 ] , where we scaled the restricted averages by plotting @xmath261 against @xmath262 . the collapse is quite good for @xmath253 and it gets worse for increasing @xmath263 . the reason why the collapse for @xmath248 , although definitely worse than the one in the main panel , is not totally lost is due to the fact that the behavior @xmath255 can be written in the scaling form ( [ scaling ] ) with @xmath264 in place of @xmath62 . since the exponents @xmath62 and @xmath264 are not too different one would need a much larger range of @xmath246 in order to resolve the different curves in the lower inset of fig . [ fig_neq_3 ] . notice that for @xmath7 small enough , when the restricted averages scale as in eq . ( [ scaling ] ) the scaling function seems to be similar to the one of the whole @xmath112 , apart from the change in sign . this statement is proven by comparing @xmath265 with @xmath266 ( dotted line in the inset ) . in this section we re - consider the numerical results of sec . [ eanumeric ] for the spin glass and we discuss whether , how , and to which extent they may be interpreted as supporting tri in the relaxation dynamics of the @xmath0 ea model . we also report , in sec . [ comparison ] the result of simulations of coarsening systems , where tri is not expected to hold , to signal the differences with the dynamics of the disordered model . a first observation regards the transformation properties of the variances @xmath110 . we have shown in sec . [ formalism ] that if tri holds the variances must all transform in the same way , analogously to what happens for the usual two - time functions @xmath76 and @xmath77 . according to the discussion of sec . [ sec : scaling ] , when scaling holds eq . ( [ eq : scalvk ] ) must be obeyed with a unique exponent @xmath267 , independently of the choice of @xmath268 . our findings , expressed in eq . ( [ scaling ] ) , show that scaling holds with an algebraic function @xmath138 , and that all the variances scale with the same exponent @xmath260 . this result , therefore , is in agreement with the predictions of tri . physically , the scaling ( [ scaling ] ) with a single exponent @xmath269 amounts to say that , fixing @xmath262 , the variances of @xmath7 and @xmath8 are proportional . this means that basically the fluctuations of @xmath8 are triggered by those of @xmath7 ( or vice versa ) . as we shall discuss in sec . [ comparison ] the situation is very different in coarsening ferromagnets . the behavior of the restricted variances may help us better understand this point . as expressed above , if tri holds the non - equilibrium fluctuations of @xmath8 and @xmath7 are deeply related , and one can modify the former by manipulating the latter . restricted averages are a tool to produce an extreme perturbation of the fluctuations of @xmath7 , because we impose @xmath270 . as discussed in sec . [ secneq ] this radically changes the behavior of the fluctuations of @xmath8 , since even the sign of @xmath271 is the opposite of the one of @xmath112 . the change in sign and the fact that @xmath254 also scales according to eq . ( [ scaling ] ) , imply that ^ > _ cv^_c p(c ) , [ proper_sg ] from eq . ( [ compos_var ] ) . this means that the most relevant source of fluctuations of @xmath8 are the fluctuations of @xmath7 , while the fluctuations of @xmath8 which are independent of the variations of @xmath7 are less important . again , this property is not found in coarsening systems , as we shall show in sec . [ comparison ] . tri is not expected to hold in coarsening systems quenched below @xmath272 @xcite where an infinite effective temperature @xcite is developed . while the features of the @xmath0 ea spin - glass discussed insofar basically agree with what one would expect in the presence of tri , in this section we show that the same analysis performed on a phase - ordering system , the ising model , leads to a striking disagreement . the behavior of the variances in the ising model quenched below @xmath272 was studied numerically in @xcite . a scaling form analogous to eq . ( [ scaling ] ) was found , with exponents compatible with @xmath273 , @xmath274 in @xmath275 ( and conjectured to hold for higher dimensionality ) . the scaling functions @xmath276 decay as @xmath277 with @xmath278 , @xmath279 , and @xmath280 . this shows that , except at the lower critical dimension @xmath281 where the anomaly of a finite effective temperature is found @xcite , the variances scale with different exponents . moreover , the behavior of the scaling functions is also very different . as discussed in sec . [ formalism ] , this excludes the possibility of tri . the restricted averages were not studied in phase ordering kinetics so far . in order to establish how generic the behavior observed in the @xmath0 ea model is , we computed these quantities in coarsening systems . we simulated the ising model on a @xmath12 square lattice , after a quench from infinite temperature to @xmath282 . the system size is @xmath283 . as compared to the spin - glass , this relatively large value is needed to avoid finite size effects , since the size of correlated regions grows much faster in this problem . as a consequence , large fluctuations of @xmath7 around the average are very unlikely . then , when computing @xmath271 we can collect a significant statistics only in a narrower time interval than in the spin - glass case . this can be seen in fig . [ fig_neq_ferro ] , where the behavior of @xmath112 and @xmath254 are shown . for @xmath112 , we basically reproduce the results in @xcite ( where however a quench to @xmath284 was considered ) . this quantity is positive and behaves as @xmath285 , where @xmath286 is a function which grows as @xmath287 for large @xmath41 . interestingly , the restricted average @xmath254 not only has the same sign as the global average , but it superimposes almost exactly on @xmath112 , for @xmath234 sufficiently large ( in the aging regime ) . this means that in the non - equilibrium regime @xmath112 is completely determined by @xmath254 . according to eq . ( [ compos_var ] ) , this implies ^ _ cv^_c p(c ) , [ proper_ferro ] which should be compared to the property ( [ proper_sg ] ) found in the @xmath0 ea model . this indicates that the non - equilibrium fluctuations of @xmath118 are entirely produced at fixed @xmath7 , and are then independent of those of @xmath7 . this behavior , which is qualitatively different from the one observed in the @xmath0 ea model , is a clear demonstration of the lack of tri in this system . this paper is devoted to the study of the properties of fluctuations in aging systems . specifically , we concentrate on the fluctuations of the autocorrelation and dynamic susceptibility , since the relevance of their relations to the issue of tri was pointed out in @xcite . in the msrjd path - integral formulation of langevin processes , which is overviewed in the first part of the paper , response functions are associated to the auxiliary field @xmath80 and hence their fluctuating parts can not be expressed in terms of the physical field @xmath15 alone . a possible way out of this limitation is to consider moments of such fluctuating parts , since these can be related to generalized response functions which are directly measurable . in particular , in this paper we focused on the _ variances _ @xmath110 , because higher moments are numerically out of reach even with the highly efficient method adopted here , based on the generalization of the fluctuation - dissipation theorem out of equilibrium and beyond linear order . the behavior of these variances fits in the general framework of two - time quantities in aging systems . for short time differences a quasi - equilibrium stationary behavior is observed , while in the aging regime time - translation invariance is lost and is substituted by a different , scaling symmetry with a unique exponent @xmath288 for all the variances . in the analytic part of the paper we derived the scaling properties of the variances and we showed that such a unique exponent is a prediction of tri . the numerical results in the @xmath0 ea therefore gives support to tri . in order to better characterize the fluctuations , we also computed the restricted average @xmath254 . the values of the restriction @xmath7 is an additional parameter that tunes the crossover from the quasi - equilibrium behavior to the fully aging regime . in particular , the latter is obtained letting @xmath7 be small ( with respect to the ea order parameter ) . in this case , the restricted average scales as the fully averaged one , but its actual value is not the main contribution to the latter . these results have to be compared to the ones found in the ising model . in this clean coarsening case the exponents @xmath267 do depend on the fluctuating observables and the restricted average @xmath254 is of the same order as the full average @xmath112 , showing that fluctuations of @xmath7 are not the main source for those of @xmath8 , at variance to what it is found in the disordered model . these qualitative and quantitative differences between these two systems add to differences found in the relation between the averaged linear response and the self - correlation , i.e. the effective temperature @xcite . our results suggest that the fluctuating out of equilibrium dynamics of the @xmath0 ea model and of coarsening systems are different and that this can be related to the fact that tri is realized in the former model while it is not in the latter . this should also be linked to the fact that the effective temperature is finite in the former and infinite in the latter cases . these ideas should be checked numerically and analytically in other out of equilibrium systems . natural cases to analyze are the dynamics of atomic glass models ( like soft sphere , lennard - jones mixtures or other ) with molecular dynamics or the evolution of kinetically constrained models @xcite . the glauber ising chain is an example that could be dealt with , possibly , analytically , exploiting the techniques derived in @xcite . studies of critical dynamics @xcite and of the relaxation of a model at its lower critical dimension should help to complete this picture . we thank e. lippiello , a. sarracino , and m. zannetti for many useful discussions . f. c. acknowledges financial support from prin 2007jhlpez ( _ statistical physics of strongly correlated systems in equilibrium and out of equilibrium : exact results and field theory methods _ ) and from universit pierre et marie curie and thanks the lpthe jussieu for hospitality during the preparation of this work . c. c. and l. f. c. thank the universidad de buenos aires , argentina , for hospitality during the last stages of this work . this work was supported in part by doe grant defg02 - 06er46316 ( c. c. ) . h. e. castillo , c. chamon , l. f. cugliandolo , and m. p. kennett , phys . lett . * 88 * , 237201 ( 2002 ) . h. e. castillo , c. chamon , l. f. cugliandolo , j. l. iguain , and m. p. kennett , phys . b * 68 * , 134442 ( 2003 ) . h. e. castillo , phys . rev . b * 78 * , 214430 ( 2008 ) . g. a. mavimbela and h. e. castillo , _ time reparametrization invariance in arbitrary range p - spin models : symmetric versus non - symmetric dynamics _ , arxiv:1011.2225 . a. parsaeian and h. e. castillo , phys . e * 78 * , 060105(r ) ( 2008 ) . h. e. castillo and a. parsaeian , nature physics * 3 * , 26 ( 2007 ) . k. e. avila , h. e. castillo , and a. parsaeian , _ mapping dynamical heterogeneity in structural glasses to correlated fluctuations of the time variables _ , arxiv:1007.0520 . l. f. cugliandolo , _ glassy dynamics _ , in slow relation and non equilibrium dynamics in condensed matter , les houches session 77 , j - l barrat , j. kurchan , m. v. feigelman , and j. dalibard eds . ( springer - verlag , berlin , 2002 ) . g. biroli , j. stat . ( 2005 ) p05014 . h. k. janssen , z. phys . b * 23 * , 377 ( 1976 ) . h. k. janssen in _ dynamical critical phenomena andrelated topics _ , ed . by c. p. enz ( springer , berlin - heidelberg , 1979 ) . h. k. janssen in from phase transition to chaos " , ed . by g. gyrgyi , i. kondor , l. sasvri , t. tl ( world scientific , singapore , 1992 ) . for colored and multiplicative noise cases see c. aron , g. biroli , and l. f. cugliandolo , j. stat . p11018 ( 2010 ) . f. corberi , e. lippiello , and m. zannetti , phys . e * 72 * , 056103 ( 2005 ) . n. andrenacci , f. corberi , and e. lippiello , phys . e * 73 * , 046124 ( 2006 ) . e. lippiello , f. corberi , and m. zannetti , phys . rev . e * 74 * , 041113 ( 2006 ) . f. corberi , e. lippiello , and m. zannetti , phys . e * 74 * , 041106 ( 2006 ) . r. burioni , d. cassi , f. corberi , and a. vezzani , phys . e * 75 * , 011113 ( 2007 ) . f. corberi , e. lippiello , and m. zannetti , j. stat . mech . p07002 ( 2007 ) . f. corberi , a. gambassi , e. lippiello , and m. zannetti , j. stat . p02013 ( 2008 ) . f.corberi , e.lippiello , a.mukherjee , s.puri , and m.zannetti , j. stat . p030016 ( 2011 ) . d. a. huse , j. appl . phys . * 64 * , 5776 ( 1988 ) ; phys . b * 43 * , 8673 ( 1991 ) . h. rieger , j. phys . a * 26 * , l615 ( 1993 ) . h. rieger , b. steckemetz , and m. schreckenberg , europhys . * 27 * , 485 ( 1994 ) . j. kisker , l. santen , m. schreckenberg , and h. rieger , phys . b * 53 * , 6418 ( 1996 ) . e. marinari , g. parisi , f. ricci - tersenghi , and j. j. ruiz - lorenzo , j. phys . a * 33 * , 2373 ( 2000 ) . t. komori , h. takayama , and h. yoshino j. phys . japan * 68 * , 3387 ( 1999 ) . f. belletti , a. cruz , l. a. fernandez , a. gordillo - guerrero , m. guidetti , a. maiorano , f. mantovani , e. marinari , v. martin - mayor , j. monforte , a. muoz sudupe , d. navarro , g. parisi , s. prez - gaviro , j. j. ruiz - lorenzo , s. f. schifano , d. sciretti , a. tarancon , r. tripiccione , and d. yllanes , j. stat . phys . * 135 * , 1121 ( 2009 ) . f. ritort and p. sollich , adv . in physics , * 52 * , 219 ( 2003 ) . s. leonard , p. mayer , p. sollich , l. berthier , and j. p. garrahan , j. stat . p07017 ( 2007 ) . kinetically constrained models j. p. garrahan , p. sollich , c. toninelli , `` dynamical heterogeneities in glasses , colloids , and granular media '' , eds . : l. berthier , g. biroli , j - p bouchaud , l. cipelletti and w. van saarloos ( oxford university press , to appear ) .

this paper studies symmetric determinantal representations ( sdr ) in characteristic , that is the representation of a multivariate polynomial by a symmetric matrix such that , and where each entry of is either a constant or a variable . we first give some sufficient conditions for a polynomial to have an sdr . we then give a non - trivial necessary condition , which implies that some polynomials have no sdr , answering a question of grenet et al . a large part of the paper is then devoted to the case of multilinear polynomials . we prove that the existence of an sdr for a multilinear polynomial is equivalent to the existence of a factorization of the polynomial in certain quotient rings . we develop some algorithms to test the factorizability in these rings and use them to find sdrs when they exist . altogether , this gives us polynomial - time algorithms to factorize the polynomials in the quotient rings and to build sdrs . we conclude by describing the case of alternating determinantal representations in any characteristic .

1210.5879

let @xmath4 be some field of characteristic @xmath0 . a symmetric determinantal representation ( sdr ) of a polynomial @xmath5 $ ] is a symmetric matrix @xmath2 with entries in @xmath6 such that @xmath7 . one can also find in the literature other definitions where for instance the symmetric matrix has linear ( degree-@xmath8 ) polynomials as entries . the two definitions are essentially equivalent , and we shall see that for our purposes , taking one or the other does not make any difference . symmetric determinantal representations have been studied at least from the beginning of the twentieth century @xcite and apparently even from the nineteenth century @xcite . definite sdrs are sdrs with the additional requirement that the matrix obtained by setting all the variables to zero is positive semi - definite . definite sdrs play an important role in convex optimization , leading to a renew of interest in these representations , definite or not , in the recent years @xcite , see also @xcite and the presentation @xcite for more perspectives on this . recently , petter brndn has given sdrs for the elementary symmetric polynomials @xcite . he uses at this end graph - theoretic constructions and considers the laplacian matrix of the graph . our constructions are also graph - theoretic but we consider the adjacency matrix of the graphs we obtain . independently , symmetric determinants in characteristic two have also been a subject of studies @xcite . symmetric determinantal representations for polynomials represented by weakly - skew circuits were given in @xcite for any field of characteristic different from @xmath0 . the authors conjectured that these representations do not always exist in characteristic @xmath0 . we prove this fact in this paper . to this end , we give a necessary condition for a polynomial to admit an sdr . we then focus on multilinear polynomials . for these polynomials , we show an equivalence between the existence of an sdr and the ability to factorize the polynomial in certain quotient rings . we develop algorithms to study the factorization in these quotient rings . altogether , we obtain polynomial - time algorithms to factorize polynomials in the quotient rings and to compute sdrs of multilinear polynomials when they exist . a polynomial @xmath5 $ ] is said _ representable _ if it has an sdr , that is if there exists a symmetric matrix @xmath2 with entries in @xmath6 such that @xmath3 . in this case , we say that @xmath2 _ represents _ @xmath1 . for instance , the polynomial @xmath9 is representable as the determinant of the @xmath10 matrix @xmath11 note that we ask the matrix to have entries in @xmath4 . a natural relaxation would be to allow entries in an extension @xmath12 of @xmath4 . actually , we shall show along the way that at least for multilinear polynomials , and most certainly for any polynomial , this relaxation is irrelevant . in the case of multilinear polynomials , corollary [ cor : basefield ] shows that if a polynomial is representable , it has an sdr which only uses elements from the field generated by its coefficients . * organization . * we begin by introducing some relevant algebraic background in section [ sec : background ] . section [ sec : representable ] is devoted to prove that sdrs exist for a large class of polynomials . then section [ sec : obstructions ] proves the main results of this paper : some polynomials are not representable , and we can characterize the multilinear representable polynomials . some partial results towards a full characterization are also given . section [ sec : factor ] is devoted to more algorithmic results . using the equivalence between representability and factorizability in certain quotient rings , we develop algorithms for these two tasks . section [ sec : alternating ] is devoted to the case of alternating determinantal representations in any characteristic . finally , we conclude in section [ sec : conclusion ] by some remaining open questions . experimentations were done using the free open - source mathematics software system _ sage _ @xcite , they allowed in return to fix a bug in its determinant method ( ticket # 10063 ) . the algorithms presented in this paper have been implemented and are available at http://perso.ens-lyon.fr/bruno.grenet/publis/symdetreprchar2.sage . let us introduce some useful notions and notations . let @xmath4 be any field of characteristic @xmath0 , and let @xmath13 $ ] be the ring of polynomials in @xmath14 indeterminates over @xmath4 . let @xmath15 , then the _ primitive monomial _ @xmath16 is defined by @xmath17 . a _ monomial _ is a polynomial of the form @xmath18 for some @xmath19 and some primitive monomial @xmath16 . the constant @xmath20 is its _ coefficient _ , and @xmath16 is its _ primitive part_. the value @xmath21 is its _ total degree _ and @xmath22 is its degree _ with respect to the variable @xmath23_. a polynomial is said to be _ multilinear _ if its monomials @xmath24 satisfy @xmath25 for all @xmath26 . working in characteristic @xmath0 causes some inconveniences , like the impossibility to halve . but , it also simplifies some computations . first , frobenius endomorphism ensures that for any polynomials @xmath27 and @xmath28 , we have @xmath29 . second , the determinant can easily be computed : [ prop : detchar2 ] the determinant of an @xmath30 symmetric matrix with entries in @xmath31 $ ] is @xmath32 where @xmath33 ranges over all involutions from @xmath34 to itself , that is permutations such that @xmath35 . the definition of the determinant is @xmath36 where @xmath33 ranges over all permutations of @xmath34 . actually , the signature of a permutation is either @xmath8 or @xmath37 , and those two elements coincide in characteristic @xmath0 . this means that the signature can be removed from the definition . consider @xmath38 for some permutation such that @xmath39 . then , @xmath40 as @xmath2 is symmetric , and @xmath41 . thus the products for a permutation and its inverse cancel out in the sum . this shows that the sum can be restricted to involutions . given some polynomials @xmath42 , we denote by @xmath43 the ideal they generate . that is , @xmath44\right\}.\ ] ] given a tuple @xmath45 , we define the ideal @xmath46 we also define the quotient ring @xmath47 as @xmath13/{\mathcal{i}}(\ell)$ ] and denote by @xmath48 or @xmath49 the canonical projection @xmath13\to{\mathcal{r}}(\ell)$ ] . the restriction of this projection to @xmath4 is one - to - one , hence @xmath4 naturally embeds into @xmath47 , and the elements of @xmath50 are called _ constants_. this morphism of rings can be extended to matrices by @xmath51 , and commutes with the determinant : @xmath52 . an element of @xmath47 is said to be _ linear _ if it is the projection of a linear polynomial . since the quotient identifies the squares of variables with constants , any element of @xmath53 has a unique multilinear representative in @xmath5 $ ] : we denote it by @xmath54 or @xmath55 . we have @xmath56 . we denote by @xmath57 or @xmath58 the map @xmath59 that sends a polynomial to the multilinear polynomial obtained by replacing each factor @xmath60 by @xmath61 . for instance , let @xmath62 then @xmath63 and @xmath64 . the square of any element of @xmath47 belongs to @xmath4 . in particular , an element of @xmath47 is invertible if , and only if , its square is not zero . for example , @xmath65 is invertible if , and only if , @xmath66 . given a tuple @xmath45 , we denote by @xmath67 the tuple @xmath68 and say that @xmath67 is a _ tuple of squares_. if @xmath67 is a tuple of squares , the square of an element @xmath69 of @xmath70 is the square of a unique element @xmath20 of @xmath4 : we denote it by @xmath71 or @xmath72 , and call it the _ absolute value _ of @xmath69 . we remark that @xmath73 and @xmath74 for all @xmath75 . furthermore , @xmath69 is invertible if , and only if , @xmath76 . we deal with some positive results . even though the main part of this paper is focused on negative results , we need to be able to represent some class of polynomials in order to give a characterization . in order to clarify some proofs , we will use the correspondence between permanents and cycle covers in graphs . we refer the reader to @xcite for the definitions concerning graphs . let @xmath77 be a weighted digraph and @xmath2 its adjacency matrix . we assume that the weights of @xmath77 are elements of @xmath13 $ ] . a _ cycle _ is a set of distinct arcs @xmath78 such that all the @xmath79 s are distinct but @xmath80 . cycle cover _ of @xmath77 is a set of disjoint cycles such that each vertex of the digraph belongs to exactly one cycle . the _ weight _ of a cycle cover is the product of the weights of all the arcs it uses . it is easily seen from the definition that the permanent of @xmath2 equals the sum of the weights of all the cycle covers of @xmath77 . since the characteristic of @xmath4 is two , the permanent of @xmath2 equals its determinant . suppose now that @xmath77 is symmetric ( that is @xmath2 is symmetric ) . proposition [ prop : detchar2 ] shows that only some special cases of cycle covers can be considered . more precisely , the determinant of @xmath2 equals the sum of the weights of the cycle covers of @xmath77 corresponding to an involution . these cycle covers are made of length-@xmath8 and length-@xmath0 cycles , and are called _ partial matchings_. as @xmath77 is symmetric , it can actually be considered as an undirected graph . length-@xmath8 cycles are _ loops _ , and length-@xmath0 cycles are _ edges_. the weight of a length-@xmath0 cycle is the product of the weights of its arcs , that is the square of the weight of the edge . thus consider a partial matching of a graph @xmath77 with ( symmetric ) adjacency matrix @xmath2 . it can be viewed as a set @xmath81 of edges such that no vertex belongs to two distinct edges . the discussion is summarized by the identity @xmath82 where @xmath83 and @xmath84 represent the weights of an edge @xmath85 and of a loop on a vertex @xmath86 respectively , @xmath87 means that the vertex @xmath86 is not covered by @xmath81 , and @xmath81 ranges over all partial matchings of @xmath77 . an example is given by figure [ fig : ex - graph ] : the adjacency matrix of the graph is given is the introduction . the only partial matchings are made of one of the three edges , to cover the central vertex , and two loops . by convention , an edge with no indicated weight has weight @xmath8 . ( 1,0.86666667 ) ( 0,0).,title="fig : " ] ( -0.0260047,-0.03313245)(0,0)[b ] ( 1.0372615,-0.04127047)(0,0)[b ] ( 0.50039057,0.88657064)(0,0)[b ] in the following , if @xmath2 is a symmetric matrix , we denote by @xmath88 the graph whose adjacency matrix is @xmath2 . conversely , given a graph @xmath77 , we denote by @xmath89 its adjacency matrix . by a slight abuse of language , we shall say that a graph _ represents _ a polynomial when its adjacency matrix is an sdr of the polynomial . in the same way , we write @xmath90 instead of @xmath91 to simplify the notations . if @xmath92 is a subset of the vertices of a graph @xmath77 , then @xmath93 represents the induced subgraph of @xmath77 obtained by deleting the vertices of @xmath94 in @xmath77 . for an edge @xmath85 of @xmath77 , the graph @xmath95 is the graph obtained by deleting @xmath85 from @xmath77 , but keeping its both extremities . [ lem : prod ] let @xmath1 and @xmath96 be two representable polynomials . then @xmath97 is representable . let @xmath2 and @xmath98 be two symmetric matrices representing @xmath1 and @xmath96 respectively . to represent the product by a graph , it is enough to consider the disjoint union of @xmath88 and @xmath99 . this means that the sdr of @xmath97 is a block - diagonal matrix with two blocks being @xmath2 and @xmath98 . the first part of the next lemma was proved in @xcite . we give here another proof which is suitable for the second part . [ lem : squares ] let @xmath5 $ ] . then @xmath100 is representable . moreover , there exists a graph @xmath77 that represents @xmath100 with two distinguished vertices @xmath101 and @xmath102 and such that @xmath103 and @xmath104 . let @xmath105 where @xmath106 . the square of a monomial @xmath107 can be represented by a graph @xmath108 of size @xmath109 . for a variable @xmath23 with exponent @xmath110 , we build @xmath110 copies of a graph with two vertices and an edge of weight @xmath23 inbetween . we also build a graph with two vertices and an edge of weight @xmath111 inbetween . these @xmath112 size-@xmath0 graphs are arranged in a line to build @xmath108 : the graphs are arranged in some arbitrary order and an edge of weight @xmath8 links two consecutive graphs ( figure [ fig : chemin ] ) . ( 1,0.02529586 ) ( 0,0 ) corresponding to some monomial @xmath107 with @xmath113 and @xmath114.,title="fig : " ] ( 0.08387192,0.04272353)(0,0)[b ] ( 0.92525382,0.04272353)(0,0)[b ] ( 0.28938381,0.04080218)(0,0)[b ] ( -0.05545727,-0.00103428)(0,0)[b ] ( 1.05847286,-0.00191051)(0,0)[b ] ( 0.7096788,0.04272353)(0,0)[b ] the extremities of @xmath108 are named @xmath101 and @xmath102 . there is no loop in @xmath108 , therefore partial matchings are perfect matchings . the only perfect matching of @xmath108 is made of all the edges of weight different from @xmath8 . the weight of such a matching is @xmath115 . the only matching of @xmath116 is made of the edges of weight @xmath8 , and has weight @xmath8 . since @xmath108 has an even number of vertices , @xmath117 and @xmath118 have no perfect matching . given a graph @xmath108 for each monomial of @xmath1 , the graph @xmath77 is the union of these graphs in which all the vertices with name @xmath101 on the one hand , and all vertices with name @xmath102 on the other hand , are merged . the perfect matchings of @xmath77 are then made of a perfect matching of some @xmath108 , and perfect matchings of weight @xmath8 of @xmath119 for all @xmath120 . the sum of the weights of the matchings of @xmath77 is @xmath121 . furthermore , the only perfect matching of @xmath122 is made of perfect matchings of @xmath116 for all @xmath123 , thus @xmath103 . by the same parity argument as before , @xmath104 . this allows us to represent in a quite simple way a large class of polynomials . [ prop : representable ] let @xmath124 , where for @xmath125 , @xmath126 for some @xmath127 $ ] . then @xmath1 is representable . by lemma [ lem : prod ] , it is sufficient to show how to represent each @xmath128 . we first prove how to represent a polynomial of the form @xmath129 where the @xmath130 s are constants from @xmath4 . ( 1,0.81578947 ) ( 0,0).,title="fig : " ] ( 0.5043773,0.81495814)(0,0)[b ] ( 0.89618235,0.67439623)(0,0)[b ] ( -0.02038761,0.24336505)(0,0)[b ] ( 1.05925323,0.2906023)(0,0)[b ] ( 0.09509331,0.67917569)(0,0)[b ] ( 0.55020256,0.44664731)(0,0)[b ] ( 0.67911895,0.37091422)(0,0)[b ] ( 0.61383141,0.23334264)(0,0)[b ] ( 0.40272114,0.23447863)(0,0)[b ] ( 0.36524817,0.36057162)(0,0)[b ] the linear polynomial @xmath131 is represented by the graph @xmath132 given on figure [ fig : lingraph ] . we prove that it effectively represents @xmath131 : a partial matching has to match the central vertex with some of its neighbors . once a neighbor is chosen , say in the direction of @xmath23 , the loop with weight @xmath23 has to be chosen . then , there is only one choice to have a partial matching which consists in covering the remaining vertices by the outside edges . thus the weight of such a partial matching is @xmath133 , and the sum over all partial matchings equals @xmath131 . now , let @xmath134 be the graph representing the polynomial @xmath135 given by lemma [ lem : squares ] , with its two distinguished vertices @xmath101 and @xmath102 . by a slight abuse of language , we call @xmath136 the edge that has weight @xmath136 in @xmath132 , and denote by @xmath137 and @xmath138 its extremities . let @xmath139 be the graph obtained from @xmath132 by removing the edge @xmath136 . we build a new graph @xmath140 in which @xmath134 _ replaces _ the edge @xmath136 : the graph @xmath140 is the disjoint union of @xmath139 and @xmath134 , in which @xmath137 and @xmath101 ( respectively @xmath138 and @xmath102 ) are merged . a partial matching of @xmath132 either is a partial matching of @xmath139 , or is made of @xmath136 and a partial matching of @xmath141 . thus @xmath142 . in @xmath140 , a partial matching can also be of two sorts : either it is made of partial matchings of @xmath143 and @xmath139 , or of partial matchings of @xmath134 and @xmath141 . indeed , no partial matching exists covering @xmath144 ( respectively @xmath145 ) . thus @xmath146 this shows that we can replace in @xmath132 each @xmath136 by the graph @xmath134 to obtain an sdr of @xmath147 . in particular , this theorem shows that if @xmath4 is a finite field of characteristic @xmath0 , every linear polynomial is representable since every element in such a field is a quadratic residue . [ gsdr ] let @xmath148 $ ] . a _ generalized symmetric determinantal representation _ ( gsdr ) of @xmath1 is a symmetric matrix @xmath2 such that @xmath7 and whose entries are polynomials of @xmath31 $ ] such that each diagonal entry is either a constant or of the form @xmath149 where @xmath150 $ ] . [ thm : representable ] a polynomial @xmath148 $ ] is representable if , and only if , it admits a gsdr . an sdr is already a gsdr . we once again work with the graph representation instead of the matrix representation . suppose we have a graph @xmath77 where the weights of the edges are any polynomials , and the weights of the loops are either constants or of the form @xmath149 . we show how we can turn this graph into an sdr . we use the same technique as in the proof of proposition [ prop : representable ] to replace each edge with weight @xmath1 in @xmath77 by the graph @xmath151 which is an sdr of @xmath100 . it remains to show how to deal with the loops . suppose some vertex @xmath86 of @xmath77 has a loop of weight @xmath152 . consider the graph @xmath132 obtained in proposition [ prop : representable ] , and let @xmath153 be the graph obtained from @xmath77 by removing the loop on @xmath86 . then @xmath77 is replaced by @xmath154 , where the central vertex of @xmath132 is merged with @xmath86 . let @xmath155 be this new graph . note that @xmath156 . then @xmath157 repeating this operation for all the loops of the graph yields the result . this section deals with negative results , showing that some polynomials have no sdr . section [ sec : nec - cdt ] is devoted to a necessary condition that holds for any polynomial . it is followed by a simple example of a polynomial with no sdr . we prove in section [ sec : multilin ] that this necessary condition is actually a characterization when applied to multilinear polynomials . finally , section [ sec : full ] gives some partial results towards a full characterization . we aim to prove in this section a necessary condition for a polynomial to be representable . we introduce a notion of factorization _ modulo _ some ideal @xmath158 to express this condition . let @xmath5 $ ] . then @xmath1 is said _ factorizable modulo @xmath158 _ if there exist some linear elements @xmath159 of @xmath47 such that @xmath160 this definition can be restated as follows . a polynomial @xmath1 is factorizable _ modulo _ @xmath158 if there exists some linear polynomials @xmath161 of @xmath31 $ ] such that @xmath162 . [ thm : nc ] let @xmath5 $ ] be a representable polynomial . then for every tuple of squares @xmath163 , @xmath1 is factorizable _ modulo _ @xmath164 . for instance , one can recall the representable polynomial @xmath165 from the introduction . then @xmath166 and @xmath167 . the proof of this theorem is of algorithmic nature . we give an algorithm that takes as inputs an sdr @xmath2 of some polynomial @xmath1 and a tuple of squares @xmath67 , and returns a factorization of @xmath1 _ modulo _ @xmath164 . the general idea is to build the projection @xmath168 of @xmath2 to get a representation of @xmath169 , and then to perform row and column operations to _ isolate _ some diagonal entry @xmath170 , that is to cancel out each entry @xmath171 for @xmath172 , keeping @xmath173 symmetric . we then show that @xmath170 is a linear element of @xmath70 . thus we can write @xmath174 where @xmath175 is obtained from @xmath173 by removing its row and column of index @xmath176 . by induction on the dimensions of @xmath173 , we can conclude that @xmath169 can be factorized as a product of linear elements . in what follows , we prove some lemmas that justify this approach . let us fix some tuple of squares @xmath67 . in the next definition , we extend the notion of gsdr , originally defined for polynomials , to elements of @xmath70 . [ gsdr : r ] let @xmath177 . a _ generalized symmetric determinantal representation _ ( gsdr ) of @xmath69 is a symmetric matrix @xmath173 such that @xmath173 has linear diagonal entries and @xmath178 . in a gsdr for a polynomial , the diagonal entries are either constants or of the form @xmath149 . the projection of such a polynomial is a linear element of @xmath70 . indeed , for all @xmath176 , @xmath179 belongs to @xmath4 . therefore , if we let @xmath180 for all @xmath176 , @xmath181 is also the projection of the linear polynomial @xmath182 . the previous remark implies in particular the following lemma : [ nc : proj ] let @xmath2 be a gsdr of some polynomial @xmath5 $ ] . then the matrix @xmath183 is a gsdr of @xmath169 . next lemma gives some structure to the gsdr of an element of @xmath70 . [ nc : diag ] let @xmath173 be a gsdr of some @xmath177 . then there exists a gsdr @xmath184 of @xmath69 whose non - diagonal entries are constants . suppose that @xmath185 , @xmath186 , for some polynomial @xmath1 . since the determinant of @xmath173 equals @xmath187 where @xmath33 ranges over the involutions ( by proposition [ prop : detchar2 ] ) , if @xmath171 divides a monomial in @xmath188 , then so does @xmath189 . thus , if @xmath169 divides a monomial , so does @xmath190 . if we replace @xmath171 and @xmath191 by the absolute value @xmath192 , the determinant of @xmath173 is unchanged as @xmath193 by definition . this proves the lemma , as @xmath184 can be obtained by replacing each non - diagonal entry by its absolute value . we now define the main tools we use to prove the theorem . these are simple algorithmic transformations that we apply on the symmetric matrix representing an element @xmath177 such that the determinant remains unchanged and the matrix becomes diagonal . all of these depend on the tuple @xmath67 , even though it is not explicitly given as an argument to simplify the notations . let @xmath194 be the algorithm that replaces each non diagonal entry @xmath171 by its absolute value @xmath195 as in lemma [ nc : diag ] . we define two other algorithms , @xmath196 ( algorithm [ algo : add ] ) and @xmath197 ( algorithm [ algo : iso ] ) . @xmath198 dimension of @xmath173 @xmath199 @xmath198 dimension of @xmath173 @xmath173 let @xmath173 be a gsdr of some @xmath177 . then @xmath200 is a gsdr of @xmath69 whose non - diagonal entries are constants . the algorithm adds @xmath123 times the @xmath176-th row to the @xmath201-th one , and then @xmath123 times the @xmath176-th column to the @xmath201-th one . these two operations do not change the determinant . furthermore , only entries of the @xmath201-th row and column are changed . but with those two operations , @xmath202 is replaced by @xmath203 while @xmath204 is replaced by @xmath205 for @xmath206 . as initially @xmath207 and @xmath208 , @xmath173 remains symmetric . furthermore , @xmath209 is first replaced by @xmath210 , and finally by @xmath211 . thus @xmath209 remains linear . this shows that @xmath173 remains a gsdr of @xmath69 after the first two operations . eventually , @xmath194 is applied to a gsdr and we obtain the second property . [ nc : iso ] let @xmath173 be a gsdr of some @xmath177 . if there exists an index @xmath176 such that @xmath212 , then @xmath213 is a gsdr of @xmath69 . furthermore @xmath214 for all @xmath172 . the matrix @xmath215 is a gsdr of @xmath69 since @xmath200 is a gsdr of @xmath69 ( for all @xmath201 and @xmath123 ) . now , let @xmath216 for some @xmath201 such that @xmath217 and consider the action of @xmath196 on the @xmath176-th row of @xmath173 . the only altered entry is @xmath171 , when the @xmath176-th column multiplied by @xmath218 is added to the @xmath201-th one , and then by @xmath199 . so @xmath171 is replaced by @xmath219 . since @xmath220 by definition , @xmath171 is replaced by @xmath221 during @xmath199 . the same is true on the @xmath176-th column . thus , if @xmath213 , @xmath222 is the only nonzero entry in the @xmath176-th row and column of @xmath175 . we recall that an element of @xmath70 is invertible if , and only if , its square is nonzero . [ nc : square ] let @xmath173 be a gsdr of some @xmath177 such that no diagonal entry is invertible . if there exists a nonzero diagonal entry , say @xmath223 , and a nonzero entry @xmath224 for @xmath225 , then one can build a new gsdr of the same dimensions @xmath226 , representing some @xmath227 such that @xmath228 , where moreover @xmath226 contains some invertible diagonal entries . let us write @xmath223 as @xmath229 . let @xmath230 where @xmath175 is obtained from @xmath173 by removing its first row and column . then @xmath231 . indeed , adding the first row of @xmath184 to the second one , and the first column to the second one yields the matrix @xmath232 whose determinant equals @xmath188 . now @xmath233 still has the same determinant . for each @xmath225 , @xmath209 is replaced by @xmath234 in @xmath235 . since no diagonal entry is invertible by hypothesis , @xmath236 for all @xmath201 and @xmath237 . thus @xmath235 contains some diagonal entries whose square is nonzero , that is some invertible diagonal entries . actually , this holds since we supposed that some @xmath224 is nonzero . now , the determinant of this matrix equals @xmath238 therefore , @xmath173 can be replaced by this new matrix @xmath226 , of the same dimensions , with some invertible diagonal entries . then @xmath226 is a gsdr for some @xmath227 such that @xmath239 . we now have all the ingredients to prove the theorem . let us first restate the theorem . we aim to prove that if @xmath5 $ ] has a gsdr , then its projection @xmath240 can be written as @xmath241 where @xmath242 , , @xmath243 are linear elements of @xmath70 . suppose we are given a gsdr @xmath2 of some polynomial @xmath1 . then we have a gsdr @xmath168 of @xmath244 by lemma [ nc : proj ] . thus we have to prove that given a gsdr @xmath173 of some @xmath177 , we can find some linear elements @xmath245 of @xmath70 such that @xmath246 . first note that if @xmath173 does not satisfy the conditions of lemma [ nc : square ] , then we can already conclude . indeed , this means that each diagonal entry is either zero , or is the only nonzero entry in its row and column . by reordering the rows and columns , we can get a block - diagonal matrix with two blocks : the first one has zero diagonal entries and the second one is diagonal . therefore , since the determinant of @xmath173 is the product of the determinants of these two blocks , we get a constant times a product of linear elements . in other words , the factorization is found . so let @xmath173 be a gsdr of some @xmath177 satisfying the hypotheses of lemma [ nc : square ] . we build a gsdr @xmath226 of some @xmath227 such that @xmath247 for some linear element @xmath248 , and such that @xmath226 has at least one invertible diagonal entry @xmath249 . if @xmath173 already satisfied the property , then @xmath250 and @xmath251 . now , the @xmath176-th row and column of @xmath252 have as only nonzero entry @xmath222 by lemma [ nc : iso ] . thus , removing the @xmath176-th row and column to @xmath175 yields a gsdr @xmath184 of some element @xmath253 such that @xmath254 . this shows that from a gsdr of dimensions @xmath30 of some @xmath177 , we can build a gsdr of dimensions @xmath255 of some @xmath253 such that @xmath256 where @xmath102 and @xmath257 are linear . we can now use induction to prove that if @xmath177 has a gsdr , then it can be written as @xmath241 for some linear elements @xmath258 . indeed , if @xmath173 is a @xmath259 gsdr of @xmath69 , then @xmath69 is linear . the proof of theorem [ thm : nc ] is of algorithmic nature . it is easily seen that the underlying algorithm runs in time polynomial in the dimensions of the input gsdr . ( more precisely , the complexity of the algorithm is @xmath260 . ) let us consider the polynomials in @xmath261 $ ] , where @xmath262 denotes the field with two elements . the ring @xmath263 has 256 elements , 136 of which can be written as the product of linear polynomials , 120 of which can not . the element @xmath264 is one of those . therefore , theorem [ thm : nc ] tells us that the polynomial @xmath265 can _ not _ be represented as the determinant of a symmetric matrix with entries in @xmath266 . in this section , we show that the necessary condition of theorem [ thm : nc ] is actually a characterization when applied to multilinear polynomials . this relies on the following structural lemma . it is valid for any polynomial , even non - multilinear . [ lem : multilin ] let @xmath5 $ ] be a representable polynomial . then there exists an sdr @xmath2 of @xmath1 such that each variable appears at most once on the diagonal . let @xmath2 be any sdr for @xmath1 , that is @xmath7 , and @xmath2 has entries in @xmath6 . suppose that for some other @xmath176 , @xmath23 appears ( at least ) twice on the diagonal , as entries @xmath267 and @xmath268 . consider the matrix obtained after adding the row of index @xmath269 to the row of index @xmath270 , and the column of index @xmath269 to the column of index @xmath270 . as already mentioned , the only altered diagonal entry is @xmath268 and it is now equal to @xmath271 . therefore , we obtain a new sdr with one occurrence of @xmath23 on the diagonal replaced by zero . we can repeat this for each variable until each variable appears at most once on the diagonal . we can use this lemma to obtain the desired characterization , when @xmath4 is a finite field of characteristic @xmath0 . [ thm : multilin ] let @xmath5 $ ] be a multilinear polynomial where @xmath4 is a finite field of characteristic @xmath0 . then the three following statements are equivalent : * @xmath1 is representable ; * for every tuple of squares @xmath163 , @xmath1 is factorizable _ modulo _ @xmath164 ; * there exists a tuple of squares @xmath163 such that @xmath1 is factorizable _ modulo _ @xmath164 . the implication @xmath272 is a special case of theorem [ thm : nc ] , and @xmath273 is evident . let us prove that @xmath274 . let @xmath163 such that @xmath275 , where each @xmath138 is a linear element of @xmath70 . for @xmath276 , @xmath277 is a linear polynomial , hence by proposition [ prop : representable ] , we know that @xmath278 has an sdr @xmath2 . by lemma [ lem : multilin ] , there exists an sdr @xmath98 of @xmath96 such that each variable appears at most once on the diagonal . hence , lemma [ nc : proj ] and lemma [ nc : diag ] ensure that @xmath279 has a gsdr @xmath173 such that each @xmath170 is linear and each @xmath171 is constant for @xmath186 . let @xmath280 be the matrix defined by @xmath281 . since each variable appears once on the diagonal , @xmath282 is a multilinear polynomial . we have @xmath283 . since both @xmath1 and @xmath282 are multilinear , we have @xmath284 , hence @xmath1 is representable . if @xmath4 is infinite , a similar characterization can be obtained . to this end , the conclusion of theorem [ thm : nc ] can be reinforced as follows : if @xmath1 is representable , then there exist linear polynomials @xmath285 , , @xmath286 _ whose coefficients are quadratic residues in @xmath4 _ such that @xmath287 . one can check that the proof of theorem [ thm : nc ] actually is a proof of this stronger statement . the converse is proved using proposition [ prop : representable ] . theorem [ thm : nc ] is valid for any polynomial . thus we have a necessary condition for all polynomials . the characterization for multilinear polynomials relies on the fact that @xmath288 in this case . if we are working with a non - multilinear polynomial @xmath1 , the projection of @xmath1 _ modulo _ some ideal @xmath158 can dramatically change the structure of the polynomial . in particular , if we have a polynomial @xmath289 for some multilinear polynomial @xmath96 , then @xmath290 but @xmath291 . thus , it is certainly impossible to go back from the projection _ modulo _ @xmath292 to @xmath1 . to come up with this issue , we look at some new specific ideal for the projection . in this section , the field @xmath4 is supposed to be finite . with the same arguments as for theorem [ thm : multilin ] , the results of this section can be extended to any field of characteristic @xmath0 . let @xmath293 be the field of fractions in @xmath14 indeterminates over @xmath4 , and @xmath294 . for @xmath5 $ ] , we can consider the multilinear polynomial @xmath295 and apply theorem [ thm : multilin ] about multilinear polynomials . in particular , @xmath296 is representable if , and only if , it is factorizable . the problem we face is that our constructions use inverse of elements in the base field . this means that we have an equivalence between factorization and sdr for multilinear polynomials in @xmath297 $ ] but the factorization or the sdr we build can use rational fractions in the @xmath298 s . to partly avoid this problem , we have to restrict the ideals we are working with to ideals of the form @xmath164 for @xmath299 . unfortunately , it is not sufficient . nevertheless , we are able to prove some partial results . [ lem : xi ] let @xmath5 $ ] . then @xmath1 is representable if , and only if , @xmath300 has an sdr with non - diagonal entries in @xmath301 $ ] . let us remark at first that @xmath302 is a bijection from @xmath31 $ ] to the set of multilinear polynomials with coefficients in @xmath303 $ ] . indeed , its inverse @xmath304 simply consists in mapping each @xmath298 to @xmath23 . using lemma [ lem : multilin ] , we can transform any ( g)sdr to an sdr such that each variable appears exactly once on the diagonal . let @xmath2 be an sdr of @xmath1 . we can apply the procedure @xmath194 to @xmath2 ( with respect to the tuple @xmath305 ) . this yields an sdr of @xmath300 as proved by lemma [ nc : proj ] . conversely , if we have an sdr @xmath306 of @xmath300 , we can replace each @xmath298 by @xmath23 to get an sdr of @xmath1 . this corresponds to applying @xmath304 to each entry . as this function is compatible with the addition and multiplication , the new matrix @xmath2 we obtain satisfy @xmath307 . let @xmath5 $ ] . * if @xmath1 is representable , then for every tuple of squares @xmath163 , @xmath300 can be factorized as a product of linear polynomials _ modulo _ @xmath164 , and the linear polynomials have coefficients in @xmath308 . * if @xmath300 can be factorized as a product of linear polynomials _ modulo _ @xmath164 for some tuple of squares @xmath67 , and if the linear polynomials have coefficients in @xmath301 $ ] , then @xmath1 is representable . to obtain a full characterization , we would need to prove that in the first statement , we can obtain linear factors with coefficients in @xmath303 $ ] . the first statement only consists in applying theorem [ thm : nc ] to @xmath309 . for the second statement , suppose @xmath310 for some @xmath67 , and each @xmath128 has coefficients in @xmath303 $ ] . using theorem [ thm : multilin ] , we can build a matrix representing @xmath300 . since the @xmath311 s have as coefficients some polynomials in the @xmath298 s , and since the transformations of lemma [ lem : multilin ] used in the proof of theorem [ thm : multilin ] use no inverse of any of the coefficients , we get an sdr of @xmath312 the non - diagonal entries of which are polynomials in the @xmath298 s . using lemma [ lem : xi ] , we conclude that @xmath1 is representable . section [ sec : multilin ] gives a characterization of representable multilinear polynomials in terms of the factorization of the polynomials into linear polynomials _ modulo _ an ideal @xmath158 . we give in this section an algorithm to decide this problem . its running time is polynomial in the number of monomials of the polynomial . in this section , @xmath4 is a finite field of characteristic @xmath0 . in the previous section , we worked with elements of the quotient ring @xmath47 for some tuple @xmath313 . the algorithms presented in this section deal with multilinear polynomials @xmath148 $ ] . theorem [ thm : multilin ] is the basic tool . since @xmath314 for any @xmath313 , it can be reformulated as follows : a multilinear polynomial is representable if , and only if , for every tuple of squares @xmath67 , there exist linear polynomials @xmath285 , , @xmath286 such that @xmath315 . it is equivalent to say that @xmath287 . moreover , as seen before this existence does not depend on the tuple @xmath67 . this motivates the following definition . we say that a multilinear polynomial @xmath1 is _ factorizable _ if there exist a tuple of squares @xmath67 and linear polynomials @xmath285 , , @xmath286 such that @xmath316 the algorithm heavily relies on the fact that the possibility to factorize a polynomial _ @xmath164 does not depend on @xmath67 . actually two tuples are used , @xmath317 and @xmath318 . to simplify the notations , these tuples are respectively denoted by @xmath221 and @xmath8 . moreover @xmath319 , @xmath320 and @xmath321 on the one hand , and @xmath322 , @xmath323 and @xmath324 on the other hand , are the functions defined in section [ sec : quotientrings ] . in the same way , let @xmath325 and @xmath326 and @xmath327 be defined by analogy . we shall sometimes write that @xmath1 is _ factorizable modulo @xmath328 _ for @xmath329 or @xmath330 instead of simply factorizable to emphasize the fact that we are working specifically with the ideal @xmath328 . let @xmath1 be a multilinear polynomial . we define its _ linear part _ @xmath331 as the sum of all its monomials of degree at most @xmath8 . for instance @xmath332 . furthermore , we write @xmath333 for the partial derivative of @xmath1 with respect to the variable @xmath23 . for a multilinear polynomial , this equals the quotient in the euclidean division of @xmath1 by @xmath23 . to show how to test the factorizability of a multilinear polynomial , we proceed in two steps . we first show how to test the factorizability of a polynomial @xmath1 whose monomial of lowest degree has degree exactly @xmath8 ( we say that @xmath1 has _ valuation _ @xmath8 ) . to this end , we show that @xmath1 is factorizable if , and only if , @xmath334 where @xmath335 is a nonzero monomial of @xmath331 ( lemmas [ lem : onefactor ] and [ lem : factor ] ) . the second step proves that given any multilinear polynomial @xmath1 , we can compute a polynomial @xmath96 of the same degree whose valuation is @xmath8 such that @xmath1 is factorizable if , and only if , @xmath96 is also . there are two cases , covered by lemmas [ lem : fullpoly ] and [ lem : linpart ] . this will allow us to describe an algorithm using alternatively those two steps to test factorizability . [ lem : onefactor ] let @xmath1 be a multilinear polynomial of valuation @xmath8 . if there exists some linear polynomials @xmath285 , , @xmath286 such that @xmath336 then there exist an index @xmath201 and a constant @xmath123 such that @xmath337 . suppose that @xmath338 and let @xmath339 . in particular , @xmath340 and @xmath341 . thus there exists @xmath201 such that @xmath342 . in other words , @xmath311 is a sum of degree-@xmath8 monomials . the linear part of @xmath1 being nonzero , the polynomial @xmath343 has a constant coefficient @xmath344 . a degree-@xmath8 monomial of @xmath96 is the product of a monomial of @xmath311 by @xmath123 . this means that @xmath345 . we now prove that we can efficiently test if some linear polynomial @xmath131 can appear in the factorization of @xmath1 _ modulo _ @xmath346 . [ lem : factor ] let @xmath1 be a multilinear polynomial and @xmath131 be a linear polynomial with no constant coefficient , having a nonzero monomial @xmath347 . if there exists a multilinear polynomial @xmath96 such that @xmath348 , then @xmath349 suppose that @xmath348 . this means that there exist polynomials @xmath350 , , @xmath351 such that @xmath352 moreover , @xmath353 for all @xmath201 . for @xmath354 , this comes from the fact that @xmath355 . since @xmath131 contains the monomial @xmath347 , @xmath356 therefore , @xmath357 since @xmath358 , the previous equality can be multiplied by @xmath359 to obtain @xmath360 since @xmath361 is a sum of squares , @xmath362 in the second step we prove that given any multilinear polynomial @xmath1 , we can find a polynomial @xmath96 of valuation @xmath8 such that @xmath1 is factorizable if , and only if , @xmath96 is also . there are two distinct cases . at first , we focus on _ full _ polynomials , that is without zero coefficient . an @xmath14-variate multilinear polynomial is full if it has @xmath363 nonzero monomials . in particular , if every coefficient equals @xmath8 , then the polynomial can be factorized as @xmath364 . in the general case , such a factorization does not necessarily exist . in the following lemma , given a full polynomial , a new polynomial is produced that is either not full , or has less variables . in any case , the number of monomials decreases . [ lem : fullpoly ] let @xmath1 be a multilinear polynomial in @xmath14 variables with exactly @xmath363 monomials . then there exists a linear polynomial @xmath131 such that @xmath365 is nonzero and has less than @xmath363 monomials . moreover , @xmath1 is factorizable if , and only if , @xmath96 is factorizable . let @xmath23 be any variable of @xmath1 and @xmath366 where @xmath367 is the coefficient of @xmath23 in @xmath1 and @xmath368 is its constant coefficient . then the constant coefficient of @xmath96 is @xmath369 and thus @xmath96 is nonzero , and the coefficient of @xmath23 in @xmath96 is @xmath370 . thus @xmath96 has less monomials than @xmath1 . by definition , @xmath96 is factorizable if @xmath1 is also . moreover , @xmath371 since @xmath1 is multilinear . thus if @xmath96 is factorizable , then so is @xmath1 . it remains to deal with the case where @xmath1 does not possess all the possible monomials . in this lemma , we consider the ideal @xmath330 instead of @xmath346 as before . [ lem : linpart ] let @xmath1 be a multilinear polynomial over @xmath14 variables with at most @xmath372 monomials . then there exists a primitive monomial @xmath16 such that @xmath373 has valuation @xmath8 . moreover , @xmath1 is factorizable if , and only if , @xmath96 is factorizable . if @xmath1 already has valuation @xmath8 , we can take @xmath374 . if @xmath1 has valuation greater than @xmath8 , let @xmath375 be a nonzero primitive monomial of @xmath1 of minimal degree . let @xmath176 be some index such that @xmath376 and define @xmath377 . then @xmath378 . moreover , @xmath379 if , and only if , @xmath380 . since @xmath381 , the coefficient of @xmath16 in @xmath1 is zero . thus @xmath16 satisfies the lemma . if @xmath1 has valuation @xmath221 , let @xmath16 be a primitive monomial of minimal degree whose coefficient in @xmath1 is zero . such a monomial exists since @xmath1 has at most @xmath372 monomials . then @xmath382 has no constant coefficient . furthermore , by the minimality of @xmath16 , every monomial of smaller degree has a nonzero coefficient in @xmath1 . this is in particular the case of the monomials @xmath383 for every variable @xmath23 that divides @xmath16 . since @xmath384 , @xmath16 satisfies the lemma . to finish the proof , we remark that @xmath385 . therefore , @xmath1 is factorizable if , and only if , @xmath96 is factorizable . for the proof of the next corollary , one needs to remark a simple fact . if a multilinear polynomial @xmath1 is representable , then @xmath333 is also representable for any variable @xmath23 . indeed , suppose that @xmath2 is an sdr of @xmath1 with each variable appearing at most once on the diagonal . suppose that @xmath386 . then the determinant of the matrix obtained by removing the row and column @xmath201 from @xmath2 equals @xmath333 . [ cor : basefield ] let @xmath148 $ ] be a multilinear polynomial and @xmath387 a field extension . if @xmath1 has an sdr with entries in @xmath388 , then it has an sdr with entries in @xmath389 . let us first consider @xmath12 as the base field . by theorem [ thm : multilin ] , @xmath1 is representable if , and only if , it is factorizable . using lemmas [ lem : fullpoly ] and [ lem : linpart ] , one can suppose that @xmath1 has valuation @xmath8 . using lemmas [ lem : onefactor ] and [ lem : factor ] and the remark before the corollary , one can deduce that @xmath1 is representable if , and only if , @xmath334 and @xmath333 is representable . now , @xmath333 is a polynomial with coefficients in @xmath4 , which has an sdr with entries in @xmath388 . moreover , it has one variable less than @xmath1 . therefore , we can prove the corollary by induction on the number of variables . the previous lemmas yield a polynomial time algorithm to decide whether some polynomial @xmath1 is factorizable . we first give an algorithm @xmath390 ( algorithm [ algo : prep ] ) corresponding to lemmas [ lem : fullpoly ] and [ lem : linpart ] . [ lem : prep ] let @xmath148 $ ] be a multilinear polynomial . then @xmath391 is either linear , or has valuation @xmath8 . moreover , @xmath1 is factorizable if , and only if , @xmath96 is also . the algorithm runs in time polynomial in the number of variables and the number of monomials of @xmath1 . the correctness is ensured by lemmas [ lem : fullpoly ] and [ lem : linpart ] . we only have to prove its termination and the running time estimate . there is a recursive call when @xmath1 is full . from lemma [ lem : fullpoly ] , @xmath392 then has at most @xmath372 monomials . either this new polynomial is full , but the number of variables decreased , or the condition `` @xmath393 is full '' is not satisfied anymore and there is no new recursive call . therefore , the number of recursive calls is bounded by the number of variables . this proves both the termination and the complexity analysis . let us now describe the algorithm @xmath394 ( algorithm [ algo : factor ] ) corresponding to lemmas [ lem : onefactor ] and [ lem : factor ] . @xmath395 [ line : linear ] [ thm : factor ] let @xmath148 $ ] be a multilinear polynomial . then the algorithm @xmath396 answers * true * if , and only if , @xmath1 is factorizable . the algorithm runs in time polynomial in the number @xmath14 of variables and the number of monomials of @xmath1 . the correctness follows from lemmas [ lem : onefactor ] and [ lem : factor ] . the termination is ensured by the fact that @xmath333 has less variables than @xmath1 . this bounds the number of iterations by @xmath14 . algorithm [ algo : factor ] only tells us if a polynomial is factorizable , but does not give us a factorization . the reason for this is that we change several times the ideal we are working with . nevertheless , we proved in section [ sec : multilin ] that given the factorization of a multilinear polynomial _ modulo _ some ideal @xmath328 , we can find a symmetric matrix representing the polynomial . we use this in the following to show how to modify algorithm [ algo : factor ] in order to get a symmetric determinantal representation . using the results of section [ sec : nec - cdt ] , we are then able to factorize any factorizable multilinear polynomial _ modulo _ any ideal @xmath328 . [ lem : merge ] given two sdrs @xmath397 and @xmath398 of two multilinear polynomials @xmath1 and @xmath96 respectively , one can build an sdr @xmath399 of @xmath400 ( @xmath401 ) in time polynomial in the dimensions of @xmath397 and @xmath398 . the algorithm is based on lemma [ lem : multilin ] . let @xmath98 be the block - diagonal matrix made of @xmath397 and @xmath398 . clearly , this matrix represents @xmath402 . using lemma [ lem : multilin ] , one can build a matrix @xmath403 such that each variable appears at most once on the diagonal and such that @xmath404 . then @xmath405 represents @xmath400 . there is an algorithm @xmath406 that given as input a multilinear polynomial @xmath5 $ ] returns an sdr of @xmath1 if one exists . this algorithm runs in time polynomial in @xmath14 and the number of monomials . the algorithm @xmath406 is made of two steps . the first one is a modification of the algorithm @xmath394 such that it returns a list of factors instead of * true * when @xmath1 is factorizable . the second one is the construction of an sdr of @xmath1 from this list of factors , using the algorithm @xmath407 of lemma [ lem : merge ] . in algorithms @xmath390 and @xmath394 , to test if @xmath1 is factorizable , it is written as @xmath408 where @xmath131 is either linear or a monomial and @xmath409 , and then one tests whether @xmath96 is factorizable . these algorithms are modified to retain the couples @xmath410 each time such an operation is performed . more precisely , we add a global variable @xmath411 containing a list of pairs of the form @xmath410 . let us now describe how @xmath390 and @xmath394 modify this variable . in @xmath390 , line [ line : full ] , the couple @xmath412 is added to @xmath411 . indeed , a recursive call is performed with the polynomial @xmath413 . but @xmath414 equals @xmath415 . in the same way , the couple @xmath416 is added to @xmath411 at lines [ line : val0 ] and [ line : val2 ] . to finish , the couple @xmath417 is added to @xmath411 at line [ line : call ] of @xmath394 . when @xmath394 answers * true * , then @xmath1 is linear . instead of this answer , the new algorithm adds the couple @xmath418 to @xmath411 ( the bit @xmath221 is arbitrary and unused ) and returns @xmath411 . at this stage , we have a list @xmath411 of couples @xmath419 , , @xmath420 . let @xmath421 and for @xmath176 from @xmath422 down to @xmath8 , @xmath423 . from the construction of @xmath411 , @xmath424 . an sdr for @xmath1 is built as follows : for all @xmath176 , an sdr @xmath425 of @xmath128 is built using proposition [ prop : representable ] . then , let @xmath426 and for @xmath176 from @xmath422 down to @xmath8 , let @xmath427 . if @xmath428 and @xmath429 , lemma [ lem : merge ] shows that @xmath430 . to conclude , the algorithm returns @xmath431 . the running time of the algorithm is controlled by the running times of @xmath390 , @xmath394 , and @xmath407 . symmetric matrices correspond to symmetric bilinear forms . we saw that in this context , there is a big difference depending on whether the characteristic of the underlying field is @xmath0 or not . as explained to us by mathieu florence @xcite , the related notion of alternating forms is known to be _ characteristic - free_. an _ alternating form _ is a bilinear form @xmath432 such that @xmath433 for any @xmath86 in the vector space @xmath434 . the matrix associated to alternating forms are the anti - symmetric matrix with zero diagonal entries . hence , we should expect an homogeneous result concerning alternating determinantal representations . it turns out to be the case : [ thm : alternating ] let @xmath4 be some field and @xmath5 $ ] be a polynomial . then , @xmath1 can be written as the determinant of an alternating matrix with entries in @xmath6 if , and only if , @xmath1 is a square . let @xmath1 be the determinant of an alternating matrix @xmath2 with entries in @xmath6 . if we consider @xmath2 as a matrix over the commutative ring @xmath13 $ ] , we see that @xmath3 is the square of the pfaffian of @xmath2 which is an element of @xmath13 $ ] ( @xcite xv , 9 , page 588 ) . conversely , let @xmath435 be the square of an element of @xmath13 $ ] . as proved in @xcite , there exists a square matrix @xmath98 with entries in @xmath6 such that @xmath436 . the matrix @xmath437 is alternate and satisfies @xmath438 . we proved in this paper that in characteristic @xmath0 , some polynomials do not admit any sdr . in the case of multilinear polynomials , we gave a complete characterization as well as algorithms to deal with these representations . we discovered some tight relations between the ability to find an sdr and to factorize the polynomial _ _ some square polynomials . thus we showed that the factorization in these quotient rings can be performed in polynomial time . a polynomial @xmath148 $ ] is representable if , and only if , for some ( equivalently any ) tuple of squares @xmath299 , @xmath439 is factorizable _ modulo _ @xmath158 into linear polynomials @xmath440[x_1,\dots , x_m]$ ] . an example of a problematic polynomial is @xmath441 . indeed , it can be factorized as @xmath442 . but once the projection _ modulo _ @xmath443 is made , it is not so clear anymore how it can be factorized . one idea could be first to factorize the polynomial and then apply our results to each factor . yet it is not clear whether this strategy can work . if the conjecture can be proved , or if some other characterization of the same kind can be found , it would also remain to see if the algorithms designed for multilinear polynomials can be extended to the general case . once again , the main difficulty is to deal with the fact that our algorithms use inverse of elements . even for multilinear polynomials , some questions remain . for instance , it could be interesting to make a quantitative study to know how many polynomials have sdrs . for instance , all polynomials in @xmath0 variables are representable , and it seems that the proportion decreases as the number of variables increases . ymmetric determinantal representation of formulas and weakly skew circuits . in _ randomization , relaxation , and complexity in polynomial equation solving _ , l. gurvits , p. pbay , j. m. rojas , and d. c. thompson , eds . , no . 556 in contemp . soc . , providence , ri , 2011 , pp . arxiv : http://arxiv.org / abs/1007.3804[1007.3804 ] , doi : http://dx.doi.org/10.1090/conm/556[10.1090/conm/556 ] , extended abstract in stacs11 . polynomials with and without determinantal representations . , 7 ( 2012 ) , 15791595 . arxiv : http://arxiv.org / abs/1008.1931[1008.1931 ] , doi : http://dx.doi.org/10.1016/j.laa.2012.04.043[10.1016/j.laa.2012.04.043 ] . symmetric determinantal representation of polynomials . , 9 ( 2012 ) , 36423660 . hal : http://hal.archives - ouvertes.fr / hal-00275615[hal-00275615 ] , doi : http://dx.doi.org/10.1016/j.laa.2012.01.004[10.1016/j.laa.2012.01.004 ] .

tree detection techniques are often used to reduce the complexity of _ a posteriori probability _ ( app ) detection in high dimensional multi - antenna wireless communication systems . in this paper , we introduce an efficient soft - input soft - output tree detection algorithm that employs a new type of look - ahead path metric in the computation of its branch pruning ( or sorting ) . while conventional path metrics depend only on symbols on a visited path , the new path metric accounts for unvisited parts of the tree in advance through an unconstrained linear estimator and adds a bias term that reflects the contribution of as - yet undecided symbols . by applying the linear estimate - based look - ahead path metric to an-algorithm that selects the best paths for each level of the tree we develop a new soft - input soft - output tree detector , called an _ improved soft - input soft - output-algorithm _ ( iss - ma ) . based on an analysis of the probability of correct path loss , we show that the improved path metric offers substantial performance gain over the conventional path metric . we also demonstrate through simulations that the iss - ma provides a better performance - complexity trade - off than existing soft - input soft - output detection algorithms .

1105.1482

the relationship between the transmitted symbol and the received signal vector in many communication systems can be expressed in the form @xmath1 where @xmath2 is the @xmath3 transmitted vector whose entries are chosen from a finite symbol alphabet , @xmath4 and @xmath5 are the @xmath6 received signal and noise vectors , respectively , and @xmath7 is @xmath8 channel matrix . as a practical decoding scheme when a code constraint is imposed , _ iterative detection and decoding _ ( idd ) has been applied to various digital communication systems including channel equalization @xcite , multi - input multi - output ( mimo ) detection @xcite , and multi - user detection @xcite . motivated by the _ turbo principle _ @xcite , an idd receiver exchanges soft information between a symbol detector and a channel decoder to achieve performance close to the channel capacity . the symbol detector computes _ a posteriori _ probabilities ( app ) on the bits comprising @xmath2 , using _ a priori _ probabilities provided by the channel decoder and the observation @xmath4 . then , the detector exchanges this soft information ( so called _ extrinsic _ information ) with a soft - input soft - output decoder , such as the max - log - map decoder @xcite . in the sequel , we refer to such a detector as an _ app detector_. direct computation of the app involves marginalization over all configurations of the vector @xmath2 , leading to exponential complexity in the system size ( e.g. , number of antenna elements in mimo systems ) . as a means of approximately performing the app detection at reduced complexity , _ tree detection techniques _ have received much attention recently @xcite . ( refer to @xcite for an overview of tree detection techniques . ) the essence of these approaches is to produce a set of promising symbol candidates via a tree search for estimating the app over this reduced set . thus far , a variety of tree detection algorithms have been proposed . in @xcite , the _ sphere decoding algorithm _ ( sda ) @xcite with a fixed radius was used to find symbol candidates . in @xcite , _ a priori _ information obtained from the channel decoder was exploited to improve the search efficiency of the sda . in @xcite , a hard sphere decoder was employed to find a single _ maximum a posteriori _ probability ( map ) symbol estimate maximizing @xmath9 and a candidate list was generated by flipping bits in the map estimate . in @xcite , the apps of all bits in @xmath2 are obtained simultaneously by modifying a bound tightening rule of a single sphere search . additionally , a more sophisticated extension of this idea was introduced in @xcite . the computational complexity of these tree detection algorithms varies depending on the channel and noise realizations , and in the worst - case the search complexity is the same as that of exhaustive search . in order to limit the worst case complexity of the tree detection approach , _ fixed - complexity tree search _ techniques @xcite have been proposed . for example , an @xmath0-algorithm was extended to soft - input soft - output detection in @xcite and an intelligent candidate adding algorithm for improving efficiency of the @xmath0-algorithm was proposed in @xcite . the stack algorithm was also exploited for list generation in combination with soft augmentation of tail bits of stack elements @xcite . other fixed - complexity soft - input soft - output detection algorithms can be found in @xcite . the @xmath0-algorithm @xcite , also known as _ @xmath10-best algorithm _ in the mimo detection literature @xcite , selects only a finite set of the @xmath0 best candidates for each layer of the detection tree . the @xmath0-algorithm is a practical candidate for soft - input soft - output detection due to its inherent nature to facilitate parallel and pipelined processing @xcite . in spite of this benefit , the @xmath0-algorithm suffers from a poor performance - complexity trade - off due to the greedy nature of the algorithm . to be specific , the algorithm checks the validity of paths in the forward direction and never traverses back for reconsideration . once the correct path is rejected , it will never be selected again in subsequent selections , resulting in wasteful search effort . moreover , these erroneous decisions often occur in early candidate selection stages where the accumulated path metric considers only a few symbol spans . one way to alleviate such error propagation is symbol detection ordering @xcite . by processing each layer in an appropriate order , the chances of errors propagating to the next stage can be reduced . nevertheless , error propagation severely limits the performance of the @xmath0-algorithm especially when the system size is large . in this paper , we pursue an improvement of the performance - complexity trade - off of soft - input soft - output @xmath0-algorithms . towards this end , we propose a new path metric capturing the contribution of the entire symbol path . while the conventional path metric accounts for the contributions of symbols along the visited path only , the new path metric looks ahead to the unvisited paths and estimates their contributions through a soft unconstrained linear symbol estimate . in fact , a _ bias term _ reflecting the information from as - yet undecided symbols is incorporated into the conventional path metric for this purpose . in order to distinguish this improved path metric from the conventional path metric and other look - ahead metrics , we henceforth refer to it as a _ linear estimate - based look - ahead ( le - la ) path metric_. we apply the le - la path metric to the soft - input soft - output @xmath0-algorithm , introducing an _ improved soft - in soft - out @xmath0-algorithm _ ( iss - ma ) . by sorting paths based on the le - la path metric , the iss - ma lessens the chance of rejecting the correct path from the candidate list and eventually improves the detection performance especially for systems of large dimension . indeed , from an analysis of the probability of correct path loss ( cpl ) , we show that the le - la path metric benefits the candidate selection process of the @xmath0-algorithm . the idea of using a look - ahead path metric has been explored in artificial intelligence search problems @xcite and can also be found in soft decoding of linear block codes @xcite . in @xcite , computationally efficient methods to obtain the bias term were investigated using semi - definite programming and @xmath11 estimation techniques . while these approaches search for a _ deterministic _ bias term ( lower - bound of future cost ) to guarantee the optimality of the sequential or depth - first search , our approach uses linear estimation to derive a bias term designed to improve candidate selection of the breadth - first search . the key advantage of using a linear estimator is that _ a priori _ information can be easily incorporated into the bias term so that the look - ahead operation benefits from the decoder output in each iteration . it is also worth emphasizing the difference between the proposed path metric and fano matric @xcite . the fano metric exploits the _ a posteriori _ probability of each path as its path metric . for a binary symmetric channel , the fano metric introduces a bias term proportional to the path length to penalize paths of short length . the extension of the fano metric to channels with memory or mimo channels is not straightforward , since it involves marginalization over the distribution of the undecided symbols . modification of the fano metric is considered for equalization of intersymbol interference ( isi ) channels in @xcite and for multi - input multi - output detection in @xcite . as a means to improve path metric of the sda , the idea of probabilistic pruning was introduced in @xcite . in @xcite , the probability density of an observed signal estimated from a separate tree search is used as a bias term . while these approaches assign an equal bias term for paths of the same length , the iss - ma provides a distinct bias term for each path in the tree , allowing for the application of a breadth - first search such as the @xmath0-algorithm . as such , our path metric can be readily combined with any type of tree - based soft - input soft - output detector . the rest of this paper is organized as follows . in section [ sec : problem ] , we briefly review the idd system and the tree detection algorithm . in section [ sec : iss - ma ] , we present the le - la path metric along with its efficient computation . we also describe the application of the le - la path metric to the soft - input soft - output @xmath0-algorithm . in section [ sec : performance ] , we present the performance analysis of the iss - ma . in section [ sec : simulation ] , we provide simulation results and conclude in section [ sec : conclusion ] . we briefly summarize the notation used in this paper . uppercase and lowercase letters written in boldface denote matrices and vectors , respectively . the superscripts @xmath12 and @xmath13 denote transpose and conjugate transpose ( hermitian operator ) , respectively . @xmath14 denotes an @xmath15-norm square of a vector and @xmath16 is a diagonal matrix that has elements on the main diagonal . @xmath17 and @xmath18 are @xmath19 matrix whose entries are all ones or zeros , respectively . the subscript is omitted if there is no risk of confusion . @xmath20 denotes a circular symmetric complex gaussian density with mean @xmath21 and variance @xmath22 . @xmath23 $ ] denotes expectation over the random variable @xmath24 . @xmath25 denotes @xmath26 - e[\mathbf{x}]e[\mathbf{y}^{h}]$ ] . for a hermitian matrix @xmath27 , @xmath28 ( or @xmath29 ) means that @xmath27 is semi - positive definite ( or positive definite ) . @xmath30 means probability of the event @xmath31 . @xmath32 denotes a joint probability density function ( pdf ) for the random variables @xmath33 . in this section , we briefly review the idd framework and then introduce the tree detection algorithms . in a transmitter , a rate @xmath34 channel encoder is used to convert a sequence of independent identically distributed ( i.i.d . ) binary information bits @xmath35 to an encoded sequence @xmath36 . the bit sequence @xmath36 is permuted using a random interleaver @xmath37 and then mapped into a symbol vector using a @xmath38-ary quadrature amplitude modulation ( qam ) symbol alphabet . we label the interleaved bits associated with the @xmath39th symbol @xmath40 by @xmath41 . due to the interleaver , we assume that these interleaved bits are mutually uncorrelated . in the system model ( [ eq : start ] ) , @xmath4 and @xmath5 are the @xmath6 received signal and noise vectors , respectively . each entry of the @xmath3 symbol vector @xmath2 is drawn from a finite alphabet @xmath42 where @xmath43 is chosen to satisfy the normalization condition @xmath44 = 1 $ ] . for example , @xmath45 for @xmath46-qam and @xmath47 for @xmath48-qam modulation , respectively . [ fig : idd ] depicts the basic structure of an idd system . the receiver consists of two main blocks ; the app detector and the channel decoder . the app detector generates the _ a posteriori _ log - likelihood ratio ( llr ) of @xmath49 using the observation @xmath4 and _ a priori _ information delivered from the channel decoder . the _ a posteriori _ llr is defined as @xmath50 where we take @xmath51 rather than @xmath52 by convention . with the standard noise model @xmath53 , ( [ eq : softmap ] ) can be rewritten @xcite @xmath54 where @xmath55 the set @xmath56 is the set of all configurations of the vector @xmath2 satisfying @xmath57 ( @xmath58 is defined similarly ) , and @xmath59 is the _ a priori _ llr defined as @xmath60 . once @xmath61 is computed , the extrinsic llr is obtained from @xmath62 . these extrinsic llrs are de - interleaved and then delivered to the channel decoder . the channel decoder computes the extrinsic llr for the coded bits @xmath36 and feeds them back to the app detector . these operations are repeated until a suitably chosen convergence criterion is achieved @xcite . the direct computation of the _ a posteriori _ llr in ( [ eq : mapprom ] ) involves marginalization over @xmath63 symbol candidates , which easily becomes infeasible for large systems employing high order modulations . a tree detection algorithm addresses this problem by searching a small set of promising symbol candidates over which _ a posteriori _ llrs are estimated . specifically , a small number of symbol vectors with large @xmath64 , equivalently , small @xmath65 , are sought . in the sequel , we refer to @xmath66 as a _ cost metric _ for tree detection , where @xmath67 is a scaling factor . the goal of the tree detection algorithm is to find symbol vectors of small cost metric , and the best ( minimum ) among them corresponds to the _ maximum a posteriori _ ( map ) solution ( denoted by @xmath68 ) . the tree detection algorithm relies on a tree representation of the search space spanned by @xmath69 . tree construction is performed from the root node as follows . first , representing the symbol realization for @xmath70 , we extend @xmath38 branches from the root ( recall that we assume @xmath38-ary qam modulation ) . for each such branch , @xmath38 child branches are extended for the possible realization of the next symbol @xmath71 . these branch extensions are repeated until all branches corresponding to @xmath72 are generated . this yields a tree of the depth @xmath73 , where each complete " path from the root to a leaf corresponds to a realization of @xmath2 . in order to find the complete paths of small cost metric , the tree detection algorithm searches the tree using a systematic node visiting rule . for notational simplicity , we henceforth denote a path associated with a set of symbols @xmath74 by a column vector @xmath75^{t}$ ] . also , we call a level of tree associated with the symbol @xmath76 the @xmath77th level " ( e.g. , the bottom level associated with @xmath78 is the first level ) . for details on tree construction , see @xcite . for a systematic search of symbol candidates , a _ path metric _ is assigned to each path @xmath79 . towards this end , we perform a qr decomposition of @xmath7 as @xmath80 \begin{bmatrix } \mathbf{r } \\ \mathbf{0}\end{bmatrix},\ ] ] where @xmath81 has an @xmath82 upper - triangular matrix whose diagonals are non - negative and @xmath83 is an @xmath8 matrix satisfying @xmath84 . using the invariance of the norm to unitary transformations , we can define the cost metric @xmath85 as @xmath86 where @xmath87 , and @xmath88^{t } = \mathbf{q}_{1}^{h}\mathbf{y}_{o}$ ] and @xmath89 . the path metric associated with the path @xmath90 can be defined as a partial sum in the cost metric @xcite @xmath91 whenever a new node is visited , the term @xmath92 , referred to as a branch metric , is added to the path metric of the parent node . since the branch metric is non - negative for all @xmath77 , the path metric @xmath93 becomes a lower bound of the cost metric @xmath85 . using @xmath93 , the tree detection algorithm compares the reliability of distinct paths and chooses the surviving paths . since the path metric is determined by the visited path , we henceforth denote @xmath93 as a _ causal path metric_. according to a predefined node visiting rule @xcite , the tree detection algorithm attempts to find the _ complete paths _ associated with smallest cost metric . denoting the set of the corresponding symbol candidates as @xmath94 , an approximate app can be expressed as @xmath95 further simplification can be achieved using max - log approximation @xcite @xmath96 since @xmath94 does not span whole symbol space , either @xmath97 or @xmath98 might be empty for some values of @xmath39 . if this case happens , the magnitude of @xmath61 is set to infinity , causing a bias in llr values . one way to cope with this event is to clip the magnitude of @xmath61 to a constant value ( e.g. , @xmath99 ) @xcite . in this section , we present the iss - ma , which improves candidate selection process of the soft - input soft - output tree detection algorithms . we first describe a genie - aided path metric that motivates our work and then introduce the new path metric that accounts for the information on unvisited paths . we also discuss an efficient way to compute the new path metric . we begin our discussion with the following path metric : a _ genie - aided path metric _ @xmath100 is defined as @xmath101 the genie - aided path metric is obtained by minimizing the sum of @xmath102 over all combinations of undecided symbols @xmath103 . this minimal term , which can be considered as a bias term , is added into the causal path metric . the genie - aided path metric can be used in the @xmath0-algorithm so that the best @xmath0 candidates with the smallest genie - aided path metric are selected in each tree level . it is easy to show that the @xmath0-algorithm employing the genie - aided path metric finds the closest ( best ) path with probability one ( even for @xmath104 ) . this can be readily shown since the genie - aided path metric provides the smallest cost metric among all tail paths . note that a similar path metric appeared in @xcite . [ theorem : map ] given the actual transmitted symbol vector @xmath105 ( i.e. , @xmath106 ) , the bias term of the genie - aided path metric is @xmath107 where the minimizer @xmath108 is the map estimate of @xmath103 , i.e. , @xmath109 see appendix [ app : map ] . _ theorem _ [ theorem : map ] implies that the bias term of the genie - aided path metric is obtained by computing @xmath110 using the map estimate of @xmath103 . this map estimate is derived under the condition that the path associated with the actual transmitted symbols , @xmath105 is given . though the genie - aided path metric offers a substantial performance gain , it is impractical to incorporate it into tree search due to the high complexity associated with the map estimation . in order to alleviate the complexity associated with map detection of @xmath103 in the genie - aided path metric , we relax the finite alphabet constraint of @xmath103 and then replace the map estimate by the linear mmse estimate @xmath111 . note that when @xmath111 is assumed to be gaussian , the map estimate is identical to the linear mmse estimate @xcite . for a particular path visited @xmath90 , we first define the le - la path metric the _ linear estimate - based look - ahead path metric _ , denoted by @xmath112 , is defined as @xmath113 where @xmath111 is the linear mmse estimate of @xmath103 . note that @xmath111 is obtained under the condition that @xmath106 . in the sequel , we denote this bias term as @xmath114 . to derive the linear mmse estimate @xmath111 , we partition the vectors @xmath115 and @xmath116 to @xmath117 and @xmath118 vectors , i.e. , @xmath119 where @xmath120 , @xmath121 , and @xmath122 are the adequately partitioned sub - matrices of @xmath81 . using ( [ eq : partition22 ] ) , @xmath112 can be expressed as @xmath123 where @xmath124 and @xmath125 . note that the term generated by _ a priori _ information @xmath126 considers only @xmath90 since the symbols @xmath103 are undecided . note also that the linear mmse estimate of the non - causal symbols @xmath103 is given by @xcite @xmath127 \right ) + e\left[\mathbf{x}_{1}^{k-1}\big|\mathbf{x}_{k}^{n } = \widetilde{\mathbf{x}}_{k}^{n}\right ] \\ \label{eq : softmmse2 } & = \mathbf{f}_{k } \left(\mathbf{y}_{1}^{k-1 } - \mathbf{r}_{11,k}\overline{\mathbf{x}}_{1}^{k-1 } - \mathbf{r}_{12,k}\mathbf{x}_{k}^{n } \right ) + \overline{\mathbf{x}}_{1}^{k-1 } , \end{aligned}\ ] ] where @xmath128 $ ] and @xmath129 @xmath130 . we can obtain @xmath131 and @xmath132 from _ a priori _ llrs as @xcite @xmath133 @xmath134 where @xmath135 and @xmath136 . the set @xmath137 includes all possible constellation points . in the first iteration of the idd where _ a priori _ llrs , @xmath138 are unavailable , @xmath139 and @xmath140 . using ( [ eq : biastm2 ] ) and ( [ eq : softmmse2 ] ) , @xmath114 can be rewritten @xmath141 where @xmath142 further , denoting @xmath143 and @xmath144 , @xmath112 can be simply expressed @xmath145 note that the bias term @xmath146 of the le - la path metric can be computed with only linear operations . note also that _ a priori _ information obtained from the channel decoder is reflected through @xmath131 and @xmath147 in the bias term . in this subsection , we discuss how the le - la path metric can be computed efficiently . recalling that the bias term is expressed as @xmath148 , computation of the path metric is divided into two steps ; 1 ) computation of @xmath149 for all @xmath39 prior to the tree search and 2 ) recursive update of the path metric for each branch extension during the search . first , using a matrix inversion formula for block matrices ( * ? ? ? * _ appendix 1.1.3 _ ) , the operators @xmath150 in ( [ eq : z2 ] ) can be computed recursively . denoting @xmath151 and @xmath152^{t}$ ] , then @xmath153 is expressed as a function of @xmath149 as @xmath154 where @xmath155 in particular , @xmath156 . see appendix [ app : recursive ] for the derivation of ( [ eq : recursive ] ) . if the _ a priori _ llrs are all zero , @xmath149 does not need to be computed for every symbol as long as the channel remains constant . if the _ a priori _ llrs are non - zero , these steps are performed for each symbol . however , the required computations can be further reduced by replacing the instantaneous covariance matrix @xmath147 by its time - average over a coherent time @xcite . next , the le - la path metric can be recursively updated for each tree extension . at the root node , a vector is defined such that @xmath157 . the vector @xmath158 is updated from that of its parent node as @xmath159 where @xmath160 is a scaler variable . using the vector for each path @xmath90 , the le - la path metric can be obtained as @xmath161 where @xmath162 . noting that the dimension of the matrix @xmath149 is @xmath163 , the number of complex multiplications for the bias term computation is proportional to @xmath164 . in order to reduce the complexity , we can look ahead only @xmath165 symbols instead of all non - causal symbols . towards this goal , we set @xmath166 and repartition the system as @xmath167 where @xmath168 and @xmath169 are the redefined sub - matrices of ( [ eq : partition22 ] ) , respectively . in this case , the bias term defined in section [ subsec : derivation ] needs to be modified based on this partitioning . in doing so , the dimension of @xmath168 and @xmath149 is reduced from @xmath170 to @xmath171 . the recursive computation of @xmath149 employing the new partitioning can be derived without matrix inversion ( see ( * ? ? ? * _ section iii . in addition , in ( [ eq : dum3 ] ) , we only need to multiply @xmath149 with the last @xmath172 elements of @xmath158 . overall , by using only @xmath173 non - causal symbols for the bias term , the number of operations for the bias computation can be reduced from @xmath174 to @xmath175 . p15 cm + * output * : @xmath176 , i=[1:q]}$ ] + * input * : @xmath115 , @xmath7 , @xmath177 , i=[1:q]}$ ] , @xmath178 and @xmath179 + + * step 1 * : ( preprocessing ) order @xmath2 and @xmath7 according to v - blast ordering @xcite or b - chase preprocessing @xcite . then , compute @xmath149 for all @xmath39 . + * step 2 * : ( initialization ) initialize @xmath180 and start the tree search from the root node . + * step 3 * : ( loop ) extend @xmath38 branches for each of @xmath0 paths that have survived at the @xmath181th level . this generates @xmath182 paths at the @xmath77th level . + * step 4 * : if @xmath183 , choose the @xmath0 best paths with the smallest @xmath184 and go to step 3 with @xmath185 . otherwise , store all @xmath182 survival candidates into the list @xmath94 and go to step 5 . + * step 5 * : ( list extension & app calculation ) for each value of @xmath39 and @xmath77 , compute @xmath186 based on @xmath94 . if the value of @xmath49 for all elements of @xmath94 is either + 1 or -1 , the value of @xmath49 of the best @xmath179 candidates ( associated with the minimum cost metric ) is flipped and these counter - hypothesis candidates are added to @xmath94 to generate the extended list @xmath187 . the app is calculated over the extended list based on ( [ eq : mapprom_dasi ] ) . + [ tb : malg ] in this section , we introduce the soft - input soft - output tree detection algorithm employing the le - la path metric . to reduce errors in early detection stages , symbol detection ordering is performed first . the v - blast ordering @xcite or b - chase preprocessing @xcite can be adopted . note that the b - chase preprocessing is preferred when @xmath0 is larger than the constellation size @xmath38 . in each level of the tree , @xmath112 of @xmath188 survival paths are compared and then the @xmath0 best paths are selected . starting from the root node , this candidate selection procedure continues to the bottom level , eventually producing @xmath188 complete paths . the symbol vectors corresponding to these complete paths generate a candidate list @xmath94 , over which the extrinsic llr for each bit is calculated . in the event that a particular bit in each of the candidates takes the same value ( all one or zero ) , the magnitude of the generated llr might become unduly large , limiting the error - correction capability of the channel decoder @xcite . in order to prevent this situation , whenever this occurs for the @xmath39th bit of the candidate list , the @xmath39th bits of the best @xmath179 candidates ( @xmath189 ) are flipped and added into the candidate list @xmath94 , generating an extended list @xmath187 . as a result , the size of @xmath187 becomes @xmath190 . a more desirable flipping method would be to flip the corresponding bit of all candidates and then select @xmath179 of them which would have the lowest cost function . since this method increases the complexity considerably , we employ the alternative that flips the best @xmath179 candidates . though the current approach would produce a slightly degraded counter - hypothesis set , we hope that it is highly likely to be , or to at least have a large overlap with , the aforementioned best counter - hypothesis set . using the list @xmath187 together with the max - log approximation , the app becomes @xmath191 a summary of the iss - ma is provided in table [ tb : malg ] . we discussed in the previous section that the transmitted symbols are always found with @xmath104 if the genie - aided path metric is used . relaxation of the finite alphabet constraint and gaussian approximation are made for undecided symbols to derive the le - la path metric . in this section , we analyze the performance of the proposed @xmath0-algorithm employing the le - la path metric . as a measure for performance , we consider the probability of a cpl event , i.e. , the probability that the tree search rejects a path associated with the transmitted symbols . in order to make the analysis tractable , we focus on the case when @xmath104 . although our analysis focuses only on the case @xmath104 , it is clear that lower cpl rate for @xmath104 implies a greater likelihood of a correct symbol being selected for @xmath192 as well . the performance analysis for @xmath192 is presented via computer simulations in section [ subsec : result ] . given the channel matrix @xmath81 and the _ a priori _ llrs , the probability of cpl can be expressed as @xmath193 where @xmath194 denotes the set of the paths selected at the @xmath39th level and @xmath195 is the probability given that @xmath196 is sent . since we consider the case of @xmath104 , @xmath197 implies that a correct path has been selected up to the @xmath198th level . with this setup and from ( [ eq : partition22 ] ) , ( [ eq : biastm1 ] ) , and ( [ eq : bb2 ] ) , one can show that @xmath112 is given by @xmath199 where @xmath200 , and @xmath201^{t}$ ] . note that @xmath202 is independent of the selection of @xmath40 . the first term in ( [ eq : ptt4 ] ) can be interpreted as the distance metric between the output of a scalar additive noise channel @xmath203}{\sqrt{\mathbf{r}_{k}^{h}\mathbf{z}_{k}\mathbf{r}_{k } + |r_{k , k}|^{2 } } } \left [ \left(\mathbf{z}_{k}\mathbf{b}_{k}\right)^{t } \;\ ; n_{k}\right]^{t}\ ] ] and a symbol candidate @xmath204 . the iss - ma chooses the @xmath0 best symbols @xmath40 according to the cost metric in ( [ eq : ptt4 ] ) . since the _ a priori _ term @xmath205 in ( [ eq : ptt4 ] ) leads to better detection , we ignore the impact of it in our discussion . if we let @xmath206 = \mathbf{\sigma}_{k}= \left(\mathbf{r}_{11,k}\mathbf{\lambda}_{k } \mathbf{r}_{11,k}^{h } + \sigma_{n}^{2 } \mathbf{i}\right)$ ] and @xmath207 , then the signal to interference plus noise ratio ( sinr ) of the scalar additive noise channel is given by @xmath208 \mathbf{z}_{k}^{2 } \mathbf{r}_{k } + \sigma_{n}^{2}|r_{k , k}|^{2 } } \\ \label{eq : rhos2 } = & \frac{1}{\sigma_{n}^{2}}\frac{\left(\mathbf{r}_{k}^{h}\left(\sigma_{n}^{4 } \mathbf{\sigma}_{k}^{-2}\right)\mathbf{r}_{k } + |r_{k , k}|^{2}\right)^{2}}{\mathbf{r}_{k}^{h}\left(\sigma_{n}^{6 } \mathbf{\sigma}_{k}^{-3}\right)\mathbf{r}_{k } + |r_{k , k}|^{2}}.\end{aligned}\ ] ] [ lemma : sinr_bound ] the sinr in ( [ eq : rhos2 ] ) is bounded by @xmath209 see appendix [ app : sinr_bound ] . taking similar steps , one can show that the sinr for the causal path metric @xmath93 is @xmath210 . hence , @xmath211 and @xmath212 can be regarded as upper and lower bounds on the sinr gain achieved by the le - la path metric , respectively . it is of interest to check the behavior of the upper and lower bound of sinr gain for high dimensional systems . suppose that @xmath213 with a fixed aspect ratio @xmath214 ( @xmath215 ) , and let @xmath216 and @xmath217 be smallest and largest diagonals of @xmath147 , respectively . then , we attain a looser bound on the sinr is @xmath218 where the upper and lower bound of the sinr gain are denoted as @xmath219 and @xmath220 , respectively . note that ( [ eq : bound2 ] ) can be shown by the relationship @xmath221 ( equivalently , @xmath222 and @xmath223 ) , where @xmath224 and @xmath225 and @xmath226 implies that the matrix @xmath227 is positive semi - definite . [ theorem : random ] for an @xmath8 matrix @xmath7 whose elements are i.i.d . random variables with zero mean and variance @xmath228 , the upper and lower bound of the sinr gain for the level @xmath229 converge to @xmath230 as @xmath213 with @xmath214 , where @xmath231 . see appendix [ app : random ] . [ corol : asymp ] as @xmath232 , we have @xmath233 in addition , as @xmath232 , @xmath234 monotonically increases and approaches @xmath235 . we can deduce from ( [ eq : stop ] ) and ( [ eq : stop2 ] ) that the actual sinr gain approaches a deterministic value between @xmath236 $ ] . one can also show that both @xmath234 and @xmath237 are an increasing functions of @xmath238 . noting that @xmath239 indicates an index for tree depth , the sinr bounds achieve their maximum at the top level of the tree @xmath240 . next , we analyze the probability of cpl using the sinr obtained . it is worth taking a close look at the term @xmath241 in ( [ eq : additive ] ) . recalling that @xmath200 and @xmath242 , @xmath241 is an mmse estimate of @xmath243 @xcite . in order to make the derivation of the cpl probability more tractable , we use a gaussian approximation for the mmse estimation error @xmath244 or equivalently , @xmath241 . under this approximation , we can assume that the interference plus noise of the scalar channel is gaussian . the validity of this approximation has been supported in many asymptotic scenarios in @xcite and @xcite . in particular , it is shown that the gaussian approximation is highly accurate for large problem size @xmath73 @xcite . using the sinr in ( [ eq : rhos2 ] ) , the probability of cpl for the @xmath39th level detection can be expressed as @xcite @xmath245 where @xmath246 . the inequality in ( [ eq : exact ] ) follows from the existence of _ a priori _ terms in ( [ eq : ptt4 ] ) , which lowers the actual cpl probability . from ( [ eq : bound ] ) , we have @xmath247 using ( [ eq : cplq ] ) , we can analyze an average probability of cpl for a random channel @xmath7 whose elements are independent complex gaussian with @xmath248 . the average probability of cpl , denoted as @xmath249 , is given by @xmath250 \\ = & \sum_{k=1}^{n } e_{\mathbf{h}}\left [ \underline{\pr } \left ( \left . \widetilde{\mathbf{x}}_{k}^{n } \notin \mathcal{l}_{k } \right|\widetilde{\mathbf{x}}_{k+1}^{n } \in \mathcal{l}_{k+1 } , \mathbf{h } \right ) \right ] + \mbox{higher order terms},\end{aligned}\ ] ] where @xmath251 $ ] denotes the expectation over @xmath7 . the average cpl probability is obtained after evaluating @xmath252 $ ] for all @xmath39 . in our analysis , we do not put our emphasis on the higher order terms since they become negligible in the high snr regime . using the relationship @xmath253 for @xmath254 and from ( [ eq : cplq ] ) , we have @xmath255 \\ & \leq 4\left(1-\frac{1}{\sqrt{2^q}}\right ) e_{\mathbf{h}}\left[q\left(\sqrt{k\frac{|r_{k , k}|^{2}}{\sigma_{n}^{2}}}\right ) \exp\left(-\frac{k\sigma_{n}^{2}\mathbf{r}_{k}^{h}\mathbf{\sigma}_{k}^{-2}\mathbf{r}_{k}}{2 } \right)\right ] \\ & = 4\left(1-\frac{1}{\sqrt{2^q}}\right ) e_{\mathbf{h}}\left[q\left(\sqrt{k\frac{|r_{k , k}|^{2}}{\sigma_{n}^{2}}}\right)\right]\label{eq : avg } e_{\mathbf{h}}\left [ \exp\left(-\frac{k\sigma_{n}^{2}\mathbf{r}_{k}^{h}\mathbf{\sigma}_{k}^{-2}\mathbf{r}_{k}}{2 } \right ) \right ] , \end{aligned}\ ] ] where ( [ eq : avg ] ) follows from independence of @xmath256 and @xmath257 . noting that @xmath256 has a chi - square distribution with @xmath258 degrees of freedom and @xmath257 has independent complex gaussian elements ( * ? ? ? * _ lemma 2.1 _ ) , we have @xcite , @xmath259 = & \left(\frac{1}{2 } - \frac{1}{2 } \sqrt{\frac{k}{k + 2\sigma_{n}^{2}}}\right)^{l - k+1 } \sum_{l=0}^{l - k}{l - k+l \choose l}\left(\frac{1}{2 } + \frac{1}{2 } \sqrt{\frac{k}{k + 2\sigma_{n}^{2}}}\right)^{l } \label{eq : first1 } % \\\label{eq : first2 } % \approx & { 2(l - k)+1 \choose l - k+1}\left(\frac{\sigma_{n}^{2}}{2k}\right)^{l - k+1}.\end{aligned}\ ] ] [ lemma : scaling ] an upper bound on the scaling gain in ( [ eq : avg ] ) is given by @xmath260 \leq & \int_{0}^{\infty } \cdots \int_{0}^{\infty } \left(\prod_{i=1}^{k-1 } \frac{1}{1+\frac{k}{2}\frac{\sigma_{n}^{2}}{\left(\lambda_{\rm max}x_{i}+\sigma_{n}^{2}\right)^{2}}}\right ) \nonumber \\ \label{eq : scale } & \times f_{\eta_{1 } , \cdots , \eta_{k-1}}\left(x_{1 } , \cdots , x_{k-1}\right ) dx_{1 } \cdots dx_{k-1},\end{aligned}\ ] ] where @xmath261 see appendix [ app : scaling ] . while @xmath262 in ( [ eq : scale ] ) tends to one as @xmath232 , ( [ eq : first1 ] ) decreases to zero with a slope @xmath263 . therefore , at high snr , the probability of cpl for the top level @xmath240 would dominate , i.e. , @xmath264 e_{\mathbf{h}}\left [ \exp\left(-\frac{k\sigma_{n}^{2}\mathbf{r}_{n}^{h}\mathbf{\sigma}_{n}^{-2}\mathbf{r}_{n}}{2 } \right ) \right],\ ] ] where the right - hand side is obtained from ( [ eq : first1 ] ) and ( [ eq : scale ] ) . following similar steps , we can also show that the upper bound of the average cpl probability for the causal path metric becomes @xmath265.\ ] ] we observe from ( [ eq : fin ] ) that the average cpl probability of the le - la path metric is smaller than that of the causal path metric by the factor of @xmath266 $ ] . since this term is strictly less than unity , it corresponds to the scaling gain obtained from the le - la path metric . in fig . [ fig : anal ] , we provide the plot of the average cpl probability versus snr for several system sizes ( @xmath267 , and @xmath268 ) . we assume uncoded qpsk transmission . the average cpl rate and its upper bound are obtained from ( [ eq : exact ] ) and ( [ eq : cplq ] ) . for a comprehensive view , we also include the average cpl rate for the causal path metric in ( [ eq : fin_causal ] ) . for all cases considered , the cpl expression in ( [ eq : exact ] ) is quite close to that obtained from the simulation results , supporting the accuracy of the analytic bound we obtained . in particular , the upper bound of the average cpl rate appears tight at high snr . [ fig : scale ] shows how the scaling gain in ( [ eq : scale ] ) varies as a function of snr and system size . we observe that the performance gain of the le - la path metric improves with system size and the maximum is achieved in low to moderate snr range ( 10 db @xmath269 20 db ) . notice that this behavior is desirable for idd , since the performance in low - to - mid snr range is critical in triggering performance improvement though iterations @xcite . in this section , we evaluate the performance of the iss - ma through computer simulations . first , we observe the performance of the soft - input soft - output @xmath0-algorithm employing the le - la path metric and that employing the conventional path metric . note that the le - la path metric is not restricted to a particular search scheme and can be extended to more sophisticated breadth - first search algorithms ( such as @xcite and @xcite ) . next , we compare the performance - complexity trade - off of the iss - ma with the existing soft - input soft - output detectors . the simulation setup for the idd system is as follows . a total of @xmath270 information bits are randomly generated . a rate @xmath271 recursive systematic convolutional ( rsc ) code with feedback polynomial @xmath272 and feedforward polynomial @xmath273 . we use a random interleaver of size of @xmath274 bits . we use a gray mapping for qam modulation . we assume fast fading channels where each entry of @xmath7 is i.i.d . complex gaussian @xmath248 and perfect knowledge of the channel state at the receiver is assumed . for the channel decoding , a _ max - log - map _ decoder @xcite is employed . the snr is defined as @xmath275 . computational complexity of detectors is measured by counting the average number of complex multiplications per symbol period and per iteration . first , we compare the performance of the causal path metric and the le - la path metric . we consider the @xmath276 16-qam mimo system , which requires high detection complexity . for fair comparison , we employ the same candidate extension strategy with @xmath277 ( described in section [ subsec : application ] ) for both algorithms . the parameter @xmath178 is set to 5 for the le - la path metric . in fig [ fig : bervsm ] , the plots of bit error rate ( ber ) versus snr are provided for several @xmath0 values ( @xmath278 and @xmath279 . each plot shows the ber curves obtained after a different number of iterations . the iss - ma outperforms the conventional @xmath0-algorithm for all @xmath0 values and after each iteration . in particular , with @xmath280 , the iss - ma shows remarkable performance gain ( more than 5 db gain ) . then , the performance gap decreases as @xmath0 increases . note that the iss - ma maintains strong performance even with small @xmath0 ( e.g. @xmath280 ) . table [ tb : comp1 ] provides computational complexity of both algorithms along with the snr required to achieve the ber of @xmath281 for the same setup . the snr is measured after the 7th iteration . in order to compare performance - complexity trade - off , it is worth looking at the performance of the iss - ma with @xmath280 and the @xmath0-algorithm with @xmath282 , where both algorithms require similar computational complexity . in these cases , the iss - ma achieves almost 1 db performance gain . we can additionally observe that the performance of the iss - ma converges faster than the conventional @xmath0-algorithm , which might also help reducing the complexity of the iss - ma by the early termination of the iterations . next , we take a look at how the performance gap between the iss - ma and the conventional @xmath0-algorithm changes in terms of different system size . table [ tb : comp2 ] presents the snr at @xmath283 ber and complexity of both algorithms for @xmath284 and @xmath285 . 16-qam is used and @xmath173 and @xmath0 are set to @xmath286 and @xmath287 for all cases . the performance is measured after the 7th iteration . we observe that the performance gain due to the le - la path metric increases with system size . in particular , the gain of the iss - ma for @xmath288 system is 0.5 db and that increases to 1.75 db for the @xmath289 system . this clearly demonstrates that future cost plays a key role for large systems . we next investigate the performance of the iss - ma as a function of the parameter @xmath173 ( see section [ subsec : efficient ] ) . in our simulations , the @xmath289 16-qam transmission is considered and @xmath0 is set to @xmath290 . [ fig : nt ] ( a ) and ( b ) show the performance and complexity of the iss - ma for different @xmath173 . note that the iss - ma with @xmath291 reduces to the conventional @xmath0-algorithm . as the parameter @xmath173 increases , the iss - ma accounts for the further future cost so that the computational complexity increases and performance improves . the iss - ma offers performance - complexity trade - off through @xmath173 . while the performance of the iss - ma improves much for small @xmath173 values , the effect of @xmath173 diminishes with larger @xmath173 . it is shown that the choice of @xmath292 is sufficient to achieve the maximal performance gain offered by the le - la path metric for the cases considered . we also take a look at the performance of the iss - ma for spatially correlated mimo channels . we model a correlated mimo channel as @xmath293 , where @xmath294 is the @xmath295 transmit correlation matrix and @xmath296 is the @xmath297 receive correlation matrix . [ fig : correlated ] shows the plot of ber vs. snr of the iss - ma and the conventional soft - input soft - output @xmath0-algorithm for correlated channels with @xmath298 note that two antennas are less correlated as the space between them increases . the @xmath289 16-qam system is considered and the parameters @xmath0 and @xmath173 are set to @xmath285 and @xmath286 , respectively . comparing the result shown in fig . [ fig : correlated ] with that in fig . [ fig : bervsm ] ( d ) , we observe that the performance of both algorithms degrades in correlated channels , but the performance gain of the iss - ma over the conventional algorithm is even larger . from this observation , we can deduce that the le - la path metric can be more powerful when channel gains are correlated . finally , we check the performance - complexity trade - off of the iss - ma along with those of the existing soft - input soft - output tree detection algorithms . for a comprehensive picture , we consider the following algorithms ; 1 . mmse - pic algorithm ; mmse parallel interference cancellation detector @xcite . this detector subtracts _ a priori _ estimates of the interfering symbols from the received vector and then applies a linear mmse estimator to obtain soft estimate of the symbols . 2 . liss algorithm ( @xmath299 ) ; list sequential stack algorithm @xcite . it is characterized by the size of stack @xmath300 and that of auxiliary stack @xmath301 . lfcsd algorithm ( @xmath302 ) ; list fixed complexity sphere decoder @xcite . a candidate list is found by the fixed complexity sphere decoder proposed in @xcite . this detector is characterized by @xmath303 and @xmath304 , which represent the size of the candidate list and the number of paths fully extended , respectively . 4 . its algorithm ( @xmath0 ) ; iterative tree search @xcite . this detector uses the conventional @xmath0-algorithm to find the candidate list . note that the parameters of the iss - ma are remarked in iss - ma ( @xmath0 , @xmath173 ) " . although the lsd @xcite and single ( parallel ) tree search ( sts ) @xcite are considered as powerful detection schemes , their complexities grow so rapidly with problem size they are infeasible for the @xmath289 system . for this reason , we only consider fixed - complexity detectors . in fig . [ fig : bervscomp ] , the performance and complexity of each algorithm are drawn in the same plot to compare the performance - complexity trade - off of the detectors . due to linear structure , the mmse - pic achieves the lowest complexity among all candidates . in addition , the performance of the mmse - pic is better than that of the liss , lfcsd , and its . this would be why the performance of the latter detectors depends on candidate size and the size is not large enough to achieve good performance in the @xmath289 system . in particular , due to the limited stack size , the stack memory used in the liss easily becomes full before reaching a leaf of the tree so that the liss often fails to find reliable candidates . [ fig : bervscomp ] shows that only the iss - ma can achieve the better performance than the mmse - pic . due to improved candidate selection process , the iss - ma finds reliable candidates only with small candidate size , thereby yielding the best ber performance while maintaining reasonable complexity . in conclusion , the iss - ma achieves the best performance - complexity trade - off among all tree detectors considered . in addition , the iss - ma provides performance gains over the mmse - pic at the expense of higher , but manageable complexity . in this paper , we discussed a new path metric , which shows great promise in terms of its performance - complexity trade - off for soft - input soft - output tree detection in an idd system . by accounting for non - causal symbols in the _ linear estimate - based look - ahead _ ( le - la ) path metric , the performance gains over the existing causal path metric are achieved . we apply the le - la path metric to the soft - input soft - output @xmath0-algorithm . by adopting the sorting mechanism exploiting the le - la path metric , we could improve the chance of selecting the correct path dramatically , thereby achieving good detection and decoding performance with fewer iterations . from cpl probability analysis , we observed that the le - la path metric reflects the reliability of selected paths much better than the causal path metric . computer simulations confirm that the proposed iss - ma can be a promising candidate for soft - input soft - output detection in high dimensional systems . the transformed vector @xmath115 can be expressed as @xmath305 , where @xmath306 . let @xmath39 be the current layer being searched then @xmath115 , @xmath2 , and @xmath307 can be partitioned into two @xmath117 and @xmath308 vectors , i.e. , @xmath309 where the upper triangular matrix @xmath81 is partitioned into four sub - matrices . given the transmitted symbol @xmath310 , _ a posteriori _ probability of @xmath103 is given by @xmath311 hence , we can show that @xmath312 where the equation ( [ eq : proofmap ] ) follows from the definition of the branch metric . hence , for @xmath106 , we have @xmath313 . let @xmath333 be a matrix generated from the first @xmath334 columns of @xmath7 . since @xmath335 , the matrices @xmath336 and @xmath337 share same eigenvalues . for an i.i.d . random matrix @xmath7 , the elements of @xmath257 are zero - mean and independent with variance of @xmath228 . according to ( * ? ? ? * _ lemma 2.29 _ ) , as @xmath213 with @xmath338 , @xmath219 converges almost surely to @xmath339 where @xmath340 is an empirical eigenvalue distribution of @xmath337 . according to the marcenko - pastur law ( * ? ? ? * _ theorem 2.35 _ ) , as @xmath213 with @xmath338 , @xmath340 converges almost surely to @xmath341 where @xmath342 . hence , from ( [ eq : vm ] ) and ( [ eq : marcenko ] ) , we obtain @xmath343 we can express @xmath153 in ( [ eq : z2 ] ) as @xmath314^{-1 } \\ = & \sigma_{n}^{2 } \left(\begin{bmatrix}\mathbf{r}_{11,k } & { \mathbf{r}}_{k+1 } \\ \mathbf{0 } & r_{k+1,k+1}\end{bmatrix } \begin{bmatrix}\mathbf{\lambda}_{k } & \mathbf{0 } \\ \mathbf{0 } & \lambda_{k+1}\end{bmatrix}\begin{bmatrix}\mathbf{r}_{11,k } & { \mathbf{r}}_{k+1 } \\ \mathbf{0 } & r_{k+1,k+1}\end{bmatrix}^{h } + \begin{bmatrix } \sigma_{n}^{2}\mathbf{i}_{k } & \mathbf{0 } \\ \mathbf{0 } & \sigma_{n}^{2 } \end{bmatrix } \right)^{-1 } \\ \label{eq : zbl3 } = & \sigma_{n}^{2 } \left(\begin{bmatrix}\sigma_{n}^{2}\left(\mathbf{z}_{k}\right)^{-1 } + \lambda_{k+1}{\mathbf{r}}_{k+1}{\mathbf{r}}_{k+1}^{t } & \lambda_{k+1 } r_{k+1,k+1 } { \mathbf{r}}_{k+1 } \\ \lambda_{k+1 } r_{k+1,k+1 } { \mathbf{r}}_{k+1}^{t } & \lambda_{k+1}r_{k+1,k+1}^{2 } + \sigma_{n}^{2 } \end{bmatrix}\right)^{-1}\end{aligned}\ ] ] to obtain the update formula , for partitioned matrices , @xmath27 given by ( * ? ? ? * _ appendix 1.1.3 _ ) @xmath315 we have @xmath316 let @xmath317 , @xmath318 , @xmath319 , and @xmath320 , then ( [ eq : zbl3 ] ) becomes @xmath321 where @xmath322 let @xmath323 be decomposed to @xmath324 , where @xmath325 are the eigenvalues of @xmath323 . then , the upper bound of the sinr is given by @xmath326 where @xmath327^{t } = \mathbf{u } \mathbf{r}_{k}$ ] and ( [ eq : cauchy1 ] ) is from the _ cauchy - schwarz _ inequality . let @xmath345 be the unordered eigenvalues of @xmath336 . the scaling gain in ( [ eq : avg ] ) can be expressed as @xmath346 & \leq e_{\mathbf{h}}\left [ \exp\left(-\frac{k}{2}\sigma_{n}^{2}\mathbf{r}_{k}^{h}\left(\sigma_{n}^{2}\mathbf{i } + \lambda_{\rm max } \mathbf{r}_{11,k}\mathbf{r}_{11,k}^{h}\right)^{-2}\mathbf{r}_{k } \right)\right ] \\ & = e_{\mathbf{h } } \left [ \prod_{i=1}^{k-1}\exp\left(-\frac{k}{2 } \frac{\sigma_{n}^{2}}{\left(\lambda_{\rm max}\eta_{i}+\sigma_{n}^{2}\right)^{2 } } |r_{i , k}|^{2}\right ) \right ] \\\label{eq : sc3 } % & = e_{\mathbf{r}_{11,k } } \left [ \prod_{i=1}^{k-1 } e_{r_{i , k}}\left[\left.\exp\left(-\frac{k}{2 } \frac{\sigma_{n}^{2}}{\left(\lambda_{\rm max}\eta_{i}+\sigma_{n}^{2}\right)^{2 } } |r_{i , k}|^{2}\right ) \right| \mathbf{r}_{11,k}\right ] \right ] \\\label{eq : sc4 } % & = e_{\mathbf{r}_{11,k } } \left[\prod_{i=1}^{k-1 } \frac{1}{1+\frac{k}{2}\frac{\sigma_{n}^{2}}{\left(\lambda_{\rm max}\eta_{i}+\sigma_{n}^{2}\right)^{2}}}\right ] , % & \leq \left(\frac{1}{k}\right)^{k-1}e_{\mathbf{r}_{11,k}}\left[{\rm det}\left(\mathbf{r}_{11,k}\mathbf{\lambda}_{k } \mathbf{r}_{11,k}^{h } + \sigma_{n}^{2 } \mathbf{i}\right)\right]\end{aligned}\ ] ] where ( [ eq : sc3 ] ) is from @xmath347 = e[e[x|y]]$ ] and ( [ eq : sc4 ] ) follows from @xmath348 being @xmath349 and independent of @xmath120 . let @xmath333 be a matrix generated from the first @xmath334 columns of @xmath7 , then the matrices @xmath336 and @xmath337 share same eigenvalues . the pdf of the unordered eigenvalues of @xmath337 is given by @xcite @xmath350 which completes the proof . d. l. milliner , e. zimmermann , j. r. barry , and g. fettweis , a fixed - compelxity smart candidate adding algorithm for soft output mimo detection , " _ ieee journal of selected topics in signal processing _ , 1016 - 1025 , dec . 2009 . l. g. barbero and t. s. thompson , extending a fixed - complexity sphere decoder to obtain likelihood information for turbo - mimo systems , " _ ieee trans . 2804 - 2814 , sep . 2008 . d. wu , j. eilert , r. asghar and d. liu , vlsi implementation of a fixed - complexity soft - output mimo detector for high - speed wireless , " _ eurasip journal on wireless commun . and networking _ , vol . 2010 , pp . 1 - 13 , 2010 . a. d. murugan , h. e. gamal , m. o. damen , and g. caire , a unified framework for tree search decoding : rediscovering the sequential decoder , " _ ieee trans . information theory _ 933 - 953 , march 2006 . j. jaldn , l. g. babero , b. ottersten , and j. s. thompson , full diversity detection in mimo systems with a fixed - complexity sphere decoder , " _ ieee international conference on acoustics , speech , and signal processing _ , april 2007 , pp . iii-49-iii-52 . p. w. wolniansky , g. j. foschini , g. d.golden , and r. a. valenzuela , v - blast : an architecture for realizaing very high data rates over the rich - scattering wireless channel , " _ proc . signals , syst . , electron . 1998 , pp . 295 - 300 . y. s. han , c. r. p. hartmann , and c. chen , efficient priority - first search maximum - likelihood soft - decision decoding of linear block codes , " _ ieee trans . information theory , _ vol . 1514 - 1523 , sep . 1993 . r. johannesson and k. s. zigangirov , _ fundamentals of convolutional coding _ , wiley - ieee press , 1999 . f. xiong , a. zerik , and e. shwedyk , sequentional sequence estimation for channels with intersymbol interference of finite or infinite length , " _ ieee trans . 795 - 804 , june 1990 . h. v. poor and s. verdu , probability of error in mmse multiuser detection , " _ ieee trans . information theory _ , 858 - 871 , may 1997 . p. li , d. paul , r. narasimhan , and j. cioffi , on the distribution of sinr for the mmse mimo receiver and performance analysis , " _ ieee trans . information theory _ 271 - 286 , jan . 2006

the frequency of low - mass pre - main sequence binary systems is significantly lower in the trapezium cluster than in taurus - auriga . we investigate if this difference can be explained through stellar encounters in dense clusters . to this effect , a range of possible models of the well observed trapezium cluster are calculated using aarseth s direct n - body code , which treats binaries accurately . the results are confronted with observational constraints . the range of models include clusters in virial equilibrium , expanding clusters as a result of instantaneous mass loss , as well as collapsing clusters . in all cases the primordial binary proportion is larger than 50 per cent , with initial period distributions as observed in taurus - auriga and the galactic field . it is found that the expanding model , with an initial binary population as in the galactic field , is most consistent with the observational constraints . this raises the possibility that the primordial group of ob stars may have expelled the cluster gas roughly 50 000 yr ago . the cluster s bulk expansion rate is thus a key observable that needs to be determined . the other models demonstrate that the rapidly decreasing binary proportion , its radial dependence and the form of the period distribution , together with structural and kinematical data , are very useful diagnostics on the present and past dynamical state of a young cluster . in particular , kinematical cooling from the disruption of wide binaries is seen for the first time . 98.10.+z ; 98.20.-d ; 97.80.-d ; 97.10.bt

astro-ph9906460

to develop an improved understanding of star formation it is very useful to empirically map variations of the binary - star properties and the initial mass function with environment , such as metallicity , density and temperature of the star - forming cloud . however , a naive interpretation of observational data may result in misleading conclusions . for instance , a comparison of binary proportions in different stellar populations , without taking into account the dynamical history of the population , is ill - fated , because the distribution of orbital elements of a population carries a memory of the dynamical history ( kroupa 1998 ) . one purpose of this paper is to further illustrate , on the one hand this danger , and on the other hand , the richness in understanding that can be gained through stellar - dynamical studies of young stellar systems . the trapezium cluster in the orion star forming region is a young , populous stellar cluster , which contains a negligible amount ( @xmath0 ) of residual gas ( wilson et al . 1997 ; odell 1999 , private communication ) and for which a large amount of observational data has been collected over decades . thus , it is an excellent target for investigating the early dynamical evolution of a young cluster with a primordial binary proportion at least as large as in the galactic field . the cluster s age is estimated to be younger than 5 myr ( prosser et al . 1994 ; hillenbrand 1997 ) , and probably @xmath1 yr . the central stellar number density is very high , with the mean inter - stellar separation being about 6000 au ( mccaughrean & stauffer 1994 ) . the trapezium cluster is a well - suited sample for the study of star formation in a dense stellar environment , as opposed to the low - density young stellar groups in the taurus - auriga star forming region ( gomez et al . 1993 ; briceno et al . 1998 ) . variations in stellar properties may occur with star formation environment . for example , the observed pre - main sequence binary population in the trapezium cluster is significantly different than in taurus - auriga . in the latter approximately all stars are in binary systems ( khler & leinert 1998 ; duchene 1999 ; ghez et al . 1997 ) , while in the trapezium cluster the binary proportion is significantly lower ( prosser et al . 1994 ; petr 1998 ; petr et al . 1998 ; but see also padgett , strom & ghez 1997 ) , being comparable to what is found in the galactic field ( duquennoy & mayor 1991 ) . this , however , does not necessarily imply that most galactic field stars stem from trapezium - cluster type assemblages , there being evidence that the proportion of long - period binary systems is significantly below the galactic field value ( scally , clarke & mccaughrean , 1999 ) . the binary population may well decrease further in the trapezium cluster , if it is a bound entity . and it may have been higher at birth . a stellar system with a half - mass radius of @xmath2 pc , containing @xmath3 stars , with a mass @xmath4 , has a relaxation time @xmath5 myr , which is the time - scale over which the cluster looses memory of its initial conditions . these are characteristic observed values for the trapezium cluster . the three - dimensional velocity dispersion in the trapezium cluster is @xmath6 km / s , the crossing time being @xmath7 yr ( an overview of the dynamical evolution of young clusters can be found in bonnell & kroupa 1999 ) . a typical star may thus have crossed the central region many times , given that the age of the cluster is @xmath8 myr . this implies that collisions between stellar systems will have been frequent . it also implies that the birth structure of the cluster will have been eradicated . indications that this may be the case comes from the apparent lack of sub - structure ( bate , clarke & mccaughrean 1998 ) . the trapezium cluster may thus be just old enough to perhaps have allowed an initial taurus - auriga binary population to have dynamically evolved to the observed reduced value . alternatively , it may well be true that the lower binary proportion of the trapezium cluster as compared to the taurus - auriga star forming region is due to different star - forming conditions . durisen & sterzik ( 1994 ) point out that binary formation from fragmentation of collapsing and rotating clouds , or from a gravitational instability of massive proto - stellar disks , is more likely in low - temperature clouds . that the binary proportion is eroded efficiently in a dense cluster has been shown by kroupa ( 1995a ) . however , whether the observed difference between the trapezium cluster and taurus - auriga can be explained with this mechanism must be investigated by constructing self - consistent models , that more closely match the properties of the trapezium cluster . if such a study shows that encounters are not efficient enough for a wide variety of models of the trapezium cluster , then the conclusion must be that the binary proportion depends on star - forming conditions . the aim of this paper is to begin tackling this problem by comparing evolving model populations with the constraints given by observations . in this paper we present the first steps towards more realistic fully self - consistent models of embedded clusters in general , and of the trapezium cluster in particular . section [ sec : obs ] summarises the available observational constraints and section [ sec : models ] details the model assumptions . in section [ sec : code ] we explain the calculation and data - evaluation programmes . the results are presented in section [ sec : res ] , and sections [ sec : disc ] and [ sec : conc ] contain the discussion and conclusions , respectively . an accompanying paper ( kroupa 1999 ) generalises these results to different numbers of stars and cluster radii , with a view to discussing the dynamical state of the orion nebula cluster and the set of viable initial conditions . the aim of this investigation is to study the evolution of those observables that allow the dynamical state of a young cluster and its birth configuration to be constrained . such observables are the velocity dispersion , the central stellar volume number density and the radial dependence of the binary proportion . in an optical high - resolution study carried out with the hubble space telescope , 319 stars in a region with size @xmath9 were identified ( prosser et al . 1994 ) . in a near infrared study covering a field of @xmath10 ( @xmath11 ) centred on the trapezium cluster , mccaughrean et al . ( 1996 ) counted 700 systems ( we call a single as well as a binary star a system ) . binary stars with separations below the 0.7 arcsec seeing limit were not resolved . this corresponds to a resolution limit of @xmath12 au , or log@xmath13 , where @xmath14 is in days , for a system with a mass of @xmath15 . a high - spatial - resolution direct - imaging near - infrared study revealed even more stars in the innermost centre of the trapezium cluster ( mccaughrean & stauffer 1994 ) , and a remarkably high stellar density of @xmath16 stars per pc@xmath17 was estimated . mccaughrean & stauffer ( 1994 ) calculated that 29 systems are within the central spherical volume with a radius of @xmath18 pc . from the analysis of the hr - diagram , based on optical photometry and spectroscopy , hillenbrand ( 1997 ) derived a mean age for the whole orion nebula cluster less than 1 myr , with an age spread of a few myr . the age determination given by prosser et al . ( 1994 ) is limited to the stars associated with the trapezium cluster and led to a medium age of @xmath19 yr , with nearly no detectable age spread . the projected one - dimensional velocity dispersion in the observational plane , within @xmath21 pc of the centre of the trapezium cluster , was derived by jones & walker ( 1988 ) to be @xmath22 km / s from their proper - motion survey using photographic plates . the data indicate no anisotropy , but these authors , van altena et al . ( 1988 ) and tian et al . ( 1996 ) note that the plate - reduction algorithms eliminate any signature due to rotation and/or expansion or contraction . thus , it is presently unknown if the trapezium cluster is expanding or contracting . the analysis of new absolute proper motions of a few dozen stars in the inner region ( @xmath23 pc ) of the orion nebula cluster ( onc ) , of which the trapezium cluster is likely to be merely the central part , however , suggests that there may be some expanding motion ( frink , kroupa & rser 1999 ) . for the present analysis we adopt the velocity dispersion obtained from the more precise relative proper motions , which we take to be corrected for radial bulk motions . photographic plates have a spatial resolution of typically 2 arcsec , which corresponds to about 1000 au in the trapezium cluster . most binary systems are therefore not resolved in these surveys . observations using the speckle holography technique in the core of the trapezium cluster by petr et al . ( 1998 ) resolve binary systems with separations in the range @xmath24 au to @xmath25 au . of the 42 systems that appear projected within the central radius , @xmath26 pc , six are ob stars and four are binary systems within the studied range of separations . the apparent binary proportion of the entire sample is @xmath27 ( equation [ eqn : binfapp ] below ) , and the binary proportion for low - mass ( @xmath28 ) stars is @xmath29 . the subscript `` app '' means that the observed @xmath30 is the apparent binary proportion that an observer deduces from projected star positions within some range of separations to which the observational apparatus is sensitive . the binary proportion of low - mass stars within the central radius @xmath31 pc is found by prosser et al . ( 1994 ) using hst imaging to be @xmath32 in the separation range @xmath33 au to @xmath34 au . the orbital periods corresponding to the respective separations are given in fig . [ fig : fappa ] . both studies thus find a binary proportion , in the respective distance ranges sampled , that is similar to the galactic field value , implying no measurable radial dependence of @xmath30 for @xmath35 pc ( for details see prosser et al . 1994 , and petr et al . in this section we construct the model clusters . the appearance of the trapezium cluster today does not reflect the initial conditions when the stars decoupled dynamically from the gas , because the velocity dispersion of 2.5 pc / myr implies significant mixing or spreading within 0.51 myr . we therefore realize a variety of initial dynamical states . these cover models in cold collapse through initially virialised systems to clusters that expand . the following parameters are chosen to specify possible different model trapezium clusters at `` birth '' , i.e. at the time when stellar dynamics starts dominating over gas dynamics : * @xmath36 stars ( point masses ) ; -1 mm * the initial mass function ( imf ) deduced from a careful analysis of star - count data in the galactic field ( kroupa , tout & gilmore 1993 ) with power - law index @xmath37 for @xmath38 ; this imf is based on scalo s ( 1986 ) imf for stars with @xmath39 ; -1 mm * lower and upper stellar mass limits of @xmath40 and @xmath41 , respectively ; -1 mm * a plummer density distribution ( aarseth , henon & wielen 1974 ) ; -1 mm * the initial position and velocity vectors of the binary - star centre - of - masses and single stars are independent of stellar mass ; -1 mm * the velocity distribution of the binary centre - of - masses and single stars is isotropic . models in virial equilibrium ( a1 , a2 ) , expanding models ( b1 , b2 ) and two cases of a cold collapse ( models c1 and c2 ) are computed . details of the models are given in table [ table : models ] , the entries of which are described below and in the following sub - sections . the resulting cluster mass is @xmath42 . the models contain no gas mass . this is a safe assumption because wilson et al . ( 1997 ) and c.r . odell ( private communication ) estimate the mass of gas to be only a few @xmath43 in the trapezium cluster . the choice of the initial number of stars is guided by the evidence presented in section [ sec : numb_age ] , remembering that not all trapezium cluster stars will have been detected , as binary systems are usually not resolved in lower spatial resolution imaging studies . a binary proportion of 60 per cent implies @xmath44 stars . however , we choose @xmath45 since rapid reduction of the number of systems in the observed area ( @xmath46 pc@xmath47 ) is to be expected owing to expansion of the cluster and ejection of stars . severe computational complications are avoided by assuming the stars have no extent . the assumption that the velocity and position vectors are not correlated with the stellar mass contradicts the observations , which show that the massive stars are concentrated in the cluster core . however , the present assumption allows investigation of whether dynamical mass - segregation in young binary - rich clusters can lead to the observed mass - segregation on time - scales of 1 myr or less . this will be the subject of another paper . the initial placement of the massive stars in the present models is also consistent with the empirical finding that ultra - compact hii regions in areas of rich and active star formation are usually distributed throughout the region over spatial scales of about 1 pc ( e.g. megeath et al . 1996 ) . the initial half - mass radius , @xmath48 , is listed in column 2 of table [ table : models ] . for models a1 to b2 it is based on the results of mccaughrean & stauffer ( 1994 ) . columns 3 and 4 list , respectively , the number of systems and stars within the central radius @xmath18 pc , and the central density ( @xmath49 $ ] , where @xmath50 is the plummer radius ) is given in column 5 . the parameters for models a1 and a2 are a compromise in that they give a central density that is larger by an order of magnitude than the observed value , and a velocity dispersion that is somewhat smaller than the observed value for virial equilibrium . the central density corresponds to an inter - system spacing of 2200 au . in this case , the initial model median relaxation and half - mass diameter crossing times are @xmath51 myr and @xmath52 myr , respectively . the central number density decreases and the half - mass radius increases within about @xmath53 yr owing to mass segregation and associated expansion , and heating through binary stars . the precise evolution can not be anticipated though , because it is not possible to easily estimate the various channels of energy exchange that exist in such a complex gravitational system , and since such realistic computations have never been been performed heretofore . column 6 of table [ table : models ] contains the initial virial ratio , @xmath54 , where the nominator and denominator are , respectively , the total kinetic and potential energy of the whole cluster ( @xmath55 and @xmath56 , where @xmath57 is the cluster mass , @xmath58 is the three - dimensional velocity dispersion , @xmath59 is a characteristic cluster radius , and @xmath60 is the gravitational constant ) . one possible state is virial equilibrium , @xmath61 . however , the cluster may be expanding or contracting ( see section [ sec : veld ] ) . to cover such possibilities , cluster models in virial equilibrium , expanding ( @xmath62 ) and cold collapse ( @xmath63 ) are studied . if the total cluster mass before mass loss is @xmath64 , and if the cluster is initially in virial equilibrium , then for a sudden mass loss of amount @xmath65 , the new @xmath66 where @xmath67 is the star - formation efficiency . assuming the cluster instantly looses a mass @xmath68 ( @xmath69 ) , @xmath62 , and the velocities are larger by a factor of @xmath70 than in a cluster in virial equilibrium with a mass @xmath71 . thus , in the case of the expanding models ( b1 , b2 ) , the initial velocities are increased by a factor of 1.55 . these models assume that the massive stars have driven out a gas mass , @xmath72 , immediately after they `` turn on '' , which is the time when the stellar - dynamical computation begins ( @xmath73 ) . this mass loss corresponds to a rather high star - formation efficiency of 42 per cent , which is nevertheless expected to lead to an expanding , unbound association ( lada , margulis & dearborn 1984 ) . it is also in - line with the masses of outflows observed around young ob stars which have short ( @xmath74 yr ) durations ( churchwell 1997 ) . in such models it is possible that most of the cluster existed in an embedded phase for a few @xmath75 yr before the hypothetical gas expulsion event . this phase , however , is not modelled here . models c1 and c2 are clusters which undergo cold collapse from two different initial @xmath48 . to achieve this , the initial velocities of all binary centre - of - masses and single stars are multiplied by a factor @xmath76 ( zero initial velocities are problematical for the n - body programme and would also be unrealistic since some clump clump motions will always be present in the proto - cluster ) . such models are interesting because dynamical friction on dense gas clumps moving in an extended gas medium is expected to lead to the contraction of a proto - cluster ( saiyadpour , deiss & kegel 1997 ; ostriker 1999 ) . furthermore , relative ammonia - clump c@xmath77o - core velocities in larger ( by a factor of 23 ) c@xmath77o cores are often observed to be significantly smaller ( @xmath78 km / sec ) than the observed velocity dispersion ( @xmath79 km / s ) of the core , which is thought to be close to the virial value of the core ( fig . 45 in benson & myers 1989 ) . in this picture , individual stellar systems remain trapped in the potential wells locally dominated by the clumps that formed them , until most of the gas is removed . clearly , relative clump clump velocities need to be measured for cores containing many clumps , to verify if this picture is correct . finally , the proper motion measurements within the trapezium cluster ( section [ sec : veld ] ) can not exclude a model in which the cluster is collapsing , or is in the phase of violent relaxation ( lynden - bell 1967 ) after a cold collapse . for each model cluster different assumptions concerning the primordial binary star population are made . in table [ table : models ] , column 7 lists the primordial total binary proportion , @xmath80 where @xmath81 and @xmath82 are the numbers of single - star and bound - binary systems , respectively . column 8 contains the maximum binary - star period . the minimum period is @xmath83 day in all cases . the form of the primordial period distribution is defined in column 9 , where k2 refers to the eigenevolved ( see below ) birth distribution , which is consistent with young systems in taurus auriga , and dm91 ( duquennoy & mayor 1991 ) refers to the gaussian log - period distribution approximating the distribution of systems in the galactic field . a @xmath15 system has , with log@xmath84 ( 11.0 ) a semi - major axis of 8200 au ( @xmath85 au ) . the model period distributions are compared with the observational data in fig . [ fig : init_p ] . note that the _ primordial _ @xmath86 is immediately reduced to the _ initial _ @xmath86 owing to disruptive crowding , which means through the immediate non - dynamical ionization of long - period binaries owing to their overlap in the high - density central region of the cluster . with @xmath87 , the assumption is made that the binary - star properties do not vary with star - forming conditions , apart from the effects of crowding , and that they are identical for the trapezium cluster to what is observed in taurus auriga . with @xmath88 , the hypothesis is that the binary proportion is lower in higher temperature star - forming clouds ( durisen & sterzik 1994 ) . additionally , the following assumptions are made : * the primordial mass - ratio distribution is obtained by random pairing of the stars . -1 mm * the primordial eccentricity distribution is thermally relaxed . the first assumption implies a mass - ratio distribution for g - dwarf systems that increases steeply towards low - mass secondaries . however , in the galactic field ( duquennoy & mayor 1991 ) , the mass - ratio distribution for g - dwarf binary systems contains fewer m dwarf companions than the initial model here . the assumption and galactic - field distribution are consistent with each other , because dynamical interactions in clusters can explain the observed properties of binary stars in the galactic field , this being part of the _ inverse dynamical population synthesis _ argument for a clustered origin of most stars ( kroupa 1995a , 1995b ) . the available observational evidence on the mass - ratio distribution of very young stars is also consistent with the assumption of random - pairing from the imf ( leinert et al . 1993 ) . the second assumption is necessary , because dynamical interactions in a galactic cluster can not evolve an arbitrary initial eccentricity distribution to the thermal distribution observed for main sequence binary systems with log@xmath89 , in which the number of systems increases linearly with the eccentricity . the primordial distribution is , however , evolved prior to the start of the @xmath90body integration to account for _ pre - main sequence eigenevolution _ ( i.e. the changes in orbital parameters owing to processes internal to a binary system such as tidal circularisation during the pre - main sequence phase ) . only binary systems with log@xmath91 are affected by this . details can be found in kroupa ( 1995b ) . [ sec : nb5 ] the computation of the dynamical evolution of star clusters is expensive because the forces have to be calculated for each pair of stars so that the cpu time scales as @xmath92 . binary systems that are perturbed by neighbours , and thus need to be integrated with very small time - steps , cause additional serious bottlenecks . an @xmath90body programme must be able to handle dynamical processes that have time - scales ranging from days to many @xmath93 yr . the state - of - the art programmes for the dynamical modelling of star clusters on commercially available personal computers have been developed by aarseth ( 1985 , 1994 , 1999 ) . for the present investigation , the successful and well tried nbody5 programme is used . it incorporates many special algorithms to ensure computational speed and efficiency . each particle is advanced with its own time - step , and the accelerations acting on it are updated with different frequencies using different techniques , depending on whether the perturbing stars are in a near - neighbour list or if the forces stem from numerous distant cluster members . binary , triple and higher order systems are treated with the regularisation technique which , with a special transformation of space - time coordinates , eliminates the time - step singularity for closely interacting stars ( see aarseth 1999 for the state - of - the art ) . the version of nbody5 employed here has been used earlier for the calculation of the dynamical evolution of binary - rich clusters by kroupa ( 1995a , b , c ) . it has extra routines that allow the inclusion of binary star populations with a variety of different initial period and eccentricity distributions . position and velocity vectors of each star are output at regular time intervals , and form the basic data from which all binary star properties are derived in subsequent data analysis . while statistical results from @xmath90body calculations correctly describe the overall dynamical evolution , the velocity and position vectors of any individual centre - of - mass particle ( e.g. a star or binary system ) diverge exponentially from the true trajectory , through the growth of errors in @xmath90body computations . goodman , heggie & hut ( 1993 ) discuss these issues in depth . @xmath94 computations with different initial random number seeds are performed for each model discussed here , giving in total 18 @xmath90body calculations . all quantities inferred from the models and presented here are averages from three models . the computations are performed for a time - span of 5 myr only , which is the upper limit for the age of the trapezium cluster from hr diagram fitting ( hillenbrand 1997 ) , although the mean age is probably closer to 1 myr . mass loss through stellar evolution can thus be neglected . an elaborate data - reduction programme reads the output from nbody5 and calculates , among many quantities , the projected velocity dispersion , the number of bound binary systems and their properties . the observational plane is taken to be the @xmath95 plane , with the @xmath96 direction being perpendicular to the galactic disk . throughout this paper , @xmath97 refers to the projected radial distance from the density maximum , whereas @xmath98 refers to the 3d distance . the choice of the projection plane is inconsequential for the present analysis , because @xmath97 is always smaller than a few pc . the apparent binary proportion , @xmath99 a hypothetical observer located at infinity sees in the projected data , is calculated from the projected positions of all stars seen within a central circle with radius @xmath97 in one model cluster at time @xmath100 . the number , @xmath101 , of apparent stellar pairs with separations in the distance range @xmath102 to @xmath103 are counted together with the number of apparently single stars , @xmath104 . binary systems with a separation @xmath105 are counted as single stars , and binary systems with @xmath106 are counted as two single stars . this is done for all stars within a central radius @xmath97 . for a comparison with proper motion data , the one - dimensional velocity dispersion of centres - of - masses ( i.e. unresolved stellar systems ) in the observational plane , @xmath20 is calculated within some @xmath97 . for consistency with the observational data , @xmath20 is corrected for the bulk radial motion giving @xmath107 . to this end , the mean projected radial velocity , @xmath108 , within @xmath97 is calculated . the corresponding radial vector quantity is subtracted from each 2d velocity vector . in the equilibrium models ( a ) , @xmath109 , while for the expanding models ( b ) , @xmath110 , and for the cold collapse models ( c ) , @xmath111 , for @xmath112 . the bulk radial motion is zero at @xmath73 in the equilibrium and expanding models , because the velocity vectors are initially isotropically distributed . in this section the results are described for each model ( virial equilibrium , expansion and cold collapse ) in turn . the overall cluster evolution and the evolution of the binary star population , as seen by a hypothetical observer , are discussed in separate subsections . throughout the discussion the binary - star orbital period is used instead of the binding energy , which is @xmath113g/2 $ ] , where @xmath114 and @xmath115 are the component masses , and @xmath60 is the gravitational constant in appropriate units . a detailed theoretical treatment of energy exchanges of binary systems with field stars is given by heggie ( 1975 ) . the models are initially over - dense by an order of magnitude ( fig . [ fig : d_vda ] ) , but the central system density agrees with the observational constraint after 2.3 myr ( model a2 ) and 4.2 myr ( model a1 ) . the cluster expands mostly through three- and four - body interactions . in a cluster consisting initially of single stars , on the other hand , the central number density would increase continuously until core collapse ( e.g. heggie & aarseth 1992 ; spurzem & takahashi 1995 ) . that the number of systems is consistently higher in model a1 than in model a2 for @xmath116 myr is a result of enhanced binary destruction in model a1 , which contains a larger number of binaries with long periods . the disruption of wide binaries cools cluster a1 , which expands more slowly than cluster a2 , leading to a larger density in the core . the one - dimensional velocity dispersion ( fig . [ fig : d_vda ] ) is similar to the observed value for @xmath117 myr , but then decreases substantially . the velocity dispersion is significantly smaller than the observational constraint when the central number density agrees with the observations . the higher density leads to a higher velocity dispersion in model a1 . this is a manifestation of the negative specific heat capacity of gravitating systems : cooling implying a hotter system ( lynden - bell 1998 ; see also bonnell & kroupa 1999 ) . in both models a1 and a2 , the mean projected radial velocity , @xmath118 km / s for @xmath119 myr , testifying to the expansion noted above . at later times @xmath120 km / s , which corresponds to the smaller rate of decrease of the central number density evident in the upper panel of fig . [ fig : d_vda ] after 2.2 myr . this is linked to the end of the main destruction phase of long - period binary systems . the small radial bulk motion , @xmath121 , has a negligible effect on the observed velocity dispersion ( @xmath107 ) . the change in slope in the upper panel of fig . [ fig : d_vda ] at @xmath122 myr comes about because binary destruction moves the truncation of the period distribution to decreasing periods and thus towards the hard / soft binary boundary ( see fig . [ fig : init_p ] ) . meanwhile binary activity ( i.e. the energy exchange between the binary systems and the cluster field ) expands the cluster , causing a shift of the hard / soft boundary to longer periods . ultimately , at time @xmath123 further binary destruction is effectively halted , at which stage the hard / soft boundary has moved beyond the truncation period of the period distribution , which in turn implies a reduction of heating of the cluster field by hardening binaries . the hard / soft , or _ thermal _ , boundary corresponds at this stage to @xmath124 days , taking the three - dimensional velocity dispersion @xmath125 km / s to be the circular velocity of a system with a total mass of @xmath15 . thus , at @xmath126 myr the cluster retains an approximately constant density , an equilibrium state between the remaining binary population and the cluster field having been established . the theoretical apparent projected binary star proportion , @xmath127 , is compared with observational constraints in fig . [ fig : fappa ] . significant evolution of @xmath128 occurs within the first 1 myr , owing to disruption of binaries and diminishing number of chance projection pairs as the cluster expands . however , @xmath129 for @xmath130 myr in the upper panel , suggesting that the initial binary proportion and period distribution can , in principle , be constrained . the above inequality is violated for times @xmath131 myr in the much smaller sample in the cluster core , and information on the initial binary proportion is lost . the model results show that the observational constraints are still too weak to allow a distinction between models a1 and a2 . both are consistent with the data for dynamical cluster ages @xmath132 myr . the total binary proportion , @xmath86 ( equation [ eqn : binf ] ) , also shown in fig . [ fig : fappa ] , decreases rapidly for @xmath117 myr , and later at a much slower rate . at @xmath73 , @xmath133 is lower than the assumed primordial proportions because wide binary systems are disrupted through superposition in the crowded inner cluster region ( see section . [ sec : prbins ] ) . that @xmath127 suffers from substantial chance - projection pairs is evident by @xmath86 decreasing at a faster rate during @xmath134 myr . binary activity leads to an overall expansion of the cluster and reduction in ambient temperature , i.e. @xmath107 . the diminished density and velocity dispersion reduce the encounter rate and rate of binary destruction . effectively , the binary hardness boundary shifts to longer periods , making further binary - system ionising events less probable because the period distribution has been truncated below the shifting boundary , as discussed in the previous subsection . at @xmath135 myr , @xmath136 and @xmath137 , and the surviving binary population is sufficiently hard that further decrease will not be significant . in the cluster , the binary proportion will increase during the subsequent cluster life - time owing to mass segregation and loss of preferentially single stars ( fig . 3 in kroupa 1995c ) . during the first 2 myr , long - period binary systems are ionised , leading to a period distribution , @xmath138 , that is truncated at progressively shorter periods . [ fig : per1 ] demonstrates that most destruction of long - period ( @xmath139 days ) binaries occurs within @xmath140 myr . there is no significant difference between the period distributions measured within @xmath141 pc and @xmath142 pc . also , the figure shows that the period distributions in models a1 and a2 are similarly truncated by @xmath143 myr . however , in model a1 a somewhat larger proportion of binary systems with @xmath144 days remains than in model a2 . this is a result of the complex interplay between ( i ) the cross section for binary - destruction being proportional to @xmath145 ( eqn . 26 in heggie & aarseth 1992 ) with the velocity dispersion @xmath58 being larger in model a1 for @xmath146 myr , ( ii ) the binary destruction rate being proportional to the square of the density of binaries ( eqn . 26 in heggie & aarseth 1992 ) , and ( iii ) the initially larger binary proportion in model a1 . as is evident from fig . [ fig : ftot_rada ] , initially there are fewer binaries near the centre in model a1 . this is a result of immediate disruption of long - period systems from crowding ( section [ sec : prbins ] ) . note that most of the cluster is initially confined to @xmath147 pc ( log@xmath148 ) , so that there is no significant difference in the period distributions sampled within @xmath149 pc and @xmath150 pc . at @xmath151 myr , the radial dependence is more pronounced , with binary - star depletion progressing further out , but the outer regions at @xmath152 pc retain a high binary proportion . at this stage of cluster evolution , the period distribution is already truncated at @xmath153 days within @xmath154 pc ( log@xmath155 ) , but remains roughly primordial at @xmath152 pc . further out , however , @xmath86 drops significantly . these are mostly single stars that have been ejected onto long - period and eccentric orbits in the cluster , and form a transient binary - deficient young halo population . after 1 myr , there remains a slight radial dependence , with @xmath86 increasing slightly with increasing @xmath156 pc ( log@xmath157 ) in model a1 . again , the decay for larger @xmath97 is a result of primarily low - mass stars being ejected during the disruption of binary systems near the cluster core . the radial dependence has vanished by @xmath158 myr due to thorough mixing of the stellar population , and @xmath159 . the lower primordial @xmath160 , together with the galactic field gaussian log - period distribution at birth , leads to no pronounced radial dependence initially in model a2 , because most binary systems are too tightly bound to be disrupted through crowding . during times @xmath161 , i.e. @xmath162 myr , however , @xmath86 is a slightly increasing function of @xmath97 , until the entire cluster is mixed . a strong decrease in @xmath86 for @xmath163 pc ( log@xmath164 ) , and a slight decrease for @xmath165 pc are evident at @xmath151 and 1 myr , respectively , for the same reason as above . however , @xmath166 for @xmath131 myr and @xmath167 pc , and little further evolution is apparent until @xmath158 myr . unfortunately the observational constraints can not ascertain that the binary proportion is lower in the central parts ( section [ sec : binf ] ) . the uncertainties remain too large . the data can be improved by pushing observational resolution to separations of about 20 au ( 0.04 arcsec at a distance of 450 pc ) in order to sample the period distribution around its likely maximum at log@xmath168 [ days ] . the models show that @xmath86 should be independent of @xmath97 for @xmath169 myr , by which time the stellar population is well mixed . if observations find evidence for @xmath170 and/or @xmath171 then this would be evidence for a taurus - auriga - like @xmath138 at birth ( model a1 ) , and that the trapezium cluster is @xmath172 old , remembering that the alternative model a2 with a galactic - field @xmath138 does not show significant variations of @xmath86 with @xmath97 . the central density is initially as large as in the equilibrium models , but decreases rapidly ( fig . [ fig : d_vdb ] ) . noteworthy is the divergence of the two models ( b1 with @xmath173 and b2 with @xmath174 initially , see fig . [ fig : init_p ] ) at @xmath175 myr . thereafter , model b1 , which initially contains more long - period binary systems , evolves with a constant stellar number density in the central region , amounting to @xmath176 stars . again , as in the equilibrium models , the break - up of long - period binaries causes cluster cooling , which reduces the velocities and thus aids a part of the expanding cluster to form a bound object . this point will be investigated in more detail in a forthcoming paper . in model b2 , which contains fewer long - period binaries , binary cooling is less effective , and a smaller part of the cluster condenses out to form a bound entity . in this case , @xmath177 stars , the decay being halted after @xmath178 myr . noteworthy is that the bound clusters that form in models b1 and b2 still have central densities @xmath179 and @xmath180 stars / pc@xmath17 , respectively . in comparison , the pleiades cluster has a central density of only about 27 systems / pc@xmath17 from fig.6 in raboud & mermilliod ( 1998a ) , or between 14 and 27 systems / pc@xmath17 from pinfield , jameson & hodgkin ( 1998 ) if the mean stellar mass is @xmath15 or @xmath181 , respectively . the projected velocity dispersion within @xmath182 pc , which is initially an order of magnitude larger than in the equilibrium models , remains equal in both models and decreases rapidly owing to the expansion against the gravitational potential and the loss of fast - moving stars from the measurement region ( @xmath183 pc ) . the evolution , shown in fig . [ fig : d_vdb ] , slows appreciably after about 0.2 myr when only the slow - moving tail of the population remains within the central radius . correction for the significant radial bulk motion is necessary , leading to a reduction in the estimated velocity dispersion by up to about 0.5 km / s at @xmath184 myr . the initial radial bulk velocity , @xmath185 , because the velocity vectors are distributed isotropically . as time progresses , the fast and slow - moving stars separate out , and a maximum bulk velocity within @xmath182 pc of @xmath186 km / s is reached at @xmath187 myr . the sharp decay thereafter comes from the loss of the fast - moving stars from the measurement region . the central number density is consistent at the three - sigma level with the observational constraint during the time interval @xmath188 myr , while the velocity dispersion , @xmath189 , is consistent at the three - sigma level with the observed value for @xmath190 myr ( fig . [ fig : d_vdb ] ) . thus this model could be a reasonable description of reality if about 60 per cent of the mass of the cluster was expelled about 50 thousand years ago , at which time the trapezium cluster would have gone into a rapid expansion phase . however , this age is significantly younger than the age inferred from hr diagram fitting ( section [ sec : numb_age ] ) . if the expanding model is correct then this age difference can be explained by the cluster having spent a few @xmath75 yr in an embedded phase prior to the postulated gas - expulsion event @xmath191 yr ago . rapid expansion halts the disruption of binary systems at an early stage . this is evident in fig . [ fig : fappb ] . from this figure it follows that the model apparent binary proportion is consistent with the observational constraint at the three - sigma level for @xmath192 myr , provided primordial @xmath160 . a smaller primordial @xmath86 would lead to earlier agreement . the total binary proportion shows little evolution after @xmath193 myr , being @xmath194 in model b1 , and @xmath195 in model b2 . the latter value is comparable to the binary proportion of late - type systems in the galactic field ( @xmath195 , kroupa 1995a ) , suggesting that expanding trapezium - cluster - type stellar assemblages could account for the majority of the galactic field population . this is corroborated by the period distribution ( fig . [ fig : per2 ] ) for log@xmath196 . however , the observed period distribution in the trapezium cluster appears to be significantly below that of the galactic field for log@xmath197 ( scally , clarke & mccaughrean 1999 ) , which may imply that the galactic field is not made of unbound trapezium - cluster - type assemblages , unless wide multiple systems form through capture in the expanding flow . for model b1 , the period distribution has a surplus of orbits with @xmath198 days . there is a distinguishable difference between the period distributions within @xmath141 pc and @xmath142 pc ( fig . [ fig : per2 ] ) . the distributions undergo no further evolution after @xmath199 myr , but within 0.5 pc , they are somewhat depleted relative to the distributions obtained for systems within 1 pc . in particular , petr s ( 1998 ) observational constraints are in nice agreement with model b2 . this depletion results from the dynamical interactions within the remaining low - mass cluster , which has a three - dimensional velocity dispersion of @xmath200 pc / myr ( fig . [ fig : d_vdb ] ) . the time - scale for population mixing within 1 pc is thus roughly 3 myr . in model b1 , formally with primordial @xmath87 , the total binary proportion increases with projected radius initially ( fig . [ fig : ftot_radb ] ) , which is a result of disruptive crowding as in model a1 . a slight remaining increase with @xmath97 is maintained at later times because the stellar population can not mix as the cluster expands . as in model a2 , there is initially no significant radial dependence of @xmath86 in model b2 . this is retained at @xmath201 myr owing to the rapid expansion , and is consistent with the presently available observational constraints suggesting no significant radial dependence within @xmath202 pc ( log@xmath203 ) . at @xmath151 myr , however , both models show a slightly smaller @xmath86 for @xmath204 pc ( log@xmath205 ) than at larger radii , reflecting the difference in @xmath138 for @xmath206 pc and @xmath142 pc noted above . cold collapse from an initial plummer density distribution leads to a rapid increase of the central number density , which is maximised at the cold - collapse ( or dynamical ) time @xmath207 myr for model c1 ( @xmath208 pc initially ) and at @xmath209 myr for model c2 ( @xmath210 pc initially ) . model c2 does not achieve as large a central density as model c1 because the inner regions virialise before the outer regions have fallen in . the evolution is shown in fig . [ fig : d_vdc ] . the velocity dispersion increases and achieves a maximum at @xmath211 when the clusters rapidly evolve into new equilibrium states through violent relaxation ( lynden - bell 1967 ) . the projected bulk radial velocity , @xmath121 , decreases as infall progresses . for model c1 the maximum radial bulk motion @xmath212 km / s , and for model c2 it is @xmath213 km / s ( fig . [ fig : d_vdc ] ) . the corrections to the projected velocity dispersion are , however , small . after virialisation , the clusters evolve on the much longer relaxation time - scale . it is longer for virialised model c2 , which has a smaller peak density , and consequently the decay of the central density proceeds at a slower rate . in both models , the central density remains significantly larger within 5 myr than the observational constraint , with an exception only for a short moment at @xmath187 myr for model c2 , when the central density briefly passes through the observational constraint . however , the velocity dispersion remains significantly smaller than the observational constraint at all times . both models c1 and c2 are assumed to have , formally , a primordial binary proportion @xmath87 . in practice the initial @xmath133 is slightly smaller than 1 ( fig . [ fig : init_p ] ) because of immediate disruption of wide systems through crowding , as in the models above . at @xmath214 , model c1 has a central density and velocity dispersion comparable to the initial ( @xmath73 ) values in model a1 ( section [ sec : evola ] ) . the binary population is thus expected to evolve similarly as in model a1 , keeping in mind that it has already undergone some evolution during the collapse . this is confirmed by comparing figs . [ fig : fappc ] and [ fig : fappa ] : the observer deduces similar apparent binary proportions , with the evolution in cluster c1 lagging behind that in cluster a1 by approximately @xmath211 . the slightly higher apparent and total binary proportion in model c1 comes about because the global density remains lower , despite the central density achieving a value close to the initial central density in the equilibrium model a1 , and because the velocity dispersion remains lower ( see also the discussion of binary disruption in section [ sec : binariesa ] ) . in model c2 the density does not increase to as high a value as in model c1 , so that fewer binary systems are ionised . this leads to an apparent binary proportion that is larger than the observational constraints at all times ( upper panel in fig . [ fig : fappc ] ) . during the initial collapse phase , the apparent binary proportion increases slightly due to the increased number of chance projection pairs . the period distribution , @xmath138 , at different times and different radii in the cluster is shown in fig . [ fig : per3 ] . by @xmath215 myr model c1 has already virialised and mixed , and there is no subsequent radial nor any time variation of @xmath138 . the evolution of @xmath216 through cold collapse is , however , evident in model c2 . at @xmath215 myr , the collapse has arrived at the maximum central density , and @xmath138 is already depleted at long periods . there is a larger depletion of @xmath138 in the more concentrated inner region ( @xmath141 pc ) . at @xmath158 myr , the cluster is virialised , and @xmath216 is even more depleted at long - periods . the maximum three - dimensional velocity dispersion is @xmath217 km / s for model c1 , and @xmath218 km / s for model c2 ( fig . [ fig : d_vdc ] ) . taking these values for the circular orbital velocity of a binary system with a mass of @xmath15 , orbital periods of @xmath219 days and @xmath220 days , respectively , are obtained . these values reflect the cutoff periods evident in fig . [ fig : per3 ] , and demonstrate the dependence of the rate of disruption of binary systems on the ambient velocity dispersion in a self - gravitating stellar system . how cold collapse influences the radial distribution of binary systems is shown in fig . [ fig : ftot_radc ] . initially @xmath221 for all @xmath97 . after one overall free - fall time , @xmath222 myr for model c2 , the binary proportion is reduced to 60 per cent within @xmath223 pc ( log@xmath224 ) , where most of the action during virialisation occurs . at larger radii , @xmath86 increases with @xmath97 and remains essentially unaffected for @xmath225 pc , i.e. in regions of the cluster that have not yet fallen in . at @xmath158 myr , however , the entire cluster has virialised , and the binary proportion is reduced at all radii , with a remaining slight increase towards the outermost less dense regions of the cluster reflecting some remaining incomplete mixing of the population . in model c1 the collapse time - scale is much shorter ( @xmath226 myr ) , even for the outermost radii , so that the binary proportion is reduced at all radii by @xmath215 myr . at this time , @xmath86 increases slightly with @xmath97 for @xmath227 pc ( log@xmath228 ) . this comes about because the cluster is still not completely mixed . during the violent relaxation phase binary systems are ionised at a larger rate owing to the increased number density . some of the liberated low - mass stars are ejected to larger radii , leading to the significant drop in @xmath86 at @xmath229 pc ( log@xmath230 , compare with model a1 in fig . [ fig : ftot_rada ] ) . such a decrease is not evident very well in model c2 because fewer binary systems are ionised in the less - concentrated core , providing fewer single stars to the outer cluster radii . a slight drop is , however , visible at @xmath231 pc ( log@xmath232 ) in model c2 . by @xmath158 myr the binary proportion is larger in the centre of model c1 than at larger radii . this is a result of dynamical mass segregation , and is observed in other young clusters ( e.g. elson et al . 1998 ; raboud & mermilliod 1998b ) . six different models of the trapezium cluster , summarised in table [ table : models ] , are evolved using aarseth s nbody5 code . they are clusters in virial equilibrium , expanding clusters assuming a star formation efficiency of 42 per cent and instant gas expulsion , and clusters in cold collapse , assuming the stars freeze out of the molecular cloud with a small relative velocity dispersion . we confront the evolving model population with observational constraints to infer if and how binary properties vary with star - forming conditions , and if the present and past dynamical state of the trapezium cluster can be inferred . the virial equilibrium ( a1 and a2 ) and initially collapsing ( c1 and c2 ) models have , in comparison with the trapezium cluster , at all times too high a central number density , while simultaneously having too small a velocity dispersion . these are thus excluded as realistic stellar - dynamical models of the trapezium cluster . however , we find that one of the expanding models is a solution that meets the observational constraints ( section [ sec : youngcl ] below ) . all models show that the variation of the binary proportion with radius , @xmath233 , is a diagnostic of the dynamical state of a cluster ( figs . [ fig : ftot_rada ] , [ fig : ftot_radb ] and [ fig : ftot_radc ] ) . for example , if the binary proportion near the centre of the cluster is lower than in its outer regions , then this indicates that the cluster is younger than a few crossing times . if the binary proportion is constant throughout most of the cluster but lower in its outermost regions , then this implies that the cluster is old enough for the stellar population to be well mixed , but young enough for binary activity in the inner regions to have ejected single stars into a halo , i.e. it is a few to ten crossing times old . if the binary proportion near the cluster centre is larger than at larger radii , then this indicates that the cluster is sufficiently old for dynamical mass segregation to have had time to occur . in particular , the collapsing models demonstrate how binary - destruction propagates to larger radii as collapse progresses , with a well - developed radial dependence of @xmath86 that is to a large extend eradicated after a few crossing times through mixing of the stellar population . despite not being possible solutions , the equilibrium and collapsing models shed important insights on the evolution of the binary population during the first few myr of a cluster s lifetime after gas dispersal ( figs . [ fig : fappa ] , [ fig : fappb ] and [ fig : fappc ] ) . the calculations show that binary destruction is very efficient in a dense star cluster . the binary proportion decreases from its initial high value to less than 40 per cent within 12 myr in our virial equilibrium cluster model , with most destruction happening within the first 0.5 myr . ultimately , expansion of the cluster shifts the hard / soft binary boundary to longer periods that have already been depleted , thus halting further reduction of the binary population , and reducing the cluster expansion rate . an equilibrium is reached between the global properties of the cluster and its remaining binary population after about 2 myr . this implies that the distribution of binary - star periods , in particular the truncation period , in a galactic cluster is a measure of its dynamical state at birth , or at the moment of maximum compression ( e.g. after a cold collapse ) . the effects of binary activity on the evolution of the central density and velocity dispersion are clearly noticeable in the equilibrium models ( fig . [ fig : d_vda ] ) . a large proportion of binaries near the hard / soft interface immediately causes overall cluster expansion , because their activity liberates kinetic energy . this stands in contrast to the immediate onset of core contraction for a single - star cluster . however , an initially larger proportion of soft binaries ( model a1 ) leads to a cooling effect , because kinetic energy is used up in ionising these binary systems . as a result , the cluster expands at a slower rate , and the central number density remains higher than in model a2 in which there are fewer such primordial binaries . the ratio of the number of binaries with periods near the hard / soft boundary to the number of soft binaries is thus a critical quantity driving the initial rate of expansion . the expanding models ( b1 and b2 ) are consistent with the observational constraints , provided expansion began @xmath234 yr ago . this might be possible if about 60 per cent of the mass in the cluster was expelled at that time . owing to the rapid expansion of models b1 and b2 , the initial binary population remains essentially unchanged . in these models , the binary proportion at a time @xmath235 ago must have been @xmath236 to be consistent with the observational constraints . this is lower than the binary proportion in taurus auriga , and , if model b2 does represent reality , may be due to dynamical evolution in the embedded cluster prior to gas expulsion , or due to a dependency on cloud temperature as suggested by durisen & sterzik ( 1994 ) . this would support the assertion that most galactic - field stars stem from trapezium - cluster like assemblages , provided the discrepancy in the proportion of wide binaries in the trapezium cluster and the galactic field can be accounted for ( section [ sec : binariesb ] ) . the finding that expanding model b2 is consistent with the observational constraints suggests that a gas mass of approximately @xmath237 may have been expelled about @xmath238 yr ago . this quantity of ejected gas is consistent with observations of outflows from young ob stars ( churchwell 1997 ) . the observations show that hundreds of @xmath43 of gas are expelled within a few to tens of @xmath239 yr per massive star . our assumption of instantaneous mass loss is thus reasonable . a neutral gas lid lies in front of the trapezium cluster as viewed from earth , but material is streaming out of the blister sw of @xmath240 ori c , where lid extinction is low and vanishes ( odell & wen 1992 ; wilson et al . 1997 ; odell 1999 ) . it is feasible that the massive outflows from the ob stars , that must have existed in the past , escaped from the cluster in that direction . the possible rather short expansion age of the trapezium cluster is interesting in the context of the existence of circum - stellar material around many of the low - mass stars in the trapezium cluster . the irradiation by the most massive star in the cluster , @xmath240 ori c , poses a very destructive environment for these objects . bally et al . ( 1998 ) model their photo - evaporation , and find that the ablation is so rapid , that the photo - ionization age of the blister in which the trapezium cluster is located may be as short as @xmath239 yr . if the age were a few @xmath75 yr , most of the circum - stellar material seen around the majority of low - mass stars should have been removed . odell ( 1999 ) , however , points out that the circum - stellar material may have quasi - equilibrium atmospheres with a much lower mass - loss rate , implying that these objects could be as old as hr diagram fitting implies the cluster to be ( section [ sec : numb_age ] ) . at present the issue is unsettled , and the present stellar - dynamical models add tantalising evidence in favour of the trapezium cluster becoming visible only a few @xmath239 yr ago by @xmath240 ori c `` turning on '' . the main shortcoming of the present study is that the trapezium cluster is treated as an isolated entity . in reality , it appears to be the core of the much more massive and extended orion nebula cluster ( onc ) . also , the onc is partially embedded in the parent elongated molecular gas cloud , the remaining parts of which lie at least about 0.2 pc behind the most massive trapezium cluster star , @xmath240 ori c ( wen & odell 1995 ) . hillenbrand & hartmann ( 1998 ) estimate that the mass in the stellar component of the onc and in the gas may together suffice to produce the observed stellar velocity dispersion . it is clear that future models will need to include the onc , and at a later stage the molecular cloud , being computationally much more expensive . however , even in this context and given the present results , it can be envisioned that the dense core of the onc , namely the trapezium cluster , may have begun expanding a few @xmath239 yr ago . a companion paper ( kroupa 1999 ) generalises the present results to different values of @xmath90 and @xmath48 , thereby identifying the candidate sets of models consistent with the observational constraints , if the onc is in virial equilibrium , expanding or undergoing a cold collapse . finally , the results presented here assume the clusters are initially plummer density distributions . this is realistic in the sense that star - forming cores are centrally concentrated , which has been used in the past to argue that plummer distributions are reasonable initial conditions for young star clusters ( lada , margulis & dearborn 1984 ) . we expect all density distributions that are centrally peaked and that have the same central densities , masses and half - mass radii as the models here to lead to a very similar evolution of the binary population , although the detailed radial dependence at some time @xmath100 may be slightly different . the greatest differences are expected for clusters that initially consist of groupings of aggregates , in the sense of a hierarchical structure . n - body calculations of the evolution of a cluster of sub - clusters , each containing approximately a dozen binary systems , have to be performed in order to address this particular set of possible models of the trapezium cluster and the onc . however , here again we expect similar evolution in the final bound cluster after the sub - clusters have merged , although the number of escaping stars will be higher during the first few global crossing times ( aarseth & hills 1972 ) . this paper presents fully self - consistent stellar - dynamical models of the evolution of the trapezium cluster over a time - span of 5 myr , assuming it consists of 1600 stars with a mass of @xmath241 . the calculations treat model clusters that are ( i ) in virial equilibrium , ( ii ) expanding and ( iii ) undergoing cold collapse . all models have a very large primordial binary population . these models constitute the hitherto most realistic stellar - dynamical calculations of a young cluster in existence , covering the time after gas expulsion but before stellar evolution becomes significant . thus , cooling of a cluster through the ionization of long - period binaries has been discovered for the first time to be a noticeable effect , if the primordial period distribution is similar to what is seen in the taurus auriga star formation complex . the calculations show that a young compact cluster , that is similar to the trapezium cluster and in virial equilibrium , undergoes significant dynamical evolution during the first few myr . the same holds true for its binary population , which is depleted rapidly , but then stabilises at @xmath242 if the initial period distribution is as in the galactic field or taurus auriga - like , respectively . this work also demonstrates which observational constraints are critical for discerning between models of very young clusters . these are the inner and outer binary proportions , and the radial bulk motion . in a young cluster the period distribution is truncated at a period that depends on the initial dynamical state of the cluster , with little further evolution after a few myr , or after violent relaxation has occurred . measurement of the period distribution , or at least of the truncation period , in a cluster such as the pleiades or hyades , thus constrains the possible range of initial configurations of these clusters . we are particularly interested to have discovered that the expanding cluster model is in best agreement with the observational constraints . this model assumes that about @xmath237 of gas was expelled from the cluster about @xmath243 yr ago , implying that the binary proportion at that time must have been significantly below the taurus auriga value , and that the trapezium cluster is unlikely to form a long - lived bound galactic cluster . the expansion rate is thus a very potent observable that urgently needs to be determined . the present models have a projected bulk radial velocity within a central radius of 0.41 pc of about 2 km / s at an expansion age of 50000 yr . while the present models are not a unique set of stellar - dynamical descriptions of the trapezium cluster , the present results will form an incentive for future observational studies ( i ) of the binary proportion in the inner and outer regions of very young star clusters , and ( ii ) of the velocity dispersion and bulk radial motions in the clusters . the results , and in particular their generalisation to a larger range of parameter space , will also be a strong incentive for conducting further , much more cpu - intensive numerical experiments , which are already being planned and partially underway . * acknowledgements * we are very grateful to sverre aarseth for freely providing nbody5 , and for his unfailing , friendly and uniquely efficient `` customer support '' . we are also thankful to bob odell for very useful correspondence , to tom megeath and phil myers for interesting conversations , and gijs nelemans for helpful comments . pk thanks ron burman and other staff of the department of physics at the university of western australia , where part of this paper was written , for their kind hospitality during march 1998 , and the staff of the harvard - smithsonian centre for astrophysics for a very pleasant visit during march june 1999 , where this paper was finished . pk acknowledges support from dfg grant kr1635 . mccaughrean m.j . , rayner j.t . , zinnecker h. , stauffer j.r . , 1996 , in beckwith s.v.w . , staude j. , quetz a. , natta a. , eds . , disks and outflows around young stars , lecture notes in physics , 465 , heidelberg , springer , p.33 * 8cl 1 & 2 & 3 & 4 & 5 & 6 & 7 & 8 & 9 + model & @xmath48 & @xmath244 & @xmath245 & log@xmath246 & @xmath247 & @xmath86 & log@xmath248 & period distribution + & ( pc ) & & & ( pc@xmath249 ) & & & ( days ) + a1 & 0.10 & 376 & 624 & 5.92 & 0.50 & 1.0 & 8.43 & k2 + a2 & 0.10 & 441 & 624 & 5.92 & 0.50 & 0.6 & 11.0 & dm91 + b1 & 0.10 & 376 & 624 & 5.92 & 1.20 & 1.0 & 8.43 & k2 + b2 & 0.10 & 441 & 624 & 5.92 & 1.20 & 0.6 & 11.0 & dm91 + c1 & 0.40 & 60 & 120 & 4.12 & 0.01 & 1.0 & 8.43 & k2 + c2 & 0.80 & 3 & 6 & 3.21 & 0.01 & 1.0 & 8.43 & k2 +

sulfur gases are common components in the volcanic and biological emission on earth , and are expected to be important input gases for atmospheres on terrestrial exoplanets . we study the atmospheric composition and the spectra of terrestrial exoplanets with sulfur compounds ( i.e. , and ) emitted from their surfaces . we use a comprehensive one - dimensional photochemistry model and radiative transfer model to investigate the sulfur chemistry in atmospheres ranging from reducing to oxidizing . the most important finding is that both and are chemically short - lived in virtually all types of atmospheres on terrestrial exoplanets , based on models of , , and atmospheres . this implies that direct detection of surface sulfur emission is unlikely , as their surface emission rates need to be extremely high ( times earth s volcanic sulfur emission ) for these gases to build up to a detectable level . we also find that sulfur compounds emitted from the surface lead to photochemical formation of elemental sulfur and sulfuric acid in the atmosphere , which would condense to form aerosols if saturated . for terrestrial exoplanets in the habitable zone of sun - like stars or m stars , earth - like sulfur emission rates result in optically thick haze composed of elemental sulfur in reducing -dominated atmospheres for a wide range of particle diameters ( 0.1 - 1 m ) , which is assumed as a free parameter in our simulations . in oxidized atmospheres composed of and , optically thick haze , composed of elemental sulfur aerosols ( ) or sulfuric acid aerosols ( ) , will form if the surface sulfur emission is 2 orders of magnitude more than the volcanic sulfur emission of earth . although direct detection of and by their spectral features is unlikely , their emission might be inferred by observing aerosol - related features in reflected light with future generation space telescopes .

1302.6603

a large number of super earths have been detected by radial velocity surveys and transit surveys . super earths are exoplanets with masses no more than 10 times the mass of earth . the atmospheres of super earths are important because characterization of super earth atmospheres is a substantial step towards eventually characterizing truly earth - like exoplanets . attempts to observe super earth atmospheres are growing ( e.g. , batahla et al . 2011 for kepler 10 b ; demory et al . 2012 and ehrenreich et al . 2012 for 55 cnc e ) , and one super earth / mini neptune gj 1214b is being observed in as much detail as possible ( e.g. , bean et al . 2010 ; croll et al . 2011 ; dsert et al . 2011 ; berta et al . 2012 ; de mooij et al . in particular , the transmission spectra of gj 1214b is nearly flat from 0.6 to 5 @xmath1 m , which has ruled out a planet with an -dominated cloud - free atmosphere . the observational push to super earth characterization has the potential to provide a handful of super earth atmospheres to study in the coming years . in the more distant future , the community still holds hope that a direct - imaging space - based mission under the terrestrial planet finder concept will come to existence , with planets and their atmospheres observed in reflected light . the concentration of trace gases in super earth atmospheres is controlled by the component gas emission from the surface and subsequent sinks in the atmosphere ( chemical reactions initiated by uv photolysis ) . even with a trace amount , some gases may leave significant footprints in the planet s spectra , for instance , , and in earth s atmosphere . in contrast to giant exoplanets where the atmospheric composition is mainly controlled by the elemental abundance , the steady - state composition of super earth atmospheres is mainly controlled by photochemical processes . we have developed a comprehensive photochemistry model that computes chemical compositions of any terrestrial exoplanet atmospheres , ranging from reducing to oxidizing , with all key non - equilibrium processes taken into consideration , including photolysis , chemical kinetics , vertical diffusion of molecules , atmospheric escape , dry and wet deposition , and condensation and sedimentation of concern condensable species ( hu et al . 2012 ; referred to as paper i hereafter ) . the effects of surface gas emission on different super earth atmospheres can be investigated with the photochemistry model . sulfur gases emitted from the surface and their photochemical products significantly shape the spectra of rocky bodies in the solar system . the most striking feature of venus atmosphere is a high planetary albedo due to thick clouds . photochemistry models of the venusian atmosphere have been developed and the formation of in the dry -dominated atmosphere have been simulated ( e.g. , yung & demore 1982 ; zhang et al . 2012 ; krasnopolsky 2012 ) . the latest photochemistry model of venus atmosphere has assumed a constant mixing ratio of ( i.e. , @xmath2 ppm ) at the altitude of 47 km ( implying significant source of from below ) and predicted the formation of at the altitudes around 66 km ( zhang et al . 2012 ; krasnopolsky 2012 ) . io , the innermost moon of jupiter , is believed to have very intensive and variable volcanic activity with emission , and the io s atmosphere is dominated by photolysis of and subsequent formation and polymerization of elemental sulfur ( e.g. , moses et al . 2002 ) . on early earth , the sulfur chemistry may have been very different from now , featuring the formation of sulfur aerosols as well as sulfate aerosols as the atmosphere was anoxic ( kasting et al . 1989 ; pavlov & kasting 2002 ; kasting & catling 2003 ; zahnle et al . the formation of insoluble sulfur aerosols is believed to be critical for the record of mass independent fractionation that timed the rise of oxygen in earth s atmosphere ( e.g. , farquhar et al , 2000 ; pavlov & kasting 2002 ; zahnle et al . in addition , organosulfur compounds , such as dimethyl sulfide ( dms ) and methanethiol ( ) , have also been suggested to be biosignatures of the early earth ( pilcher 2003 ) . the greenhouse effect of has been suggested to have contributed to the warming of early mars ( e.g. , halevy et al . 2007 ) , a proposition that has been challenged by photochemistry studies that predict sulfur and sulfate aerosol formation as a result of emission on early mars and the anti - greenhouse effect of these aerosols ( tian et al . 2010 ) . terrestrial exoplanets could have a wide range of sulfur gas emission . sulfur gases are common volcanic gases on earth , and in some scenarios may be more prevalent on exoplanets . sulfur is a major building block for rocky planets and the abundance of sulfur is about one sixteenth that of carbon in the solar system ( lodders 2003 ) . on present - day earth , sulfur compounds , mainly in the form of and , are dominant volcanic gases in additional to hydrogen , oxygen and carbon emissions . in the earth s magma , the dissolved sulfur has a weight percentage ranging from 10@xmath3 to 10@xmath4 ( e.g. , wallace & edmonds 2011 ) and degassing of sulfur compounds as the magma decompresses when rising to the surface provides a global volcanic sulfur flux of about @xmath5 s cm@xmath4 s@xmath7 ( seinfeld & pandis 2006 ) . 90% of current earth s sulfur emission is in the form of , whereas the / ratios for individual volcanoes vary widely between 0.01 and 1 ( see holland 2002 and references therein ) . an intriguing fact is that the amounts of sulfur compounds in earth s atmosphere is extremely low despite the substantial emission rates , due to very short chemical timescales . on earth , the lifetime of and in the troposphere is only 2 days , which makes the mixing ratio of these gases in the atmosphere very small ( seinfeld & pandis 2006 ) . in earth s troposphere , the main sink of is the reaction with the hydroxyl radical ( lelieveld et al . 1997 ) , and the main sink of is the removal by dry and wet deposition . another reason that and photochemistry is interesting is that earth - based biological processes involve sulfur compounds . there are multiple ways that life can produce , including the reduction of sulfate ( e.g. , watts 2000 ) and the disproportionation of sulfur compounds of intermediate oxidation states ( e.g. , finster 2008 ) . in general , 10 to 50 percent of the global emission on earth is as a metabolic byproduct , whereas over 70% percent of natural emission is volcanic , although on modern earth 90% of atmospheric is anthropogenic ( seinfeld & pandis 2006 ) . understanding the atmospheric response to and emission will allow us to examine whether or not is a potential biosignature gas on a terrestrial exoplanet . previous investigation of sulfur photochemistry in the context of super earth characterization has been very limited . sulfur compounds are generally not considered in most models of terrestrial exoplanet atmospheres . des marais et al . ( 2002 ) describe the spectral features of certain molecules such as , , and in terrestrial exoplanets . miller - ricci et al . ( 2009 ) present spectra of super earths under equilibrium chemistry with photochemistry estimates and without considering sulfur compounds . zahnle et al . ( 2009 ) investigate sulfur photochemistry in hot jupiters and suggest that and can be generated photochemically from . kaltenegger & sasselov ( 2010 ) study the effect of and on the terrestrial planetary spectra and suggest that 1 - 10 ppm has potentially detectable spectral features that could indicate active volcanism . kaltenegger & sasselov ( 2010 ) , however , do not link the surface emission of sulfur gases to the mixing ratio of sulfur species via photochemistry involving sulfur compounds , or consider formation of aerosols . domagal - goldman et al . ( 2011 ) study the chemistry of organic sulfur compounds that are strongly linked to biology ( e.g. , ) in anoxic atmospheres and suggest that the most detectable feature of organic sulfur gases is their indirect photochemical product , ethane . moreover , the fate of surface emission of sulfur compounds , e.g. , and , is yet to be explored for -dominated atmospheres . -dominated atmospheres , like or atmospheres with water vapor , could also maintain a habitable temperature at the surface through collision - induced absorption ( pierrehumbert & gaidos 2011 ; wordsworth 2012 ) . in all , it is still largely unknown whether or not and spectral features can be observed in the future in the atmosphere of an exoplanet and whether or not the emission rate of sulfur compounds on a terrestrial exoplanet can be inferred . in this paper we investigate the atmospheric chemistry resulting from and surface emission in atmospheres having very different oxidation states ranging from reducing to oxidizing . here , emission " means the mass flux from the planetary surface to the atmosphere that may include volcanic sources and biogenic sources . we focus on terrestrial exoplanet atmospheres that include super earths , since those planets hold the most interest on the path to finding and characterizing planets that might harbor life . in 2 we briefly describe our photochemistry model and radiative transfer model , with a focus of treatments of sulfur chemistry and aerosols . in 3 we describe the key chemistry pathways involving and in both reducing and oxidizing atmospheres . in 4 we present the main results on sulfur photochemistry in terrestrial exoplanet atmospheres and the spectral features of , , and photochemical aerosols . in 5 we discuss on whether or not can be a biosignature gas on a planet with atmospheric conditions different from earth s . we present our conclusions in 6 . we have developed a comprehensive photochemistry model to investigate atmospheres of terrestrial exoplanets , and validated the model by simulating the atmospheric compositions of current earth and mars ( see paper i ) . we now describe briefly the main features of the photochemistry model and the specifics for this work . the purpose of the photochemistry model is to compute the steady - state chemical composition of an exoplanetary atmosphere . the system is described by a set of time - dependent continuity equations , one equation for species at each altitude . each equation describes : chemical production ; chemical loss ; eddy diffusion and molecular diffusion ( contributing to production or loss ) ; sedimentation ( for aerosols only ) ; emission and dry deposition at the lower boundary ; and diffusion - limited atmospheric escape for light species at the upper boundary . starting from an arbitrary initial state , the system is numerically evolved to the steady state in which the number densities no longer change . because the removal timescales of different species are very different , the implicit inverse euler method is employed for the numerical time stepping . the generic model computes chemical and photochemical reactions among 111 molecules and aerosols made of o , h , n , c , s elements , and formation of sulfur ( ) and sulfate ( ) aerosols . the numerical code is designed to have the capacity of treating both reducing and oxidizing atmospheres . for the chemical and photochemical reactions , we use the reaction rates data from both the nist database ( http://kinetics.nist.gov ) and the jpl publication ( sander et al . we have also adopted relevant reaction rates from kasting ( 1990 ) , yung & demore ( 1999 ) , and moses et al . sulfur polymerization reaction rates still lack consistent experimental measurements , and we adopt the reaction rates proposed by kasting ( 1990 ) and those proposed by moses et al . both sets of sulfur polymerization reaction rates are speculative and they are widely discrepant , the effect of which will be discussed later in section 3.1 . ultraviolet and visible radiation in the atmosphere is computed by the @xmath8-eddington 2-stream method , with molecular absorption , rayleigh scattering and aerosol mie scattering contributing to the opacity . for the stellar input spectrum we used the air mass zero ( am0 ) reference spectrum produced by the american society for testing and materials for sun - like stars , and used the simulated non - active m star spectrum from allard et al . ( 1997 ) for quiet m stars . in this paper we use the photochemistry model to study the sulfur chemistry in atmospheres ranging from reducing to oxidizing on terrestrial exoplanets . we use dominated atmospheres as the representative cases for reducing atmospheres , and we use , and dominated atmospheres as the representative cases for oxidized atmospheres that could be both reducing and oxidizing . with the photochemistry model , we simulated the chemical composition of , , and -dominated atmospheres , with sulfur compounds emitted from the surface at various rates . parameters of the atmospheric models are tabulated in table [ atmospara ] and details of water , carbon , and oxygen chemistry have been described in paper i. we will focus on the sulfur chemistry and photochemistry in terrestrial exoplanet atmospheres in this paper . llll parameters & reducing & oxidized & oxidized + main component & 90% , 10% & & 90% , 10% + mean molecular mass & 4.6 & 28 & 42.4 + + stellar type & g2v & g2v & g2v + semi - major axis & 1.6 au & 1.0 au & 1.3 au + mass & @xmath9 & @xmath9 & @xmath9 + radis & @xmath10 & @xmath10 & @xmath10 + + surface temperature & 288 k & 288 k & 288 k + surface pressure & @xmath11 pa & @xmath11 pa & @xmath11 pa + tropopause altitude & 120 km & 13.4 km & 8.7 km + temperature above tropopause & 160 k & 200 k & 175 k + maximum altitude & 440 km & 86 km & 51 km + + in the convective layer & @xmath12 @xmath13 s@xmath7 & @xmath14 @xmath13 s@xmath7 & @xmath15 @xmath13 s@xmath7 + minimum & @xmath16 @xmath13 s@xmath7 & @xmath17 @xmath13 s@xmath7 & @xmath18 @xmath13 s@xmath7 + altitude for the minimum & 107 km & 17.0 km & 11.6 km + near the top of atmosphere & @xmath19 @xmath13 s@xmath7 & @xmath20 @xmath13 s@xmath7 & @xmath21 @xmath13 s@xmath7 + + liquid water ocean & yes & yes & yes + water vapor boundary condition & @xmath22 & @xmath22 & @xmath22 + rainout rate & earth - like & earth - like & earth - like + + & @xmath23 cm@xmath4 s@xmath7 & @xmath23 cm@xmath4 s@xmath7 & n / a + & n / a & @xmath24 cm@xmath4 s@xmath7 & @xmath24 cm@xmath4 s@xmath7 + & @xmath25 cm@xmath4 s@xmath7 & @xmath25 cm@xmath4 & @xmath25 cm@xmath4 + & vary & vary & vary + & vary & vary & vary + + & + & + & + & + & + & + & + & + & + & + & + & + & + [ atmospara ] we consider in our atmospheric chemistry models the o , h , and s bearing species and a subset of c bearing species . the gaseous molecules considered in this paper are , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , and , and the aerosols considered are aerosols and aerosols . this set of species is comprised of common h , o , and c bearing species and photochemical products of and emission . we assume a constant relative humidity at the surface of 60% to mimic the supply of water vapor from a liquid water ocean . to reduce the stiffness of the system and improve the numerical stability , fast " species with relatively short chemical loss timescales are computed directly from the photochemistry equilibrium . we consider in this work , , , , and as fast varying species . as such , the photochemistry model rigorously finds steady - state composition of the atmosphere starting with initial compositions without any sulfur compounds . once the model converged to the steady state , we checked explicitly the mass conservation of o , h , c , n , s atoms and verified the choice of fast species to have been appropriate . we have required all models to balance mass flux within @xmath26 for convergence , and typically our models balance mass flux to @xmath27 . we have also explicitly checked the redox ( i.e. , hydrogen budget ) balance ( see our definition of the redox number , flux , and balance in paper i ) , and required the models to balance redox flux to @xmath26 . one of the most significant controlling factors that determine the steady - state composition of atmospheres on terrestrial exoplanets is the dry deposition velocities of emitted gases and their major photochemical byproducts . of particular importance in this paper is the dry deposition velocities of and , which could vary by orders of magnitude ( see table [ atmospara ] for the fiducial values of key dry deposition velocities ) . the dry deposition velocities depend on the properties of the lower atmosphere and the surface . for example , in a model of the early cold martian atmosphere , the deposition velocities are assumed to be reduced by an artificial factors of up to 1000 compared with those in warm current earth ( e.g. , tian et al . 2010 ) , in order to account for less efficient deposition at a lower temperature . for another example , in modeling the early martian atmosphere , because the putative mars ocean is believed to be saturated with dissolved and other sulfur species , deposition is assumed to be balanced by the an equivalent return flux from the ocean ( e.g. halevy et al . 2007 ; tian et al . 2010 ) and then @xmath28 of is assumed to be zero . in this paper , we explore the effect of varying the dry deposition velocities of and , which can be scaled down by a factor of 100 if the surface is saturated with sulfide or sulfite . another important factor that strongly influences the atmospheric sulfur chemistry is the formation and sedimentation of aerosols . photochemically produced and may condense to form aerosols if their concentrations exceed their vapor saturation concentrations . the parameterization of condensation and sedimentation of aerosols in our photochemistry model is described in detail in paper i. saturation vapor pressure of is taken as recommended by seinfeld & pandis ( 2006 ) for atmospheric modeling , with a validity temperature range of 150 - 360 k. is the stable form of elemental sulfur because the molecule has a crown - shape ring structure that puts least strain on the s - s bond among sulfur allotropes and the crown structure allows for considerable cross - ring interaction between nonbonded atoms ( meyer 1976 ) . the saturation pressure of is then taken as the total sulfur saturation pressure against liquid sulfur at @xmath29 k and solid ( monoclinic ) sulfur at @xmath30 k tabulated by lyons ( 2008 ) . we consider the average aerosol particle diameter , a key parameter that determines the aerosols dynamical and optical properties , to be a free parameter . on earth , the ambient aerosol size distribution is dominated by several modes corresponding to different sources . the condensation submode " , formed from vapor condensation and coagulation , has an average diameter of @xmath31 @xmath1 m ( seinfeld & pandis , 2006 ) . on venus , the mode 1 particles with an an average diameter of @xmath32 @xmath1 m dominate the upper cloud ( e.g. , carlson et al . 1993 ) . on titan , the photochemical aerosols in the stratosphere have mean diameters in the range of 0.1 - 1 @xmath1 m ( rages et al . we treat the particle diameter a free parameter , and explore the effects of varying the particle diameter from 0.1 to 10 @xmath1 m . the dry deposition velocity of aerosols is assumed to be 0.2 cm s@xmath7 , a sensible deposition velocity of particles having diameters between 0.1 and 1 @xmath1 m ( sehmel 1980 ; seinfeld & pandis 2006 ) . before leaving this section we provide the physical rationale of specifying temperature - pressure profiles and eddy diffusion coefficients for our atmospheric photochemistry models . we modeled the atmospheric composition from the 1-bar pressure level up to the altitudes of about 10 scale heights . we chose appropriate vertical resolution for each scenario so that there are 4 layers per scale height . we adopted a temperature profile for our atmosphere models , without considering feedback on temperature of the atmospheric composition . the surface temperature is assumed to be 288 k. the semi - major axis of a terrestrial exoplanet around a sun - like star implied by this surface temperature is 1.6 au , 1.0 au , and 1.3 au , for , , and dominated atmosphere , estimated based on a similar procedure as kasting et al . the temperature profiles are then assume to follow appropriate dry adiabatic lapse rate ( i.e. , the convective layer ) until 160 k ( atmosphere ) , 200 k ( atmosphere ) , and 175k ( atmosphere ) and to be constant above ( i.e. , the radiative layer ) . the adopted temperature profiles are consistent with significant greenhouse effects in the convective layer and no additional heating above the convective layer for habitable exoplanets . we did not consider the climate feedback of or sulfur aerosols in the atmosphere , which could be important to determine the surface temperature ( e.g. halevy et al . 2007 ; tian et al . 2010 ) . while not ideal , these temperature profiles yield the same results discussed in the below as temperature profiles varied by several tens of k. the precise temperature - pressure structure of the atmosphere is less important than photochemistry for the investigation of sulfur chemistry because the most important photolysis and chemical reactions are not significantly affected by minor deviations in the temperature profile . we found that variation of temperature profiles by a few tens of k has minor impact on the atmospheric composition . vertical transport of gases in the atmosphere is parameterized by eddy diffusion , and the coefficients are assumed to be those of earth s atmosphere scaled by the atmospheric scale height to account for , , and being the dominant species . the current earth s eddy diffusion coefficient profile has been derived from mixing ratio profiles of several long - lived gases ( massie & hunten 1981 ; also shown in figure 1 of paper i ) . we use the empirical eddy diffusion coefficient profile for current earth as a template , and scale the coefficient inversely with the mean molecular mass for , , and dominated atmospheres . the justification for such scaling is that the eddy diffusion coefficient is proportional to the mixing length which is in turn a portion of the atmospheric scale height ( e.g. , smith 1998 ) . the pressure surface to pressure surface projection also ensures that the eddy diffusion coefficient profile features a eddy diffusion minimum near the tropopause for atmospheres with different mean molecular masses . our approach to parameterize vertical transport is of course an approximation . an accurate representation of vertical transport would likely involve circulation on the global scale ( e.g. , holton 1986 ) . we explore the effect of eddy diffusion coefficients ranging one or two orders of magnitude from the nominal value in a sensitivity study ( see section 4.1 ) . we compute synthetic spectra of the modeled exoplanet s atmospheric transmission , reflection and thermal emission with a line - by - line radiative transfer code ( seager & sasselov , 2000 ; seager et al . 2000 ; miller - ricci et al . 2009 ; madhusudhan & seager 2009 ) . opacities are based on molecular absorption with cross sections computed based from the hitran 2008 database ( rothman et al . 2009 ) , molecular collision - induced absorption when necessary ( e.g. , borysow 2002 ) , rayleigh scattering , and aerosol extinction are computed based on the mie theory ( e.g. , van de hulst 1981 ) . the transmission is computed for each wavelength by integrating the optical depth along the limb path , as outlined in seager & sasselov ( 2000 ) . the reflected stellar light and the planetary thermal emission are computed by the @xmath8-eddington 2-stream method ( toon et al . we used the refractive index of aerosols from tian et al . ( 2010 ) for the uv and visible wavelengths and from sasson et al . ( 1985 ) for infrared ( ir ) wavelengths . we used the refractive index of aerosols ( assumed to be the same as 75% sulfuric acid solution ) from palmer & william ( 1975 ) for uv to ir wavelengths , and jones ( 1976 ) for far ir wavelengths . the particle size distribution of aerosols controls their optical properties . we adopt the lognormal distribution as @xmath33,\ ] ] where @xmath34 is the number of particles per volume in the diameter bin @xmath35 , @xmath36 is the total number density of particles , @xmath37 is the median diameter of the particles , and @xmath38 is the particle size dispersion ( defined as the ratio of the diameter below which 84.1% of the particles lie to the median diameter ) . the lognormal distribution is a reasonable assumption because it provides a good fit to the particle size distribution measured in earth s atmosphere ( e.g. , seinfeld & pandis 2006 ) , and a sensible particle size dispersion parameter for photochemically produced aerosols is in the range of @xmath39 ( seinfeld & pandis 2006 ) . what is important for the radiative transfer model and the photochemistry model is the surface area mean diameter @xmath40 and the volume mean diameter @xmath41 , respectively . the mean diameters are related to the median diameter as @xmath42 @xmath43 we use the surface area mean diameter @xmath40 ( referred to as mean diameter " in the following ) as the free parameter for specifying a particle size distribution , as it is relevant to the radiative properties of the particle population . the volume mean diameter @xmath41 is useful in the conversion from mass concentration of the condensed phase ( computed in the photochemistry model ) to the number of aerosol particles for radiative transfer computation . the extinction cross sections of and molecules in aerosols for various mean particular diameters are shown in figure [ cross_compare ] . elemental sulfur aerosols and sulfuric acid aerosols have different optical properties at the visible and infrared ( ir ) wavelengths . in the visible , aerosols have a larger cross section than aerosols ( figure [ cross_compare ] ) . for wavelengths less than 400 nm aerosols are both reflective and absorptive . in the infrared , the cross section of aerosols drops significantly with increasing wavelength unless the mean diameter is in the order of 10 @xmath1 m . in contrast , aerosols have an enhancement of absorption at the mir wavelengths ( 5 - 10 @xmath1 m ) for all particle sizes ( figure [ cross_compare ] ) . we now briefly describe the most important processes of sulfur chemistry that occur in reducing and oxidizing atmospheres on rocky exoplanets . the primary sulfur emission from the planetary surface would be and ; they are either deposited back to the surface via dry or wet deposition , or converted into other forms of sulfur compounds in the atmosphere by photochemical reactions . one of the main purposes of this paper is to study the fate of sulfur gases emitted from the surface and their possible photochemical byproducts in the atmosphere . the fate of sulfur gases emitted from the surface is mainly controlled by the redox power of the atmosphere - the ability to reduce or oxidize a gas in the atmosphere . the redox power , in turn , is controlled by both the main component in the atmosphere ( e.g. , , , and ) and the surface emission and deposition of trace gases ( i.e. , , , and ) , as shown in table [ atmosredox ] . in the extreme cases of the atmospheric redox state , i.e. , the -dominated atmospheres and the -rich atmospheres , the atmospheric redox power is surely reducing or oxidizing , regardless of the nature of surface emission or deposition . however , for an intermediate redox state , the atmosphere would be composed of redox - neutral species like and , and the redox power of the atmosphere can be mainly controlled by the emission and the deposition fluxes of trace gases from the surface . the higher the emission of reducing gases is , the more reducing the atmosphere becomes . we already know that sulfur gas emissions are effectively oxidized into sulfate , the most oxidized form of sulfur , in the oxic atmospheres such as the earth s ( e.g. , seinfeld & pandis 2006 ) . in anoxic atmospheres , which include the reduced atmospheres and the oxidized atmospheres , previous studies have shown that both elemental sulfur and sulfate could be formed ( kasting 1990 ; pavlov & kasting 2002 ; zahnle et al . 2006 ; hu et al . 2012 ) . for this paper , we use dominated atmospheres as the representative cases for reducing atmospheres , and we use , and dominated atmospheres as the representative cases for oxidized atmospheres that could be both reducing and oxidizing . we now describe the key sulfur chemistry processes in these atmospheres as the follows . .redox power of atmospheres on rocky exoplanets . [ cols="<,<,<,<,<,<",options="header " , ] [ atmosredox ] both and emitted from the surface are efficiently converted into elemental sulfur in reducing atmospheres . the major chemical pathways for sulfur compounds in the -dominated atmosphere and the results of photochemistry model simulations are shown in figure [ h2_chem ] . atomic hydrogen produced from photodissociation of water vapor and itself is the key reactive species that converts and into elemental sulfur . the primary chemical loss for in the atmosphere is via and the produced can then react with again or with itself to produce elemental sulfur . can also react with to produce . the primary chemical loss for in the atmosphere is photodissociation that produces . can be either photodissociated to elemental sulfur , or be further reduced to via by h or and then converted to elemental sulfur ( see figure [ h2_chem ] ) . the and molecules produced in the atmosphere will polymerize to form , and will condense to form aerosols if it is saturated in the atmosphere . due to its ring structure , is stable against photodissociation . is a strong uv absorber ( kasting 1990 ) therefore , aerosols , if produced in the atmosphere , can effectively shield uv photons so that and may accumulate beneath the aerosol layer ( see the case for a sulfur emission rate 300 times higher than earth s current volcanic sulfur emission rate shown in figure [ h2_chem ] ) . the primary source of atomic hydrogen is the photolysis of , which occurs above the altitudes of @xmath44 pa pressure level . the atomic hydrogen can be then transported by eddy diffusion to the pressure level of @xmath45 pa to facilitate the removal of and and the production of elemental sulfur . additional numerical simulations show that an increase of the eddy diffusion coefficient by one order of magnitude can increase the yield of elemental sulfur by about 20% , because the transport of atomic hydrogen becomes more efficient . a secondary source of atomic hydrogen is photolysis of ( reaction [ h2sloss1 ] ) . this secondary source for atomic hydrogen is particularly important when the host star is a quiet m dwarf , because a quiet m dwarf produces few photons that could dissociate water . for planets around quiet m dwarfs the photolysis of could be the main source of atomic hydrogen in their atmospheres . additional numerical simulations show that in the habitable zone of a quiet m dwarf having an effective temperature of 3100 k , photolysis alone can produce enough atomic hydrogen to drive the formation of elemental sulfur in the atmosphere . we here comment on the uncertainty of photochemistry models regarding the yield of . in our model , we have assumed polymerization of elemental sulfur proceeds via we also include photodissociation for , , and . however , the reaction rates of sulfur polymerization ( reactions [ sp0 ] - [ sp4 ] ) have not been well established by laboratory studies , and previous authors have adopted different rate constants for these reactions . in particular , kasting ( 1990 ) and pavlov & kasting ( 2002 ) have used 3-order - of - magnitude lower rates for reactions ( [ sp0 ] - [ sp1 ] ) and 1-order - of - magnitude lower rates for reactions ( [ sp2 ] - [ sp4 ] ) , compared with moses et al . ( 2002 ) . in this work , we have adopted the reaction rates of moses et al . ( 2002 ) for elemental sulfur reactions . our sensitivity tests show that adopting the reaction rates of kasting ( 1990 ) would result in about 3 to 10 times less . we have chosen a higher sulfur polymerization rates for nominal models because : ( 1 ) the chemical pathways of reactions ( [ sp0 ] - [ sp4 ] ) are probably not complete and there may be other pathways to form ; ( 2 ) , , and may condense as suggested by lyons ( 2008 ) and the polymerization may still proceed in the condensed phase to . experimental studies are encouraged to settle this important uncertainty . and gases emitted from the surface can be converted into both elemental sulfur ( ) and sulfuric acid ( ) in the oxidized ( but anoxic ) atmospheres such as and dominated atmospheres . the major chemical pathways that lead to formation of both elemental sulfur and sulfuric acid , and the results of photochemistry model simulations are shown in figure [ n2_chem ] . the production of elemental sulfur aerosols involves uv photons and atomic hydrogen , as does in reducing atmospheres ; whereas the production of sulfuric acid requires oxidizing species , notably and . these reactive species , either reducing or oxidizing , are produced from photolysis of water and . in particular , the source of , responsible for converting to sulfuric acid in the atmosphere , is the photodissociation of . as a result , the amount of uv photons that are capable of dissociating water controls the yield of . for example in the habitable zone of a quiet m dwarf the yield of is much reduced compared with solar - like stars by at least one order of magnitude . the photochemically produced and may condense to form aerosols in the atmosphere if saturated . as a result , aerosols in the atmospheres provide a uv shield that enables the accumulation of and beneath the layer of aerosols . in particular for an earth - sized planet in the habitable zone of a sun - like star , when the surface emission rate is more than two - orders - of - magnitude higher than the current earth s volcanic sulfur emission rate , photochemical aerosols in the atmosphere lead to substantial uv shielding for accumulation of and , as shown in figure [ n2_chem ] . we find that only when sulfur emission is highly elevated with respect to current earth could or accumulate to the order of parts per million mixing ratio in the and atmospheres . the relative yield between elemental sulfur and sulfuric acid is controlled by the redox power of the atmosphere . in general , more sulfuric acid aerosols and less elemental sulfur aerosols are anticipated in a more oxidizing atmosphere . the redox power of the atmosphere , in turn , is determined by both the main constituent and the reducing gas emission . dominated atmospheres are mosre oxidizing than dominated atmospheres as photodissociation of leads to atomic oxygen . therefore the primary sulfur emission is more likely to be converted to sulfuric acid in dominated atmospheres than in dominated atmospheres ( see figure [ n2_chem ] ) . surface emission of reducing gases , including , , , and , alters the redox budget of the atmosphere and therefore increases the relative yield of elemental sulfur versus sulfuric acid aerosols ( e.g. , zahnle et al 2006 ) . as shown in figure [ n2_chem ] , when the sulfur emission rate increases , both and atmospheres become more and more reducing ( because is reducing ) , which results in a dramatic increase of elemental sulfur production in the atmosphere . furthermore , the / ratio in the sulfur emission affects its contribution to the redox power of the atmosphere and then the relative abundances of the two types of aerosols in the atmosphere significantly . as a result of the increase in the / emission ratio , the amount of aerosol in the atmosphere increases , and the amount of in the atmosphere decreases dramatically ( see figure [ n2_ratio ] ) . for an earth - like planet having an atmosphere , if the / emission ratio is less than 0.1 ( as is the case for current earth ; holland 2002 ) , the dominant type of aerosols in the atmosphere is sulfate ; whereas elemental sulfur aerosols become the dominant type if the / emission ratio is larger than 1 . the main finding is that on terrestrial exoplanets having atmospheres ranging from reducing to oxidizing , the primary sulfur emission from the surface ( e.g. , and ) is chemically short - lived . the sulfur emission leads to photochemical formation of elemental sulfur ( ) and sulfuric acid ( ) , which would condense to form aerosols if saturated in the atmosphere . in reducing atmospheres ( e.g. , atmospheres ) , aerosols are photochemically formed based on and emission ; and in oxidized atmospheres ( e.g. , and atmospheres ) , both and aerosols may be formed ( see figure [ sulfur_s ] ) . in general , the higher the surface sulfur emission , the more aerosols exist in the atmosphere ( see figure [ sulfur_s ] ) . as a result of photochemical production of elemental sulfur and sulfuric acid , terrestrial exoplanets with a habitable surface temperature ( e.g. , @xmath46 k ) and substantial sulfur emission from the surface are likely to have hazy atmospheres . in this paper , we use hazy " to describe an atmosphere that has significant aerosol opacities at visible wavelengths ( e.g. , 500 nm ) . we find that even with an earth - like surface sulfur emission , 1-bar dominated atmospheres on habitable rocky exoplanets are hazy with aerosols ( see figure [ sulfur_s ] ) . we also find that if the sulfur emission rate is @xmath47 times more than the earth s current volcanic sulfur emission rate , photochemical aerosols become optically thick at visible wavelengths in oxidized atmospheres including and atmospheres ( see figure [ sulfur_s ] ) . the key parameters that determine the aerosol opacity in the atmosphere are the surface sulfur emission rate , the dry deposition velocity , and the aerosol particle size . first , a higher surface sulfur emission rate leads to more sulfur and sulfate aerosols in anoxic atmospheres ( e.g. , figure [ sulfur_s ] and figure [ sulfur_gms ] ) . second , larger dry deposition velocities of and cause more rapid removal of these sulfur compounds from the atmosphere , which reduces the chance of converting them into condensable molecules ( i.e. , and ) . therefore , larger dry deposition velocities of and result in lower aerosol loading and aerosol opacities in the atmospheres , as shown in figure [ sulfur_vb ] . third , we find that the particle size has only secondary effects on the chemical composition of the atmosphere ( i.e. , by increasing the penetration of ultraviolet radiation ) , but has a primary effect on the aerosol optical depth . for mean particle diameter varying in the range of @xmath48 @xmath1 m ( i.e. , typical particle sizes of photochemical aerosols on earth ( e.g. , seinfeld & pandis 2006 ) and titan ( e.g. , rages et al . 1983 ) ) , we do not see a notable variation in the yield of elemental sulfur , but we see an enhancement of production with large particles ( figure [ sulfur_gms ] ) . even with the same aerosol abundances , however , micron - sized particles cause lower opacities at the visible wavelengths and higher opacities in mir compared with submicron - sized particles ( figure [ sulfur_gms ] ) . we capture the effects of the three key parameters on the aerosol opacity in anoxic atmospheres on terrestrial exoplanets by fitting the following power - law formula , i.e. , @xmath49 where @xmath50 is the vertical optical depth due to aerosols at 1 bar , @xmath51 is the total sulfur emission rate , @xmath52 is the dry deposition velocities of and with respect to current earth values , @xmath53 is the mean particle diameter of aerosols , @xmath54 , @xmath55 , and @xmath56 are positive numbers , and @xmath57 is a constant that covers other uncertainties . we have fit the empirical relation ( [ eqnsyn ] ) through an extensive parameter exploration using photochemistry models ( see figure [ sulfur_s ] - [ sulfur_vb ] for examples ) and determined the values of @xmath57 , @xmath54 , @xmath55 and @xmath56 for dominated reducing atmospheres and for and dominated oxidized atmospheres . we summarize graphically the parameter regime in which sulfur emission leads to a hazy atmosphere in figure [ syn ] . here we use @xmath58 and @xmath59 as the representatives for aerosol opacities at visible wavelengths and mir wavelengths ; due to the complex nature of the extinction cross sections of aerosol particles ( figure [ cross_compare ] ) , it is not practical to fold the full wavelength dependency into the empirical formula . also , we find that it is always true that @xmath60 for mean particle diameter in the order of 10 @xmath1 m and @xmath61 for mean particle diameter in the order of 0.1 @xmath1 m . for atmospheres , and mean particle diameter @xmath62 in the the range of @xmath48 @xmath1 m , @xmath63 and for @xmath62 in the the range of @xmath64 @xmath1 m , @xmath65 for and atmospheres , and @xmath62 in the the range of 0.1 and 1 @xmath1 m , @xmath66 and for @xmath62 in the the range of @xmath64 @xmath1 m , @xmath67 the constant @xmath57 in equations ( [ h2_syn1 ] - [ n2_syn2 ] ) spans about one order of magnitude , which covers the variation of the following model inputs : * the / ratio in the surface sulfur emission , ranging from 0.01 to 10 ; * temperature profiles deviating from the adopted temperature profile by @xmath68 k that controls the mixing ratio of water vapor in the atmosphere by the cold trap ; * stellar ultraviolet radiation received by the planet , ranging from the habitable zone of solar - like stars to the habitable zone of quiet m dwarfs with an effective temperature of 3100 k ; * eddy diffusion coefficients ranging from 0.1 to 100 times the values of earth s atmosphere ; * sulfur polymerization reaction rates ( reactions [ sp0 ] - [ sp4 ] ) ranging by one order of magnitude . to summarize , we find that the emission of and from the surface is readily converted into sulfur ( ) and sulfate ( ) in anoxic atmospheres of terrestrial exoplanets . the photochemical sulfur and sulfate would condense to form aerosols if saturated in the atmosphere , which is likely to occur on a planet in the habitable zone of either a sun - like star or a quiet m star . the aerosol layer is optically thick at the visible and nir wavelengths if the surface sulfur emission is comparable to earth s volcanic sulfur emission in the atmosphere , and more than @xmath47 times of the earth s volcanic sulfur emission in other anoxic atmospheres , depending on the dry deposition velocities of sulfur compounds and particle size of the aerosols . the sulfur emission from surface shapes the spectra of terrestrial exoplanets at the visible and nir wavelengths , mostly through the photochemical formation of and aerosols . we use the model outputs from the photochemistry models to compute the transmission , reflection , and thermal emission spectra of a terrestrial exoplanet with various levels of sulfur emission , and show examples of the computed spectra in figure [ spec ] . we see that submicron - sized aerosols dominate the transmission and reflection spectra at wavelengths from visible up to 3 @xmath1 m , if the sulfur emission is more than about two orders of magnitude higher than earth s volcanic sulfur emission . in general , an atmosphere with high sulfur emission and therefore high aerosol loading generally exhibits a flat transmission spectrum ( the features at nir muted ) , and a high visible albedo ( see figure [ spec ] ) . notably , aerosols are purely reflective at 500 nm but absorptive at 300 nm . the absorption edge of aerosols in 300 - 400 nm is evident in the reflection spectra for planets with enhanced sulfur emission ( figure [ spec ] ) , which is a potential diagnostic feature for aerosols . although opaque at visible wavelengths , the atmospheres with enhanced sulfur emission are likely to be transparent in the mir wavelengths ( @xmath69 @xmath1 m ) . the spectral features of aerosols depend on their particle sizes , so we now consider two possibilities : if the particles are submicron - sized , the aerosol molecules have negligible cross sections at mir ( see figure [ cross_compare ] ) ; or if the particles are micron - sized , the falling velocity of aerosol particles is large enough to rapidly remove aerosols from the atmosphere , as implied by equation ( [ h2_syn2 ] ) and equation ( [ n2_syn2 ] ) that are applicable for micron - sized particles . therefore in both cases the aerosol opacities at mir are minimal even for very high sulfur emission rates ( see figure [ spec ] for examples of atmospheres , and atmospheres are qualitatively similar ) . the only exception , in which aerosols indeed affect mir spectra , is the case of abundant aerosols . the main spectral effect of aerosols is absorption at mir wavelengths ( @xmath70 @xmath1 m ; figure [ spec ] ) . however , the column - average mixing ratio of needs to be larger than 0.1 ppm in order to produce significant aerosol absorption at mir . we find with numerical exploration that such a high abundance of aerosols is only possible in highly oxidizing dominated atmospheres without reducing gas emission ( see paper i for an example of such atmospheres ) . with reducing gas emission ( i.e. , and ) , it is unlikely that mixing ratio exceeds 0.01 ppm in anoxic atmospheres for a wide range of sulfur emission rates ( see figure [ sulfur_s ] ) . in summary we expect the spectral effects of and aerosols to be minimal at mir for most cases . we now turn to consider the direct spectral features of and . it has been previously proposed that and can be detectable on terrestrial exoplanets by their spectral features ( kaltenegger & sasselov 2010 ) . however , our photochemistry models show that both and are chemically short - lived in the atmospheres , which implies that that substantial surface emission is required to maintain a detectable level of either or in the atmosphere . has diagnostic absorption features at 7.5 @xmath1 m and 20 @xmath1 m ( see figure [ spec ] ) . for these features to be detectable the mixing ratio of needs to be larger than 0.1 ppm , which corresponds to sulfur emission rates 1000 times more than current earth s sulfur emission rates for , , and atmospheres ( see figure [ sulfur_s ] ) . the spectral feature of is the pseudo - continuum absorption at wavelengths longer than 30 @xmath1 m , which coincides with the rotational bands of . we find that the only scenario in which may be directly detected is the case with extremely high sulfur emission rates ( i.e. , 3000 times higher than the current earth s sulfur emission rate ) on a highly desiccated planet without liquid water ocean so that there is no water vapor contamination . we therefore conclude that direct detection of and is tricky : they are chemically short - lived so that extremely large surface emission is required for a detectable mixing ratio in the atmosphere , and their spectral features may be contaminated by other gases in the atmosphere . finally , we suggest that the emission of sulfur compounds might be indirectly inferred by detecting sulfur and sulfate aerosols . our numerical exploration reveals a monotonic relationship between the abundance of aerosols in the atmosphere and the emission rates of sulfur compounds ( see figure [ sulfur_s ] ) , and the composition of aerosols is correlated with the / ratio of the surface emission ( see figure [ n2_ratio ] ) . a combination of featureless low atmospheric transmission ( large planet radius viewed in transits ) and high planetary albedo ( large planetary flux at the visible wavelengths viewed in occultations ) may establish the existence of aerosols in the atmosphere . in particular , elemental sulfur ( ) aerosols are absorptive at wavelengths shorter than 400 nm and therefore might be identified by the absorption edge ( see figure [ spec ] ) . sulfate aerosols ( ) , if abundant in the atmosphere , lead to absorption features at the mir wavelengths ( @xmath71 - 10 @xmath1 m ) . however , none of these features are uniquely diagnostic of certain types of aerosols . the identification of aerosol composition , therefore , is by no means straightforward . we learn from the solar system exploration that the discriminating piece of information for aerosol identification comes from polarization of reflected stellar light . historically , the bright clouds on venus were identified to be mainly composed of droplets after the phase curve of the planet in polarized light had been observed ( e.g. , young 1973 ; hansen & hovenier 1974 ) . we therefore postulate that aerosol identification on terrestrial exoplanets and the inference of surface sulfur emission might require observation of polarized reflected light as a function of planetary illumination phase . biosignatures are gases in an exoplanet s atmosphere produced by life . in order to confirm a certain gas to be a plausible biosignature , one has to verify that the gas can accumulate in the atmosphere and that the amount of gas detected can not be produced through plausible abiotic processes . it has been proposed and widely discussed that ( and its photolytic product ) , , and are exoplanet biosignatures ( e.g. , sagan et al . 1993 ; des marais et al . 2002 ; segura et al . 2005 ; segura et al . 2007 ; hu et al . organosulfur compounds , such as dimethyl sulfide ( dms ) and methanethiol ( ) , have also been suggested to be biosignatures of the early earth ( pilcher 2003 ) and anoxic exoplanets ( domagal - goldman et al . 2011 ) . can be produced from several metabolic origins on earth , and so is a candidate biosignature gas . life on earth can produce through sulfate reduction ( when the environment is reduced ) and sulfur disproportionation . microorganisms can disproportionate sulfur compounds of intermediate oxidation states , including thiosulfate , sulfite , and elemental sulfur , into and sulfate ( finster 2008 ) . for example , disproportionation of sulfite in the ocean is described by in which the gibbs free energy released is 58.9 kj mol@xmath7 sulfite . the sulfite reducers , including microorganisms in genus _ desulfovibrio _ and _ desulfocapsa _ , extract energy from the disproportionation ( kramer and cypionka , 1989 ) . the effect of biotic production is the increase of the / ratio of the surface sulfur emission . if the / ratio in the volcanic sulfur emission is low ( i.e. , less than 0.1 ) , sulfur disproportionation and sulfate reduction by life may increase the / ratio significantly , which may lead to a change in the redox input to the atmosphere and therefore the dominant aerosol species in the atmosphere , as suggested by figure [ n2_ratio ] . specifically , for a habitable terrestrial exoplanet having a weakly oxidizing atmosphere , biotic production of in excess of the geological emission could result in a higher amount of aerosols and a much lower amount of aerosols in the atmosphere compared with a planet without life . although it is currently not possible to distinguish different types of aerosols , could be a biosignature gas in the long term . the geological production of , and the consequent risk of a false positive mis - identification of geological for biological , will be a major obstacle of confirming to be a biosignature gas . sulfur is believed to be present in the mantle of all terrestrial planets , and what determines the / ratio in the volcanic outgassing is the oxygen fugacity of the upper mantle , temperature of the location where magma degassing happens , water content in the conduit of magma , and gas content dissolved in the magma ( e.g. , holland 1984 ; kasting et al . 1985 ; holland 2002 ; burgisser & scaillet 2007 ) . the current earth volcanic emissions are oxidized , dominated by , and with minor contributions of , and . this volcanic gas composition is consistent with a magma buffered by the quartz - fayalite - magnetite ( qfm ) equilibrium , i.e. , a relatively oxidized upper mantle ( holland 1984 ) . as a global average , the volcanic / emission ratio on earth is 0.1 ( holland 2002 ) . however , if the mantle of a rocky exoplanet is much more reducing than that of the current earth , significant geological source of can be expected ( holland 1984 ; kasting 1993 ) . as a result , it would be very hard to rule out a geological contribution to the emission flux by remote sensing . in summary , although can be produced by energy - yielding metabolism , it is very unlikely to be a useful biosignature gas for three reasons . firstly , itself is unlikely to be detectable directly by remote sensing because of its weak spectral features , and their contamination by the spectral features of water . secondly , has a very short atmospheric lifetime , and so unrealistic emission rates are required to build up significant levels in any atmosphere . this second point could be overcome , in principle , by detecting aerosols in an anoxic atmosphere , and discriminating them from aerosols . discrimination between sulfur and sulfuric acid aerosols is not possible with current equipment , but may be possible in the future through analysis of reflected light . thirdly , however , suffers from a significant false positive risk , as geological sources can also produce , and the ratio of / in geological emissions depends on mantle chemistry , the physical structure of the outgassing events , and the extent of surface reprocessing of vented sulfur gases . to infer that life was generating on a planet , this study shows that the observer would have to determine the / aerosol ratio and have knowledge of the geological outgassing ratio of / and have knowledge of the surface chemistry that might modulate the primary outgassing rate . this seems an unreasonable requirement . we studied the effect of and surface emission on anoxic atmospheres of terrestrial exoplanets . with a newly established one - dimensional photochemistry model that treats all relevant chemical reactions and photochemical processes of o , h , c , and s bearing species , as well as formation and sedimentation of sulfur and sulfate aerosols , we find that and gases emitted from surface are chemically short - lived in both reducing and oxidizing atmospheres . the sulfur emission results in photochemical production of elemental sulfur ( ) and sulfuric acid ( ) , which would condense to form aerosols if they are saturated in the atmosphere . for a planet in the habitable zone of a sun - like star or a m star , earth - like sulfur emission rates would result in optically thick aerosol layers in -dominated atmospheres ; and a sulfur emission rate 2-orders - of - magnitude higher than the earth s volcanic sulfur emission rate would result in optically thick aerosol layers in and -dominated atmospheres . the composition of the photochemically produced aerosols mostly depends on the redox state of the atmosphere : aerosols are formed in the reducing atmospheres ( e.g. , atmospheres ) , and both and aerosols are formed in the oxidized atmospheres that could be both reducing and oxidizing ( e.g. , and atmospheres ) . based on extensive numerical simulations , we provide empirical formulae that show the dependency of the aerosol optical depth on the surface sulfur emission rates , the dry deposition velocities of sulfur compounds , and the aerosol particle sizes . we find that direct detection of and is unlikely due to the rapid photochemical conversion from and to elemental sulfur and sulfuric acid in atmospheres having a wide range of redox powers . for a terrestrial exoplanet with sulfur emitted from the surface at an enhanced rate , it is likely that at visible wavelengths the planet s atmosphere appears to be opaque due to the aerosol loading and that the planet has high visible albedo . however , for earth - like planets with 1-bar atmospheres ranging from reducing to oxidizing , we find the effect of photochemical sulfur and/or sulfate aerosols in the mir wavelengths is minimal , because micron - sized particles that interact with mir photons have large gravitational settling velocities and therefore short atmospheric lifetime . finally , as the aerosol composition is tightly related to the ratio of the versus emission , although direct detection of and by their spectral features is unlikely , their existence might be inferred by observing aerosol - related features in reflected light with future generation space telescopes . we thank shuhei ono and jenny suckale for helpful discussions about the sulfur - bearing gases release by the volcanism . we thank feng tian for providing the refractive index data of aerosols , and antigona segura for discussions regarding atmospheres . we thank james kasting for helpful suggestions about the photochemical model . we thank linda elkins - tanton for helpful suggestions on the mantle degassing . we thank vikki meadows for discussions about detecting clouds on venus . we thank the anonymous referee to the improvement of the manuscript . rh is supported in part by the nasa earth and space science fellowship ( nessf / nnx11ap47h ) . carlson , r. w. , kamp , l. w. , baines , k. h. , pollack , j. b. , grinspoon , d. h. , encrenaz , th . , drossart , p. , tayler , f. w. , 1993 , planet . space sci . , 41 , 477 croll , b. , albert , l. , jayawardhana , r. , et al . , 2011 , apj , 736 , 78 farquhar , j. , bao , h. , thiemans , m. , 2000 , science , 289 , 756 finster , k. , 2008 , journal of sulfur chemistry , 29 , 281 halevy , i. , zuber , m. t. , schrag , d. p. , 2007 , science , 318 , 1903 hansen , j. e. , hovenier , j. w. , 1974 , journal of atmospheric sciences , 31 , 1137 hauglustaine , d. a. , granier , c. , brasseur , g. p. , mgie , g. , 1994 , journal of geophysical research , 99 , 1173 holland , h. d. , 1984 , chemical evolution of the atmosphere and oceans , princeton university press , princeton , new jersey kasting , j. f. , holland , h. d. , pinto , j. p. , 1985 , journal of geophysical research , 90 , 10497 kasting , j. f. , zahnle , k. j. , pinto , j. p. , young , a. t. , 1989 , origins of life and evolution of the biosphere , 19 , 95 kasting , j. f. , whitmire , d. p. , reynolds , r. t. , 1993 , icarus , 101 , 108 kasting , j. f. , catling , d. , 2003 , ara&a , 41 , 429 kramer , m. , cypionka , h. , 1989 , arch . microbiol . , 151 , 232 krasnopolsky , v. a. , 2012 , icarus , 218 , 230 lelieveld , j. , roelofs , g. j. , ganzeveld , l. , feichter , j. , rodhe , h. , 1997 , phil . lond . , b352 , 149 lodders , k. , 2003 , apj , 591 , 1220 lyons , j. r. , 2008 , journal of sulfur chemistry , 29 , 269 madhusudhan , n. , seager , s. , 2009 , apj , 707 , 24 moses , j. i. , zolotov , m. y. , fegley , b. , 2002 , icarus , 156 , 76 palmer , k. f. , williams , d. , 1975 , applied optics , 14 , 208 pavlov , a. a. , kasting , j. f. , 2002 , astrobiology , 2 , 27 pierrehumbert , r. , gaidos , e. , 2011 , apj , 734 , l13 rothman , l. s. , et al . , 2009 , journal of quantitative spectroscopy & radiative transfer , 110 , 533 sander , s. p. , et al . 2011 , chemical kinetics and photochemical data for use in atmospheric studies , evaluation number 17 , jpl publication 10 - 6 sagan , c. , thompson , w. r. , carlson , r. , gurnett , d. , hord , c. , 1993 , nature , 365 , 715 sasson , r. , wright , r. , arakawa , e. t. , khare , b. n. , sagan , c. , 1985 , icarus , 64 , 368 seager , s. , sasselov , d. d. , 2000 , apj , 537 , 916 sehmel , g. a. , 1980 , atmospheric environment , 14 , 983 seinfeld , j. h. , pandis , s. n. , 2006 , atmospheric chemistry and physics - from air pollution to climate change , 2nd edition , published by wiley , new jersey segura , a. , kasting , j. f. , meadows , v. , cohen , m. , scalo , j. , crisp , d. , butler , r. a. h. , tinetti , g. , 2005 , astrobiology , 5 , 706 segura , a. , meadows , v. s. , kasting , j. f. , crisp , d. , cohen , m. , 2007 , a&a , 472 , 665 smith , m. d. , 1998 , icarus , 132 , 176 tian , f. , claire , m. w. , haqq - misra , j. d. , smith , m. , crisp , d. c. , catling , d. , zahnle , k. , kasting , j. f. , 2010 , earth and planetary science letters , 295 , 412 toon , o. b. , mckey , c. p. , ackerman , t. p. , 1989 , jgr , 94 , 16287 van de hulst , h. c. , 1981 , light scattering by small particles , new york : dover zahnle , k. , claire , m. , catling , d. , 2006 , geobiology , 4 , 271 zahnle , k. , marley , m. s. , freedman , r. s. , lodders , k. , fortney , j. , 2009 , apj , 701 , l20 zhang , x. , liang , m. c. , mills , f. p. , belyaev , d. a. , yung , y. l. , 2012 , icarus , 217 , 714 ( solid lines ) and @xmath73 ( dashed lines ) cm@xmath4 s@xmath7 , i.e , 3 and 300 times higher than the earth s volcanic sulfur emission rate . the / ratio in the sulfur emission is 0.5 and the particle mean diameter is 0.1 @xmath1 m . other model parameters are tabulated in table [ atmospara ] . uv photons and atomic hydrogen effectively convert and into elemental sulfur , and elemental sulfur aerosols shield uv photons so that and may accumulate below the aerosol layer if the sulfur emission is more than 300 times higher than the earth s volcanic emission rate . , scaledwidth=50.0% ] , the upper panel schematically illustrates the chemical pathways from the primary sulfur emissions ( i.e. , and ) to elemental sulfur and sulfuric acid . for double arrows the label above the arrow indicates the oxidizing agents , and the label below the arrow indicates the reducing agents . the lower two panel shows the results of photochemistry simulations for and dominated atmospheres , respectively . the total surface sulfur emission is @xmath74 ( solid lines ) and @xmath73 ( dashed lines ) cm@xmath4 s@xmath7 , i.e , 3 and 300 times higher than the earth s volcanic sulfur emission rate . the / ratio in the sulfur emission is 0.5 and the particle mean diameter is 0.1 @xmath1 m . other model parameters are tabulated in table [ atmospara ] . both elemental sulfur aerosols and sulfuric acid aerosols are formed in the oxidized and anoxic atmospheres . the origins of the principle reducing agents ( and ) and the principle oxidizing agents ( , and ) is photodissociation of and . the apparent depletion of at the pressure level of 10 - 100 pa in the atmosphere ( the blue solid line in the middle panel ) is due to the production of atomic hydrogen by methane photodissociation at this pressure level . , scaledwidth=50.0% ] cm@xmath4 s@xmath7 , or 30 times the earth s volcanic sulfur emission rate , and the particle mean diameter is 0.1 @xmath1 m . other model parameters are tabulated in table [ atmospara ] . as a larger fraction of surface sulfur emission is in the form of , the amount of aerosols in the atmosphere increases , and the amount of aerosols in the atmosphere decreases dramatically . , scaledwidth=50.0% ] m ( dashed lines ) are shown in the lower panel . the planet is an earth - sized rocky planet orbiting a sun - like star , with reducing ( -dominated ) , weakly oxidizing ( -dominated ) , or highly oxidizing ( -dominated ) atmospheres . the aerosol particle mean diameter is assumed to be 0.1 @xmath1 m , and the / ratio of the surface emission is 0.5 . other model parameters are tabulated in table [ atmospara ] . sulfur emission 2-orders - of - magnitude larger than current earth s volcanic sulfur emission ( @xmath75 s cm@xmath4 s@xmath7 ) leads to substantial aerosol opacities in the visible wavelengths in and atmospheres , and sulfur emission comparable with current earth s volcanic sulfur emission leads to substantial aerosol opacities in the visible wavelengths in atmospheres . the wiggle in the concentration of sulfuric acid aerosols reflects the competition between two effects : more sulfur is available to be converted into sulfuric acid as the sulfur emission increases , but the atmosphere becomes more reducing and less oxidizing as the sulfur emission increases . , scaledwidth=100.0% ] m as a function of total sulfur emission rates , for an earth - sized rocky planet orbiting a sun - like star at 1 au ( black lines ) , a habitable planet around quiet m dwarf having effective temperature of 3100 k ( red lines ) , and an earth - sized rocky planet orbiting a sun - like star at 1 au with particle mean diameter of 1 @xmath1 m ( blue lines ) . the left column shows the case of weakly oxidizing atmospheres , and the right column shows the case of reducing atmospheres . the / ratio in the sulfur emission is 0.5 and other model parameters are tabulated in table [ atmospara ] . we see that decreasing uv photon flux has little effect on the formation , but results in a decrease of the amount of sulfuric acid aerosols , and therefore a decrease of mir optical depth . we also see that particle diameter variation in @xmath48 @xmath1 m has little effect on the chemical composition , but for similar mass abundance a larger particle size results in a smaller optical depth in the visible wavelengths and a larger optical depth in the mir wavelengths . , scaledwidth=50.0% ] ) , for reducing ( ) and oxidized ( and ) atmospheres . the upper - left corner of the parameter regime ( small sulfur emission rates , large deposition velocities ) leads to clear atmospheres ; whereas the lower - right corner of parameter regime ( large sulfur emission rates , small deposition velocities ) leads to optically thick aerosols in the atmosphere composed of sulfur ( ) and sulfate ( ) . the widths of the shaded boundary regime between clear atmospheres and optically thick atmospheres contain the uncertainties of : ( 1 ) the mean aerosol particle size ranging from 0.1 to 1 @xmath1 m , ( 2 ) the / ratio of the sulfur emission ranging from 0.01 to 10 ( i.e. , more leads to thicker haze ) , ( 3 ) the spectral type of the host star ranging from g2v to m5 , ( 4 ) the strength of vertical mixing in the atmosphere by eddy diffusion ranging from 0.1 to 10 times earth s value , and ( 5 ) the surface temperature ranging from 270 to 320 k. earth and venus are shown for a reference in the solar system : earth s volcanic sulfur emission and deposition velocity are plotted ; and venus equivalent upward flux and deposition velocity at the altitude of 58 km are plotted ( adapted from krasnopolsky 2012 ) . we note that the equivalent flux is a transfer rate across the 58-km altitude , and does not imply a surface emission rate . the features at 7.5 @xmath1 m and 20 @xmath1 m requires a mixing ratio on the order of ppm to be spectrally significant , which corresponds to a sulfur emission flux of more than @xmath76 cm@xmath4 s@xmath7 due to rapid photochemical removal of in the atmosphere . , scaledwidth=50.0% ]

the schwarzschild spacetime metric of negative mass is well - known to contain a naked singularity . in a spacelike slice , this singularity of the metric is characterized by the property that nearby surfaces have arbitrarily small area . we develop a theory of such `` zero area singularities '' in riemannian manifolds , generalizing far beyond the schwarzschild case ( for example , allowing the singularities to have nontrivial topology ) . we also define the mass of such singularities . the main result of this paper is a lower bound on the adm mass of an asymptotically flat manifold of nonnegative scalar curvature in terms of the masses of its singularities , assuming a certain conjecture in conformal geometry . the proof relies on the riemannian penrose inequality . equality is attained in the inequality by the schwarzschild metric of negative mass . an immediate corollary is a version of the positive mass theorem that allows for certain types of incomplete metrics . * authorship remarks * most of the definitions and results of this paper were the work of the first author , originally presented at a conference in 2005 under the heading `` negative point mass singularities '' . the second author wrote the majority of the paper , with individual contributions cited throughout .

0909.0522

the first metrics one typically encounters in the study of general relativity are the minkowski spacetime metric and the schwarzschild spacetime metric , the latter given by @xmath0 where @xmath1 is the time coordinate , @xmath2 are spatial coordinates , and @xmath3 is some positive number . this represents the exterior region @xmath4 of a non - rotating black hole of mass @xmath3 in vacuum . a spacelike slice of this lorentzian metric can be obtained by taking a level set of @xmath1 ; under a coordinate transformation @xmath5 , the resulting 3-manifold is isometric to @xmath6 minus the ball of radius @xmath7 about the origin , with the conformally flat metric @xmath8 where @xmath9 is the usual flat metric on @xmath6 and @xmath10 . its boundary is in fact a minimal surface which represents the apparent horizon of the black hole . we will refer to this riemannian metric as the _ schwarzschild metric _ ( of mass @xmath3 ) . consider instead the metric @xmath11 with @xmath12 . this gives a riemannian metric on @xmath6 minus a closed ball of radius @xmath13 about the origin that approaches zero near its inner boundary . one may loosely think of this manifold as a slice of a spacetime with a single `` black hole of negative mass . '' in fact , this metric has a naked singularity , as the singularity on the inner boundary is not enclosed by any apparent horizon . in this paper we introduce a theory of such `` zero area singularities '' ( zas ) , modeled on the `` schwarzschild zas metric '' ( i.e. , @xmath11 with @xmath12 ) , yet far more general . some of the interesting problems we address are : when can such singularities be `` resolved '' ? what is a good definition of the mass of such a singularity ? can the adm mass of an asymptotically flat manifold of nonnegative scalar curvature be estimated in terms of the masses of its singularities ? to say another word about the third question , recall the positive mass theorem of schoen and yau @xcite which states that , under suitable conditions , the adm mass of an asymptotically flat 3-manifold is nonnegative , with zero mass occurring only for the flat metric on @xmath6 . the riemannian penrose inequality @xcite improves this to give a lower bound on the adm mass in terms of the masses of its `` black holes . '' here , the case of equality is attained by the schwarzschild metric with @xmath14 . see appendix a for details on asymptotic flatness and adm mass . the main theorem of this paper is the riemannian zas inequality ( see theorems [ thm_zas_ineq ] and [ thm_zas_ineq_full ] ) , introduced by the first author @xcite . it is an analog of the riemannian penrose inequality , but for zero area singularities instead of black holes . specifically , this inequality gives a lower bound for the adm mass of an asymptotically flat manifold in terms of the masses of its zas . its proof assumes an unproven conjecture ( conjecture [ conj_conformal ] ) regarding the outermost minimal area enclosure of the boundary , with respect to a conformal metric . while the conjecture is known to be true in a number of cases , proving it remains an open problem . although we shall write `` the '' riemannian zas inequality in this paper , we remark that other similar inequalities may be discovered in the future that also deserve this title . table [ table_thms ] illustrates how this theorem fits together with the positive mass theorem and riemannian penrose inequality . .the schwarzschild metric as a case of equality [ cols="<,<,<",options="header " , ] + by the above discussion of capacity , we see that @xmath15 if and only if @xmath16 . by theorem [ thm_pos_cap ] , if @xmath17 , then @xmath18 , since we saw above that the capacity of @xmath19 is positive . we now determine the mass of @xmath19 assuming @xmath20 . the first step is to compute the regular masses of the concentric spheres @xmath21 : @xmath22 we can use lhopital s rule to evaluate this limit ( of the form @xmath23 ) as @xmath24 : @xmath25 this limit is @xmath26 if @xmath27 , finite but nonzero if @xmath28 , and zero if @xmath29 . for @xmath29 , this shows that the mass of @xmath19 is zero . ( it is at least zero by our computation , but it is a priori at most zero . ) for @xmath30 , we claim the mass of @xmath19 is @xmath26 . given any sequence of surfaces @xmath31 converging in @xmath32 to @xmath19 , we will compare them to a sequence of round spheres . let @xmath33 be the minimum value of the @xmath34-coordinate that @xmath35 attains ; then @xmath35 encloses @xmath36 , so @xmath37 . also , since @xmath35 converges in @xmath32 , the ratio of areas @xmath38 converges to 1 . using the proof of theorem [ thm_pos_cap ] , we can estimate the regular mass of @xmath35 : @xmath39 where the last equality comes from equations @xmath40 and @xmath41 . taking @xmath42 of both sides ( and using @xmath43 ) , we have @xmath44 . ( we showed above that this limit is @xmath26 for concentric round spheres . ) thus , @xmath18 if @xmath27 . if @xmath45 , we will see below that @xmath19 is harmonically regular . by corollary [ cor_masses_agree ] , its mass is given by @xmath46 , which evaluates to @xmath47 . we now determine the values of @xmath48 for which @xmath19 is a regular zas . the zas @xmath19 of the metric @xmath49 with @xmath50 given by @xmath51 is regular if and only if @xmath28 . in this case , @xmath19 is harmonically regular . we find a necessary condition for @xmath19 to be regular . assume @xmath52 is some local resolution ; by spherical symmetry we may assume @xmath53 depends only on @xmath34 . by applying a conformal transformation , we may assume that the spheres @xmath54 have constant area @xmath55 in @xmath56 ( i.e. , @xmath57 ) . ( in other words , the metric @xmath56 is that of a round cylinder with spherical cross sections . ) then @xmath58 a necessary condition for a local resolution is that @xmath59 is positive and finite on @xmath19 . compute this normal derivative : @xmath60 this is positive and finite in the limit @xmath61 if and only if @xmath28 . thus , @xmath45 is necessary for the existence of a local resolution . if @xmath45 , then @xmath62 . a calculation shows that the arc length parameter for @xmath56 is given as @xmath63 . thus , @xmath64 , which is smooth on @xmath65 ; moreover , @xmath53 is @xmath56-harmonic . thus , @xmath52 is a local harmonic resolution . evidently the case @xmath28 is special . it is left to the reader to show that if the schwarzschild zas metric is written in the form @xmath66 , then to first order , @xmath67 is given by a constant times @xmath68 , where @xmath69 is the distance to @xmath19 . for our purposes a `` removable singularity '' is a point deleted from the interior of a smooth riemannian manifold . such singularities can be viewed as zas . for example the manifold @xmath70 with the flat metric @xmath71 has its interior `` boundary '' as a zas . to see this , let @xmath72 with the flat metric @xmath73 on its interior , which is isometric to @xmath70 . clearly the boundary @xmath74 is a zas . it is not difficult to see that the zas @xmath19 of the spherically symmetric metric @xmath51 is removable if and only if @xmath75 , which corresponds to a deleted point in flat space . ( see lemma 4.1 of chapter 1 in @xcite . ) in fact , it is straightforward to show that a collection of removable singularities has zero mass ; see @xcite . ( the key tool is the existence of a harmonic function that blows up at the removable singularities . ) however , not all zas with zero mass are removable , as seen in earlier examples . there is also a notion of a `` removable @xmath76 singularity '' ; this occurs when an embedded circle is removed from the interior of a smooth riemannian manifold . the resulting space has a zero area singularity that is topologically a 2-torus . it is left to the reader to verify that removable @xmath76 singularities have mass equal to @xmath26 ( at least for a circle deleted from @xmath6 ) . thus , our definition of mass is not well - adapted to studying these types of singularities . in this section , we prove by example that there exists a zas that has a local resolution but no local harmonic resolution ; we begin by showing that the problem of finding a local harmonic resolution is equivalent to solving a linear elliptic pde . let @xmath52 be some local resolution of a zas @xmath77 . suppose for now that there exists a local harmonic resolution @xmath78 of @xmath77 defined on a neighborhood @xmath79 . set @xmath80 , which is smooth and positive on @xmath81 by lemma [ lemma_smooth_extension ] . also , @xmath82 . apply formula @xmath83 from appendix b ( with @xmath84 playing the role of @xmath85 ) . this leads to the equation @xmath86 since @xmath84 is assumed to be @xmath87-harmonic . then @xmath88 satisfies @xmath89 , where @xmath90 . then @xmath88 satisfies the linear elliptic pde @xmath91 where @xmath92 . on the other hand , if @xmath88 is some positive solution to @xmath93 that is smooth up to and including the boundary , then the above discussion implies that @xmath78 is a local harmonic resolution of @xmath77 , where @xmath94 and @xmath95 . note that solutions of @xmath93 may lose regularity at the boundary if , for some local resolution @xmath52 , the function @xmath96 is not smooth at the boundary . now we exhibit such an example . consider a spherically symmetric metric @xmath56 given by @xmath97 and a function @xmath98 , both for @xmath99 . then @xmath52 is a local resolution of the zas of @xmath100 . we assume that if a local harmonic resolution exists , it is also spherically symmetric . by equation @xmath101 , the laplacian on functions of @xmath34 is given by @xmath102 thus , for our choice of @xmath103 , we have @xmath104 , so equation @xmath93 becomes @xmath105 where @xmath88 is the quotient of @xmath53 and the unknown harmonic resolution function . the general solution of this second order linear ode on @xmath106 is given by @xmath107 where @xmath108 and @xmath109 are arbitrary constants . to satisfy the condition @xmath110 , it must be that @xmath111 . however , in this case , @xmath112 does not extend smoothly to zero : @xmath113 which diverges at @xmath114 . thus , there exists no solution of @xmath93 for this choice of @xmath115 obeying the necessary boundary conditions and extending smoothly to the boundary . in other words , the zas @xmath19 admits no local harmonic resolution . the above discussion also appears in @xcite . in this section we prove the two main theorems of this paper , stated below . both were introduced and proved in @xcite ( except the case of equality in theorem [ thm_zas_ineq ] , which is a new result ) . the proofs rely on a certain unproven conjecture , explained in section [ sec_conf_conj ] . suppose @xmath115 is an asymptotically flat metric on @xmath116 of nonnegative scalar curvature such that all components of the boundary @xmath117 are zas . assume there exists a global harmonic resolution @xmath52 of @xmath19 with @xmath56 complete . also assume conjecture [ conj_conformal ] ( below ) holds . then the adm mass @xmath3 of @xmath118 satisfies @xmath119 where @xmath120 is as in definition [ def_zas_mass ] . equality holds in @xmath121 if and only if @xmath118 is a schwarzschild zas of mass @xmath12 . [ thm_zas_ineq ] if @xmath19 has components @xmath122 with respective regular masses @xmath123 , then by equation @xmath124 and proposition [ prop_masses_agree ] , the statement @xmath125 can be written @xmath126 suppose @xmath115 is an asymptotically flat metric on @xmath116 of nonnegative scalar curvature such that all components of the boundary @xmath117 are zas , and that @xmath81 has no other singularities . assume that conjecture @xmath127 holds . then the adm mass @xmath3 of @xmath118 satisfies @xmath125 . [ thm_zas_ineq_full ] ( we mean that @xmath81 has no singularities other than @xmath19 in the following sense : if a sequence of surfaces @xmath128 converges in @xmath129 to @xmath19 , then the closed region exterior to @xmath128 is a complete manifold - with - boundary with respect to @xmath115 for each @xmath130 . ) we have not attempted to characterize the case of equality in the general case ; this issue is discussed in section [ sec_coe_general ] . * remarks * robbins @xcite showed theorem [ thm_zas_ineq_full ] in the case @xmath131 is connected by applying the technique of weak inverse mean curvature flow ( imcf ) due to huisken and ilmanen @xcite . however , this approach gives no bound on the adm mass if more than one zas is present ; the proof relies crucially on the monotonicity of the hawking mass under imcf , but monotonicity is lost when the surfaces travel past a zas . note that the positive mass theorem does not apply to @xmath118 as above , since the latter can not be extended to a smooth , complete manifold . the following special case with @xmath132 deserves attention , since it is a new version of the positive mass theorem that allows for certain types of singularities and incomplete metrics . [ positive mass theorem with singularities ] suppose @xmath118 satisfies the hypotheses of theorem [ thm_zas_ineq_full ] with @xmath132 . then @xmath133 , where @xmath3 is the adm mass of @xmath118 . the above result is a nontrivial statement , since we have shown that there exist singularities of zero mass that are not removable . our tool in the proof of theorem [ thm_zas_ineq ] will be the riemannian penrose inequality . the latter applies to manifolds whose boundary is a minimal surface , so it will be necessary to transform our manifold into such ( by a conformal change , for example ) . this motivates the following problem , independent of zas theory . * problem : * given a smooth asymptotically flat manifold @xmath134 with boundary ( and no singularities ) , is it possible to find a positive harmonic function @xmath88 such that the boundary has zero mean curvature in the metric @xmath135 ? ( we require @xmath88 to be harmonic so that the sign of scalar curvature is preserved ; this will be discussed further below . ) it turns out that * the answer is no * , even if superharmonic functions are allowed . as a counterexample , consider the spherically symmetric metric @xmath136 , with @xmath137 given on @xmath138 $ ] as @xmath139 , with @xmath140 . smoothly extend @xmath50 to @xmath141 so that it 1 ) remains positive and 2 ) is given by @xmath142 outside of a compact set . ( the geometric significance of this asymptotically flat metric is 1 ) its boundary has positive mean curvature ( since @xmath143 ) and 2 ) the ratio of the areas of the sphere at @xmath114 and at @xmath144 gets arbitrarily large as @xmath145 . ) the general spherically symmetric harmonic function can be written as @xmath146 where @xmath147 and @xmath148 are constants . consider the boundary sphere at @xmath114 . by the formula for the first variation of area , its mean curvature is given by @xmath149 by formula @xmath150 in appendix b , the mean curvature of the boundary in the metric @xmath151 is given by @xmath152 if this is to equal zero , we must have @xmath153 . then the function @xmath88 can be written as @xmath154 if @xmath155 is sufficiently small , then by inspection , @xmath156 must vanish on @xmath138 $ ] , regardless of value of @xmath148 . in particular , this metric provides the desired counterexample . ( by the maximum principle , superharmonic functions only tend to increase @xmath157 , and thus would not change the outcome . ) see @xcite for a further discussion on obstructions to the above problem . the above example shows that one can not in general force the boundary to be a minimal surface via a global harmonic conformal factor . however , we predict in the following conjecture that we can `` almost '' do so in the following sense : given @xmath134 as above , there exists a positive @xmath158-harmonic function @xmath88 such that @xmath159 contains a compact surface @xmath160 of zero mean curvature ( with respect to @xmath161 ) such that @xmath160 and @xmath19 have approximately the same area ( with respect to @xmath161 ) . the following conjecture was first given in @xcite . ( for details on the outermost minimal area enclosure of the boundary of an asymptotically flat manifold , see section [ sec_omae ] in appendix a. ) let @xmath134 be a smooth asymptotically flat 3-manifold with compact , smooth , nonempty boundary @xmath19 , with @xmath158 extending smoothly to @xmath19 . given @xmath140 , there exists a smooth , positive function @xmath88 on @xmath159 and metric @xmath162 satisfying the following conditions . let @xmath160 be the outermost minimal area enclosure of @xmath19 with respect to @xmath163 . 1 . @xmath88 is harmonic with respect to @xmath158 and tends to 1 at infinity . 2 . in the metric @xmath163 , the areas of @xmath160 and @xmath19 differ by at most @xmath164 . 3 . @xmath160 is disjoint from @xmath19 . [ conj_conformal ] the third condition guarantees that @xmath160 is @xmath165 with zero mean curvature measured in @xmath163 ( see section [ sec_omae ] in appendix a ) . we remark that the conjecture is known to be true in the spherically symmetric case . the effect of applying the conformal conjecture to an asymptotically flat manifold with boundary is illustrated in figure [ fig_conf_conj ] . [ fig_conf_conj ] now we explain how this conjecture relates to theorems [ thm_zas_ineq ] and [ thm_zas_ineq_full ] . we wish to obtain a lower bound on the adm mass of @xmath118 . for the first theorem , our strategy is to take a global harmonic resolution and apply the conformal conjecture ; this produces a manifold - with - boundary to which the riemannian penrose inequality applies . ( note that because the conformal factor is harmonic , the property of nonnegative scalar curvature is preserved . see equation @xmath166 in appendix b. ) this gives a lower bound on the adm mass of the latter manifold ; in the next section we explain how to transform it to the desired bound . finally , theorem [ thm_zas_ineq_full ] will be a consequence of theorem [ thm_zas_ineq ] and the definition of mass . if true , conjecture [ conj_conformal ] implies the following statement , which is better adapted for the proof of theorem [ thm_zas_ineq ] . let @xmath118 have boundary @xmath19 consisting of zas and admitting a global harmonic resolution of @xmath19 . given @xmath140 , there exists a global harmonic resolution @xmath167 of @xmath19 such that @xmath168 ( the outermost minimal area enclosure of @xmath19 in the metric @xmath56 ) is minimal and @xmath169 . [ conformal_conj_cor ] apply conjecture [ conj_conformal ] to some global harmonic resolution @xmath52 , obtaining a metric @xmath170 and a function @xmath88 that is harmonic with respect to @xmath56 , such that @xmath171 . from equation @xmath83 in appendix b , the function @xmath172 is harmonic with respect to @xmath170 . it is readily checked that @xmath167 is a global harmonic resolution of @xmath19 obeying the desired properties . we prove the first part of theorem @xmath173 . note that nonnegativity of the scalar curvature is preserved under global harmonic resolutions and applications of conjecture [ conj_conformal ] , since the conformal factors are harmonic ( see equation @xmath166 in appendix b ) . this fact will allow the use of the riemannian penrose inequality . [ proof of first part of theorem [ thm_zas_ineq ] ] by hypothesis , we may assume the existence of a global harmonic resolution @xmath52 of @xmath19 as in conjecture [ conformal_conj_cor ] , for any @xmath140 . as above , we let @xmath160 be the outermost minimal area enclosure of @xmath19 in the metric @xmath56 . ( we reiterate the points that the riemannian penrose inequality applies to @xmath174 and @xmath175 . ) now we make a series of estimates : @xmath176 where @xmath177 as @xmath145 ; therefore inequality @xmath121 holds . before proving the second part of theorem [ thm_zas_ineq ] , we show that theorem [ thm_zas_ineq_full ] easily follows from the above . if the capacity of @xmath19 is positive , then @xmath18 by theorem [ thm_pos_cap ] . therefore @xmath125 follows trivially . now assume @xmath19 has zero capacity . let @xmath31 converge to @xmath19 in @xmath32 , and let @xmath178 be @xmath115-harmonic , vanishing on @xmath35 and tending to 1 at infinity . as before , @xmath179 gives a global harmonic resolution of the zas @xmath35 of the metric @xmath180 . by theorem @xmath173 , we have @xmath181 where @xmath182 is the adm mass of the metric @xmath180 . formula @xmath183 in appendix b allows us to compute the adm mass of a conformal metric in terms of the original metric . in the case at hand it gives @xmath184 since @xmath178 is harmonic and @xmath185 is homologous to @xmath35 , stokes theorem shows that the second term on the right - hand side is equal to @xmath186 , which equals @xmath187 . since @xmath19 has zero capacity , @xmath188 . now , taking @xmath189 of @xmath190 and applying @xmath191 , we have @xmath192 by taking the supremum of this expression over all such @xmath31 , we have that @xmath125 . now , we prove the second part of theorem [ thm_zas_ineq ] ( characterizing the case of equality ) . [ proof of second part of theorem [ thm_zas_ineq ] ] it is left to the reader to show that the schwarzschild zas metric @xmath193 ( with @xmath194 and @xmath71 the flat metric ) on @xmath195 has both adm mass and zas mass equal to @xmath3 . ( for the latter , apply the definition of regular mass to the resolution @xmath196 of @xmath19 . ) assume all components of @xmath117 are harmonically regular zas . let @xmath52 be a global harmonic resolution of @xmath19 as in conjecture [ conformal_conj_cor ] ; in particular @xmath197 is minimal and @xmath198 , ( where , as before , @xmath197 is the outermost minimal area enclosure of @xmath19 in the metric @xmath56 ) . assume that the adm mass @xmath3 of @xmath118 equals @xmath120 , so that equality holds at each step in the proof of inequality @xmath121 . in particular , the following must hold : 1 . @xmath199 , so equality holds in the riemannian penrose inequality . thus , @xmath174 is isometric to the schwarzschild metric of mass @xmath200 outside of @xmath197 . ( this does not complete the proof , since we have not yet classified the region inside of @xmath197 . ) @xmath59 is constant on @xmath19 ( by the case of equality of hlder s inequality ) . the minimum of the quadratic @xmath201 is attained , so @xmath202 . thus , @xmath203 squaring @xmath204 , dividing by @xmath205 , and applying the definition of regular mass gives @xmath206 out strategy is to show @xmath207 and @xmath19 have the same capacity ( in the @xmath56 metric ) ; since the former encloses the latter by definition , lemma @xmath208 would imply that these surfaces are equal . first , @xmath209 to compute the capacity of @xmath207 , let @xmath210 be the @xmath56-harmonic function that equals @xmath211 on @xmath207 and tends to 1 at infinity . in fact , @xmath212 is given by @xmath213 . ( here , we have used the identification of @xmath56 outside of @xmath207 with the schwarzschild metric . ) @xmath214 therefore , @xmath215 . but we know from lemma @xmath208 that the reverse inequality holds as well ( since @xmath207 encloses @xmath19 by definition ) . by the second part of lemma @xmath208 , we have that @xmath216 . therefore @xmath217 , so @xmath115 is given by @xmath218 which is the schwarzschild zas metric of mass @xmath219 . we now give a global example of a manifold of positive adm mass that contains a zero area singularity ; the construction begins with a schwarzschild space with a cylinder appended to its horizon . for fixed parameters @xmath220 , let @xmath81 be the manifold with boundary @xmath221 . give @xmath81 a metric @xmath56 according to : @xmath222 where @xmath34 parametrizes @xmath223 and @xmath224 is the round metric of radius 1 on @xmath225 . the first region is a round cylinder of length @xmath226 with spherical cross sections of area @xmath227 . the second region , diffeomorphic to @xmath228 , has the schwarzschild metric of mass @xmath229 . by construction , the sphere @xmath230 is the apparent horizon , and this metric is @xmath231 . for a diagram , see figure [ fig_extended_schwarz ] . let @xmath53 be the unique @xmath56-harmonic function that equals @xmath211 on @xmath232 and tends to 1 at infinity . then @xmath19 is a harmonically regular zas of @xmath233 with global harmonic resolution @xmath52 . in this example , we shall compute the mass of the manifold @xmath118 as well as mass of the zas @xmath19 and find that they 1 ) respect inequality @xmath121 and 2 ) are equal if and only if @xmath234 . [ fig_extended_schwarz ] by spherical symmetry , @xmath53 depends only on @xmath34 . such harmonic functions on a round cylinder and schwarzschild space are readily characterized ; @xmath53 takes the form : @xmath235 where @xmath236 is a parameter to be determined and @xmath237 is chosen so that @xmath53 is continuous ( and so @xmath115 is continuous ) . for @xmath115 to be a @xmath32 metric , the unit normal derivative @xmath238 of @xmath53 at @xmath160 must be continuous . inside @xmath160 , @xmath239 . outside , @xmath240 , and a short computation shows @xmath241 on @xmath160 . equating these interior and exterior boundary conditions on @xmath242 allows us to solve for @xmath236 : @xmath243 so that @xmath244 furthermore , @xmath245 outside @xmath160 , so the adm mass @xmath3 of @xmath118 is in fact @xmath246 . on the other hand , the mass of the harmonically regular zas @xmath19 can be computed from definition [ def_reg_mass ] and proposition [ prop_masses_agree ] as @xmath247 where we have used the fact that @xmath248 now , inequality @xmath121 ( the riemannian zas inequality ) is satisfied with @xmath118 ; by the above computations , this is stated as : @xmath249 note that equality holds if and only if the length @xmath226 of the cylinder is zero , in which case @xmath118 is the schwarzschild zas . interestingly , if @xmath250 , then @xmath118 has positive mass , even though @xmath19 has negative mass . this is indicated in figure [ fig_zas_cylinder ] . in fact , @xmath19 is not enclosed by an apparent horizon , since @xmath115 is spherically symmetric , and the area of spheres is strictly increasing with distance from @xmath19 . the increase in mass with the length of the cylinder is explained by the region of positive scalar curvature introduced by the cylinder . indeed , this phenomenon supports the heuristic that positive scalar curvature tends to increase adm mass . finally , we note that @xmath115 is merely @xmath231 , so the zas theory of this paper does not technically apply . however , @xmath115 could be perturbed to a @xmath165 metric exhibiting the qualitative properties above . these values obey inequality ( [ zas_ineq ] ) for all @xmath251 , with equality only for @xmath234 . [ fig_zas_cylinder ] discussions in this paper raise a number of interesting questions . to complete the arguments of this paper , conjecture [ conj_conformal ] must be proved . it is readily verified in a spherically symmetric setting establishing this conjecture as a theorem ( or providing a counterexample ) will be crucial to furthering the theory of zas . for zas that are not harmonically regular , the case of equality of theorem [ thm_zas_ineq_full ] is not necessarily the schwarzschild zas metric . for instance , @xmath6 minus a finite number of points gives an asymptotically flat manifold of zero scalar curvature ( and zero adm mass ) whose boundary consists of zas of zero mass . however we conjecture that deleted points are the only obstruction to uniqueness of the schwarzschild zas as the case of equality : if , in theorem [ thm_zas_ineq_full ] , it happens that @xmath252 , then @xmath118 is isometric to a schwarzschild metric with finitely many points deleted , if @xmath253 , or the flat metric on @xmath6 with finitely many points deleted if @xmath254 . [ conj_coe ] the first author conjectured the following inequality for the situation in which black holes and zas exist simultaneously @xcite . recall that a surface is area outer - minimizing if every surface enclosing it has equal or greater area . let @xmath115 be an asymptotically flat metric on @xmath81 with nonnegative scalar curvature . assume @xmath81 has compact smooth boundary @xmath131 that is a disjoint union @xmath255 , where @xmath256 is an area outer - minimizing minimal surface and @xmath19 consists of zas . then the adm mass @xmath3 of @xmath118 obeys @xmath257 ( note that @xmath256 and @xmath19 need not be connected , and @xmath120 is nonpositive . ) the heuristic behind this conjecture is simply that the potential energy between a black hole and a zas is positive ( since their masses have opposite sign ) , which would make a positive contribution to the adm mass . by this logic and the riemannian penrose and zas inequalities , the right hand side of @xmath258 underestimates the adm mass . it is expected that equality would hold only in the cases of flat space , the schwarzschild metric , and the schwarzschild zas metric , all with points possibly deleted ( which correspond to removable zas of zero mass ) . see appendix b of @xcite for further discussion . a special case to consider is when @xmath131 consists of exactly two components one connected zas and one connected area - outer - minimizing minimal surface whose masses are equal in magnitude but opposite in sign . the conjecture would predict that the adm mass of @xmath118 is nonnegative . alternatively , such a manifold may be the most obvious source of a counterexample to the conjecture . recently , j. streets proved the existence and uniqueness of a ( weakly defined ) inverse mean curvature flow of surfaces `` out '' of a zero area singularity @xmath19 ( see @xcite ; full details available in unpublished version ) . this gives rise to a canonical family of surfaces @xmath259 converging to @xmath19 as @xmath260 . moreover , this family has the property that the limit of the hawking mass is optimal in the following sense : @xmath261 where the supremum is taken over all sequences of surfaces @xmath31 converging in @xmath262 to @xmath19 . ( this result assumes that the background metric has nonnegative scalar curvature . ) in other words , the existence of the canonical family obviates the need for taking the sup of a limsup . perhaps @xmath263 is a `` better '' definition of mass than that which we have adopted . however , it is unknown how to bound the adm mass in terms of this quantity ( unless @xmath19 consists of regular zas see corollary [ cor_reg_mass ] ) . we make the following related conjecture , which would show that the mass of zas defined via hawking masses underestimates the definition of mass we have chosen . it is motivated by the first part of proposition [ reg_mass_thm ] . for sequences @xmath31 converging in @xmath32 to a zas @xmath19 of a metric of nonnegative scalar curvature , @xmath264 the left hand side can be computed by equation @xmath263 above . in the case of @xmath19 connected , the conjecture trivially holds for @xmath19 of positive capacity , since both sides equal @xmath26 ( theorems [ thm_capacity ] and [ thm_pos_cap ] ) . along another line of thought , recall the notion of removable @xmath76 singularities from section [ sec_removable ] . our definition of mass is unsatisfactory for such singularities , since it always @xmath26 . a more sophisticated definition of mass would give a finite quantity for removable @xmath76 singularities ( and would ideally give a lower bound on the adm mass in the same sense as the riemannian zas inequality . ) in this section we define the bartnik mass of a collection of zas . let @xmath118 be an asymptotically flat manifold with boundary @xmath131 such that @xmath115 has nonnegative scalar curvature . let @xmath265 be a surface in @xmath81 that is a graph over @xmath266 ( see section [ sec_surfaces ] ) . recall the _ outer bartnik mass _ of @xmath256 is defined by @xcite @xmath267 where the infimum is taken over all `` valid extensions '' @xmath268 of @xmath256 and @xmath269 is the adm mass of @xmath268 extension _ of @xmath256 is an asymptotically flat manifold with boundary @xmath134 of nonnegative scalar curvature containing a subset that is isometric to @xmath270 and has no singularities outside of @xmath270 ; it is said to be _ valid _ if @xmath256 is not enclosed by a surface of less area . if @xmath271 is a collection of zas , we define the ( outer ) _ bartnik mass _ of @xmath19 as @xmath272 where @xmath35 is some sequence of surfaces converging in @xmath129 to @xmath19 . the limit exists ( possibly @xmath26 ) and is independent of the choice of sequence because of the monotonicity of the outer bartnik mass ( i.e. , if @xmath273 encloses @xmath274 , then @xmath275 ) . a priori , the bartnik mass of @xmath19 could be positive , negative , or zero . in fact , there may exist examples of zas of positive bartnik mass in a scalar - flat manifold with no apparent horizons . however , the riemannian zas inequality gives a lower bound on the bartnik mass in terms of the mass of the zas ( assuming conjecture [ conj_conformal ] ) . given a surface @xmath256 , one version of the difficult and open `` bartnik minimal mass extension problem '' is to determine whether there exists some valid extension @xmath268 whose adm mass equals the outer bartnik mass of @xmath256 . however , the same problem for a collection of zas @xmath19 may in fact be more tractable . this is because any extension can be viewed as a valid extension , explained as follows let @xmath276 in @xmath129 ; without loss of generality , we may assume that each surface enclosing @xmath35 has area at least that of @xmath35 , and the areas of @xmath35 decrease monotonically to zero . then , given any extension @xmath268 of @xmath35 ( for some @xmath130 ) , @xmath268 gives a valid extension for some @xmath277 with @xmath278 . thus , for the purposes of computing the bartnik mass of zas @xmath19 , it is not necessary to restrict only to valid extensions . the definitions of zas , resolutions , mass , and capacity extend naturally to higher dimensions ( with suitable changes made to certain constants ) . also , the riemannian penrose inequality has recently been generalized to dimensions less than eight by the first author and d. lee @xcite . thus , the riemannian zas inequality in dimensions less than eight will follow if conjecture @xmath127 is proved ( in dimensions less than eight ) . ultimately , it would be desirable to go beyond the time - symmetric case and develop a theory of zas for arbitrary spacelike slices of spacetimes . one might hope to prove a version of the zas inequality , @xmath125 , in this setting . in figure [ fig_mass_theorems ] , we illustrate how such an inequality would fit in with the positive mass theorem and the penrose inequality . [ fig_mass_theorems ] .... \section{appendix 2 : scalar curvature and potential energy of zas } the schwarzschild zas metric is scalar flat , but it is nevertheless interesting to ask how its scalar curvature behaves distributionally at the singularity . to study this , we approximate this singular metric ( with $ m = -2 $ ) with a $ c^0 $ metric defined on all of $ \r^3$. for $ \epsilon > 0 $ small , define the radial function \begin{equation } u_\epsilon(r ) = \left\{\begin{array}{cc } \epsilon , & 0 \leq r \leq 1\\ 1-\frac{1}{r}+\epsilon , & r \geq 1 \end{array}\right . \end{equation } and set $ g_\epsilon = u_{\epsilon}^4 \delta$. this metric is smooth and scalar flat , except on the sphere of radius 1 . we compute the integral of the distributional scalar curvature over $ \r^3$. let $ a_\delta$ be the region bounded by the spheres of radii $ 1-\delta$ and $ 1+\delta$ about 0 . then \begin{align * } \int r_{g_\epsilon } dv_{g_\epsilon } & = -8 \lim_{\delta \to 0 } \int_{a_\delta } u_\epsilon^{-5}\delta u_{\epsilon } \ : u_\epsilon^6 dv & & \text{(equation $ ( \ref{eq_conf_scalar_curv})$)}\\ & = -8\epsilon \lim_{\delta \to 0 } \int_{r=1-\delta}^{1+\delta } \left(u''(r ) + \frac{2}{r}u'(r)\right)4\pi r^2 dr & & \text{(laplacian on $ \r^3$)}\\ & = -32 \pi \epsilon\lim_{\delta \to 0 } \left(r^2u'(r)\big|_{1-\delta}^{1+\delta}\right ) & & \text{(integration by parts)}\\ & = -32 \pi \epsilon . \end{align * } \noindent taking the limit $ \epsilon \to 0 $ , we conclude that the schwarzschild zas has zero distributional scalar curvature at its singularity . next , we recall that if $ ( m , g)$ is a time - symmetric , spacelike slice of a spacetime obeying einstein 's equation , then the energy density $ \mu$ is given by $ r/16\pi$ , where $ r$ is the scalar curvature of $ g$. we now estimate the distributional ( gravitational ) potential energy of the schwarzschild zas by the approximation $ g_\epsilon$. classically , if $ \mu$ represents a spherically symmetric matter distribution in a bounded region , then its gravitational potential energy is given by \begin{equation } u:=-\int \frac{m(r ) 4\pi r^2 \mu(r)}{r } dr , \end{equation } where $ m(r)$ is quantity of matter at radius $ \leq r$. to utilize this formula for $ g_\epsilon$ , we must use the volume form for $ g_\epsilon$ and replace the distance $ r$ with $ \epsilon^2 r$. in this case , $ m(r)$ vanishes for $ r < 1 $ and equals $ -32\pi$ for $ r \geq 1$. the integral for potential energy is easy to compute by observing that near 1 , $ u''(r)$ behaves as a delta function . \begin{align } u & = -\int \frac{m(r ) 4\pi r^2 \mu(r)}{\epsilon^2 r } u_\epsilon^6 dr\\ & = -\int \frac{m(r ) 4\pi r^2 \frac{-8}{16\pi } u_\epsilon^{-5}\left(u''(r ) + \frac{2}{r } u'(r)\right)}{\epsilon^2 r}u_\epsilon^6 dr & \text{(equation $ ( \ref{eq_conf_scalar_curv})$)}\\ & = -\frac{(-32\pi\epsilon)(4\pi ) \left(\frac{-8}{16\pi}\right)\epsilon^{-5}}{\epsilon^2}\epsilon^6\\ & = -64\pi . \end{align } this indicates that in the distributional sense , the schwarzschild zas has finite , but non - zero , potential energy . .... the essence of this paper was to associate a quantity `` mass '' to the class of zero area singularities . the primary application of the theory was to estimate the adm mass of an asymptotically flat manifold of nonnegative scalar curvature in terms of the masses of its singularities . in 1997 , the first author , just out of graduate school , sat next to barry mazur at a conference dinner at harvard , who , quite characteristically , asked the first author a series of probing questions about his research , which at the time concerned black holes . one of the questions was `` can a black hole have negative mass , and if so , what properties would it have ? '' contemplating this natural question marked the beginning of an enjoyable journey leading to this paper . the second author would like to thank mark stern and jeff streets for helpful discussions . global geometric problems in general relativity are often studied on asymptotically flat manifolds . topologically , such spaces resemble @xmath6 outside of a compact set , and thus have well - defined notions of `` infinity . '' geometrically , they approach euclidean space at infinity . a connected riemannian 3-manifold @xmath118 ( with or without boundary ) is * asymptotically flat * ( with one end ) if there exists a compact subset @xmath279 and a diffeomorphism @xmath280 ( where @xmath281 is the closed unit ball about the origin ) , such that in the coordinates @xmath282 on @xmath283 induced by @xmath284 , the metric satisfies , for some @xmath285 and @xmath286 , @xmath287 for all @xmath288 , where @xmath289 , @xmath290 , and @xmath291 is the scalar curvature of @xmath115 . the @xmath292 are called * asymptotically flat coordinates*. ( this definition can easily be generalized to allow for several ends and higher dimensions . ) unless otherwise stated , it is assumed that @xmath115 is smooth and extends smoothly to the boundary . for asymptotically flat manifolds , there is an associated quantity called the adm mass that measures the rate at which the metric becomes flat at infinity @xcite . physically , it represents the mass of an isolated gravitational system . the * adm mass * of an asymptotically flat 3-manifold @xmath118 is the number @xmath293 where the @xmath292 are asymptotically flat coordinates , @xmath185 is the coordinate sphere of radius @xmath294 , @xmath295 is the euclidean unit normal to @xmath185 ( pointing toward infinity ) , and @xmath296 is the euclidean area form on @xmath185 . [ def_adm_mass ] due to a theorem of bartnik @xcite , the mass is finite and depends only on @xmath81 and @xmath115 ( and not the choice of coordinates ) . it is left to the reader to show that the schwarzschild metric @xmath297 on @xmath298 is asymptotically flat with adm mass equal to @xmath3 , for any real number @xmath3 . when there is no ambiguity , we sometimes refer to the adm mass as simply `` mass . '' .... a special class of asymptotically flat manifolds : \begin{definition } an asymptotically flat manifold $ ( m , g)$ is \textbf{harmonically flat at infinity } if there exists a compact subset $ k \subset m$ and a positive function $ v:\r^n \sm \ol{b_1(0)}\to \r$ , harmonic with respect to the flat metric $ \delta$ and approaching a positive constant at infinity , such that $ ( m \sm k , g)$ is isometric to $ ( \r^n \sm \ol{b_1(0)},v^{\frac{4}{n-2 } } \delta)$. we will refer to $ v$ as the \textbf{harmonic function associated to $ ( m , g)$}. \end{definition } note that $ v$ has an expansion \begin{equation } v(x ) = a + \frac{b}{|x|^{n-2 } } + \ocal\left(|x|^{1-n}\right ) \label{expansion } \end{equation } for real constants $ a>0 $ and $ b$. also , recall that if $ g_1 = w^\frac{4}{n-2 } g_0 $ for a smooth , positive function $ w$ , then scalar curvatures of $ g_1 $ and $ g_0 $ are related by $ $ r_1 = w^{-\frac{n+2}{n-2 } } \left(-4\frac{n-1}{n-2 } \delta_0 w + r_0 w\right).$$ from this , we immediately deduce that manifolds harmonically flat at infinity have vanishing scalar curvature outside of a compact set . additionally , we have the following lemma , which gives an explicit expression for the mass of a manifold that is harmonically flat at infinity : \begin{lemma } let $ ( m , g)$ be harmonically flat at infinity with associated harmonic function $ v:\r^n\sm \ol{b_{1}(0)}\to \r$. then the adm mass $ m$ of $ ( m , g)$ is equal to $ 2(n-2)ab$ , where $ a$ and $ b$ are the constants determined by the expansion $ ( \ref{expansion})$. \label{hfmass } \end{lemma } .... a complete asymptotically flat manifold @xmath118 without boundary possesses a unique _ outermost minimal surface _ @xmath256 ( possibly empty , possibly with multiple connected components ) . each component of @xmath256 is minimal ( i.e. , has zero mean curvature ) and is not enclosed by any other minimal surface . see figure [ fig_omae ] for an illustration . some well - known ( but not obvious ) results are that 1 ) @xmath256 is a disjoint union of smooth , immersed 2-spheres ( that are in general embedded ) , 2 ) the region of @xmath81 outside of @xmath256 is diffeomorphic to @xmath6 minus a finite number of closed balls , and 3 ) @xmath256 minimizes area in its homology class outside of @xmath256 . see @xcite and section 4 of @xcite for further details and references . related to the above is the concept of outermost minimal area enclosure : let @xmath118 be an asymptotically flat manifold with smooth , compact boundary @xmath131 . then there exists a unique embedded surface @xmath160 ( possibly of multiple connected components ) enclosing @xmath131 such that 1 ) @xmath160 has the least area among all surfaces that enclose @xmath131 and 2 ) @xmath160 is not enclosed by a surface of equal area ; such @xmath160 is called the _ outermost minimal area enclosure of @xmath131 _ ( see figure [ fig_omae ] ) . in general , all of the following cases may occur : i ) @xmath299 , ii ) @xmath300 but @xmath301 is nonempty , and iii ) @xmath302 is empty . note that @xmath160 may only have @xmath231 regularity and may not have zero mean curvature ( see theorem 1.3 of @xcite ) . however , @xmath303 is @xmath165 with zero mean curvature ( from standard minimal surface theory ) . [ fig_omae ] .... although the schwarzschild zas metric is scalar flat , it is interesting to ask what its `` distributional '' scalar curvature is at the singularity . to study this , we approximate this singular metric with a $ c^1 $ metric defined on $ \r^3$. for $ \epsilon > 0 $ small , define the radial function \begin{equation } % u_\epsilon(r ) = \left\{\begin{array}{cc}\frac{2\epsilon}{\pi } % \arctan\left(\frac{\pi}{2\epsilon}r - \frac{\pi}{2\epsilon}\right ) + % \epsilon , & 0 \leq r\leq 1\\ % 1-\frac{1}{r}+\epsilon , & r \geq 1 % \end{array}\right . u_\epsilon(r ) = \left\{\begin{array}{cc } \epsilon e^{\frac{r-1}{\epsilon } } , & 0 \leq r \leq 1\\ 1-\frac{1}{r}+\epsilon , & r \geq 1 \end{array}\right . \end{equation } which we observe is $ c^1$. set $ g_\epsilon = u_\epsilon^4 \delta$ , which converges uniformly to the schwarzschild zas ( of mass $ -2 $ ) as $ \epsilon \to 0$. although $ g$ is merely $ c^1 $ , we can nonetheless compute the integral of its scalar curvature in the ball $ b$ of radius $ 1$. \begin{align * } \int_{b } r_{g_\epsilon } dv_{g_\epsilon } & = \int_{b } ( -8u_\epsilon^{-5}\delta_\delta u_\epsilon)(u_\epsilon^6)dv_\delta\\ & = -32\pi\int_0 ^ 1 ( u''(r ) + \frac{2}{r } u'(r))u(r)r^2 dr\\ & = -32\pi\left(\left.u(r)u'(r)r^2\right|_0 ^ 1 + \int_0 ^ 1 -r^2 u'(r)^2 - 2ru(r)u'(r ) + 2ru(r)u'(r)\right)\\ & = -32\pi\left(\epsilon - \int_0 ^ 1 r^2 e^{2(r-1)/\epsilon}dr\right)\\ & = -32\pi\left(\epsilon/2+\epsilon^2/2-\epsilon^3/4 + \epsilon^3e^{-2/\epsilon}/4\right)\\ & = \ocal(\epsilon ) . \end{align * } \noindent taking the limit $ \epsilon \to 0 $ , we conclude that the schwarzschild zas has zero distributional scalar curvature at its singularity . next , we recall that if $ ( m , g)$ is a time - symmetric , spacelike slice of a spacetime obeying einstein 's equation , then the energy density $ \mu$ is given by $ r/16\pi$ , where $ r$ is the scalar curvature of $ g$. we now compute the potential energy $ u$ of $ g_\epsilon$ in $ b$ as \begin{equation } u = -\int_0 ^ 1 \frac{e(r)4\pi r^2 \mu(r ) u_\epsilon(r)^4}{d(0,r ) } u_\epsilon(r)^2dr . \end{equation } this corresponds to the integral of the classical gravitational potential $ -\frac{m_1 m_2}{r}$. here , $ 4\pi r^2 \mu(r ) u(r)^4 $ is the energy contained in the shell of radius $ r$ , $ u(r)^2 dr$ is the line element , $ d(0,r)$ is the distance ( in $ g_\epsilon$ ) from $ r$ to the origin , and $ e(r)$ is the integral of energy density over the ball or radius $ r$ : \begin{align } e(r ) & = \int_0^r 4\pi \mu(x)x^2 u(x)^6 dx . \end{align } now , we compute the potential energy : \begin{align } u & = -4\pi(-8 ) \int_0 ^ 1 \frac{e(r)r^2 \left(u''(r ) + \frac{2}{r } u'(r)\right)}{d(0,r ) } u(r ) dr % \frac{\epsilon}{4}\int_0 ^ 1 \frac{r^2 u'(r)^2 \cdot r^2 r(r)}{r}u(r)^6dr\\ % & = \frac{-8\epsilon}{4}\int_0 ^ 1 r^3 u'(r)^2 \left(u''(r ) + \frac{2}{r } u'(r)\right)u(r)dr\\ % & = -2\epsilon\int_0 ^ 1 r^3 u'(r)^2 \left(u''(r ) + \frac{2}{r } u'(r)\right)u(r)dr\\ \end{align } .... here , we present several formulas that describe the behavior of certain geometric quantities under conformal changes . assume @xmath304 and @xmath305 are riemannian metrics on a 3-manifold such that @xmath306 , where @xmath307 . for any smooth function @xmath85 , direct calculation shows @xmath308 where @xmath309 and @xmath310 are the ( negative spectrum ) laplace operators for @xmath304 and @xmath305 , respectively @xcite . in particular , if @xmath311 , then @xmath312 . more generally , if @xmath313 , then @xmath314 . if @xmath315 and @xmath316 are the respective scalar curvatures of @xmath304 and @xmath305 , then @xmath317 in particular , if @xmath88 is harmonic with respect to @xmath304 , then @xmath304 has nonnegative scalar curvature if and only if @xmath305 does . if @xmath256 is a hypersurface of mean curvature @xmath318 with respect to @xmath319 ( and outward unit normal @xmath320 ) , @xmath321 , then @xmath322 our convention is such that the mean curvature of a round sphere in flat @xmath6 is positive . if @xmath323 are asymptotically flat with respective adm masses @xmath324 and @xmath325 , and if @xmath326 at infinity , then @xmath327 where , in some asymptotically flat coordinate system @xmath292 , @xmath185 is the coordinate sphere of radius @xmath294 , @xmath295 is the euclidean unit normal to @xmath185 ( pointing toward infinity ) , and @xmath296 is the euclidean area form on @xmath185 . .... \section{editorial : outstanding issues and questions } \subsection{major issues } \begin{itemize } % \item what would a kerr black hole of negative mass look like ( in a spacelike slice ) ? for % simplicity , start with the non - relativistic limit . % \item for the conformal conjecture , we know that the o.m.a.e . is $ c^{1,1}$ , as well as of zero mean curvature . % can we figure out if its smooth ? if not , we may not even be able to apply the penrose inequality to it ! % \item make sure the two places where i solved a pde are in fact correct . remember to discuss with hugh ! % \item how do we define completeness for a manifold with zas , and how can we be sure that it carries over to completeness in the manifold % to which we apply the rpi ? i claim completeness for af manifolds is redundant . % what do lee - parker have to say - no mention of completeness . \item discuss modification of bartnik mass with hugh - not allowing extensions to have more zas . \end{itemize } \subsection{minor issues } \begin{itemize } % \item did we actually need $ c^2 $ convergence of surfaces , besides the one proposition for the regular case ? \item is the description of the hierarchy picture ok ? \item keep references current : streets ' new note , bray - khuri , bray - lam \end{itemize } \subsection{deferred to thesis } \begin{itemize } \item right now , the proof of the proposition which shows $ m = m_{\text{reg}}$ for regular zas is incorrect / incomplete . it can be fixed for globally harmonically regular zas ( see $ 6/17/07 $ , $ 8/6/08 $ notes ) . the rest will be taken up in my thesis . to simplify the discussion greatly , we 1 ) need a new name for regular mass , and 2 ) need to know if global harmonic resolutions exist . \item proposition 11 might be false - showing that global harmonic resolutions exist . on july 25 , i showed it is true in the spherically symmetric case . if it turns out to be false in general , i 'll need to rework the first main theorem by assuming the existence of a global harmonic resolution ( and the corollary to the conformal conjecture ) . \item potential energy computations . \item showing that if $ \frac{\ol \delta \varphi}{\varphi}$ extends smoothly to the boundary , then a regular zas is harmonically regular . \end{itemize } ....

we present interplanetary network ( ipn ) detection and localization information for 211 gamma - ray bursts ( grbs ) observed as untriggered events by the burst and transient source experiment ( batse ) , and published in catalogs by kommers et al . ( 2001 ) and stern et al . ( 2001 ) . ipn confirmations have been obtained by analyzing the data from 11 experiments . for any given burst observed by batse and one other distant spacecraft , arrival time analysis ( or `` triangulation '' ) results in an annulus of possible arrival directions whose half - width varies between 14 arcseconds and 5.6 degrees , depending on the intensity , time history , and arrival direction of the burst , as well as the distance between the spacecraft . this annulus generally intersects the batse error circle , resulting in a reduction of the area of up to a factor of 650 . when three widely separated spacecraft observed a burst , the result is an error box whose area is as much as 30000 times smaller than that of the batse error circle . because the ipn instruments are considerably less sensitive than batse , they generally did not detect the weakest untriggered bursts , but did detect the more intense ones which failed to trigger batse when the trigger was disabled . in a few cases , we have been able to identify the probable origin of bursts as soft gamma repeaters . the vast majority of the ipn - detected events , however , are grbs , and the confirmation of them validates many of the procedures utilized to detect batse untriggered bursts .

astro-ph0409053

this paper presents the 8th catalog of gamma - ray burst ( grb ) localizations obtained by arrival time analysis , or `` triangulation '' between the missions in the 3rd interplanetary network ( ipn ) , which began operations in 1990 and continues to operate today . two of these catalogs ( hurley et al . 1999a , b ) were supplements to the batse 3b and 4br burst catalogs ( meegan et al . 1996 ; paciesas et al . 1999 ) . the others involved bursts observed by numerous other spacecraft ( laros et al . 1997 , 1998 ; hurley et al . , 2000a , b , c ) . in this paper , we present ipn data on 211 _ untriggered bursts which occurred throughout the entire _ compton gamma - ray observatory ( cgro ) mission ( 1991 april through 2000 may ) . the batse data on these events , such as durations , fluxes , fluences , and coarse location information , appear in two catalogs , kommers et al . ( 2001 ) and stern et al . a final ipn supplement catalog , to the batse 5b catalog , is in preparation ( hurley et al . 2004 , briggs et al . , 2004 ) . _ _ the purpose of searching the batse data for untriggered events was mainly to extend the number - intensity ( log n - log s ) distribution to weaker bursts than those that could trigger the detector , and thus to gain more information on the burst distribution , particularly at the weak end . other objectives included the detection of bursts from known and unknown soft gamma repeaters , and very soft transients which could constitute a previously unknown phenomenon . ( one significant outcome of this effort was the detection of the bursting pulsar ) . the purpose of searching the ipn data for these events was to confirm as many of them as possible , reduce the sizes of their error circles , and validate the procedures used to identify these untriggered events . we have used the same procedures as those employed in the other batse catalog supplements , and refer the reader to hurley et al . ( 1999a , b ) for the detailed descriptions . generally speaking , using the arrival time and direction of a burst at batse , and its time history , we searched the data of the near - earth spacecraft for a confirmation at the same time ; for the spacecraft which were far from earth , we searched for a confirmation ( i.e. , an event with a matching time history ) in the appropriate crossing time window . although more than 15 separate gamma - ray burst experiments were operating on over a dozen missions throughout the duration of the cgro mission , confirmations were obtained from the data of just 11 experiments : the _ bepposax gamma - ray burst monitor ( frontera et al . 1997 ; feroci et al . 1997 ) , the _ defense meteorological satellite program ( dmsp , terrell et al . 1992 ) , ginga ( murakami et al . 1989 ) , konus - a ( aptekar et al . 1997 ) , konus -_wind ( aptekar et al . 1992 ) , the near earth asteroid rendezvous mission ( near , goldsten et al . 1997 ) , phebus ( terekhov et al . 1994 ) , _ pioneer venus orbiter ( klebesadel et al . 1980 ) , sross c-2 ( kasturirangan et al . 1997 ) , _ ulysses ( hurley et al . 1992 ) , and watch - granat ( brandt , lund , & rao 1990 ) . we note here , however , two important differences in the procedures and results between the triggered and untriggered events . _ _ _ _ _ first , the untriggered burst catalogs contain a much higher proportion of weak events than the batse triggered burst catalogs . because the ipn instruments are generally much less sensitive than batse , they detected a smaller fraction of the untriggered than the triggered ones . second , the untriggered event time histories were recorded in the 1.024 s resolution batse data , while the triggered event time histories were recorded with much higher time resolution . thus when an untriggered event was detected only by batse and another near - earth spacecraft , the low time resolution and the proximity of the two spacecraft results in a very wide annulus which is consistent with , but does not constrain the batse error circle . twenty - one events fell into this category , and it is only possible to confirm their detection , but not to obtain a meaningful annulus or error box for them . there are 873 untriggered bursts in the kommers et al . ( 2001 ) catalog and 1838 untriggered bursts in the stern et al . ( 2001 ) catalog . the two sets are not mutually exclusive ( stern et al . 2001 ) , and the total number of untriggered bursts is approximately 2000 , depending on the exact acceptance criteria . their peak fluxes range from 0.06 to 25 photons @xmath1 s@xmath2 . figure 1 gives the ipn efficiency for detecting untriggered bursts as a function of their peak fluxes . this is defined as the number of bursts detected by the ipn divided by the total number of untriggered bursts in a particular flux range . there are many factors which determine whether a burst is detected by an ipn spacecraft . in addition to the burst intensity and time history , solar activity , earth - blocking for spacecraft in low earth orbit , the number of spacecraft active in the ipn , and data return all play important , time - variable roles . figure 1 therefore gives time - averaged efficiencies . approximately one out of nine untriggered batse bursts was observed by at least one spacecraft in the ipn . their fluxes range from 0.15 to 25 photons @xmath1 s@xmath2 . for comparison , approximately one out of every three triggered batse bursts was observed by ipn spacecraft ( hurley et al . 1999b ) . of the 211 ipn events , only 90 could be localized ( 85 to annuli only , and 5 to error boxes ) . for each confirmed untriggered burst , table 1 lists the spacecraft which observed the event . ( a list of _ all grbs and the ipn spacecraft which detected them may be found at http://ssl.berkeley.edu/ipn3/masterli.html or http://heasarc.gsfc.nasa.gov / w3browse/. ) _ for those bursts which can be localized , either to a single annulus whose width is comparable to or less than the diameter of the batse error circle ( an example is shown in figure 2 ) , or to an error box ( an example is shown in figure 3 ) , the 6 columns in table 2 give : \1 ) the date of the burst , in yymmdd format , 2 ) the universal time of the burst at earth in seconds , 3 ) the right ascension of the center of the ipn annulus , epoch j2000 , in the heliocentric frame , in degrees , 4 ) the declination of the center of the ipn annulus , epoch j2000 , in the heliocentric frame , in degrees , 5 ) the angular radius r@xmath3 of the first ipn annulus , in the heliocentric frame , in degrees , and 6 ) the half - width @xmath4 r@xmath5 of the first ipn annulus , in degrees ; the 3 @xmath6 confidence annulus is given by r@xmath3 @xmath7 @xmath4 r@xmath5 . if the burst was detected by a third , distant spacecraft , and a non - degenerate second annulus could be derived for it , the information in columns 4 , 5 , and 6 is repeated for this annulus . for the bursts in table 2 , table 3 gives the batse error circles , from kommers et al . ( 2001 ) and stern et al . ( 2001 ) , and either a ) the intersection points of the ipn annulus with the error circle , or b ) for the three - spacecraft localizations , the four corners of the ipn error box . for each entry , the first line contains : \1 ) the date of the burst , in yymmdd format , 2 ) the universal time of the burst at earth , in seconds , 3 ) the right ascension of the center of the batse error circle , in degrees , 4 ) the declination of the center of the batse error circle , in degrees , and 5 ) the radius of the batse error circle , in degrees ; this is the combination of the one sigma statistical error and a 1.6 degree systematic error , summed in quadrature . the four following lines contain the right ascension and declination , in degrees , of the error box . for those bursts which were observed by batse and a single ipn spacecraft ( e.g. figure 2 ) , the coordinates are those of the intersection of the 3 @xmath6 ipn annulus with the 1 @xmath6 ( statistical plus systematic ) batse error circle . although all of the annuli are statistically consistent with the positions of their respective 1 @xmath6 batse error circles , in some cases part or all of the annulus does not actually intersect the error circle . in those cases , the coordinates are set to zero . for those bursts which were observed by two distant ipn spacecraft ( e.g. figure 2 ) , and for which an ipn - only error box can be derived , the coordinates given are those of the ipn error box . all coordinates are j2000 , and all event times are the ones used to identify the bursts in the stern et al . ( 2001 ) and kommers et al . ( 2001 ) catalogs . we note here a number of unusual circumstances surrounding some of the bursts in the stern et al . ( 2001 ) and kommers et al . ( 2001 ) catalogs . * some of the bursts in the two catalogs in fact correspond to batse triggers . in some cases , the triggers were not caused by the bursts , but the bursts were nevertheless recorded in triggered mode . * the kommers et al . ( 2001 ) catalog was divided into two parts : high energy ( he ) events , and low energy ( le ) events . initially , there were 125 le events , but 75 of them were intentionally eliminated from the final catalog because their origin was suspected to be either magnetospheric , x - ray binaries in outburst , or activity from soft gamma repeaters ( sgrs ) . we have identified four of the 75 eliminated events as bursts from sgr1806 - 20 . these four can be found in the complete list of sgr bursts identified in the kommers et al . ( 2001 ) search , available at http://space.mit.edu/batse/data.html . * a total of 9 of the untriggered events probably originated from soft gamma repeaters . in some cases , they had in fact triggered batse and were recorded in triggered mode . the ipn localizations of these sgr bursts serve as a good calibration of the techniques and data used here , however . they verify , for example , that the 1 s resolution batse data files in the stern et al . ( 2001 ) and kommers et al . ( 2001 ) catalogs have the correct timing , and that the localization procedures used in these catalogs , and by the ipn , are accurate . the following list gives the details . grb920903 , 05728 s. this burst was observed by watch - granat ( sazonov et al . both the ulysses / watch and ulysses / batse annuli are consistent with the watch error circle , but do not intersect the batse error circle . the batse error circle lies @xmath0 7 degrees away from the watch error circle , but is consistent with it , given the statistical and systematic uncertainties . the intersection of the narrower ulysses / batse annulus with the watch error circle is given here . grb920920 , 04415 s. this event was recorded in triggered mode following batse trigger 1948 . the trigger occurred due to a different grb . grb930209 , 15737 s. this event was recorded in triggered mode following batse trigger 2177 . the trigger occurred due to a solar flare . grb930702 , 68333 s. this burst corresponds to batse trigger 2426 , which is believed to be a cygnus x-1 fluctuation . grb931005 , 82288 s. this burst is a kommers et al . ( 2001 ) le event which was eliminated from the final catalog . it is batse trigger 2565 , from sgr1806 - 20 . grb940710 , 35477 s. this event occurred 251 s before batse trigger 3071 , whose duration is t90=71 s. the location of the kommers et al . ( 2001 ) event is ra , decl . = 99.4@xmath8 , -33.3@xmath8 , with uncertainty 3.4@xmath8 , and that of batse 3071 is ra , decl . = 96.42@xmath8 , -36.59@xmath8 , with uncertainty 1.33@xmath8 . the centers of the two circles are therefore 4.1@xmath8 apart . the ipn annulus is consistent with both error circles . thus the kommers et al . ( 2001 ) event may be a precursor to 3071 . grb950730 , 76147 s. this event corresponds to batse trigger 3720 , which was due to a cygnus x-1 fluctuation . the burst occurred 25 s after the trigger . grb950904 , 52777 s. this event occurred about 75 s after batse trigger 3776 . its duration is given as 108 s in stern et al . ( 2001 ) , but this duration included the triggered event , which is unrelated to it . the correct duration is @xmath0 30 s. grb961119 , 21322 s. this burst is a kommers et al . ( 2001 ) le event which was eliminated from the final catalog . the ipn annulus is consistent with the position of sgr1806 - 20 . grb961119 , 21536 s. this burst is a kommers et al . ( 2001 ) le event which was eliminated from the final catalog . it is probably from sgr1806 - 20 , but it can not be triangulated with any precision . grb961119 , 26961 s. this burst is a kommers et al . ( 2001 ) le event which was eliminated from the final catalog . it is probably from sgr1806 - 20 , but it can not be triangulated with any precision . grb980622 , 51085 s. this event is batse trigger 6861 . it originates from sgr1627 - 41 . grb980728 , 64911 s. this event is batse trigger 6954 . it originates from sgr1627 - 41 . grb980801 , 12920 s. this event is batse trigger 6959 . it originates from sgr1627 - 41 . grb980913 at 19983 s. this is batse trigger 7087 . grb981022 , 56447 s. this event is batse trigger 7171 , from sgr1900 + 14 . grb990110 , 31141 s. this is batse trigger 7315 . this burst originates from sgr1900 + 14 . grb990429 , 35555 s. this event is batse trigger 7536 . it originates from sgr1900 + 14 . grb990507 , 71334 s. this is batse trigger 7552 . grb991101 , 54480 s. this is batse trigger 7835 . it is probably a grb observed with particle contamination . the kommers et al . ( 2001 ) and stern et al . ( 2001 ) studies of untriggered batse bursts pointed to different conclusions about the grb population . the sample of stern et al . provides evidence for a grb number - intensity relation which continues to increase at low intensities , while the sample of kommers et al . provides evidence for a flattening . the analysis which we have presented here indicates only that many of the events with peak fluxes above @xmath00.15 photon @xmath1 s@xmath2 are likely to be real , and that relatively few of them have been misclassified . the likelihood of reality increases with peak flux ( figure 1 ) . as there are hundreds of untriggered bursts below the ipn threshold , the possibility exists that the different conclusions about the number - intensity relation are due to the differences in classifying weak untriggered events . a recent study of untriggered batse bursts by mitrofanov et al . ( 2004 ) reinforces and quantifies this idea . while this study is a preliminary one and does not draw any conclusions about the weak events , it should eventually lead to a clearer classification of them . a definitive statement about the weak burst population may also be forthcoming after the launch of the swift mission ( gehrels et al . support for the _ ulysses grb experiment is provided by jpl contract 958056 . joint analysis of _ ulysses and batse data was supported by nasa grants nag 5 - 1560 and nag5 - 9701 . near data analysis was supported under nasa grants nag 5 - 3500 and nag 5 - 9503 . we are also grateful to the near team for their modifications to the xgrs experiment which made gamma - ray burst detection possible . the konus - wind experiment was supported by russian space agency contract and rbrf grant # 03 - 02 - 17517 . _ _ 910601 & 62220 & _ ulysses + 910830 & 05148 & _ ulysses + 910908 & 33924 & _ ulysses + 910910 & 14747 & _ ulysses + 911029 & 17453 & _ ulysses , pvo , _ ginga + 911120 & 43957 & _ ulysses + 920109 & 23306 & _ ulysses + 920216 & 58688 & _ ulysses + 920303 & 24523 & _ ulysses + 920622 & 23828 & _ ulysses + 920626 & 64276 & _ ulysses + 920717 & 57852 & _ ulysses + 920903 & 05728 & _ ulysses , watch / granat + 930118 & 64426 & _ ulysses + 930209 & 15737 & _ ulysses , phebus + 930408 & 06847 & phebus + 930424 & 38888 & _ ulysses + 930506 & 55245 & _ ulysses + 930626 & 07023 & _ ulysses + 930710 & 13810 & _ ulysses + 930909 & 45100 & _ ulysses + 940213 & 07260 & _ ulysses + 940222 & 08083 & _ ulysses + 940311 & 44967 & _ ulysses + 940710 & 35477 & _ ulysses + 940712 & 00070 & _ ulysses + 940727 & 40865 & _ ulysses + 940730 & 39690 & _ ulysses , dmsp , sross - c + 940930 & 23017 & _ ulysses + 941104 & 35178 & _ ulysses + 950104 & 32438 & konus-_wind + 950111 & 46528 & _ ulysses + 950131 & 78592 & _ ulysses , konus-_wind + 950203 & 08456 & _ ulysses , sross - c + 950207 & 72568 & konus-_wind + 950211 & 15919 & _ ulysses , konus-_wind + 950224 & 33800 & konus-_wind + 950603 & 21257 & konus-_wind + 950611 & 21122 & _ ulysses + 950614 & 00779 & konus-_wind + 950615 & 12104 & _ ulysses + 950622 & 71470 & _ ulysses , konus-_wind + 950625 & 09685 & konus-_wind + 950722 & 64127 & _ ulysses , konus-_wind + 950723 & 73608 & konus-_wind + 950728 & 45743 & konus-_wind + 950730 & 76147 & _ ulysses , konus-_wind + 950904 & 52777 & _ ulysses + 951001 & 41868 & _ ulysses , konus-_wind + 951005 & 14826 & konus-_wind + 951013 & 57096 & konus-_wind + 951112 & 67850 & _ ulysses + 951124 & 25132 & _ ulysses , konus-_wind + 951213 & 32675 & konus-_wind + 951215 & 73379 & konus-_wind + 951218 & 28745 & _ ulysses , konus-_wind + 951231 & 77068 & konus-_wind + 960107 & 68607 & konus-_wind + 960115 & 31956 & _ ulysses , konus-_wind + 960123 & 43643 & konus-_wind + 960201 & 82195 & _ ulysses + 960202 & 05968 & _ ulysses , konus-_wind + 960207 & 65033 & _ ulysses , konus-_wind + 960304 & 48776 & _ ulysses + 960321 & 19663 & konus-_wind + 960418 & 08267 & konus-_wind + 960504 & 18779 & konus-_wind + 960602 & 42667 & konus-_wind + 960603 & 60930 & konus-_wind + 960614 & 83621 & konus-_wind + 960715 & 58326 & konus-_wind + 960725 & 63535 & _ bepposax + 960817 & 24647 & konus-_wind + 960826 & 58072 & konus-_wind + 960905 & 02568 & konus-_wind + 961017 & 23648 & _ ulysses + 961023 & 07747 & _ bepposax + 961106 & 43031 & _ bepposax + 961107 & 12691 & konus-_wind + 961110 & 26976 & konus-_wind , _ bepposax + 961113 & 80523 & konus-_wind + 961119 & 21322 & _ ulysses , konus-_wind + 961119 & 21536 & konus - a + 961119 & 26961 & konus - a + 961120 & 30433 & _ bepposax + 961123 & 59316 & konus-_wind + 961208 & 68232 & _ bepposax + 961209 & 74677 & _ ulysses + 961213 & 49966 & _ ulysses , konus-_wind + 961222 & 43207 & _ bepposax + 961224 & 36648 & konus-_wind , _ bepposax + 970116 & 58238 & _ ulysses , konus-_wind + 970119 & 42607 & konus-_wind + 970221 & 13750 & _ bepposax + 970223 & 64885 & _ bepposax + 970311 & 30254 & _ bepposax + 970406 & 25471 & _ ulysses , konus-_wind , _ bepposax + 970525 & 31783 & _ bepposax + 970610 & 36151 & _ bepposax + 970617 & 61459 & _ bepposax + 970720 & 68515 & konus-_wind + 970801 & 29048 & _ ulysses , konus-_wind + 970817 & 69692 & _ bepposax + 970825 & 40632 & _ bepposax + 970827 & 25872 & _ bepposax + 970926 & 79655 & konus-_wind + 971015 & 20356 & konus-_wind + 971017 & 01897 & konus-_wind + 971019 & 57427 & konus-_wind + 971027 & 09808 & _ ulysses , bepposax + 971028 & 75126 & _ bepposax + 971101 & 23483 & _ ulysses , konus-_wind + 971102 & 05581 & _ bepposax + 971103 & 27090 & _ bepposax + 971121 & 43992 & _ ulysses , konus-_wind + 971207 & 67900 & _ ulysses , konus-_wind + 971207 & 75492 & _ ulysses , _ bepposax + 971228 & 53605 & _ bepposax + 971228 & 79012 & konus-_wind , near + 980106 & 44231 & konus-_wind + 980205 & 19783 & _ ulysses , konus-_wind , _ bepposax , near + 980207 & 58212 & konus-_wind + 980223 & 76640 & _ bepposax + 980226 & 41332 & _ bepposax + 980304 & 52863 & konus-_wind , _ bepposax + 980329 & 55486 & _ bepposax + 980429 & 20493 & konus-_wind + 980518 & 67488 & _ bepposax + 980520 & 52002 & _ bepposax + 980523 & 31208 & _ ulysses , konus-_wind + 980602 & 46528 & konus-_wind + 980605 & 51131 & konus-_wind , _ bepposax + 980613 & 17465 & _ bepposax + 980622 & 51085 & _ ulysses + 980626 & 70184 & konus-_wind + 980629 & 32377 & _ ulysses , konus-_wind + 980705 & 23165 & _ bepposax + 980706 & 63987 & konus-_wind , _ bepposax + 980709 & 16963 & _ bepposax + 980712 & 18577 & _ bepposax + 980713 & 13301 & konus-_wind , _ bepposax + 980715 & 35282 & _ bepposax + 980728 & 53879 & konus-_wind + 980728 & 55355 & konus-_wind + 980808 & 78791 & _ bepposax + 980810 & 15944 & _ bepposax + 980812 & 17640 & _ bepposax + 980812 & 18950 & _ ulysses , konus-_wind , _ bepposax + 980907 & 40388 & _ bepposax + 980908 & 02480 & _ bepposax + 980913 & 19983 & _ ulysses , near + 980916 & 73322 & _ bepposax + 980917 & 35279 & _ bepposax + 980923 & 30178 & _ bepposax + 981002 & 05466 & _ bepposax + 981018 & 01612 & _ bepposax + 981019 & 79603 & _ ulysses , konus-_wind , _ bepposax + 981022 & 21682 & _ bepposax + 981022 & 56447 & _ ulysses , konus-_wind + 981101 & 26940 & _ ulysses , konus-_wind , near + 981106 & 38479 & _ bepposax + 981215 & 80709 & konus-_wind , near + 981216 & 19755 & _ bepposax + 990104 & 39597 & _ bepposax + 990109 & 41054 & _ ulysses , konus-_wind + 990110 & 31141 & _ ulysses + 990128 & 37252 & _ ulysses , konus-_wind , _ bepposax + 990204 & 30169 & _ ulysses , konus-_wind + 990305 & 34451 & konus-_wind + 990421 & 65775 & _ ulysses + 990504 & 40929 & _ bepposax + 990509 & 74345 & _ ulysses + 990526 & 47273 & _ bepposax + 990603 & 66686 & _ bepposax + 990606 & 11124 & konus-_wind + 990618 & 37636 & _ bepposax + 990621 & 43943 & _ bepposax + 990705 & 57685 & _ ulysses , konus-_wind , _ bepposax , near + 990707 & 54801 & _ ulysses , konus-_wind , _ bepposax + 990711 & 49110 & _ bepposax + 990719 & 79380 & _ bepposax + 990720 & 00025 & _ bepposax + 990725 & 41016 & _ bepposax + 990727 & 48288 & _ bepposax + 990803 & 57565 & _ bepposax + 990806 & 60168 & konus-_wind + 990828 & 70019 & _ ulysses , konus-_wind + 990917 & 52494 & _ bepposax + 990917 & 71095 & _ bepposax + 990919 & 49338 & konus-_wind , _ bepposax , near + 990919 & 86038 & konus-_wind + 990926 & 32653 & near + 991002 & 15031 & _ bepposax + 991004 & 22825 & _ bepposax + 991005 & 15265 & _ ulysses + 991011 & 35968 & konus-_wind , _ bepposax + 991120 & 27069 & near + 991205 & 82651 & _ bepposax + 991217 & 21782 & _ bepposax + 000102 & 27709 & _ ulysses + 000205 & 45486 & _ ulysses , _ bepposax + 000206 & 09183 & _ ulysses , _ bepposax + 000210 & 14030 & konus-_wind + 000211 & 45217 & _ bepposax + 000224 & 82209 & _ bepposax + 000318 & 12931 & konus-_wind + 000403 & 13199 & _ ulysses , konus-_wind , near + 000405 & 77386 & _ ulysses + 000420 & 61374 & _ ulysses , konus-_wind + 000502 & 54060 & _ bepposax + 000511 & 66298 & _ ulysses , konus-_wind + _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ 910601 & 62220 & 307.008 & -20.544 & 31.555 & 0.602 + 910830 & 05148 & 152.409 & 12.566 & 48.690 & 0.306 + 910908 & 33924 & 154.638 & 11.730 & 84.755 & 0.092 + 910910 & 14747 & 155.049 & 11.574 & 19.634 & 0.072 + 911029 & 17453 & 344.478 & -7.880 & 29.352 & 0.050 + 911120 & 43957 & 167.141 & 6.825 & 86.994 & 0.049 + 920109 & 23306 & 347.750 & -6.711 & 47.721 & 0.173 + 920216 & 58688 & 342.389 & -8.336 & 32.244 & 0.135 + 920303 & 24523 & 338.881 & -8.709 & 21.076 & 0.083 + 920622 & 23828 & 150.168 & 6.668 & 48.411 & 0.135 + 920626 & 64276 & 330.525 & -6.378 & 65.379 & 0.045 + 920717 & 57852 & 152.612 & 4.858 & 77.394 & 0.050 + 920903 & 05728 & 338.736 & -0.545 & 54.228 & 0.063 + 930118 & 64426 & 163.857 & -13.874 & 54.969 & 0.035 + 930209 & 15737 & 159.547 & -14.829 & 70.983 & 0.050 + 930424 & 38888 & 144.212 & -12.262 & 78.603 & 0.549 + 930506 & 55245 & 323.364 & 11.717 & 66.872 & 0.992 + 930626 & 07023 & 325.198 & 11.491 & 67.667 & 0.374 + 930710 & 13810 & 326.814 & 12.090 & 82.054 & 0.024 + 930909 & 45100 & 336.355 & 17.978 & 87.746 & 0.038 + 940213 & 07260 & 151.385 & -51.418 & 39.877 & 0.189 + 940222 & 08083 & 146.406 & -52.260 & 86.434 & 0.118 + 940311 & 44967 & 136.417 & -52.192 & 85.580 & 0.439 + 940710 & 35477 & 128.502 & -39.575 & 25.746 & 0.750 + 940712 & 00070 & 128.902 & -39.677 & 44.911 & 0.058 + 940727 & 40865 & 313.160 & 41.197 & 47.661 & 0.206 + 940730 & 39690 & 134.062 & -41.599 & 35.990 & 0.072 + 940930 & 23017 & 341.171 & 58.806 & 65.096 & 0.510 + 941104 & 35178 & 26.375 & 73.600 & 57.376 & 0.187 + 950111 & 46528 & 322.648 & -44.901 & 66.079 & 0.119 + 950131 & 78592 & 331.984 & -31.567 & 29.273 & 0.527 + 950203 & 08456 & 332.839 & -30.167 & 89.992 & 0.030 + 950211 & 15919 & 335.834 & -25.077 & 30.200 & 0.130 + 950611 & 21122 & 196.445 & -57.824 & 26.966 & 1.033 + 950615 & 12104 & 198.715 & -60.874 & 56.631 & 0.518 + 950622 & 71470 & 204.112 & -66.756 & 68.801 & 0.141 + 950722 & 64127 & 98.551 & 83.396 & 57.905 & 0.379 + 950730 & 76147 & 320.422 & -82.230 & 77.858 & 0.039 + 950904 & 52777 & 190.636 & 67.767 & 36.463 & 0.144 + 951001 & 41868 & 202.666 & 60.024 & 80.559 & 0.068 + 951112 & 67850 & 216.222 & 54.930 & 39.289 & 0.202 + 951124 & 25132 & 219.201 & 55.029 & 38.194 & 1.044 + 951218 & 28745 & 44.049 & -57.317 & 39.343 & 0.089 + 960115 & 31956 & 45.389 & -63.277 & 45.746 & 0.090 + 960201 & 82195 & 41.130 & -68.152 & 88.655 & 0.146 + 960202 & 05968 & 221.086 & 68.185 & 49.051 & 0.164 + 960207 & 65033 & 218.206 & 69.749 & 73.961 & 0.089 + 960304 & 48776 & 192.220 & 74.434 & 75.937 & 0.404 + 961017 & 23648 & 174.945 & 32.043 & 15.768 & 0.302 + 961209 & 74677 & 179.842 & 31.294 & 2.542 & 5.564 + 961213 & 49966 & 179.845 & 31.446 & 24.486 & 0.129 + 970116 & 58238 & 176.897 & 33.808 & 84.760 & 0.143 + 970406 & 25471 & 335.943 & -35.323 & 42.658 & 0.277 + 970801 & 29048 & 337.672 & -22.111 & 38.624 & 0.222 + 971027 & 09808 & 169.177 & 13.952 & 58.403 & 0.013 + 971101 & 23483 & 349.656 & -13.606 & 27.793 & 0.014 + 971121 & 43992 & 351.033 & -12.487 & 73.989 & 0.115 + 971207 & 67900 & 171.412 & 11.889 & 75.290 & 0.226 + 971207 & 75492 & 171.412 & 11.887 & 62.197 & 0.051 + 971228 & 79012 & 83.395 & 20.899 & 81.666 & 2.305 + 980205 & 19783 & 165.363 & 12.132 & 53.914 & 0.040 + & & 183.317 & -72.020 & 38.790 & 2.212 + 980523 & 31208 & 329.478 & -11.682 & 36.270 & 0.183 + 980622 & 51085 & 330.752 & -9.840 & 77.221 & 0.029 + 980629 & 32377 & 331.305 & -9.351 & 80.020 & 0.019 + 980812 & 18950 & 336.276 & -5.670 & 38.173 & 0.061 + 980913 & 19983 & 340.559 & -2.546 & 31.980 & 0.278 + & & 67.928 & 25.235 & 56.955 & 0.483 + 981019 & 79603 & 344.985 & 1.273 & 45.039 & 0.029 + 981022 & 56447 & 345.260 & 1.560 & 58.633 & 0.012 + 981101 & 26940 & 346.143 & 2.571 & 61.551 & 0.180 + & & 275.722 & -23.305 & 44.973 & 0.085 + 981215 & 80709 & 301.383 & -18.295 & 16.297 & 0.223 + 990109 & 41054 & 345.757 & 8.751 & 60.734 & 0.016 + 990110 & 31141 & 345.649 & 8.805 & 58.037 & 0.004 + 990128 & 37252 & 342.818 & 9.649 & 62.261 & 0.007 + 990204 & 30169 & 161.450 & -9.832 & 73.846 & 0.169 + 990421 & 65775 & 146.070 & -7.792 & 55.096 & 0.026 + 990509 & 74345 & 144.820 & -7.244 & 81.947 & 0.375 + 990705 & 57685 & 147.515 & -8.031 & 76.306 & 0.004 + & & 167.925 & -19.482 & 71.632 & 0.008 + 990707 & 54801 & 147.737 & -8.132 & 56.409 & 0.010 + 990828 & 70019 & 155.115 & -12.504 & 67.522 & 0.024 + 990919 & 49338 & 149.773 & 12.594 & 74.222 & 0.082 + 990926 & 32653 & 335.529 & -9.725 & 14.580 & 1.019 + 991005 & 15265 & 340.866 & 17.520 & 18.198 & 0.110 + 991120 & 27069 & 198.957 & -13.840 & 45.024 & 0.095 + 000102 & 27709 & 165.147 & -34.791 & 51.204 & 0.236 + 000205 & 45486 & 156.800 & -40.955 & 70.209 & 0.128 + 000206 & 09183 & 336.592 & 41.031 & 51.005 & 0.395 + 000403 & 13199 & 314.499 & 40.072 & 63.855 & 0.040 + & & 308.246 & 19.750 & 46.194 & 0.167 + 000405 & 77386 & 133.665 & -39.692 & 57.653 & 0.304 + 000420 & 61374 & 310.666 & 37.563 & 86.354 & 0.181 + 000511 & 66298 & 309.117 & 34.756 & 75.388 & 0.712 + 910601 & 62220 & 297.900 & 8.300 & 2.330 + & & 296.058 & 9.756 & + & & 298.589 & 10.529 & + & & 295.545 & 8.284 & + & & 299.858 & 9.600 & + 910830 & 5148 & 202.100 & 14.700 & 2.130 + & & 202.699 & 12.651 & + & & 202.991 & 16.650 & + & & 202.060 & 12.570 & + & & 202.358 & 16.815 & + 910908 & 33924 & 230.800 & -32.000 & 2.260 + & & 230.053 & -34.172 & + & & 231.638 & -29.857 & + & & 229.842 & -34.113 & + & & 231.434 & -29.806 & + 910910 & 14747 & 150.700 & 30.300 & 2.000 + & & 148.384 & 30.285 & + & & 152.795 & 31.170 & + & & 148.393 & 30.134 & + & & 152.864 & 31.031 & + 911029 & 17453 & 11.300 & -18.700 & 2.130 + & & 11.838 & -20.769 & + & & 13.048 & -17.368 & + & & 11.712 & -20.794 & + & & 12.966 & -17.277 & + 911120 & 43957 & 74.300 & 35.100 & 3.050 + & & 76.296 & 32.540 & + & & 75.564 & 37.976 & + & & 76.405 & 32.601 & + & & 75.694 & 37.937 & + 920109 & 23306 & 345.500 & 33.800 & 11.410 + & & 332.965 & 39.182 & + & & 357.208 & 40.384 & + & & 332.726 & 38.745 & + & & 357.521 & 39.973 & + 920216 & 58688 & 322.900 & 18.200 & 2.260 + & & 321.367 & 16.478 & + & & 325.055 & 19.170 & + & & 321.597 & 16.313 & + & & 325.162 & 18.913 & + 920303 & 24523 & 318.400 & -10.900 & 4.220 + & & 318.255 & -15.118 & + & & 317.616 & -6.752 & + & & 318.434 & -15.120 & + & & 317.788 & -6.724 & + 920622 & 23828 & 147.100 & -39.400 & 2.720 + & & 148.636 & -41.858 & + & & 145.100 & -41.656 & + & & 149.203 & -41.600 & + & & 144.594 & -41.337 & + 920626 & 64276 & 1.000 & 50.900 & 2.060 + & & 0.000 & 0.000 & + & & 0.000 & 0.000 & + & & 0.000 & 0.000 & + & & 0.000 & 0.000 & + 920717 & 57852 & 159.300 & -71.600 & 2.000 + & & 153.631 & -72.582 & + & & 165.497 & -72.133 & + & & 153.477 & -72.483 & + & & 165.564 & -72.026 & + 920903 & 05728 & 299.100 & 28.800 & 1.890 + & & 295.016 & 35.387 & + & & 295.583 & 36.120 & + & & 295.067 & 35.214 & + & & 295.799 & 36.158 & + 930118 & 64426 & 219.200 & -32.600 & 2.560 + & & 220.633 & -34.866 & + & & 221.151 & -30.652 & + & & 220.541 & -34.904 & + & & 221.075 & -30.599 & + 930209 & 15737 & 239.300 & -58.200 & 1.750 + & & 237.240 & -59.590 & + & & 237.460 & -56.756 & + & & 237.051 & -59.508 & + & & 237.272 & -56.830 & + 930424 & 38888 & 230.100 & -57.600 & 2.260 + & & 233.507 & -58.979 & + & & 232.932 & -55.956 & + & & 231.473 & -59.744 & + & & 230.887 & -55.382 & + 930506 & 55245 & 259.300 & 34.400 & 11.710 + & & 252.549 & 24.267 & + & & 253.061 & 45.092 & + & & 254.773 & 23.377 & + & & 256.039 & 45.844 & + 930626 & 07023 & 21.800 & -40.200 & 10.320 + & & 8.784 & -43.846 & + & & 21.353 & -29.886 & + & & 8.494 & -42.978 & + & & 20.326 & -29.950 & + 930710 & 13810 & 316.500 & -69.200 & 2.060 + & & 322.028 & -69.921 & + & & 310.697 & -69.198 & + & & 322.071 & -69.873 & + & & 310.705 & -69.149 & + 930909 & 45100 & 244.000 & 9.400 & 2.000 + & & 245.880 & 8.658 & + & & 245.084 & 11.092 & + & & 245.923 & 8.773 & + & & 245.187 & 11.023 & + 940213 & 07260 & 191.900 & -28.100 & 2.130 + & & 194.111 & -27.262 & + & & 192.498 & -26.038 & + & & 194.303 & -27.915 & + & & 191.751 & -25.974 & + 940222 & 8083 & 200.200 & 22.500 & 4.400 + & & 203.083 & 19.022 & + & & 195.460 & 23.007 & + & & 202.862 & 18.872 & + & & 195.441 & 22.750 & + 940311 & 44967 & 105.200 & 26.400 & 2.130 + & & 0.000 & 0.000 & + & & 0.000 & 0.000 & + & & 103.541 & 27.936 & + & & 105.460 & 28.518 & + 940710 & 35477 & 99.400 & -33.300 & 3.760 + & & 95.439 & -35.148 & + & & 98.463 & -29.626 & + & & 96.866 & -36.433 & + & & 100.452 & -29.648 & + 940712 & 00070 & 66.700 & -73.100 & 2.800 + & & 66.982 & -75.899 & + & & 63.191 & -70.522 & + & & 67.472 & -75.892 & + & & 63.527 & -70.480 & + 940727 & 40865 & 312.500 & -1.900 & 2.260 + & & 0.000 & 0.000 & + & & 0.000 & 0.000 & + & & 0.000 & 0.000 & + & & 0.000 & 0.000 & + 940730 & 39690 & 79.700 & -61.400 & 1.750 + & & 83.263 & -61.843 & + & & 82.831 & -60.531 & + & & 0.000 & 0.000 & + & & 0.000 & 0.000 & + 940930 & 23017 & 78.700 & 32.000 & 3.760 + & & 74.499 & 30.863 & + & & 81.554 & 34.911 & + & & 74.267 & 31.968 & + & & 80.379 & 35.492 & + 941104 & 35178 & 345.300 & 18.500 & 5.920 + & & 351.494 & 19.349 & + & & 340.023 & 21.742 & + & & 351.427 & 19.743 & + & & 340.265 & 22.072 & + 950111 & 46528 & 24.300 & -5.800 & 1.790 + & & 25.708 & -6.916 & + & & 23.085 & -4.481 & + & & 25.546 & -7.092 & + & & 22.919 & -4.654 & + 950131 & 78592 & 319.400 & -58.800 & 6.790 + & & 332.034 & -61.367 & + & & 307.773 & -56.126 & + & & 332.489 & -60.312 & + & & 308.789 & -55.212 & + 950203 & 08456 & 274.600 & 42.700 & 2.000 + & & 273.162 & 41.011 & + & & 276.802 & 43.897 & + & & 273.232 & 40.979 & + & & 276.851 & 43.846 & + 950211 & 15919 & 325.600 & 3.300 & 1.750 + & & 323.883 & 2.949 & + & & 327.178 & 4.064 & + & & 323.955 & 2.698 & + & & 327.274 & 3.822 & + 950611 & 21122 & 269.800 & -69.500 & 2.480 + & & 0.000 & 0.000 & + & & 0.000 & 0.000 & + & & 0.000 & 0.000 & + & & 0.000 & 0.000 & + 950615 & 12104 & 310.200 & -48.500 & 6.110 + & & 317.991 & -52.050 & + & & 302.386 & -45.503 & + & & 316.943 & -52.875 & + & & 301.692 & -46.428 & + 950622 & 71470 & 213.400 & -1.500 & 4.680 + & & 216.922 & 1.582 & + & & 210.342 & 2.043 & + & & 217.169 & 1.275 & + & & 210.055 & 1.773 & + 950722 & 64127 & 74.200 & 25.800 & 3.580 + & & 78.163 & 25.551 & + & & 70.224 & 25.950 & + & & 78.144 & 26.310 & + & & 70.337 & 26.703 & + 950730 & 76147 & 218.300 & -13.400 & 2.193 + & & 216.499 & -14.090 & + & & 220.702 & -13.533 & + & & 216.511 & -14.168 & + & & 220.711 & -13.611 & + 950904 & 52777 & 197.100 & 33.300 & 2.190 + & & 196.048 & 31.299 & + & & 198.492 & 31.452 & + & & 195.522 & 31.561 & + & & 198.968 & 31.778 & + 951001 & 41868 & 127.100 & -0.300 & 2.970 + & & 129.732 & 1.076 & + & & 126.889 & 2.663 & + & & 129.600 & 1.304 & + & & 127.153 & 2.670 & + 951112 & 67850 & 225.700 & 16.900 & 1.940 + & & 223.970 & 15.896 & + & & 227.667 & 16.437 & + & & 223.782 & 16.281 & + & & 227.727 & 16.858 & + 951124 & 25132 & 157.300 & 47.600 & 2.190 + & & 157.526 & 45.416 & + & & 154.289 & 48.462 & + & & 160.092 & 46.514 & + & & 156.796 & 49.765 & + 951218 & 28745 & 20.700 & -20.700 & 2.000 + & & 19.312 & -22.227 & + & & 22.801 & -21.085 & + & & 19.491 & -22.354 & + & & 22.748 & -21.288 & + 960115 & 31956 & 132.300 & -50.200 & 4.030 + & & 137.695 & -52.409 & + & & 127.935 & -47.375 & + & & 137.535 & -52.562 & + & & 127.746 & -47.504 & + 960201 & 82195 & 170.700 & -15.600 & 6.400 + & & 177.126 & -17.328 & + & & 164.395 & -13.665 & + & & 177.042 & -17.609 & + & & 164.306 & -13.944 & + 960202 & 05968 & 104.400 & 54.600 & 2.060 + & & 107.037 & 53.246 & + & & 100.892 & 54.993 & + & & 107.379 & 53.511 & + & & 101.036 & 55.318 & + 960207 & 65033 & 9.800 & 35.800 & 2.260 + & & 8.257 & 33.928 & + & & 12.029 & 34.464 & + & & 8.022 & 34.072 & + & & 12.193 & 34.665 & + 960304 & 48776 & 72.800 & 23.400 & 3.050 + & & 75.184 & 21.292 & + & & 69.613 & 22.567 & + & & 75.734 & 21.995 & + & & 69.476 & 23.427 & + 961017 & 23648 & 162.800 & 42.400 & 4.030 + & & 157.992 & 40.590 & + & & 164.912 & 46.136 & + & & 158.495 & 40.001 & + & & 165.805 & 45.805 & + 961209 & 74677 & 188.600 & 33.400 & 5.630 + & & 188.241 & 27.778 & + & & 184.992 & 38.211 & + & & 183.130 & 30.222 & + & & 181.856 & 33.794 & + 961213 & 49966 & 186.300 & 57.300 & 2.720 + & & 189.274 & 55.140 & + & & 181.933 & 56.019 & + & & 188.793 & 54.961 & + & & 182.254 & 55.744 & + 970116 & 58238 & 181.300 & -49.300 & 2.560 + & & 178.432 & -51.084 & + & & 184.499 & -50.830 & + & & 178.071 & -50.803 & + & & 184.795 & -50.520 & + 970406 & 25471 & 287.300 & -28.200 & 4.120 + & & 284.199 & -31.319 & + & & 287.704 & -24.096 & + & & 284.742 & -31.673 & + & & 288.335 & -24.186 & + 970801 & 29048 & 311.600 & -54.200 & 2.260 + & & 312.760 & -56.361 & + & & 307.838 & -53.740 & + & & 313.585 & -56.156 & + & & 308.119 & -53.267 & + 971027 & 09808 & 194.600 & 67.500 & 1.630 + & & 0.000 & 0.000 & + & & 0.000 & 0.000 & + & & 0.000 & 0.000 & + & & 0.000 & 0.000 & + 971101 & 23483 & 359.000 & 13.100 & 1.840 + & & 0.531 & 12.026 & + & & 357.113 & 13.202 & + & & 0.513 & 12.003 & + & & 357.112 & 13.173 & + 971121 & 43992 & 288.000 & -71.000 & 3.310 + & & 0.000 & 0.000 & + & & 0.000 & 0.000 & + & & 0.000 & 0.000 & + & & 0.000 & 0.000 & + 971207 & 67900 & 91.200 & 56.000 & 5.150 + & & 89.818 & 50.916 & + & & 89.893 & 61.105 & + & & 90.537 & 50.865 & + & & 90.833 & 61.146 & + 971207 & 75492 & 131.400 & 71.100 & 5.540 + & & 125.571 & 65.980 & + & & 148.573 & 72.644 & + & & 125.820 & 65.945 & + & & 148.619 & 72.537 & + 971228 & 79012 & 50.600 & -58.100 & 3.580 + & & 56.348 & -60.131 & + & & 44.681 & -56.488 & + & & 0.000 & 0.000 & + & & 0.000 & 0.000 & + 980205 & 19783 & 146.400 & -37.400 & 2.449 + & & 141.408 & -36.889 & + & & 149.002 & -39.630 & + & & 141.537 & -36.855 & + & & 149.169 & -39.591 & + 980523 & 31208 & 336.100 & -50.600 & 5.440 + & & 328.652 & -48.129 & + & & 341.734 & -46.631 & + & & 328.989 & -47.767 & + & & 341.252 & -46.362 & + 980622 & 51085 & 247.000 & -49.900 & 5.350 + & & 248.967 & -55.115 & + & & 249.002 & -44.724 & + & & 249.068 & -55.100 & + & & 249.083 & -44.739 & + 980629 & 32377 & 271.700 & 53.600 & 1.840 + & & 270.855 & 51.833 & + & & 273.931 & 54.899 & + & & 270.917 & 51.822 & + & & 273.978 & 54.870 & + 980812 & 18950 & 337.300 & 33.900 & 3.140 + & & 340.558 & 32.346 & + & & 333.941 & 32.499 & + & & 340.474 & 32.231 & + & & 334.018 & 32.381 & + 980913 & 19983 & 10.800 & 6.500 & 3.580 + & & 12.306 & 3.250 & + & & 10.380 & 10.056 & + & & 11.775 & 3.054 & + & & 9.818 & 9.945 & + 981019 & 79603 & 318.700 & -36.700 & 3.220 + & & 322.038 & -38.539 & + & & 315.772 & -34.531 & + & & 322.079 & -38.490 & + & & 315.821 & -34.489 & + 981022 & 56447 & 287.900 & 8.000 & 6.110 + & & 286.585 & 2.032 & + & & 287.253 & 14.077 & + & & 286.609 & 2.027 & + & & 287.278 & 14.079 & + 981101 & 26940 & 290.700 & 18.400 & 2.260 + & & 285.480 & 20.739 & + & & 285.453 & 20.570 & + & & 285.856 & 20.658 & + & & 285.829 & 20.489 & + 981215 & 80709 & 298.000 & -2.800 & 2.720 + & & 295.278 & -2.895 & + & & 300.530 & -1.796 & + & & 295.333 & -3.352 & + & & 300.663 & -2.236 & + 990109 & 41054 & 319.500 & -49.200 & 1.940 + & & 322.111 & -48.305 & + & & 319.282 & -47.265 & + & & 322.068 & -48.254 & + & & 319.369 & -47.262 & + 990110 & 31141 & 287.100 & 9.700 & 3.223 + & & 287.081 & 6.477 & + & & 286.582 & 12.883 & + & & 287.088 & 6.477 & + & & 286.589 & 12.884 & + 990128 & 37252 & 304.900 & -41.900 & 1.680 + & & 306.488 & -43.105 & + & & 303.203 & -40.804 & + & & 306.503 & -43.094 & + & & 303.216 & -40.793 & + 990204 & 30169 & 131.600 & 63.200 & 3.490 + & & 130.012 & 59.793 & + & & 138.763 & 62.040 & + & & 131.050 & 59.720 & + & & 138.262 & 61.569 & + 990421 & 65775 & 93.000 & -44.000 & 2.640 + & & 0.000 & 0.000 & + & & 0.000 & 0.000 & + & & 0.000 & 0.000 & + & & 0.000 & 0.000 & + 990509 & 74345 & 234.000 & -25.300 & 7.280 + & & 230.251 & -31.794 & + & & 229.206 & -19.522 & + & & 229.333 & -31.312 & + & & 228.459 & -20.114 & + 990705 & 57685 & 79.100 & -72.300 & 1.750 + & & 77.451 & -72.112 & + & & 77.488 & -72.150 & + & & 77.460 & -72.090 & + & & 77.497 & -72.127 & + 990707 & 54801 & 102.700 & -53.000 & 1.840 + & & 105.730 & -53.289 & + & & 103.292 & -51.196 & + & & 105.736 & -53.261 & + & & 103.336 & -51.202 & + 990828 & 70019 & 221.000 & -66.600 & 2.560 + & & 215.426 & -67.990 & + & & 219.021 & -64.176 & + & & 215.346 & -67.937 & + & & 218.889 & -64.195 & + 990919 & 49338 & 69.400 & 74.000 & 2.720 + & & 71.601 & 71.360 & + & & 74.556 & 76.382 & + & & 72.134 & 71.404 & + & & 75.176 & 76.285 & + 990926 & 32653 & 350.600 & -6.100 & 3.760 + & & 351.358 & -9.784 & + & & 349.452 & -2.518 & + & & 349.287 & -9.628 & + & & 347.741 & -3.646 & + 991005 & 15265 & 329.000 & 37.900 & 1.840 + & & 0.000 & 0.000 & + & & 0.000 & 0.000 & + & & 0.000 & 0.000 & + & & 0.000 & 0.000 & + 991120 & 27069 & 153.400 & -28.700 & 2.060 + & & 153.311 & -30.759 & + & & 152.564 & -26.777 & + & & 153.536 & -30.757 & + & & 152.770 & -26.717 & + 000102 & 27709 & 213.200 & -13.900 & 2.720 + & & 0.000 & 0.000 & + & & 0.000 & 0.000 & + & & 0.000 & 0.000 & + & & 0.000 & 0.000 & + 000205 & 45486 & 203.000 & 14.600 & 2.000 + & & 204.808 & 13.637 & + & & 201.462 & 15.941 & + & & 204.665 & 13.421 & + & & 201.297 & 15.740 & + 000206 & 09183 & 28.100 & 18.200 & 2.130 + & & 29.289 & 16.398 & + & & 30.195 & 17.451 & + & & 27.932 & 16.076 & + & & 30.257 & 18.792 & + 000403 & 13199 & 275.100 & -10.700 & 1.710 + & & 272.977 & -10.966 & + & & 274.144 & -11.737 & + & & 272.758 & -10.720 & + & & 273.909 & -11.487 & + 000405 & 77386 & 226.900 & -52.500 & 2.640 + & & 0.000 & 0.000 & + & & 0.000 & 0.000 & + & & 0.000 & 0.000 & + & & 0.000 & 0.000 & + 000420 & 61374 & 104.100 & 54.200 & 1.840 + & & 102.483 & 52.632 & + & & 107.083 & 53.652 & + & & 101.930 & 52.887 & + & & 107.232 & 54.064 & + 000511 & 66298 & 48.700 & 38.000 & 2.260 + & & 47.337 & 36.019 & + & & 50.491 & 39.779 & + & & 46.089 & 37.093 & + & & 48.679 & 40.260 & +

we study the quantum - mechanical evolution of the nonrelativistic oscillator , rapidly moving in the media with the random vector fields . we calculate the evolution of the level probability distribution as a function of time , and obtain rapid level diffusion over the energy levels . our results imply a new mechanism of charmonium dissociation in qcd media .

hep-ph0201235

the propagation of the charged particles , both quantum and classical , through different types of media had long ago become an important branch of theoretical physics . ( see e.g. an excellent review @xcite for the discussion on the present status of the subject ) . however , much less is known about the propagation of neutral bound systems and the wave packets with the zero total charge but nonzero higher multipole moments . this is especially true for quantum systems . recently there was a revived interest in the subject . this is due to numerous possible applications in nuclear and particle physics , especially in the study of exclusive processes and the creation and the diagnostics of the quark - gluon plasma @xcite . such systems can be also studied in electrodynamics , where it is possible to observe the propagation of @xmath0 positronium through the random media . although a significant advance in the study of the propagation of these objects was made recently , very little is known about the evolution of the fast wave the packets and the dipoles in the random media , and it s influence on the particle and nuclear cross - sections , diffractive processes and energy losses . the aim of this paper is to study the propagation of the highly energetic , but nonrelativistic ( in it s c.m . rest reference frame ) electrically neutral dipoles and wave packets , with nonzero dipole moment , and with the internal interactions , through the random media . for simplicity , and in order to have an almost explicit solution we shall consider the simplest example of such kind @xmath1 a harmonic oscillator moving through the media with the random vector fields . the latter can be any media , that where there are random vector fields , i.e. the fields transforming as vectors under the lorentz group transformations . in nature these are electromagnetic and yang - mills fields . we shall call below any such media a vector media . the examples of the vector random media include amorphous solid state media @xcite , usual electromagnetic plasma @xcite , dense qcd media @xcite , and quark - gluon plasma @xcite . it is less clear what happens in the nuclear media or nuclei , where the fields are colorless and it is not obvious how they transform under lorentz transformations . in our paper we consider the nonrelativistic oscillator , whose center of mass moves with the relativistic speed through the random media , which is characterized by the static random electric fields . in our simple model , the particles inside the dipole interact via harmonic oscillator potential ( no coulomb interactions between them ) , while the interaction with the external field is electromagnetic . however we shall argue below that at least qualitatively our results also hold for the interaction with the qcd colored fields , and for the general law of interaction between the constituents of the dipole . our main result will be the development of the general formalism that will allow us to determine the density matrix , expansion rate and level diffusion for nonrelativistic harmonic oscillator moving with high velocity in the random media . we shall derive the effective action and the effective transition operator for the bound system propagation in the random media . this effective action will be used to study two closely connected problems : the propagation of the small initial size dipole / wave packet through the random media and of the oscillator in the ground state through the random media . we shall determine the energy level diffusion rate of the small wave packet for small and big times , as well as the probability that it occupies level n after time t. we shall also determine the expansion rate of the ground state and the probabilities that after time t the oscillator will stay in the ground state or will move into the state with the energy level n. taking into account that the electric fields transform as the components of the rank 2 tensor during the lorentz transformation , while the time as the component of the 4-vector we shall obtain for a certain interval of times the superdiffusion expansion law for the wave packet density radius : ^2=(a_0t^2+a_1t^3)/^2 [ super ] here @xmath2 describes the quantum mechanical expansion rate of the wave packet and@xmath3 is the coefficient depending on the properties of the media , @xmath4 is the lorentz factor and,@xmath5 is the radius of the dipole , and d is the inverse squared radius of the initial state ( for the ground state d=1 ) . thus , random media may strongly influence the cross - sections and energy losses . the physical reason for `` superdiffusion '' is the extraction of the resonant ( i.e. with oscillator proper frequency ) mode from the spectrum of the random electric field fluctuations . this superdiffusion effect is very similar to the one discussed recently in refs . @xcite in the context of the random walk of the classical particles in the time dependent random potential in statistical mechanics . our results imply a new mechanism for charmonium dissociation in vector media : it is excited to higher energy levels due to the scattering on random electric fields , and then dissociates into @xmath6 pairs @xcite . the calculation will be made in the eikonal approximation , i.e. we shall assume that the center of mass of the dipole moves along the straight line with the ultrarelativistic constant velocity @xmath7 , where @xmath8 is the speed of light . we shall neglect in this approximation both the deviations of the center of mass from the straight line , and the velocity changes due to the interaction with the random fields . the latter corrections will be taken into account elsewhere . we shall assume that the size of our dipole is small relative to the inhomogeneity scale of the media @xmath9 . as it was mentioned above , all our calculations will be carried in the simplified model of the dipole with the oscillator interaction between the constituents and electromagnetic interaction with the external field . we neglect the coulomb interaction between internal constituents of the dipole . we expect that our result will not change qualitatively if we generalize the external field to a color one , and take into account coulomb interaction between the constituents . we shall see that the influence of the random media on the oscillator has very transparent mechanical analogue in the classical problem of the excitation of the oscillator by the external resonance force @xcite . the random media obviously has the role of this external force . on the other hand previous investigations , using this analogy , correspond to the study of the proper oscillations of the oscillator . some of the aspects of the problem of the dipole propagation were considered before @xcite , @xcite,@xcite , but the latter authors did not take into account the influence of the media on the propagation of the dipole and it s expansion , and the change of it s quantum state during the propagation . they only considered the interaction free dipole , neglecting the influence of the media on the dipole structure and the internal interactions . although as we shall see the internal interactions do not lead to new qualitative effects ( as it was noted in ref . @xcite ) , the interaction with the media does play a major role if time is not too small . the only previous attempt to take into account the influence of the media on the propagation of the oscillator in the media was made in ref . @xcite , but in the different context . the latter authors considered not vector , but scalar media , in particular nuclei , and parameterized it s influence by taking into account absorption . the problem of harmonic oscillator propagation naturally arises in the context of the propagation of the bound systems like charmonium through the quark - gluon plasma . note that there are strictly speaking two different problems that can be discussed in the connection with the oscillator : one is the propagation of the two - particle bound state - dipole through the media , another is the propagation of the wave packet . in the approximation used in this paper we shall see that technically this is the same problem . the quantum dynamics of the bound state can be reduced to the quantum dynamics of the wave packet . so below , if not stated otherwise we shall not distinguish between them . however , beyond the approximation of the homogeneous electric field discussed below , these are two different problems , with different lagrangians . the paper is organized as follows . in the second chapter we shall develop the general formalism for the description of the quantum dipole moving through the random media . first , we shall derive the effective lagrangian for the propagating oscillator in this media , which we shall model by the usual plasma . we shall neglect the radiation . the discussion of this important phenomena will be considered in a separate paper . next we shall consider the simplified model , that leads to the same qualitative results . we shall show that in the eikonal approximation it is sufficient to consider the behavior of the oscillator in the time - dependent homogeneous electric field . we shall calculate using the functional integral method the effective evolution operator and the probability to find an oscillator at particular levels as a function of time . in the third chapter we shall find the averaged density matrix and study the expansion of the small wave packets and see the influence of the media on the wave packet expansion . in the fourth chapter we shall consider the probability evolution for the ground state wave packets and apply our results to the simplest model of charmonium . we summarize our main results in conclusion . some details of the calculations will be discussed in appendices . the most widespread model of the random media where the bound system can propagate is the conventional model of plasma @xcite . such model can be used qualitatively to describe the amorphous material . the media is represented by a random set of ions that generate electric potentials with the debye screening . v(r)=g(-r)/r [ potential ] here g is the electric charge . if we neglect in this model three particle correlations , the potential@xmath1potential correlation function in such system is ( @xcite ) l(r -r ) = < v(r)v(r)>=a^2/r -r[correlation ] a is the constant dependent on the properties of the media . we will assume that the correlations are gaussian , i.e.that all correlation functions can be expressed through the two - point correlation function ( [ correlation ] ) . this is the usual assumption in the statistical mechanics . more complicated systems with nongaussian correlations are usually highly nontractable . for the qcd media or the quark - gluon plasma the most wide - spread model is the model discussed in ref . this model is very similar to the conventional model of plasma , discussed above , except that instead of the conventional electric charges the particles have the color charges @xmath10 , that are the generators of the @xmath11 color group , while the interactions contain factors @xmath12 that are very similar to the charge products . if we take into account only the binary interactions , the plasma becomes color - diagonal , and we can neglect the color factors , that will not make qualitative influence on our results.the potentials and the correlation functions are the same as in eq . ( [ correlation ] ) ( except some numerical factors whose influence will be discussed in a separate publication ) . note that in terms of the language of the statistical mechanics for all cases ( plasma , qcd media , quark - gluon plasma ) our problem is the well known problem of the description of the propagation of the wave packet in the random gaussian media characterized by given correlation functions ( see appendix a for a more detailed discussion ) . consequently we consider in this paper all types of plasma as an electromagnetic plasma . consider now the fast moving neutral system of two particles ( dipole ) with a harmonic interaction between it s components in the center of mass reference frame : w(r - r)=nr - r^2 [ interpotential ] the system is moving in a random media discussed above with a velocity @xmath7 , where @xmath8 is the speed of light . for this ultrarelativistic speed we can use the eikonal approximation . we assume that the center of mass moves along the straight line with a constant velocity . radiation will be neglected . according to our assumption the oscillator is nonrelativistic , consequently we can separate the motion of two constituents of the dipole into the motion of the center of mass and the relative motion of the two particles ( see ref . @xcite and appendix b for detailed discussion ) : l = l_c.m . + l_n.r.[sum1 ] in the eq . ( [ sum1 ] ) the first term describes the free motion of the center of mass with the speed @xmath13 , while the second is the lagrangian of the relative nonrelativistic motion of the two components of the dipole in the c.m . reference frame . the latter lagrangian can be written for the case of the small dipole with the size @xmath14 , l_n.r.=m(u^2-_0 ^ 2 u^2)- g(v(r_1,t)-v(r_2,t ) ) [ nr ] here v is the electric potential in the oscillator c.m . frame , @xmath15 are the coordinates of the dipole components in this frame , and @xmath16 is the relative coordinate of two constituents of the dipole . the derivation of the lagrangian ( [ nr ] ) is discussed in appendix b. for the small size dipole we may use a taylor expansion : v(r_1,t)-v(r_2,t)=(r_1-r_2 ) [ tay ] here v(0,t ) is the electric potential in the c.m . reference frame at the moment t ( t here is the laboratory time ) that corresponds to the random potential in the point where the center of mass is located , it s derivative is obviously the electric field in this point . thus for the small size dipole , with the size @xmath17 , we can neglect inhomogeneity of the field and reduce the problem of the dipole motion to the problem of the oscillator in the homogeneous electric field e=- [ e1 ] ( measured in the oscillator c.m . frame in the point corresponding to the oscillator c.m . ) . in this approximation , and this is an approximation that we will use in the present paper , the problem of the moving oscillator is equivalent to the problem of the propagation of the wave packet in the electric field @xmath18 . for the larger dipole we must use the full second term in the lagrangian ( [ nr ] ) , taking into account the inhomogeneity of the field . the same separation can be made for the nonrelativistic neutral wave packet , whose center of mass moves with a relativistic speed . in this case the nonrelativistic lagrangian that describes the space - time evolution of the dipole in it s rest frame can be written as l_n.r.=m(^2-_0 ^ 2 r^2)- gv(r , t ) [ nr2 ] here @xmath19 is the distance between two particles in the rest frame of the dipole or the radius vector in the c.m . at rest reference frame for a wave packet . v is the potential in the wave packet c.m . reference frame . the lagrangian ( [ nr ] ) in the approximation ( [ tay ] ) is evidently a particular case of eq . ( [ nr2 ] ) . let us go to the rest frame of the dipole . in the latter frame our problem is evidently a problem of the nonrelativistic oscillator with the internal interaction given by eq . ( [ interpotential ] and in the external electromagnetic field . electromagnetic field acting on the particles that built up the oscillator can be easily found by making the lorentz transformation @xcite . ( note that within the present approximation , since the relative motion is nonrelativistic , the relative speeds of the two particles can be neglected to perform the lorentz transformation according to c.m . speed @xmath13 ) . after the lorentz transformation ( see appendix b for details ) , the laboratory frame 4-potential @xmath20 of the electromagnetic fields becomes @xmath21 : @xmath22 the electromagnetic tensor @xmath23 ( @xmath24 in the laboratory reference frame ) becomes : @xmath25 here @xmath26 is the lorentz factor : = 1/ [ gamma1 ] we see that the transverse electric field is greatly enhanced , while the longitudinal field remains the same . we also see that in addition to electric fields there arises a magnetic field equal to b = v_0e [ b ] fortunately , this magnetic field can be neglected . first , the magnetic field is of the same order as the electric one , but the lorentz force acting on the dipole is suppressed as @xmath27 , where @xmath28 is the relative speed of the dipole constituents , since we assume dipole to be a nonrelativistic system . second , the magnetic field is the field with very short period and high frequency @xmath29 . if we consider the classical motion in such periodic field , we shall come to the system of differential equations for the particle moving in combined oscillator - lorentz force field , where we can use the floque theorem . the floque theorem @xcite says that the system of the ordinary differential equations with periodic coefficients has the solution : @xmath30 where b is the periodic matrix function and a is the diagonal matrix , with entries @xmath31 , the coefficients @xmath32 are called the eigenvalues of the problem . the matrixes are @xmath33 and include both the 3 coordinates and its derivatives in time . due to the energy conservation law and the fact that the classical energy of the oscillator is @xmath34 , it is clear that all eigenvalues of the problem must be imaginary and appear in pairs ( since the solution is real ) . thus magnetic fields may contribute only to high frequency small fluctuations along the classical trajectory corresponding to the motion without the magnetic field . consequently , it s effect can be safely neglected . clearly the same is true for the quantum mechanical case , since the corresponding path integral is saturated by the classical trajectories . we have seen that the fields that act on the oscillator in it s rest frame are time - dependent . in the eikonal approximation the coordinate of the bound state mass center is @xmath35 where t is the proper time in the oscillator frame , and @xmath4 is a lorentz factor : t = t/[propertime ] using the time and frequency transformation and field transformation laws we can write the correlation function of the potentials in the oscillator rest frame . the correlation function will become time - dependent : = a^2 ^ 2(-q)/q [ q ] q= [ q1 ] in the latter equations x and y are the transverse coordinates that are the same in both reference frames , while z - s are the z - coordinates of the oscillator in the c.m . frame . due to the lorentz contraction , we can neglect z dependence and therefore assume @xmath36 ( this can be done in quantum mechanics , where we neglect radiation ) . recall that @xmath37 and @xmath38 are the proper time of the oscillator . thus we see that our problem of the moving oscillator wave packet has been reduced to the problem of the nonrelativistic oscillator at rest in the high frequency external electric field that is strong in the transverse plane , and it s longitudinal component has the same strength as in the media at rest , with the random field correlator ( [ q1 ] ) . in the small oscillator limit discussed above , the electric field can be considered as homogeneous , and we get the correlation function = a^2 ^ 2(-q)/q [ q ] q = v_0t - tt - t [ q3 ] here and below except in the @xmath4 factors we assume @xmath39 , and use the units where @xmath40 . we shall argue below that the results obtained in the homogeneous field approximation hold qualitatively even for the large scale oscillator . ( [ q1 ] ) holds also for the dipole , since in this case we work from the beginning in the homogeneous field approximation ( [ e1 ] ) . now we have to consider the problem of the nonrelativistic oscillator in the random media , with the correlation function of the scalar potentials defined by eq . ( [ q ] ) . this problem is very similar to the problem of the propagation of the electromagnetic wave in the random media characterized by the gaussian distribution function ( see e.g. ref . @xcite ) . our main goal is to find the level distribution of the oscillator at the moment of time t if at the 0 moment he was in the state characterized by the wave function @xmath41 , or equivalently by a set of coefficients @xmath42 : ( r,0)=a_n_n ( r ) [ expansion ] here @xmath43 are the wave functions of the stationary states of the pure oscillator without the random fields : _ n ( x)=(m_0/)^1/4(-m_0 x^2/2 ) [ u ] ( we shall choose the units where the planck constant @xmath44 ) . the probability that at the time t the oscillator will be at the level n is given by p(n , t)=<a_n(t)^2>[pr ] the average is over all field configurations . the simplest way to calculate the latter average for the oscillator is to calculate the average product of the green functions : f(x_0,y_0,x_1,y_1,t)=<g(x_0,x_1,t)g^*(y_0,y_1,t ) > [ product ] here g is the oscillator green function in the external field v , that describes time evolution of the oscillator in the given realization of the field v. the average is taken over all external fields v with a gaussian weight . once we know the function @xmath45 , the time evolution of the @xmath46 coefficients can be easily calculated : a_n(t)^2=d^3x_0d^3y_0d^3x_1d^3y_1f(x_0,y_0,x_1,y_1,t)(x)(y)_n(x)_n(y ) [ evolution ] here we assumed that at t = o the oscillator was in the state described by the wave function @xmath47 . one possible way to carry the calculations is to use the functional integrals method . the green function of the 3d quantum oscillator is well known and can be represented as ( see e.g. ref . @xcite,@xcite,@xcite and references therein ) : g(x , y , t)=dx(t)(i^t_0m(x^2-_0 ^ 2 x^2))ds [ greenosc1 ] here the integration is over all paths with x(t)=x_1,x(0)=x_0 [ no ] consider now the arbitrary wave packet . the full functional integral we are interested in is @xmath48 the operator k is just the inverse of the correlation function of the potentials v , and is determined by the properties of the media . the integration is over all paths with x(t)=x_1,y(t)=y_1;x(0)=x_0,y(0)=y_1 [ boundary ] note that in the eq . ( [ funct ] ) the integration over the oscillator paths is in minkowski space , with the usual minkowski time t , and i before the action . however , in the part of the functional integral connected with the integration over v there is no i before the action , and the integration is purely statistical one . this integration is analogous to euclidean space functional integral in the quantum field theory , and it is evident that it has the same nature as the coupling of oscillator to the finite temperature thermal bath . it is very hard to carry out the exact functional integration in eq . ( [ funct ] ) even though the corresponding functional integrals over v and x , y are gaussian . indeed , integration over v can be carried out . we can proceed in the standard way , as in the theory of the random fields , by looking for extremum v trajectory , and then substituting the result into the functional integral ( [ funct ] ) that will be now the integral only over x , y . in order to carry the integration , we write the term v(x(t),t ) ) in the action as @xmath49 . varying the action over v(x , t ) , we obtain : i(x - x(t)-i(y - y(t))=-^t_0 k(x , t;xt)v(x,t ) [ extr1 ] this is the equation for the extremal trajectory . the latter equation ( except the factor i due to the fact that we carry the average in the minkowski space ) looks like the potential created by two opposite charges moving over the trajectories x(t ) and y(t ) . the equation can be inverted giving v(x , t)=-i^t ds ( r(x , t;x(s),s)-r(x , t , y(s),s)/2 [ v ] here r is the correlation function ( [ q1 ] ) , that is inverse to the operator k. substituting this expression for v in the functional integral ( [ funct ] ) we get f(x , y , x,y,t)=d x(t)dy(t ) [ intout ] here the effective action a is @xmath50 note that the above action looks quite similar to the famous feynman action for the polaron @xcite . it describes two particles with the singular interaction potential , that includes both interaction with it s own trajectory and the interaction with the trajectory of the second particle - the situation quite similar to the problem of electrostatic dipole . it is possible to deal with the latter action using mean field approximation or perturbation theory , however for fast particle the problem can be significantly simplified . in order to simplify the problem further , first note that the main contribution into the action comes from the region @xmath51 , due to the exponential yukawa cutoff in g. in the area @xmath51 , we can expand @xmath52 then the radius in the yukawa correlation function [ q1 ] will have the form q= [ mun ] since the system is nonrelativistic , the second term in the brackets in eq . ( [ mun ] ) can be neglected , and the system reduces to the usual quantum mechanics , i.e. we can neglect both the longitudinal and the transverse self - motion . concerning the interaction with the trajectory of the second particle , the contribution is clearly maximal when x(t ) is close to y(t ) , and in this case also we can neglect the transverse motion . this means that for a relativistic particle in the first approximation we can neglect the space dependence of the electric field and assume that the system is in the homogeneous time dependent random electric field . in fact it was clear from the very beginning that we can take our electric field as homogeneous . indeed , if we look at the action of the electric field , we see that the term due to the space inhomogeneity contributes into the action as @xmath53 , relative to @xmath54 due to the time evolution , and thus can be neglected , even if the initial wave packet is not of the small radius . we shall see below , that the inhomogeneity of the electric field must be taken into account only at the times of the order of the minimum of two scales : @xmath55 , when the longitudinal expansion of the wave packet becomes significant@xmath1of the order of @xmath56 . thus our results can be applied even for the more general case when @xmath57 . concerning the small dipole , we deal with it in the approximation of the homogeneous electric field from the beginning , the field being equal to the field in the c.m . of the dipole . next , we need to deal with the coulomb singularity of the yukawa potential . we shall follow the line of ref . the contribution of the singular part of the interaction to the functional integral is zero : the probability of such configuration is zero . moreover , it is known that such singularity really does not appear due to quantum fluctuations , and it s appearance is due to the quasiclassical method of the path integral calculation , since in the classical case there can be fall on the center . consequently , if we are interested in the qualitative results , we can substitute the singular correlation function with a nonsingular one , omitting the coulomb part and dealing directly with the electric fields . as for the strength of the electric field we can take the strength of the field it is clear from eqs . ( [ nr]),([e1 ] ) that we have to take the field in the reference frame where the center of mass of the system is at rest . thus we come to the conclusion that both qualitatively in the leading eikonal approximation and quantatively we can describe our model by an oscillator coupled to the homogeneous time dependent random electric field with the correlation function in the longitudinal direction given by = b^2(-v_0t - s ) [ corl ] and in the transverse plane with the correlation function = ( b^2 ^ 2)(-v_0t - s)_ij [ cor ] here t and s denote the proper time ( in the moving oscillator frame of reference ) , and the frequency is increased greatly by @xmath4-a lorentz factor . the corresponding potential is @xmath58 , and the corresponding lagrange density in the action can be rewritten as s_a = a^2dt e(t)(^2+^2 ^ 2)e(t)dt [ action ] note that the simplifications we did are very similar to the feynman trick for polaron , where he substituted the complicated nonlinear action of a polaron by an action of the harmonic oscillator . now we shall calculate the 4-point function f in the above model . in order to average over the field configurations we shall use the action s = a^2((e)^2+^2 ^ 2e^2)dt [ action ] the constant a is determined using the relation with the correlation function of the oscillator : = b^2(1,^2 ) [ al ] ( the brackets depend on whether the field is in longitudinal or transverse direction ) . then for the longitudinal field we have : e_l^2(0)=b_l^2= 1/(a_l^2(2 ) ) [ kl1 ] for the transverse field we have@xcite : e_t^2(0)=b_t^2 ^ 2= 1/(a_l^2(2 ) ) [ kl ] here both electric fields @xmath59 and @xmath60 are the fields in the frame moving with the velocity @xmath7 where the harmonic oscillator is at rest , so that b is the @xmath61 , where index l means the laboratory system , and the time t here and below refers to the frame moving with the oscillator . hence for the longitudinal field a_l=1/(b_l^(1/2 ) ) [ rel1 ] for the transverse field : a_t=1/(b_t^(3/2 ) ) [ rel ] let us now repeat the derivation of the 4-point function g for the concrete example of the averaging weight @xmath62 . the derivation is slightly different from the previous chapter , since there we dealt with potential v and varied over it , while for the final simplified model we can deal directly with electric field and vary over it that simplifies the problem . due to the obvious factorization we can start from the one - dimensional problem . the whole calculation is carried in the rest frame of the moving oscillator . more explicitly , the functional integral we need to calculate is f(x , y , x,y,t)=dx(t ) dy(t ) de(t ) [ aver ] here s(x , y , e ) is the action s=^t_0 dt(x^2-_0 ^ 2 x^2-y^2+_0 ^ 2 y^2)-ex+ey [ first ] the statistical weight s(e ) is given by eq . ( [ action ] ) . the boundary conditions for x , y are evident . note that the first term in the functional integral ( [ aver ] ) is multiplied by i while the second is not . we first will carry the integration over the random fields , and then consider the effective action for x and y. the integral over the electric field e is gaussian and can be taken . we first need to vary the action s over the fields e and then substitute the result into the action . varying over e we get : i ( -x_i(t)+y_i(t))=a^2(^2-^2 ^ 2)e_i(t ) [ vary ] the latter equation can be immediately solved using the green function of the operator @xmath63 : ( ^2-^2)(-t - t)/(2)=(t - t ) [ green ] consequently , we write e_i(t)=-(i/(2a^2))^t_0)(-t - s)(x_i(s)-y_i(s))ds [ equation ] we now substitute the above expression for e into the action in order to get effective action for x and y. we have for the terms that previously contained the electric field s_2=-^t_0^t_0 dtds(-t - s ) ( x(t)-x(s ) ( y(t)-y(s))/(2a^2 ) [ final ] note that the internal integration over s in the action goes up to t and not t , like in the minkowski field case . the reason is that the averaging over field configurations takes part only after the concrete trajectory in x(t ) is realized . physically , we first calculate functional integral for given e realization , and then average over all e. we now have the effective action @xmath64 , where s_1=im(x^2-y^2-_0 x^2+_0 y^2)/2 [ ac1 ] the calculation of the functional integral proceeds in the standard way . we seek the solution of the classical equation of motion , giving the exponent , and then calculate the preexponential factor . let us start from the equations of motion . we get im(+_0 x)=-^t_0(-t - s ( x(s)-y(s))/(a^2 ) [ eq1 ] im(+_0 y)=-^t_0 ( -t - s ) ( x(s)-y(s)/(a^2 ) [ eq2 ] we see that the equations for x and y are identical , thus + _ 0 ^ 2 ( x - y)=0 [ minus ] the boundary conditions are : _ 1=x(t)-y(t)=x_1-y_1;_0= x(0)-y(0)=x_0-y_0 [ bound ] the corresponding solution is evidently x(t)-y(t)=(_0 ( _ 0 ( t - t))+_1_0t)/(t ) [ hom ] once we know x@xmath1y , we can substitute it to eq . ( [ eq2 ] ) and get the equation for y , which is just the equation for the oscillator under the action of the external imaginary force . @xmath65 note that the external force that appears in the r.h.s . of the eq . ( [ eq3 ] ) has three parts . two of them are exponentially suppressed and are different from zero only close to t=0 or t = t . however , the third term is not suppressed , and , moreover , is a resonance force applied to the harmonic oscillator with a frequency @xmath66 . this is of course what was to be expected , since the random field distribution contains all possible frequencies , including the resonant one . the solution for x for given boundary conditions will be a sum of the solution of the homogeneous equation plus the solution of the nonhomogeneous one : x(t)=^t_0(_0 ( s - t ) ) u(s)ds+x_0(_0 t)+b(_0 t ) [ u ] here u(s ) is the r.h.s . of eq . ( [ eq3 ] ) . the coefficient b(t ) is determined from the boundary conditions . the general solution of the nonhomogeneous equation is a sum of five terms : x(t)=x_a(t)+x_b(t)+x_c(t)+x_d(t)+x_e(t ) [ sum ] the first term is due to the resonant part of the external field and is given by x_a(t)=il(_1t(_0 t)-_0 t ( _ 0 ( t - t)))/_0 [ res ] the second part includes the terms that come from the integration of the resonant part that are not linearly enhanced as the term above : x_b(t)=il(_0(_0 t ) ( _ 0 t)-_1(_0 t ) ) [ nr5 ] here @xmath67 the third part comes from the integration of the nonresonant force , but is not exponentially suppressed , however it is suppressed by additional power of @xmath68 relative to the previous term : x_c(t)=(il/(^2+_0 ^ 2 ) ) ( _ 0_1-_0(_0(_0 t)+(_0t))((_0t)-(_0 t ) ) [ nr6 ] finally the 4th term is the exponentially suppressed part coming from the nonresonant part of the external force : @xmath69 finally there is a solution of the homogeneous equation that can be rewritten as : @xmath70 note that if we are interested in time scales @xmath71 we can neglect the terms that are exponentially suppressed . the terms in @xmath72 have the same structure essentially as in @xmath73 but are enhanced by at least @xmath74 . hence we can write for @xmath75 x(t)=x_a(t)+x_b(t)+x_e(t ) [ f ] here in @xmath76 we can in the expression for b(t ) neglect all the terms except coming from @xmath77 . then for b(t ) we have b(t)=(x_1-x_0(_0 t)-il ( t ( _ 1(_0 t)-_0)+(t)(_0 ( _ 0 t)-_1)/_0)/(_0 t ) [ b ] we see that the classical trajectory of the oscillator in random field corresponds to the sum of two terms : first , the usual harmonic oscillations ( due to the homogenic part of the solution ) , second the linear expansion ( although suppressed as @xmath78 ) in the imaginary direction . we now calculate the action as the function of the boundary conditions using the solution ( [ f ] ) . due to the equations of motion the action is just i(x ( t)x_1-y(t)y_1-x_0x(0)+y(0 ) y_0 ) , [ ac ] the other terms in the action disappear due to the equations of motion . it is easy to obtain : @xmath79 this action gives us the transition operator as a function of boundary conditions . note that it contains three types of terms : 1 ) real part ; 2 ) imaginary part with linearly enhanced terms due to quantum fluctuations 3 ) imaginary part with only trigonometric terms . note that we neglected for all the terms that are suppressed by @xmath80 and by the powers of @xmath4 . the transition operator in the quasiclassical approximation will be f=(is ) [ tr ] we need now to consider the preexponential factor , and normalize to the evolution operator of the harmonic oscillator without electric fields . in order to calculate the preexponent we need to expand : x(t)=x_cl(t)+^_0 c_n ( 2n t / t ) [ ex ] y(t)=y_cl ( t ) + ^_0 b_n(2n t / t ) [ y ] since the cross terms with @xmath81 evidently cancel we need to consider the integral in the action due to the sum over n. it is equal to : @xmath82 we can now move to the integration over the variables @xmath83 and @xmath84 . then we evidently get after the integration over @xmath85 the product of delta - functions @xmath86 then we can evidently let @xmath87 . the preexponential factor @xmath88 will be the same as the corresponding factor for the oscillator without the electric field which is evidently : ^n=_n=0 = ( _ 0t)[ds ] . now we can calculate the transition operator g , whose matrix elements give the transition probabilities we are looking for . we obtain @xmath89 note that the first term is just the usual evolution operator for the harmonic oscillator , while the second term contains the action of the random field . the operator is written in terms of the proper time , in the oscillator reference frame . in order to work in the laboratory frame we need to substitute @xmath90 , where @xmath91 is the laboratory time . we can use the latter transition operator to calculate transition amplitudes as the functions of time . once we know the transition operator ( [ g ] ) we can calculate the evolution of the initial distribution . if in the initial state t=0 the oscillator is in the state @xmath92 , the probability that after time t the oscillator will be in the state k is @xmath93 here @xmath94 is the normalization factor : n_n= ( m_0/)^1/4 [ norm ] in order to carry the actual calculation we need to go to the variables @xmath95 and carry out the corresponding gaussian integral over the four variables . note that due to the cross - term between @xmath96 in the transition operator f we have ^2 [ ne ] the easiest example to calculate is the transition probability @xmath97 for the oscillator entering the media in the initial state , and then staying in the initial state . the corresponding gaussian integral is a^0_0 ( t)^2=1/ [ h ] where h is given by h=1 + 2_0 dt_l+4d^2(_0 ^ 2t_l^2-^2 ^ 2 ( _ 0 t_l/ ) ) [ answer ] here d_l=2b_l^2/ ( m_0 ^ 2(^2 ^ 2 + _ 0 ^ 2 ) ) [ d ] for the longitudinal motion and d_t=2b_t^2 ^ 2/(m_0 ^ 2 ) ( ^2 ^ 2 + _ 0 ^ 2 ) ) [ d ] for the transverse motion . we see that for @xmath98 the dependence is @xmath99 the result for the arbitrary n , k is given by the derivative : @xmath100 here the integral i is i = ds_0ds_1d_0d_1 ( is+_1s_0+_2_0 + _ 1_0+_2_1 -m_0(_1 ^ 2+_0 ^ 2+s_1 ^ 2+s_0 ^ 2)/4 ) /(_0 t ) [ i ] the integral i is a gaussian integral that can be easily taken : i=(-z / h-(_1 ^ 2+_1 ^ 2)/m_0 ) ) ) [ i ] here @xmath101 the problem is of course 3dimensional , and the explicit expression for 3d coefficients are just a direct product of one - dimensional coefficients in @xmath102 plane and in z direction a^n_1n_2n_3_k_1k_2k_3=_i=1,2 a^n_i_k_i^n_3_k_3 [ pro ] here we took into account that for sufficiently big @xmath4 the transverse fields are strongly enhanced and we need only to take into account the transverse transitions . the evolution of the arbitrary initial distribution ( [ expansion ] ) can be calculated in the same way . in the previous section we considered the time evolution of the arbitrary initial states . in this section we shall consider the evolution of the small wave packet that has at t=0 the form in the transverse plane : ( y)=(2dm_0/)^1/4(-dm_0y^2 ) [ small ] we first calculate explicitly the density matrix @xmath103 where the average means the averaging over all random field configurations , and then use this expression to calculate the probabilities . substituting eq . ( [ small ] ) into the equations of the previous section , we get for the density matrix @xmath104 here @xmath105 is given by eq . ( [ d ] ) and the function @xmath106 is given by d(t)=(d^2(_0 t/)^2+(_0 t/))^2/4.+dd(_0 t - sin(2_0 t/)/2 ) [ diff3 ] we can determine density radius of the dipole / wave packet . the density radius characterizes the decrease of @xmath107 as a function of @xmath108 . note , that this is completely different physical quantity than the effective radius of the dipole / wave packet , that determines the effective cross - sections . the expansion of the wave packet density is determined by @xmath109 x^2d(t)/d=(d(_0 t/)^2+(1/d)(_0 t/))/4.+d(_0 t - sin(_0 t2/)/2 ) [ diff ] note that for the small t@xmath110 we have x^2(t)(1/d+(d-1/d)_0 t^2/^2 + 16d_0 ^ 2t^3/^2)/4 [ small1 ] we see that there are three pieces in the latter equation : first , the initial radius of the wave packet , second - the geometrical factor , that is equal to zero when the particle is in the ground state , and is bigger , the smaller is the wave packet radius ( in correspondence with the uncertainty principle ) . this factor describes the wave - packet diffusion , that is present when there are no interactions . finally , the third term is wave - packet radius independent and describes the density expansion due to a strong field . we see that the latter for small t goes as @xmath111 , and for the case of the ground state @xmath112 this is the only term that contributes to the expansion . before the time @xmath91 reaches @xmath113 the density radius squared increases according to a square law , while for times @xmath114 it increases according to the linear law . we can determine , till what time scale we can consider the field @xmath18 as homogeneous . it is well known in the theory of random systems @xcite that the field can be considered homogeneous if the inhomogeneity scale in transverse direction is much bigger than time ( and z ) correlation scale.in our case the latter scales are @xmath115 and @xmath116 correspondingly . however , due to the longitudinal diffusion the longitudinal - time correlation scale increases . due to the fact that the longitudinal electric field does not change strength due to lorentz boost , the expansion for the small packet in the longitudinal direction is determined by the geometrical factor . for the big packet with the radius of the ground state or bigger the expansion rate will still be determined by the diffusion . the longitudinal scale will reach the radius @xmath115 at the minimum of the times @xmath117 , @xmath118 . this means that for all realistic values of @xmath119 and for sufficiently large @xmath4 , our results , derived for the homogeneous fields , certainly hold up to the times of @xmath120 , that is the characteristic time in the interaction of the dipole and the target . for the case of the oscillator the same result can be obtained in the different way , using the erenfest theorem . indeed + ^2 x = e(t)/m [ e ] this operator equation is easily solved : x(t)=a(t_l/)+^t_0 ds((t - s ) ) e(s ) /m [ pitaron ] then = ( b^2))^t_0^t_0dsds((t - s))(t - s)<e_z(s)e_z(s ) > [ averl ] for longitudinal diffusion , and = ( b^2 ^ 2)^t_0^t_0dsds((t - s))(t - s)<e_t(s)e_t(s ) > [ aver ] for transverse diffusion . in this way we obtain the law of expansion : b^2((2t_l-(2t/)/(2 ) ) ) ^2 ^ 2/(^2 ^ 2+^2 ) [ diffusion ] we depicted the characteristic dependence of the square radius of the oscillator in the initial ground state and in the initial small wave packet state in figures [ ret8],[ret9],[ret10 ] and [ ret12],[ret13],[ret14 ] respectively . note that in this derivation it is irrelevant if we consider classical or quantum problem . the coincidence of the results of two approaches is of course due to the erenfest theorem . once we know the laws of quantum expansion we can calculate the probability distribution for the system to be in the n - th state . the probability is given by = s(t)^n/2/r(t)^(n+1)/2 p_n(q(t)/(2d(t ) ) [ prob ] here @xmath121 @xmath122 are the legendre polynomials , and d(t ) is given by eq.([diff3 ] ) ; @xmath123 @xmath124 the sample graphs are given below ( see figures [ ret3],[ret4],[ret5 ] ) . consider now the oscillator entering the random media in the ground state . then we can put d=1 in the formulae of the preceding chapter and all results will be significantly simplified : r(t)=h(t)/(4d(t ) ) , s(t)=d^2(_0 ^ 2t^2-^2(_0 t/)^2/d(t)[simple ] the probability is given by @xmath125 the sample graphs are given below ( see figs . [ ret1],[ret2],[ret11 ] ) . these results can be applied to the simplest model of charmonium moving through the quark - gluon plasma . this picture of charmonium is of course not realistic , since it does not contain absorption @xcite , coulomb interactions @xcite , color and radiation . it is given here for illustrative purposes only . the more realistic picture of charmonium will be discussed elsewhere . in our model the charmonium enters the media ( say quark - gluon plasma or nuclear matter ) as the dipole in the ground state . the charmonium is scattered by the random electric fields in the media and is excited to the states with @xmath126 or higher states , where it decays via @xmath127 pair decay mechanism . we assume for simplicity that the decay probability is one , i.e. once the charmonium crosses the @xmath127 threshold @xcite , the total probability that charmonium will still be charmonium and not decay will be given by the sum p(t)=_i=0,j=0,k=0^i+j+kn_0a(0ijk)^2 [ char ] here @xmath128 is the energy level corresponding to the @xmath127 threshold . it is enough to take into account only transitions in transverse directions , so the sum is given by p(t)=_p+sn_0 a^p_0a^s_0 ^2 [ c ] the result is depicted in figure [ ret6 ] . in the calculations we have taken the realistic @xmath66 = 1 gev . for the realistic @xmath129 we take 0.12 gev , for @xmath130 ( that corresponds to quick charmonium produced in the fermilab experiment ) , and for the field b we take the field in the qcd vacuum @xcite , that is say @xmath131 gev@xmath132 , and for illustrative purposes we took @xmath133 . our results imply that this mechanism can make a significant contribution in the dissociation of the rapid charmonium in the qcd media . we also did not take into account that there exist , for the small wave packet , backward transitions from higher levels to low ones , that must be discarded in the realistic model ( since charmonium will dissociate ) . this question will be dealt separately . we have studied the qualitative influence of the random media on the relativistic propagation of the nonrelativistic dipoles and wave packets with internal interaction ( see eqs . ( [ density]),([small1 ] ) ) . we showed that the media : 1 ) enhances the expansion rate and leads to the new regime of the superdiffusion expansion at the rate @xmath134 for a ground state and for the small wave packet for a sufficiently large time intervals if a field is sufficiently large ; 2 ) leads to the diffusion over levels , leading to new types of diffractive events . we have seen that the wave packet expansion rate depends on the field strength and can be used to study the properties of the media . moreover , our results strongly suggest that the energy losses and radiation of the dipole or wave packet propagating through the random media will strongly depend on the media ( compare with the corresponding calculation for a color parton @xcite ) , and can be used for the diagnostics of the quark - gluon plasma . our results also imply an additional mechanism of the charmonium output suppression relative to the known ones @xcite . our results imply the possibility of an additional mechanism of dissociation of rapid charmonium by excitation to the higher levels due to scattering on the random fields , and consequent dissociation into @xmath6 pairs . we have calculated the transition operator ( [ g]),([finaction ] ) ) for the oscillator , and calculated the time dependence of the probabilities @xmath135 of the oscillator occupying the n - th level if it was in some initial state . the corresponding formula is given by eqs . ( [ s]),([d]),([s]),([r]),([prob ] ) . we used these results to study the simple illustrative model of charmonium , and were able to determine how many charmoniums will stay after time t ( or , equivalently , if a charmonium beam goes through the shell of the width @xmath136 ) ( see eqs . ( [ prob1 ] ) , ( [ char ] ) . the results obtained in this paper were derived for a particular case of the moving oscillator . we expect that the same results still hold , at least qualitatively , for more general system , such as a coulomb one . and . indeed , the mechanism of superdiffusion expansion is the pick - up by the system of the resonance frequencies from the media field frequency spectrum . in the system other than oscillator , say in the coulomb system , it will pick the frequencies corresponding to the distances between nearby levels , and consequently will quickly expand , as in the simple oscillator case . it will be very interesting and necessary for the study of realistic systems to take into account absorption @xcite . we have calculated the effective action and transition operator ( [ final ] ) for the propagation of the general relativistic dipole in the media . this action , obtained after integration out of the electric field is quite general , and can be used for the arbitrary bound systems . however , for the cases other than harmonic oscillator it is hard to carry the exact calculations and one may resort to numerical methods . it will be very interesting to use our method to study a number of problems not considered here : the propagation of positronium through amorphous media and it s ionization rate , and the more realistic charmonium model that includes color and absorption , as well as to the detailed analysis of the influence of the nonhomogeneity . it will be also very interesting to carry the calculation of the radiation loss and it s influence on the dynamics of the dipole . at this point , due to the appearance of the traces of the several color matrices , we shall expect real technical difference between calculations in the usual plasma and quark gluon plasma . the author is indebted to prof . l. frankfurt for numerous very fruitful discussions and careful reading the paper . the author is also thanks dr . a. kamenev for useful discussions . the aim of this section is to connect the usual language of the ref . @xcite model with the language of the statistical mechanics of the random media . let us recall briefly the model of ( @xcite ) . the model describes the qcd media as the usual debye plasma , with a potential v^a_i ( q)=g ( t^a_i)_cc(-iqx_i)/(q^2+^2 ) [ pot ] here @xmath137 is the debye screening mass . the parton - parton interactions are proportional to @xmath138 . the averaged potential vanishes everywhere since @xmath139 , while the parton - parton interactions are proportional to tr(t^a_it^b_j)=_ab_ij(d_i / d_a)c_2i [ wang ] here @xmath140 is the dimension of the su(n ) representation , @xmath141 is the second casimir , ( @xmath142 for quarks ( @xmath143 ) and n for gluons ( @xmath144 ) . thus after averaging we see that such model looks like multicomponent plasma , with a particular interacting charge for each set of color partons , with different types of partons not interacting in the leading order . thus , we expect that apart from some numerical factors , connected with the casimirs , the propagation of the wave packet in such a plasma will be qualitatively the same as it s propagation through the usual electromagnetic plasma described in the article . the true color effects will appear only when we shall take into account multiparticle correlations . we now turn to the description of the charged plasma in the language of the statistical mechanics ( @xcite ) . in order to do it , we can just rewrite the statistical sum of the plasma as a euclidean functional integral . the choice of the action must be such that it reproduces the correlation function ( [ correlation ] ) . clearly , the functional integral is gaussian and can be represented as z = de(-d^3qv ( q^2+^2)v ) [ connection ] moving into the oscillator rest frame and taking into account that the time derivative term will be dominant , we come to the action ( [ ac ] ) . this description is just the usual language of the statistical mechanics of the random media . let us introduce the center of mass coordinates ( for simplicity let us assume that both particles have the same mass ) @xmath28 and the relative motion coordinate @xmath145 : @xmath146 we shall denote m the reduced mass of the oscillator in c.m . reference frame . then the lagrangian can be written as a sum : l = ds ( v ) + l_n.r.(u)[sum ] indeed , the full lagrangian is l = ds ( r_1)+ds ( r_2 ) -(v(r_1)-v(r_2)+v(u))dt [ sto ] here ds ( r)=dt [ par]we then write : ds(r_1)=dt [ w1 ] ds(r_1)=dt [ w2 ] in these two expressions we expand their r.h.s . in taylor series in u around v. then we immediately obtain : ds(r_1)+ds(r_2)=ds(v)+dt(^2)/(/2 = ds(v)+dt(^2)/2 [ sf ] here t is the oscillator proper time . adding the interaction term to the eq . ( [ sf ] ) we immediately obtain the decomposition ( [ sum ] ) . let us now discuss briefly the lorentz transformation @xcite . in the laboratory system we have only the electric field described by the static potential v. the lorentz transformation says : v=(v-va_3 ) a_3=(a_3-vv)[a ] in the laboratory frame @xmath147 , hence @xmath148 . substituting the latter formula in eq . [ a ] we get v = v(1-v^2)=v[f ] finally we get v=v a_3=vv [ fa ] differentiating the latter equations in the moving reference frame we obtain the transformation laws in the text . here for completeness we shall write the formulae for the double integral used in the text : @xmath149 the integral was calculated using ref . @xcite the legendre polynoms are defined as @xcite:2^nn!p_n(x)=(x^2 - 1)^n [ leg ] a.i . akhieser and n.f . shulga , high energy electrodynamics in matter , gordon and breach , 1996 . l. frankfurt , j . miller , m. strikman , annual review on particle and nuclear physics , 44 ( 1994 ) 501 . i. m. lifshits , s. a. gredeskul and l. a. pastur , introduction to the theory of the disordered media , new york , wiley , 1988 . l. d. landau , e. lifshits , statistical mechanics , pergamon press , 1980 . h. heiselberg , m. hjorten - jensen , phys . rept.,328 ( 2000 ) 237 ; k. rajagopal and f. wilczek , in `` at the frontier of particle physics , handbook of qcd '' , feitschrift in honour of b. ioffe , m. shifman ed . , w. scientific , 2001 , v.3 , pp . 206 - 251 ; hep - ph/001333 . d. gross , b. svetitski and l. yaffe , rev . phys . , 53 ( 1981 ) 43 . j. cleymens , r.v . gavai , e. suhenen , phys . , 130 ( 1986 ) 217 . s.feng , l. golubovich and y .- c . zhang , phys . 65 ( 1990 ) 1028 . l. golubovic , s. feng and f. zeng , phys . rev . lett . , 67 ( 1991 ) 2115 . j. kogut and l. susskind , phys . lett . , 34 ( 1975 ) 767 . c. quigg and j. rosner , phys . reports , c56 ( 1979 ) 167 . landau , e. m. lifshits , mechanics , pergamon press , 1976 . farrar , l. frankfurt , m. strikman , phys . lett . , 61 ( 1988 ) 686 . b. kopeliovich and b.g . zakharov , physical review , d44 ( 1991 ) 3466 . m. gyulassy and x. wang , nuclear physics b , 420 ( 1994 ) 583 . m. danos , v. gillet , m. gauvin , methods in relativistic nuclear physics , elsevier publishers , 1984 . l. landau , e. lifshits , field theory , pergamon press , 1978 . see e.g. a.a . adronov , a.a . vitt and s.e . khaikin , theory of oscillations , pergamon press , 1966 ; n. erugin , linear systems of ordinary differential equations with periodic and quasiperiodic coefficients , academic press , 1966 ; c.a . swanson , comparison and oscillation theory of linear differential equations , academic press , 1968 . g. samelsohn and r. mazar , physical review e , 58 ( 1998 ) 1094 . r. p. feynman and a. r. hibbs , _ quantum mechanics and path intgrals _ , ( mcgraw - hill , new - york , 1965 ) , p.357 . h. kleinert , path integrals , world scientific , 1995 . l. schulman , technics and applications of path integration , john wiley and sons , 1981 . r. feynman , statistical mechanics , w.a . bouchaud , europhys . ( 1990 ) 505;j.p . bouchaud , d. touati and d. sornette , phys . rev . 68 ( 1992 ) 1787 . d. kharzeev , invited talk international conference on statistical qcd , bielefeld , germany , 26 - 30 august , 2001 . w. poschl and b. muller , phys . rev . d60 ( 1999 ) 114505 . l. gerland , l. frankfurt , m. strikman , h. stocker , w. greiner , in quark matter 2000 ( berkeley , california , 20 - 25.7.2000 ) , j. of physics g27 , 695 - 702 ( 2001 ) . r. baier , yu . dokshitser , a.h . mueller , s . peigne , d. schiff , in proceedings of future physics in hera , v. 2 ( 1996 ) 1003 , hamburg , 1995/1996 , and references therein . , edited by m. abramowitch and i. stegun , ( dover , new - york , 1970 ) . m. shifman , a. vainshtein and v. zakharov , nuclear physics b147 ( 1979 ) 1 . a. p. prudnikov , yu.a . brychkov , o.i . marichev , inteegrals and series , v.2 : special functions , gordon and breach science publishers , 3d printing , 1992 . h. bateman , a. erdel , higher transcendental functions , mc graw hill book company , 1953 .

a random-lift of a base graph is its cover graph on the vertices\times v(g)$ ] , where for each edge in there is an independent uniform bijection , and has all edges of the form . a main motivation for studying lifts is understanding ramanujan graphs , and namely whether typical covers of such a graph are also ramanujan . let be a graph with largest eigenvalue and let be the spectral radius of its universal cover . friedman ( 2003 ) proved that every `` new '' eigenvalue of a random lift of is with high probability , and conjectured a bound of , which would be tight by results of lubotzky and greenberg ( 1995 ) . linial and puder ( 2008 ) improved friedman s bound to . for-regular graphs , where and , this translates to a bound of , compared to the conjectured . here we analyze the spectrum of a random-lift of a-regular graph whose nontrivial eigenvalues are all at most in absolute value . we show that with high probability the absolute value of every nontrivial eigenvalue of the lift is . this result is tight up to a logarithmic factor , and for it substantially improves the above upper bounds of friedman and of linial and puder . in particular , it implies that a typical-lift of a ramanujan graph is nearly ramanujan . [ multiblock footnote omitted ]

0911.4148

over the last quarter of a century , _ expander _ graphs have played a vital role in a remarkable variety of areas , ranging from combinatorics to discrete geometry to theoretical computer science , while exhibiting deep connections to algebra and number theory . notable applications of expanders , to name just a few , include the design of efficient communication networks , explicit error - correcting codes with efficient encoding and decoding schemes , derandomization of randomized algorithms , compressed sensing and the study of metric embeddings . see the expository article of sarnak @xcite on these intriguing objects , as well as the comprehensive survey of hoory , linial and wigderson @xcite demonstrating their many applications . informally , an expander is a graph where every small subset of the vertices has a relatively large edge boundary ( see section [ sec : intro - exp ] for a formal definition ) . most applications utilize @xmath12-regular sparse expanders ( @xmath20 fixed ) , where it is well - known that expansion is related to the ratio between @xmath12 and @xmath17 , the second largest eigenvalue in absolute value of the adjacency matrix . the smaller @xmath17 is , the better the graph expansion becomes . as a consequence of the alon - boppana bound @xcite ( see also @xcite ) @xmath21 where the @xmath22-term tends to @xmath23 as the graph size tends to @xmath24 . graphs for which @xmath25 are in that respect optimal expanders and are called _ ramanujan _ graphs . a proof that @xmath12-regular expanders exist for any @xmath26 was given by pinsker @xcite in the early 70 s via a simple probabilistic argument . however , constructing good expanders _ explicitly _ is far more challenging and particularly important in applications ( see @xcite and the references therein ) , a task that was first achieved by margulis @xcite . thereafter ramanujan graphs were constructed explicitly in the seminal works of lubotzky - phillips - sarnak @xcite and margulis @xcite , relying on deep number theoretic facts . till this date ramanujan graphs remain mysterious : not only are there very few constructions for such graphs , but for instance it is not even known whether they exist for any @xmath26 . a striking result of friedman @xcite shows that almost every @xmath12-regular graph on @xmath0 vertices is _ nearly _ ramanujan it has @xmath27 ( the @xmath22-term tends to @xmath23 as @xmath28 ) . what proportion of these graphs satisfy @xmath25 remains an intriguing open problem . the useful connection between expanders and the topological notion of _ covering maps _ was extensively studied by many authors over the last decade . various properties of random covers of a given graph were thoroughly examined ( see e.g. @xcite ) , motivated in part by the problem of generating good ( large ) expanders from a given one . given two simple graphs @xmath1 and @xmath2 , a _ covering map _ @xmath29 is a homomorphism that for every @xmath30 induces a bijection between the edges incident to @xmath31 and those incident to @xmath32 . in the presence of such a covering map we say that @xmath2 is a _ lift _ ( or a cover ) of @xmath1 , or alternatively that @xmath1 is a _ quotient _ of @xmath2 . the _ fiber _ of @xmath33 is the set @xmath34 , and if @xmath1 is connected then all fibers are of the same cardinality , the _ covering number_. one well - known connection between covers and expansion is the fact that the universal cover of any @xmath12-regular graph is the infinite @xmath12-regular tree @xmath35 , whose spectral radius is @xmath14 , the eigenvalue threshold in ramanujan graphs . in fact , greenberg and lubotzky @xcite ( cf . @xcite*chapter 4 ) extended the alon - boppana bound to any family of general graphs in terms of the spectral radius of its universal cover ( also see @xcite*theorem 4.1 ) . it is easy to see that any lift of a @xmath12-regular base graph @xmath1 is itself @xmath12-regular and inherits all the original eigenvalues of @xmath1 . one hopes that the lift would also inherit the expansion properties of its base graph , and in particular that almost every cover of a ( small ) ramanujan graph will also be ramanujan . since our focus here is on lifts of ramanujan graphs ( regular by definition ) we restrict our attention to base graphs that are @xmath12-regular for @xmath26 . a random uniform @xmath0-lift of a base graph @xmath1 ( a uniform cover of @xmath1 with covering number @xmath0 ) has the following convenient description : it is the graph @xmath2 on the vertices @xmath3\times v(g)$ ] , where for each edge @xmath4 in @xmath1 there is an independent uniform bijection @xmath5 , and @xmath2 has all edges of the form @xmath6 . the random lift of a complicated base - graph is thus a hybrid between the complex geometry of the quotient and the randomness due to the bijections . in an important development in the study of the spectrum of random lifts friedman @xcite showed in 2003 that with high probability ( w.h.p . ) every `` new '' eigenvalue of an @xmath0-lift ( one that is not inherited from the base graph ) is at most @xmath36 , where @xmath7 is the largest eigenvalue of the base - graph and @xmath8 is the spectral - radius of its universal cover . when the base - graph @xmath1 is @xmath12-regular , @xmath14 and friedman s result implies that in its random @xmath0-lift @xmath2 the largest absolute value of all nontrivial eigenvalues is w.h.p . @xmath37 where @xmath38 denotes @xmath39 . conversely , @xmath40 and by alon - boppana it is also at least @xmath41 . this lower bound was conjectured by friedman @xcite to be tight ( for general graphs he conjectured that all new eigenvalues are at most @xmath10 as in the greenberg - lubotzky bound ) . in a recent paper @xcite , linial and puder were able to significantly improve friedman s bound and show that w.h.p . all the new eigenvalues of @xmath2 are at most @xmath42 . consequently , an @xmath0-lift @xmath2 of a @xmath12-regular @xmath1 w.h.p . satisfies @xmath43 when @xmath1 is a @xmath12-regular expander with nontrivial eigenvalues of @xmath44 as is the case for ramanujan graphs , this translates to @xmath45 . our main result in this work is the new near optimal upper bound of @xmath46 when @xmath1 is @xmath12-regular with all nontrivial eigenvalues at most @xmath17 in absolute value . for @xmath47 it substantially improves the known bounds , , and when @xmath48 as in ramanujan graphs it is tight up to a logarithmic factor , giving @xmath49 . [ thm-1 ] let @xmath1 be a @xmath12-regular graph with all nontrivial eigenvalues at most @xmath17 in absolute value and let @xmath14 be the spectral radius of its universal cover . let @xmath2 be a random @xmath0-lift of @xmath1 . for some explicit absolute constant @xmath50 , every nontrivial eigenvalue of @xmath2 is at most @xmath51 in absolute value except with probability @xmath52 . [ cor-2 ] let @xmath1 be a @xmath12-regular ramanujan graph vertices and let @xmath2 be a random @xmath0-lift of @xmath1 . with probability @xmath53 every nontrivial eigenvalue of @xmath2 is at most @xmath54 in absolute value , where @xmath50 is an explicit absolute constant . note that the above corollary implies that typical random @xmath0-lifts of ramanujan graphs are nearly ramanujan . no attempt was made to optimize the explicit constant in theorem [ thm-1 ] . finally , the statement of theorem [ thm-1 ] holds even when the size of the base - graph @xmath55 is allowed to grow with @xmath0 provided that @xmath0 is large enough in comparison ( e.g. , @xmath56 ) . the previous bounds on the spectra of random @xmath0-lifts of a fixed graph @xmath1 due to friedman @xcite and linial and puder @xcite were both obtained via wigner s trace method . the fact that the universal cover @xmath57 of a connected graph @xmath1 is the infinite tree of non - backtracking walks from an arbitrarily chosen vertex makes the trace method particularly useful for relating the new eigenvalues of the lift with @xmath8 , the spectral - radius of @xmath57 . even when the geometry of a graph is very well understood , bounding its nontrivial eigenvalues can be extremely challenging . for instance , a line of papers ( cf . @xcite ) established various bounds for the second eigenvalue of certain random regular graphs , culminating in the optimal bound @xmath58 for a uniformly chosen @xmath12-regular graph on @xmath0 vertices , proved by friedman @xcite using highly sophisticated arguments . it turns out that this model is essentially the special case of an @xmath0-lift of a graph comprising a single vertex with self - loops : it is easy to see that for @xmath12 even , the random @xmath12-regular graph obtained by @xmath59 independent uniform permutations in @xmath60 is equivalent to an @xmath0-lift of the base - graph @xmath1 that has a single vertex with @xmath59 loops ( this model is in fact contiguous to the uniform random @xmath12-regular graph for @xmath61 , cf . e.g. @xcite ) . unfortunately , when the base - graph features a complex and rich structure ( e.g. the lps - expanders , whose expansion properties hinge on a deep theorem of selberg ) it becomes significantly harder to control the spectrum of its lifts . indeed , there are many examples of geometric properties that have been pinpointed precisely for the random regular graph yet remain unknown for arbitrary expanders ( see @xcite for a recent such example ) . estimating the number of closed walks in lifts of arbitrary ramanujan graphs thus appears to be a formidable task . in this work , the bounds obtained for the spectra of lifts of arbitrary expanders rely on an approach introduced by kahn and szemerdi @xcite , which is quite different from wigner s trace method . this approach was originally used to control the spectrum of a random regular graph , and several new ideas are required to adapt it to the more complicated geometry of the lifts considered here . another related problem in the study of spectra of lifts , yet of a rather different nature , considers the @xmath62-lift of a base - graph ( rather than @xmath0-lifts of a small fixed graph ) . bilu and linial @xcite showed that for any @xmath12-regular graph @xmath1 there _ exists _ a @xmath62-lift with all new eigenvalues at most @xmath63 . this was shown by means of the lovsz local lemma , combined with the crucial observation of @xcite whereby the new eigenvalues correspond precisely to the eigenvalues of a _ signing _ of the adjacency matrix of @xmath1 ( the matrix obtained by replacing a subset of its @xmath64 entries by @xmath65 ) . in the absence of such a characterization when the covering number @xmath0 is large , different tools are needed for the problem studied here , where we seek a bound that holds for almost every @xmath0-lift with @xmath0 sufficiently large . as stated above , while a random @xmath12-regular graph @xmath1 has second eigenvalue @xmath66 w.h.p . ( the @xmath22-term tending to @xmath23 as @xmath67 ) , the probability that @xmath1 is ramanujan is unknown . see @xcite for some experimental results suggesting that this probability is bounded away from @xmath23 and @xmath64 . as this is essentially the simplest special case of a random lift ( the quotient being a single vertex with self - loops ) , it is natural to conjecture the following : [ conj-3 ] for any ramanujan graph @xmath1 there exist some @xmath68 such that its random @xmath0-lift @xmath2 satisfies @xmath69 , where the @xmath22-term tends to @xmath23 as @xmath28 . note that the limiting constant in the above conjecture depends on the base - graph @xmath1 , as it is plausible that its structure may affect the probability of being ramanujan . for instance , a random cover of a complete graph on @xmath70 vertices might behave quite differently compared to lifts of a sparse @xmath12-regular ramanujan graph . however , as we next elaborate , experimental results lead us to suspect that up to normalization this is not the case . figure [ fig : lifts : a ] shows the cumulative distribution function ( c.d.f . ) of @xmath71 where @xmath2 is the @xmath72-lift of 3 different @xmath73-regular ramanujan base - graphs : @xmath74 ( complete graph on @xmath75 vertices ) , the @xmath76-vertex petersen graph and the @xmath77-vertex dodecahedral graph . each curve was evaluated from @xmath78 random lifts . in these simulations , the probability of a random lift being ramanujan for each of these three base - graphs was bounded between @xmath79 and @xmath80 . somewhat surprisingly , aligning the number of vertices of the graph cover @xmath2 to be the same ( via @xmath72-lifts of the dodecahedral graph , @xmath81-lifts of the petersen graph and @xmath82-lifts of @xmath74 , giving @xmath83-vertex covers for each graph ) resulted in the curves of the individual c.d.f.s coinciding fairly accurately . this is demonstrated in figure [ fig : lifts : b ] , in light of which we speculate that the following stronger version of the statement of conjecture [ conj-3 ] holds . first , it seems plausible that for any integer @xmath20 the limiting distribution of the second eigenvalue of the random cover is independent of the base - graph . namely , there exists a distribution @xmath84 on @xmath85 $ ] such that for any @xmath12-regular ramanujan graph @xmath1 on @xmath55 vertices , the distribution of @xmath71 for its random @xmath0-lift @xmath2 converges to @xmath84 as @xmath28 . second , the strong fit between the curves after aligning the total graph sizes suggests that even the rate of convergence to @xmath84 depends on @xmath86 rather than on the geometry of the base - graph or even its relative size ( in case @xmath55 is allowed to depend to @xmath0 ) . of - course , one clearly needs some level of `` burn - in '' for the covering number @xmath0 compared to @xmath55 since the cover @xmath2 starts as ramanujan at @xmath87 . for example , it may be that for any @xmath88 the total - variation distance between the distribution of @xmath71 and @xmath84 decays as a function of @xmath89 alone , namely that @xmath90 where @xmath91 depends only on @xmath92 and tends to @xmath23 as @xmath93 . the base - graph @xmath1 from theorem [ thm-1 ] corresponds to the algebraic definition of an expander known as an @xmath94-graph . an alternative closely - related criterion is the traditional definition of an expander graph in terms of its combinatorial edge or vertex expansion . let @xmath1 be a @xmath12-regular graph on @xmath55 vertices . the cheeger constant of @xmath1 ( also referred to as the edge isoperimetric constant ) is defined as @xmath95 where @xmath96 denotes @xmath97 and @xmath98 is the set of edges with exactly one endpoint in @xmath99 . we say that @xmath1 is a @xmath100-edge - expander for some fixed @xmath101 if it satisfies @xmath102 . similarly , one defines a @xmath100-vertex - expander by replacing @xmath98 with the vertex boundary . for @xmath1 as above the eigenvalues of the corresponding adjacency matrix are @xmath103 by perron - frobenius . we say that @xmath1 is an @xmath94-graph if @xmath104 for all @xmath105 . this notion was introduced by alon in the 1980 s , motivated by the fact that when @xmath17 is much smaller than @xmath12 such graphs exhibit strong pseudo - random properties , resembling a random graph with edge density @xmath106 . a notable example of this is captured by the _ expander mixing lemma _ : if @xmath107 are ( not necessarily disjoint ) subsets of vertices of an @xmath94-graph then @xmath108 where @xmath109 ( @xcite*chapter 9 ) . relating the above two notions of expansion is the following well - known discrete analogue of cheeger s inequality bounding the first eigenvalue of a riemannian manifold ( alon @xcite , alon - milman @xcite , dodziuk @xcite , jerrum - sinclair @xcite ) : @xmath110 see the survey @xcite for further information on expanders . we begin by describing the kahn - szemerdi @xcite approach for obtaining an @xmath44 bound for random @xmath12-regular graphs . following broder and shamir @xcite , the actual random graph model studied by @xcite is the @xmath111-regular graph obtained from the union of @xmath12 permutations , contiguous to the lift of a single vertex with @xmath12 loops . let @xmath2 be the random graph in mention and let @xmath112 denote its adjacency matrix . by the rayleigh quotient principle , the second ( in absolute value ) eigenvalue of the graph @xmath2 can be written as @xmath113 where @xmath114 is the trivial eigenvector of @xmath112 . to bound @xmath71 , the authors of @xcite analyzed the maximal possible value of @xmath115 separating the contribution of the pairs @xmath116 to the bilinear form into two cases : 1 . _ heavy pairs _ : the contribution from those pairs @xmath117 where @xmath118 is suitably large . here it is shown that w.h.p . the total contribution to @xmath119 by any pair of unit vectors @xmath120 is at most @xmath44 . 2 . _ light pairs _ : the remaining pairs @xmath116 . here it was shown that two fixed vectors @xmath121 are unlikely to contribute more than @xmath44 to the bilinear form , and an @xmath122-net argument was used to extend this result to any unit vectors @xmath123 . adapting this method to lifts of general graphs requires several additional ingredients . even in the simpler setting of @xcite , some of the arguments are only sketched and might prove difficult to complete in detail . more crucially , in our case we have little knowledge of the base - graph @xmath1 , hence the study of both the `` heavy '' and `` light '' parts becomes significantly more involved . first , our only input on @xmath1 is the magnitude of its second eigenvalue , which turns the analysis of the heavy part into a delicate optimization problem , requiring two levels of dyadic expansions of the potential contributions to the final bilinear form . second , the consideration of the light part relies on a non - trivial martingale argument which may be useful in other applications : in the absence of sufficient control over the expectation ( due to the unknown contribution of the heavy part ) we resort to an @xmath124 analysis of the increments in the corresponding doob s martingale and apply a bernstein - kolmogorov type large deviation inequality due to freedman . throughout the paper we use @xmath1 to denote the base - graph , a @xmath12-regular graph on @xmath55 vertices , and let @xmath2 denote its random @xmath0-lift . the asymptotic notation is used under the assumption that @xmath125 . for the sake of clarity , when addressing a vertex in @xmath126\times v(g)$ ] we will typically denote it either by @xmath127 or by @xmath128 using indices @xmath129 $ ] and @xmath130 $ ] . whenever @xmath131 are vertices in some graph whose identity is clear from the context , the abbreviation @xmath132 will denote that these two vertices are adjacent . for example , @xmath133 will usually stand for @xmath134 , which in turn implies that @xmath135 in @xmath1 by the definition of the lift . unless stated otherwise , all logarithms are using base @xmath62 and @xmath136 denotes the @xmath124 norm in the appropriate euclidean space . the rest of this paper is organized as follows . section [ sec : heavy ] deals with the contribution of the heavy pairs to the bilinear form @xmath119 . section [ sec : light ] deals with the contribution of the light pairs . in the final section , section [ sec : mainproof ] , we combine these results to conclude the proof of theorem [ thm-1 ] . let @xmath1 be an @xmath94-graph ( that is , a @xmath12-regular graph where all nontrivial eigenvalues are at most @xmath17 in absolute value ) with adjacency matrix @xmath137 , and @xmath2 be a random @xmath0-lift of @xmath1 with adjacency matrix @xmath112 . as mentioned before , the largest nontrivial eigenvalue of @xmath2 in absolute value is precisely @xmath138 where @xmath139 denotes the trivial eigenvector . we first analyze the typical contribution to @xmath140 from pairs @xmath141 with fairly large products . more precisely , we say that a pair @xmath141 is _ heavy _ if @xmath142 , and otherwise it is _ light_. for @xmath143 , define @xmath144 to be the random variable @xmath145 the next theorem estimates the contribution of the heavy pairs along the edges of @xmath2 . the exponent of @xmath55 in the requirement @xmath146 was selected to simplify the exposition and can be replaced by @xmath147 for any @xmath148 . [ thm - heavy ] let @xmath1 be an @xmath94-graph with @xmath149 and let @xmath2 be a random @xmath0-lift of @xmath1 for @xmath56 . then with probability at least @xmath53 every @xmath143 with @xmath150 and @xmath151 satisfy @xmath152 and moreover @xmath153| \leq 7000 \ , \lambda \log d$ ] . the main ingredient in the proof of theorem [ thm - heavy ] is the following lemma , which provides an upper bound on the number of edges in a cut between subsets of vertices in @xmath2 . [ prop - e(a , b)-lift - bound ] let @xmath2 be a random @xmath0-lift of an @xmath94-graph @xmath1 with @xmath154 and @xmath56 . then except with probability @xmath52 , every two subsets of vertices @xmath155 with @xmath156 satisfy @xmath157 we will next show how to derive theorem [ thm - heavy ] from this lemma , whose proof is postponed to subsection [ sec - cuts - between - large - subsets ] . consider the following dyadic expansion of @xmath121 : @xmath158 and identify any element @xmath159 in @xmath160 or @xmath161 with the vertex @xmath127 in @xmath2 . these definitions , together with the assumption on @xmath162 and @xmath163 , implies that @xmath164 furthermore , if @xmath165 and @xmath166 then a necessary condition for @xmath167 is that @xmath168 , and so @xmath169 where we @xmath170 is the number of edges between the two subsets of vertices in @xmath2 corresponding to @xmath160 and @xmath161 . to prove that @xmath171 , we set @xmath172 and analyze the sum in according to whether or not @xmath173 . first , consider @xmath174 . as @xmath2 is @xmath12-regular , trivially @xmath175 , yielding @xmath176 where the second inequality used . similarly , @xmath177 and since @xmath174 occurs if and only if @xmath178 , altogether in this case @xmath179 we now focus on @xmath180 . consider the case where @xmath181 . repeating the above argument , we now get @xmath182 by symmetry , the same argument holds for the case @xmath183 , and we infer that @xmath184 it remains to treat @xmath185 with @xmath180 . this will be achieved with the help of proposition [ prop - e(a , b)-lift - bound ] , which estimates the size of the cut between two subsets @xmath107 in case @xmath156 . indeed , for @xmath186 we have @xmath187 and therefore @xmath188 . thus , proposition [ prop - e(a , b)-lift - bound ] gives that with probability @xmath53 , @xmath189 for the first expression in the right - hand - side of , note that there are at most @xmath190 integers @xmath191 such that @xmath192 ( here we used the fact that @xmath193 ) . for each such value , we can combine with cauchy - schwartz to get that @xmath194 and summing over @xmath191 it follows that @xmath195 for the second expression in , again recall that @xmath154 , and so @xmath196 and the same applies to the analogous quantity for @xmath197 . altogether , these two estimates for sum up to @xmath198 and combining this with and gives that @xmath199 recalling , we deduce that @xmath200 to obtain the statement on @xmath201|$ ] first note that since @xmath2 is @xmath12-regular @xmath202 with probability @xmath64 for any two unit vectors @xmath121 . we have already established that , except with probability @xmath52 , every pair of vectors @xmath121 with norm at most @xmath64 satisfies @xmath203 . hence , @xmath204 where the last inequality holds for any sufficiently large @xmath0 . reapplying the result on @xmath144 ( along with the triangle inequality ) now completes the proof of the theorem ( with room to spare ) . write @xmath205 and @xmath206 where @xmath207 . our assumption on @xmath208 then translates into @xmath209 we aim to show that , except with probability @xmath53 , for any such @xmath107 we have @xmath210 , or in terms of @xmath211 , that @xmath212 define the following partition of the fibers according to a dyadic expansion of their proportion that is included in @xmath213 . @xmath214 \times \{v\})\right|}{n } \leq 2^{-i } \right\}~(i=0,1,\ldots,\log n)\,,\\ a_i & = a \cap \bigcup_{v \in s_i } ( [ n ] \times \{v\ } ) \,,\quad s_i = |s_i|\,,\quad \alpha_i = s_i 2^{-i}\,.\end{aligned}\ ] ] notice that by these definitions , @xmath215 is the number of fibers with about @xmath216 vertices from @xmath213 , and so @xmath217 . in other words , there are about @xmath218 vertices of @xmath213 in fibers of type @xmath219 , and more precisely , @xmath220 similarly , we perform an analogous dyadic expansion for @xmath221 : @xmath222\times\{v\})\right|}{n } \leq 2^{-j } \right\}~(j=0,1,\ldots,\log n)\,,\\ b_j & = b \cap \bigcup_{v \in t_j } ( [ n]\times\{v\ } ) \,,\quad t_j = |t_j|\,,\quad \beta_j = t_j 2^{-j}\,,\end{aligned}\ ] ] and again have that @xmath223 clearly , @xmath224 and our bound on @xmath225 will follow from an analysis of the number of edges between the various types of @xmath226 s and @xmath227 s . first , consider the case @xmath228 . here we have @xmath229 and @xmath230 . since there are @xmath0 edges in @xmath2 between any pair of fibers that correspond to adjacent vertices in @xmath1 , the expander mixing lemma ( see ) applied to the base - graph @xmath1 gives that @xmath231 recalling that @xmath232 ( see ) it follows that @xmath233 where the last inequality used the fact that @xmath149 and so @xmath234 . next , consider @xmath235 in case @xmath236 . there is a total of @xmath215 fibers in @xmath226 , thus by definition of the @xmath0-lift of a @xmath12-regular graph there are at most @xmath237 fibers , where @xmath227 may have vertices that contribute to @xmath235 . since @xmath227 has at most @xmath238 vertices in each fiber , and each vertex has @xmath12 neighbors in @xmath2 , we deduce that @xmath239 where the last inequality followed from . similarly , we have @xmath240 and conclude that @xmath241 it remains to treat the case @xmath242 for all @xmath243 , where the required bound will only hold w.h.p . consider a prescribed set of @xmath191 pairs of vertices @xmath244 ( @xmath245 $ ] ) in @xmath2 . we wish to bound the probability that @xmath246 by @xmath247 . by the independence of the different pairs of fibers in the lift , it clearly suffices to show this when all the @xmath248 s are on one fiber and all the @xmath249 s are on another , i.e. , for some @xmath250 and all @xmath251 we have @xmath252 and @xmath253 . when @xmath254 then it is straightforward that this probability is indeed at most @xmath247 . to see this , expose the pairings of @xmath255 one by one , and note that for @xmath256 , the probability to match @xmath257 to @xmath258 , given that so far we succeeded in matching all the @xmath259 previous pairs , is @xmath260 . further note that , when considering potential edges between @xmath226 and @xmath227 , the case @xmath261 can only arise when @xmath262 , otherwise no two fibers have more than @xmath263 points of @xmath226 and @xmath227 respectively . since we excluded the case @xmath262 , the above estimate holds for any of our sets @xmath264 . write @xmath265 and recall that @xmath226 and @xmath227 are contained in the fibers corresponding to @xmath219 and @xmath266 resp . , and have at most @xmath267 and @xmath268 vertices on each of these respective fibers . suppose first that the identity of the fibers @xmath219 and @xmath266 are given ( we will account for these later ) . in this case , the number of configurations of the vertices of @xmath226 on the fibers @xmath219 can be upper bounded by @xmath269 if @xmath270 and by @xmath271 if @xmath272 , here using the well - known inequality @xmath273 . similarly , the number of configuration of @xmath227 on the fibers @xmath266 is at most @xmath274 . for each such configuration of the vertices of @xmath264 there are at most @xmath275 pairs which may potentially be connected in @xmath2 . from the above estimate on the probability of @xmath191 pairs being adjacent in @xmath2 , it now follows that for any @xmath276 and choice of @xmath219 and @xmath266 , @xmath277 defining @xmath278 we then get @xmath279 n } \,.\label{eq - prob - wij}\end{aligned}\ ] ] we will next establish a threshold for @xmath280 such that the above probability would be at most @xmath281 and then translate this bound to the cut @xmath235 via the corresponding @xmath282 s . consider the equation @xmath283 and note that for @xmath284 it has a unique solution @xmath285 monotone increasing in @xmath286 . let @xmath287 be the solution to @xmath288\,,\ ] ] and define its counterpart ( similar to the relation between @xmath282 and @xmath280 ) @xmath289 combining these definitions with the probability bound , while noting that this bound in that equation is monotone decreasing in @xmath280 ( and hence in @xmath282 ) in the range @xmath290 , we deduce that for any @xmath291 , @xmath292 since @xmath293 with @xmath294 we can sum @xmath191 over all possible values that @xmath295 can accept and infer that @xmath296 next recall that the above estimate was for @xmath264 with a given choice of the fibers @xmath297 . summing the above probability over all possible choices for such fibers ( using a trivial bound of @xmath298 options for each of the sets ) and then further summing over at most @xmath299 pairs of @xmath300 we deduce that @xmath301 with the last inequality valid for any sufficiently large @xmath0 since @xmath302 . collecting and this yields that , except with probability @xmath303 , any two sets @xmath107 with @xmath304 ( as per ) satisfy @xmath305 it thus suffices to bound @xmath306 in order to complete the proof . [ lem - wij - star ] let @xmath287 be the solution to and let @xmath307 be its counterpart as given in . then for any @xmath308 we have @xmath309 we first consider pairs @xmath300 such that @xmath310 . by the definition of @xmath287 as the solution of , the right - hand - side of that equation , which we denote by @xmath311 , necessarily then satisfies @xmath312 \geq 2\,.\ ] ] furthermore , since for any @xmath313 the solution of @xmath314 satisfies @xmath315 ( this is easy to verify using the monotonicity of @xmath316 ) we can infer an upper bound on @xmath287 in the form of @xmath317\,,\end{aligned}\ ] ] and as a consequence @xmath318\,.\end{aligned}\ ] ] immediately by the last denominator is at least @xmath64 and so @xmath319 where @xmath320 by definition . next , note that for the same reason @xmath321 using the fact that @xmath322 and @xmath323 as given in , . the combination of the last three equations implies that for large enough @xmath0 , @xmath324 before we further analyze the expressions @xmath325 we wish to narrow down the range of pairs @xmath159 . first , we can quickly move to @xmath326 . indeed , if for instance @xmath327 then using @xmath328 an analogous calculation holds for @xmath329 , yielding that @xmath330 the case @xmath331 is treated similarly : @xmath332 plugging the last two equations in and defining @xmath333 it follows that @xmath334 recalling the definition of @xmath311 , it now remains to bound @xmath335 with @xmath336 given by @xmath337}{9e(s_i , t_j)}\right ] } \leq \frac{i\alpha_i + j\beta_j } { \log\left [ \frac{2^{i+j}(i\alpha_i + j\beta_j)}{9e(s_i , t_j)}\right]}\,.\ ] ] note that the last inequality ( where we reduced the argument of the @xmath338 ) is only legitimate provided that @xmath339 in what follows we will show that this is indeed the case and then proceed to bound @xmath340 . this will be achieved by splitting the analysis into two cases , according to the structure of @xmath341 in the base graph @xmath1 . recall that we have @xmath342 as @xmath1 is an @xmath94-graph . = 1em * case ( i ) : * @xmath343 since @xmath344 and @xmath345 , in this case we have @xmath346 with the regime of @xmath159 as in in mind , suppose first that @xmath347 . it follows that @xmath348 } \leq \sum_{\substack{i\geq 4\log d\\ j\geq 20\\ |i - j|\leq2\log d } } \frac{i\alpha_i + j\beta_j}{\log\frac{m j}{18 d \alpha_i } } \,.\end{aligned}\ ] ] as @xmath349 we have @xmath350 and plugging in the fact that @xmath351 , @xmath352 where we used the fact that @xmath353 for the above @xmath300 . note that we have just verified eq . by showing that its left - hand - side is at least @xmath354 . similarly , if @xmath355 then @xmath356 altogether , in this case we have @xmath357 * case ( ii ) : * @xmath358 rewriting the assumption on @xmath341 in terms of @xmath359 and @xmath360 , we have @xmath361 which gives that @xmath362 where the second inequality was derived from the fact that @xmath363 for any @xmath364 , and the last one by the fact that @xmath365 . notice that if @xmath366 then @xmath367 and thus @xmath368 . we therefore have @xmath369 and so @xmath370 this verifies and further implies that @xmath371 . altogether , @xmath372 combining with the two cases , for the @xmath336 s proves that @xmath373 where in the last inequality we plugged in the fact that @xmath308 . it remains to consider the case @xmath374 where by definition @xmath375 since the @xmath94-graph @xmath1 satisfies @xmath342 , we have the following two cases : = 1em * case ( i ) : * @xmath343 the above bound on @xmath376 then translates into @xmath377 and summing over all such @xmath300 while recalling that @xmath232 we get @xmath378 where we used the inequalities @xmath379 , @xmath323 and @xmath193 . * case ( ii ) : * @xmath358 here we have @xmath380 and so @xmath381 by cauchy - schwartz , @xmath382 and similarly , @xmath383 we deduce that in this case @xmath384 adding the bounds obtained for the two cases , gives @xmath385 the proof of lemma [ lem - wij - star ] is now concluded by combining the above bound with . notice that for proving proposition [ prop - e(a , b)-lift - bound ] we can assume that @xmath308 , since otherwise @xmath386 and the statement immediately follows from the trivial bound @xmath387 for @xmath308 we can apply lemma [ lem - wij - star ] ( recalling the discussion preceding this lemma ) combined with and obtain that , except with probability @xmath303 , every two subsets @xmath107 with @xmath304 satisfy @xmath388 when the subsets @xmath107 satisfy in addition @xmath389 then @xmath390 and the above bound ( for @xmath391 ) translates to @xmath392 altogether , we have established the statement of proposition [ prop - e(a , b)-lift - bound ] under the additional assumption @xmath393 for the subsets @xmath107 in mention . the separate case of @xmath394 is much simpler to handle , and is treated by the next claim using a standard first moment argument . [ cl - small - a , b ] let @xmath1 be an arbitrary graph on @xmath55 vertices and let @xmath2 be a random @xmath0-lift of @xmath1 with @xmath88 . then with probability @xmath53 , every two subsets @xmath395 of size @xmath396 have @xmath397 . suppose that @xmath107 are two subsets that satisfy @xmath396 and @xmath398 , and consider their union @xmath399 . clearly , @xmath400 whereas the number of edges in the induced subgraph on @xmath401 satisfies @xmath402 as argued below eq . , if @xmath403 are @xmath191 arbitrary distinct pairs of vertices in @xmath2 of which no @xmath404 vertices share the same fiber , @xmath405\right ) \leq ( 3/n)^k\ ] ] ( that argument applies to any base - graph @xmath1 by definition of the @xmath0-lift ) . consider @xmath191 distinct pairs of vertices in @xmath401 that may potentially be adjacent in @xmath2 . clearly , for large enough @xmath0 these do not contain any @xmath404 points on the same fiber since @xmath406 , hence the probability that they are all adjacent is at most @xmath247 . it now follows that the probability there exists a subset @xmath407 of size @xmath408 with @xmath409 is at most @xmath410 where @xmath411 is an absolute constant . as @xmath412 and @xmath413 , the base of the exponent in the last expression is at most @xmath414 whereas @xmath415 necessarily to allow @xmath416 . summing this error probability over the @xmath417 possible values of @xmath418 completes the proof . as the bound given in the above claim is clearly smaller than the bound for @xmath391 , together they imply that holds with probability @xmath53 for any two sets @xmath107 with @xmath419 . this concludes the proof of proposition [ prop - e(a , b)-lift - bound ] . we now move on to estimating the expected total of all light pairs along the edges of @xmath2 . recall that a pair @xmath141 is _ light _ if @xmath420 . to bound the bilinear form @xmath421 with respect to the light pairs , we will approximate each such vector using an @xmath122-net , where @xmath422 more precisely , we consider the @xmath89-dimensional lattice @xmath423 , and show that the required statement on the bilinear form holds for any two vectors @xmath121 with norm at most @xmath64 in this lattice . [ thm - light ] let @xmath1 be an @xmath94-graph for @xmath26 and @xmath149 and let @xmath2 be a random @xmath0-lift of @xmath1 . for @xmath143 , let @xmath424 be the random variable @xmath425 let @xmath426 denote the @xmath89-dimensional lattice @xmath427 . then except with probability @xmath428 , every @xmath429 with @xmath150 and @xmath151 satisfy @xmath430| \leq 250\ , \lambda \log d$ ] . in order to establish the above concentration result , we must first estimate the variance of @xmath424 . [ lem - variance - rl ] let @xmath1 be an @xmath94-graph , and let @xmath143 be two fixed vectors satisfying @xmath150 and @xmath431 . then @xmath432 as in the treatment of the heavy pairs , we consider the following dyadic expansion of @xmath31 and @xmath433 : @xmath434 and the assumptions @xmath150 and @xmath431 translate into @xmath435 further note that , if @xmath165 and @xmath166 then a necessary condition for @xmath436 is that @xmath437 . consider the graph @xmath438 where every two fibers that are connected in @xmath1 have a complete bipartite graph between them in @xmath438 . that is , @xmath439 in @xmath438 if @xmath440 . it follows that @xmath441 and we aim to bound the sum in the right - hand - side by at most @xmath442 . the adjacency matrix of @xmath438 is therefore precisely @xmath443 , where @xmath444 is the all - ones matrix of order @xmath0 and @xmath445 denotes tensor product , and so by the definition of @xmath1 ( and the properties of tensor products ) it follows that @xmath438 is an @xmath446-graph . as such , for any two subsets @xmath107 of its vertices , @xmath447 we now separate the sum in into two cases , comparing @xmath448 to @xmath449 to justify the last inequality , note first that we may assume that @xmath450 otherwise the statement of the lemma holds trivially . indeed , since there are @xmath451 edges in @xmath1 , summing over @xmath452 pairs , each of which contributes at most @xmath453 , gives at most @xmath454 . for @xmath455 we have @xmath456 and so this is clearly at most @xmath457 and we are done . assume therefore that @xmath458 , in which case @xmath459 and the above inequality follows from the fact that @xmath149 . in case @xmath174 we have @xmath460 . here , applying the trivial bound @xmath461 gives @xmath462 where the inequality between the two lines used . performing the same calculation for the sum over @xmath463 gives the same bound . altogether these two bounds sum up to @xmath464 and we get that for @xmath174 @xmath465 ( the factor of @xmath466 in the first inequality due to the fact that @xmath467 ) . thus , eq . translates this bound to @xmath468 and confirms the statement of the lemma for the case @xmath174 . it remains to handle @xmath180 . = 1em * case ( i ) : * @xmath469 in this case , we use the trivial bound @xmath461 , giving @xmath470 where we used the facts that @xmath471 for @xmath472 , and that @xmath473 . by , it now follows that @xmath474 and adding the symmetric case where we sum over @xmath475 , we get @xmath476 * case ( ii ) : * @xmath477 here we have two bounds according to the two expressions in the upper bound . that is , we break @xmath478 into the sum of the two expressions corresponding to @xmath479 and to @xmath480 and bound each of them separately . first , by the sum corresponding to @xmath479 contributes @xmath481 where the last inequality is due to the facts @xmath482 and @xmath483 . second , the sum corresponding to @xmath484 contributes @xmath485 where the first inequality of the last line followed from cauchy - schwartz and the last one used the fact @xmath486 . the last two inequalities now give a combined bound of @xmath487 . as the same holds for the sum over @xmath475 , altogether we have @xmath488 adding together cases ( i),(ii ) implies that when @xmath489 @xmath490 and now translates this bound to @xmath491 , confirming the statement of the lemma ( with room to spare ) for the case @xmath180 as required . next , we need to address the support of @xmath31 . a vector @xmath492 is called _ sparse _ if it has at most @xmath263 non - zero entries on each fiber , that is , if @xmath493 : x_{ij } \neq 0\}\right| \leq n/2\quad\mbox { for all } j\in v(g)\,.\ ] ] the next lemma establishes concentration for @xmath424 provided that @xmath31 is sparse . [ lem-2-fixed - sparse ] let @xmath143 be two fixed vectors such that @xmath31 is sparse , @xmath150 and @xmath431 . let @xmath1 be an @xmath94-graph for @xmath26 and @xmath149 and let @xmath2 be a random @xmath0-lift of @xmath1 . for any @xmath494 , @xmath495\right| > a \lambda \log d \right ) \leq 2d^{-amn/12}\,.\ ] ] for any @xmath440 and @xmath496 $ ] , define @xmath497 by this definition , @xmath498 and we can now expose the values of @xmath499 sequentially by going over the pairs of fibers @xmath440 one - by - one ( in an arbitrary order ) , and for each such pair revealing the relevant part of the bijection between the fibers . more precisely , when processing a given pair of fibers @xmath135 , we proceed as follows : 1 . without loss of generality , suppose @xmath500 , and let @xmath501 $ ] be the largest index such that @xmath502 . sequentially go over @xmath503 and expose the neighbor of @xmath127 in the bijection between the two fibers , i.e. , @xmath504 $ ] such that @xmath505 , thereby determining @xmath506 . crucially , since the vector @xmath31 is sparse , it contains at most @xmath263 non - zero entries in any given fiber , and so in the above defined process @xmath507 . denote by @xmath508 the filter corresponding to this process ( that is , @xmath509 is the @xmath510-algebra generated by the first @xmath511 exposed edges ) , and let @xmath512 be doob s martingale corresponding to the function @xmath424 with respect to @xmath508 : @xmath513 \,.\ ] ] as usual , @xmath514 $ ] whereas at the end of the process the martingale equals @xmath424 . we wish to analyze the increment @xmath515 . suppose that in step @xmath511 we are now exposing an edge between the fibers @xmath135 . clearly , if @xmath516 and our process already exposed the edges between the fibers @xmath517 , then their contribution is canceled in @xmath518 . furthermore , if the edges between @xmath517 are to be exposed in the future , then @xmath519 = { \mathbb{e}}\bigg[\sum_i x_{kk'}(i){\ , \big| \,}{\mathcal{f}}_{t-1}\bigg]\,,\ ] ] since the bijections between distinct pairs of connected fibers are independent . that is to say , @xmath515 contains only terms that arise from the effect of the edge exposed at time @xmath511 on @xmath520 $ ] . in light of this , it suffices to treat the case where @xmath135 is the first pair of fibers processed , and the analysis of @xmath515 will hold analogous for any other pair . in what follows , since we are now concentrating solely on the two fibers @xmath135 , we omit the subscripts @xmath521 from @xmath31 and @xmath433 to simplify the notation . similarly , we use the abbreviation @xmath522 at step @xmath523 we are therefore exposing the match of @xmath524 . for simplicity we will analyze @xmath525 and by merely changing the indices the same argument would carry to all other values of @xmath511 . recall that @xmath526 and that given the event that @xmath527 is matched to some @xmath528 $ ] we have @xmath529 let @xmath530 and @xmath531 be two possible values after revealing the match of @xmath527 , denoting its index by @xmath532 and @xmath533 respectively . we can now couple the distributions over the remaining entries of @xmath5 and @xmath534 via switching @xmath535 . that is , if we let @xmath536 then @xmath537 agree everywhere except possibly on @xmath538 and there we have @xmath539 clearly , in any pair of coupled @xmath537 all summands of the form @xmath540 in eq . cancel from @xmath541 except when @xmath542 , hence @xmath543 by definition each of the above terms @xmath540 is at most @xmath544 in absolute value , thus repeating this argument for any step @xmath511 gives that with probability @xmath64 , @xmath545 obtaining an @xmath124 bound on the increments @xmath518 is slightly more delicate then the above @xmath546 bound . to this end , we will write @xmath525 explicitly : recall from , that @xmath547 averages over permutations @xmath5 on @xmath3 $ ] whereas @xmath530 averages over all such permutations @xmath534 that have @xmath548 for some @xmath528 $ ] , which is exposed in @xmath549 and identifies the match of @xmath527 . in other words , @xmath547 is the mean of sums analogous to @xmath530 with all possible values @xmath550 replacing @xmath551 . each such value has equal probability and the case @xmath552 does not contribute to @xmath525 . in the remaining cases we can go over the possible values of @xmath553 ( each with equal probability ) and couple @xmath537 using the switching that was used to establish the @xmath546 bound , letting them agree everywhere except on @xmath538 . altogether we obtain that @xmath554 since the expressions in the last line do not depend on @xmath551 , we conclude that @xmath555 where @xmath556$}\,.\ ] ] estimating @xmath557 requires extra care due to the indicators @xmath558 . there are two possible cases : a. if @xmath559 then by our ordering of the @xmath560 s according to decreasing absolute values we have @xmath561 for all @xmath562 and so @xmath563 moreover , @xmath564 is at least the average of the @xmath565 s for @xmath566 , and so in this case @xmath567 b. otherwise , @xmath568 and there exists some @xmath569 such that @xmath561 for all @xmath570 . in this case @xmath571 by cauchy - schwartz , in this case we thus have @xmath572 combining the cases , since @xmath551 is uniform on @xmath3 $ ] it now follows that @xmath573 applying the same analysis to a general @xmath574 $ ] , while assuming without loss of generality that the remaining unmatched @xmath575 s are @xmath576 , yields @xmath577 at this point our assumption that @xmath507 due to the fact that @xmath31 is sparse plays its important role . for some @xmath578 consider the total coefficient of @xmath579 after summing over @xmath580 . the first expression in contributes at most @xmath581 whereas the second one adds up to @xmath582 altogether we conclude that @xmath583 and by extending this analysis to all @xmath584 pairs of connected fibers we get @xmath585 where the last step was by lemma [ lem - variance - rl ] . we can now apply the following large deviation inequality for martingales , which is a special case of a result of freedman @xcite ( see also @xcite for a variant of this inequality ) . let @xmath586 be a martingale with respect to a filter @xmath587 and let @xmath588 denote its increments . suppose that @xmath589 for all @xmath590 and that @xmath591 . then for any @xmath592 we have @xmath593$}\right)\leq 2\exp\left[-\frac{s^2}{2(\sigma^2+ms)}\right]\,.\ ] ] in our case by and we have @xmath594 and @xmath595 . since the final value of @xmath512 is @xmath424 whereas @xmath514 $ ] , we now get @xmath596\right| \geq s\right ) & \leq 2\exp\bigg(-\frac{s^2 mn}{8\lambda(125\lambda\log d + s)}\bigg)\,.\end{aligned}\ ] ] in particular , for @xmath597 with @xmath494 we have @xmath598 and so @xmath599\right| \geq s\right ) & \leq 2\exp\left(-\frac{s mn}{16 \lambda}\right ) \\ & = 2d^{-a mn/(16 \ln 2 ) } < 2d^{-amn/12 } \,,\end{aligned}\ ] ] as required . as a corollary , we can now infer the concentration result of lemma [ lem-2-fixed - sparse ] without requiring that @xmath121 should be sparse . [ cor-2-fixed ] let @xmath143 be two fixed vectors with @xmath150 and @xmath431 . let @xmath1 be an @xmath94-graph for @xmath26 and @xmath149 and let @xmath2 be a random @xmath0-lift of @xmath1 . then for any @xmath600 , @xmath495\right| > a \lambda \log d \right ) \leq 2 d^{-amn/24}\,.\ ] ] let @xmath143 satisfy @xmath150 and @xmath431 . define the sparse vectors @xmath601 by @xmath602 since for every @xmath300 we have that @xmath603 is precisely one of @xmath604 while the other is @xmath23 , we deduce that @xmath605 clearly , by the triangle inequality if @xmath606 then at least one of the variables @xmath607 must deviate from its mean by at least @xmath608 . for each of the pairs @xmath609 and @xmath610 we may apply lemma [ lem-2-fixed - sparse ] for a choice of @xmath611 and obtain that @xmath612\right| > ( a/2 ) \lambda \log d \right ) & \leq 2d^{-amn/24}\,,\ ] ] and the same applies to @xmath613 . the required result immediately follows . to carry the result from the above corollary to every pair of vectors in the lattice @xmath614 we need the following simple claim : there are at most @xmath615 vectors @xmath616 such that @xmath617 . let @xmath618 , set @xmath619 and consider @xmath620 , the @xmath89-dimensional ball centered at @xmath23 with radius @xmath418 . for each @xmath621 , define the set @xmath622 clearly , each @xmath623 satisfies @xmath624 , and so @xmath625 . furthermore , for any @xmath626 we have @xmath627 , and altogether , if we let @xmath628 denote the lebesgue measure on @xmath629 then @xmath630 with the last inequality following from the fact that @xmath631 for all @xmath191 . since for every @xmath31 we have @xmath632 , we now deduce that @xmath633 applying corollary [ cor-2-fixed ] to all @xmath634 with a choice of @xmath635 and then taking a union bound over all @xmath636 possible pairs now completes the proof of the theorem [ thm - light ] . first , note that since @xmath17 is an upper bound on all nontrivial eigenvalues in absolute value , we can always increase it and take @xmath149 , and this would not effect the result ( recall that the bound we target for is @xmath637 where @xmath14 ) . in this case , as @xmath638 , it suffices to show that every nontrivial eigenvalue of @xmath1 is @xmath639 except with probability @xmath52 . clearly , since the @xmath0-lift is comprised of a uniform perfect matching between any two fibers that are adjacent in @xmath1 , we have @xmath643 where @xmath644 are defined by @xmath645 and @xmath646 for @xmath647 $ ] . the assumptions on @xmath121 give that @xmath648 and furthermore , by cauchy - schwartz , @xmath649 and similarly @xmath650 . altogether , as @xmath651 are orthogonal to the trivial eigenvector , and since every nontrivial eigenvalue of @xmath137 is at most @xmath17 in absolute value , @xmath652 as required . to prove theorem [ thm-1 ] , assume the events described in theorem [ thm - heavy ] and theorem [ thm - light ] let @xmath143 satisfy @xmath640 and @xmath150 , @xmath653 . consider @xmath654 , the closet vectors to @xmath121 respectively among all vectors in @xmath655 , where @xmath614 . theorem [ thm - heavy ] and theorem [ thm - light ] now give that @xmath656 | & \leq 7000 \ , \lambda \log d\,,\\ | r_l(\tilde{x},\tilde{y } ) - { \mathbb{e}}[r_l(\tilde{x},\tilde{y } ) ] | & \leq 250 \ , \lambda \log d\,,\end{aligned}\ ] ] and since @xmath657 by definition , we get @xmath658\right| + \left|{\mathbb{e}}\left[\tilde{x}^{\textsc{t}}a_h\ , \tilde{y}\right]\right| \\ & \leq \left|r_l(\tilde{x},\tilde{y } ) - { \mathbb{e}}[r_l(\tilde{x},\tilde{y})]\right| + \left|r_h(\tilde{x},\tilde{y } ) - { \mathbb{e}}[r_h(\tilde{x},\tilde{y})]\right| + \left|{\mathbb{e}}\left[\tilde{x}^{\textsc{t}}a_h\ , \tilde{y}\right]\right| \\ & \leq \lambda+7250 \,\lambda \log d\,.\end{aligned}\ ] ] finally , by the definition of the lattice @xmath426 , both @xmath659 and @xmath660 satisfy @xmath661 and so @xmath662 and @xmath663 . therefore , for instance , @xmath664 and similarly @xmath665 and @xmath666 . combining these inequalities , it now follows that @xmath667 completing the proof .

we study symplectic properties of monodromy map of second order linear equation with meromorphic potential having only simple poles on a riemann surface . we show that the canonical symplectic structure on the cotangent bundle implies the goldman poisson structure on the corresponding character variety under monodromy map , thereby extending the recent results of the paper of m.bertola , c.norton and the author from the case of holomorphic to meromorphic potentials with simple poles .

1702.04705

the goal of this paper is to study symplectic aspects of monodromy map for second order linear equation on a riemann surface of genus @xmath1 with meromorphic potential admitting @xmath2 simple zeros . our results generalize results of the recent paper by m.bertola , c.norton and the author @xcite where the case of holomorphic potential was treated in detail ( see also an earlier paper by s.kawai for another approach to this problem @xcite ) . the condition of coordinate invariance of the equation @xmath3 on a riemann surface @xmath4 of genus @xmath1 implies that the function @xmath5 transforms as a projective connection under a coordinate change while the solution @xmath6 locally transforms as @xmath7-differential @xcite . thus any meromorphic potential @xmath8 with simple poles can be represented as @xmath9 where @xmath10 is a fixed holomorphic projective connection on @xmath4 and @xmath11 is a meromorphic quadratic differential with @xmath2 simple poles . this is a parameterization of the space of meromorphic potentials on a given riemann surface is natural since meromorphic quadratic differentials with @xmath2 simple poles form the cotangent space @xmath0 to moduli space of riemann surfaces of genus @xmath1 with @xmath2 marked points . however , there remains a freedom in the choice of the `` base '' holomorphic projective connection @xmath10 for a given riemann surface . in this paper , following @xcite we assume that @xmath10 holomorphically depends on moduli of @xmath4 ( in particular , this requirements rules out the use of fuchsian projective connection as @xmath10 ) . furthermore , following @xcite we choose @xmath10 to be the bergman projective connection @xmath12 ( up to the factor @xmath13 this projective connection is the zeroth order term in the expansion of the canonical normalized meromorphic bidifferential @xmath14 near the diagonal @xmath15 @xcite ; @xmath16 is the prime - form ) . therefore , we are going to study the equation written in the form + ( s_b+q)=0 . the ratio @xmath17 of two linearly independent solutions of ( [ sint ] ) solves the schwarzian equation @xmath18 where @xmath19 is an arbitrary local parameter on @xmath4 and @xmath20 denotes schwarzian derivative . the schwarzian equation determines a @xmath21 monodromy representation of the fundamental group @xmath22 which turns out to be liftable to an @xmath23 representation @xcite ( the lift to an @xmath24 representation is a non - trivial fact due to spinorial nature of solutions @xmath25 ) . following @xcite , we define an @xmath23 monodromy representation of equation ( [ sint ] ) directly , by an appropriate change of dependent variable @xmath6 . denote the standard generators of the fundamental group by @xmath26 ; these generators obey the relation @xmath27 . since all poles @xmath28 of the potential of equation ( [ sint ] ) are simple then both eigenvalues of monodromy matrices @xmath29 equal to @xmath30 for each @xmath31 ; all other monodromies are @xmath32 matrices satisfying certain genericity conditions @xcite . our goal is to study symplectic properties of the monodromy map for equation ( [ sint ] ) . since monodromy matrices depend on the choice of normalization point of solutions of ( [ sint ] ) it is natural to work with the corresponding character variety which we denote by @xmath33 ; a point of @xmath33 is an equivalence class of monodromy representations which differ by a simultaneous conjugation with the same matrix . the index @xmath34 indicates that the monodromies around poles @xmath35 are not generic : all of their eigenvalues are equal to @xmath30 . the main result of this paper states that the canonical poisson structure on @xmath0 implies the goldman bracket on the character variety @xmath36 under the monodromy map of equation ( [ sint ] ) , therefore generalizing the result of @xcite to potentials with simple poles . moreover , similar to @xcite and @xcite , we prove that the same statement holds for equation ( [ sint ] ) where the bergman base projective connection is replaced either by schottky or quasi - fuchsian ( bers ) projective connection . we follow the same strategy as in @xcite ; it is based on the use of the `` abelian periods '' of the quadratic differential @xmath11 as coordinates on the underlying moduli space @xmath0 . these coordinates are periods of the abelian differential @xmath37 on the canonical covering @xmath38 of @xmath4 ( the genus of @xmath38 equals @xmath39 ) . we call them `` abelian periods '' or simply `` periods '' to distinguish from monodromy matrices of equation ( [ sint ] ) which are also called sometimes `` periods of quadratic differentials '' @xcite ; following the terminology of @xcite the monodromy matrices should probably be rather called `` non - abelian periods '' . in the theory of dynamical systems the abelian periods of the quadratic differential @xmath11 are known under the name of `` homological coordinates '' ( see @xcite and numerous recent papers ; a substantial reference list can be found in @xcite ) . the phase space associated to equation ( [ sint ] ) is the space of pairs ( riemann surface @xmath4 of genus @xmath1 , meromorphic quadratic differential @xmath11 on @xmath4 with @xmath2 simple poles ) . this phase space is nothing but the cotangent bundle @xmath0 ( up to subspaces of codimension one and higher ; these subspaces contain differentials with multiple zeros ) . the natural symplectic structure on @xmath0 is defined by @xmath40 , where @xmath41 are local coordinates on the moduli space @xmath42 while @xmath43 are corresponding momenta ( coefficients in the decomposition of a cotangent vector in the basis @xmath44 ) . the set of holomorphic local coordinates @xmath45 on @xmath42 can be chosen as follows . to determine locally the conformal structure of @xmath4 we pick ( outside of hyperelliptic locus and for @xmath46 ) a set of @xmath47 entries of the period matrix @xmath48 of @xmath4 ; in different neighbourhoods of the moduli space these entries might have to be chosen differently . the quadratic differentials corresponding to cotangent vectors @xmath49 are products @xmath50 of normalized holomorphic differentials . an additional set of @xmath2 coordinates which determine the positions of punctures @xmath51 on @xmath4 we choose to be @xmath52 where @xmath53 and @xmath54 form a pair of normalized holomorphic 1-forms on @xmath4 corresponding to some torelli marking ( these coordinates are also local : in different coordinate charts on @xmath42 one might need to choose another pair of normalized holomorphic differentials and/or different torelli markings ) . the quadratic differential corresponding to cotangent vector @xmath55 is the meromorphic quadratic differential ( given by the formula ( [ qk ] ) below ) whose only simple pole is at @xmath56 . the momenta @xmath43 are then defined to be coefficients of decomposition of an arbitrary meromorphic quadratic differential with simple poles in the basis described above . an alternative set of darboux coordinates on @xmath0 ( more precisely , on the subset @xmath57 which contains quadratic differentials with all simple zeros ) is given by abelian periods of the quadratic differential @xmath11 which are defined as integrals of @xmath37 over odd part of homology group @xmath58 of canonical covering @xmath38 . the canonical two - sheeted covering @xmath38 is defined by equation @xmath59 in @xmath60 ; the branch points of @xmath38 lie at zeros and poles of @xmath11 ( in modern literature @xmath38 is sometimes attributed to seiberg - witten @xcite or hitchin @xcite , although this two - sheeted covering was extensively used starting from early days of teichmller theory @xcite ) . the genus of @xmath38 equals @xmath39 ; it admits a natural holomorphic involution which we denote by @xmath61 . the homology group @xmath58 can be decomposed into direct sum @xmath62 of even and odd subspaces under the action of @xmath61 ; the dimension of @xmath63 equals @xmath64 while @xmath65 . choosing a symplectic basis @xmath66 in @xmath67 with the intersection matrix @xmath68 we define periods of @xmath11 by a_i=_a_i^-v,0.7 cm b_i=_b_i^-v . the intersection pairing in @xmath67 defines the natural symplectic form @xmath69 which turns out to coincide with the canonical symplectic form @xmath70 on @xmath71 restricted to the space of quadratic differentials with simple zeros . moreover , we show that the function generating the transformation from canonical darboux coordinates @xmath72 to darboux coordinates given by periods @xmath73 is given by @xmath74 which generalizes the formula obtained in @xcite to the case @xmath75 . the symplectic form @xmath76 induces a poisson structure on the character variety @xmath33 via the monodromy map of equation ( [ sint ] ) . we emphasize that this monodromy map essentially depends on the choice of the base projective connection . the choice of bergman projective connection @xmath12 as the base is not unique since @xmath12 transforms non - trivially under the change of torelli marking of @xmath4 . the bers projective connection chosen as the base in @xcite carries even more freedom , since it depends on a choice of a point in the teichmller space as a parameter . nevertheless , as it was shown in @xcite , the poisson structure induced on the character variety is the same for the bergman projective connection ( independently of torelli marking used ) and bers projective connection used in @xcite ( independently of the choice of the initial point in the teichmller space ) . technically , it is convenient to work with the matrix first order equation constructed by introducing functions @xmath77 where @xmath25 are two linearly independent solutions of ( [ sint ] ) . denote by @xmath78 the wronskian matrix of @xmath79 and @xmath80 . the matrix @xmath78 satisfies the first order matrix equation = ( 0 & v + uv & 0 ) where the meromorphic function @xmath81 on @xmath4 is given by u= --1 , and @xmath82 is the schwarzian derivative of the coordinate @xmath83 with respect to a local coordinate @xmath19 . notice that the coefficients @xmath84 and @xmath85 in ( [ lsi ] ) are ( holomorphic and meromorphic respectively ) differentials on @xmath38 , not on @xmath4 itself . the poisson bracket between @xmath86 and @xmath87 ( assuming that the coordinates @xmath88 and @xmath89 are independent of moduli ) is given by the following expression : \{u(z ) , u ( ) } = _z- _ , where the bimeromorphic function @xmath90 on @xmath91 is given by @xmath92 ; the differential operator @xmath93 is known as the `` lenard s operator '' in the theory of integrable systems . the computation of the poisson bracket ( [ pbpot ] ) from the fundamental poisson bracket @xmath94 ( and , therefore , also from fundamental poisson bracket on @xmath0 ) is based on variational formulas for the canonical bidifferential @xmath95 proved in @xcite . the monodromy map for equation ( [ lsi ] ) gives an @xmath23 representation of @xmath96 . a technical computation originally performed in @xcite allows to find the poisson bracket between traces of monodromy matrices of equation ( [ lsi ] ) along two arbitrary loops @xmath97 and @xmath98 . the result is the goldman s bracket @xcite : \{trm_,m_}=_p trm__p - trm__p ^-1 where the monodromy matrices @xmath99 ; cycles @xmath100 and @xmath101 are cycles obtained by resolving the intersection point @xmath102 in two different ways ( see @xcite ) . the paper is organized as follows . in section [ cancov ] we describe main objects associated to the canonical two - sheeted covering . in section [ seceq ] we define the @xmath23 monodromy representation for equation ( [ sint ] ) via an appropriate matrix reformulation . in section [ varfo ] we derive variational formulas for the main objects : the potential of the matrix equation and for monodromy matrices . in section [ canper ] we show that the period coordinates are darboux coordinates for the canonical symplectic structure on @xmath0 . in section [ cango ] we outline the modifications which have to be made in the scheme of @xcite to cover the case of meromorphic differentials with first order poles . denote the moduli space of meromorphic quadratic differentials on riemann surface of genus @xmath1 with @xmath2 simple poles and @xmath103 simple zeros by @xmath104 . here we list a few basic facts about the canonical covering of a riemann surface determined by any quadratic differential @xmath105 . the presentation is parallel to the case of canonical covering defined by a holomorphic quadratic differential which was considered in detail in @xcite . let @xmath4 be a riemann surface of genus @xmath1 . the torelli marking is a choice of canonical basis @xmath106 in @xmath107 with the intersection index @xmath108 . let @xmath109 be the dual basis in @xmath110 normalized by @xmath111 ; integrals of @xmath53 over @xmath112-cycles give the period matrix @xmath113 . introduce the canonical meromorphic bidifferential @xmath114 , where @xmath115 is the prime - form ( sometimes @xmath95 called `` bergman kernel '' following the paper by hawley - schiffer @xcite of 1966 , although this bidifferential was already used by klein @xcite ) . the bidifferential @xmath116 is symmetric , @xmath117 and its only singularity is on the diagonal : as @xmath118 in a local coordinate @xmath19 one has b(x , y)=(+ s_b((x))+ )d(x ) d(y ) ; moreover , @xmath116 is normalized by the requirement that all of its @xmath119-periods vanish with respect to each variable ( its @xmath120-period with respect , say , to @xmath121-variable , equals @xmath122 ) . therefore , @xmath116 depends on torelli marking of @xmath4 ; under the change of torelli marking it transforms according to formula given at page 21 of @xcite . the term @xmath12 in ( [ singb ] ) transforms as projective connection under a change of coordinate @xmath19 ; it is called the `` bergman projective connection '' ( this name is inherited from the `` bergman kenel '' terminology adopted in @xcite although , probably , it would be historically more appropriate to call it `` klein s projective connection '' ) . the projective connection @xmath12 is important both in physics ( it equals to the @xmath123-component of the energy - momentum tensor of free bosons on a riemann surface @xcite ) and in mathematics ( it describes variations of a holomorphic section of determinant of hodge vector bundle over various moduli spaces @xcite , and also variations of the determinant of laplacian on a riemann surface @xcite ) . let @xmath124 be a meromorphic quadratic differential on @xmath4 with @xmath2 simple poles ( denoted by @xmath35 ) and @xmath103 simple zeros ( denoted by @xmath125 ) . the canonical covering @xmath38 is defined by equation @xmath59 in @xmath126 . the two - sheeted covering @xmath127 is branched at all poles and zeros of @xmath11 ; thus the total number of branch points is @xmath128 and the genus of @xmath38 equals @xmath129 . the abelian differential @xmath84 is holomorphic on @xmath38 ; it has zeros of order 2 ( on @xmath38 ! ) at the branch points @xmath130 and no other zeros . denote the natural involution on @xmath38 by @xmath61 and decompose the homology group @xmath58 into direct sum of the even and odd subspaces @xmath62 , where @xmath131 and @xmath65 . the holomorphic part of cohomology group @xmath132 is similarly decomposed as @xmath133 where @xmath134 and @xmath135 . the differential @xmath84 belongs to @xmath136 . introduce a set of generators of @xmath67 denoted by @xmath137 with the intersection index @xmath68 . the integrals of @xmath84 over @xmath137 a_i=_a_i^- v0.7 cm b_i=_b_i^- v0.7 cm i=1, 3g-3+n are called the ( abelian ) periods of the quadratic differential @xmath11 . the periods can be used as local coordinates on the space @xmath138 which are called `` homological coordinates '' in the theory of dynamical systems @xcite . and @xmath139 and are attributed to seiberg and witten , see for example @xcite and references therein . ] another way of looking at coordinates @xmath140 is to consider them as combinations of integrals of the differential @xmath84 between different branch points of @xmath38 . the differential @xmath84 can be used to introduce a special system of local coordinates on @xmath38 and @xmath4 . namely , if @xmath141 is a point of @xmath38 which does not coincide with branch points @xmath142 then the local parameter ( the `` flat coordinate '' ) on @xmath38 in a neighbourhood of @xmath141 is given by z(x)=_x_1^x v where @xmath143 is a chosen `` first '' zero of @xmath84 . near branch points @xmath142 of @xmath38 the local coordinates on @xmath38 are given by ( these coordinates are called `` distinguished '' ) : _ i(x)=^1/3,0.7 cm i=1 , , 3g-3+n ; near @xmath56 the distinguished local coordinate is @xmath144 the `` flat '' local coordinates on @xmath4 near every point except zeros or poles of @xmath11 are the same as on @xmath38 ( also defined up to a sign ) . on the base curve @xmath4 the distinguished local coordinate near point @xmath145 is given by _ i(x)=^2/3 , and near @xmath56 : _ k(x)=^2 . to work with the canonical covering @xmath38 it is convenient ( following lemma 3.1 of @xcite ) to choose a special set of generators @xmath146 , @xmath147 of the fundamental group @xmath148 which satisfy the relation ( _ i=1 ^ 4g-4+n_x_i ) ( _ i=1^n _ y_i ) _ i=1^g_i_i_i^-1_i^-1=id . the topology of the covering @xmath38 determines a representation @xmath149 of @xmath150 in the symmetric group @xmath151 . in parallel to lemma 3.1 of @xcite one can prove that the generators @xmath152 can be chosen in such a way that h(_i)=h(_i)=(21)0.7 cm i=1 , , g while h(_x_i)=h(_y_j)=id,0.7 cm i=1, ,4g-4+n,0.7 cm j=1, ,n . then one can cut @xmath38 along loops representing generators @xmath153 and @xmath154 on both sheets of the covering ; in this way one gets the two - sheeted covering @xmath155 of the fundamental polygon @xmath156 of @xmath4 branched at @xmath128 points @xmath142 and @xmath51 . it is convenient to connect the branch points by the following system of branch cuts : , 0.7 cm i=1, ,n , , 0.7 cm i=1, ,2g-2 . under this choice of branch cuts the covering @xmath155 becomes a two - sheeted branched covering of the fundamental polygon @xmath156 with branch cuts chosen as shown in fig.[chat0 ] . with loops @xmath157 and @xmath158.,width=340 ] 3.0 cm denote by @xmath159 the corner of @xmath156 . to study the monodromy group of the equation ( [ sint ] ) we are going to introduce a system of @xmath2 loops @xmath160 on @xmath155 ( fig.[chat0 ] ) chosen such that the loop @xmath157 goes around the branch cut @xmath161 $ ] . projections of these loops to @xmath156 will be denoted by the same letters . we are going to write the second order linear equation on a riemann surface ( the `` schrdinger equation '' ) in the form [ schrint1 ] + ( s_b+ ) = 0 where the derivative is taken with respect to some local coordinate @xmath19 on @xmath4 . the solution @xmath162 of ( [ schrint1 ] ) is locally a @xmath7-differential which we write as @xmath163 . choosing two linearly independent solutions of ( [ schrint1 ] ) construct the wronskian matrix ( ) = ( _ 1 & _ 2 + _ 1 _ & _ 2 _ ) . this matrix satisfies the equation = ( 0 & 1 + - s_b()- ( ) & 0 ) where we use the notation @xmath164 and @xmath165 . let the quadratic differential @xmath166 have @xmath2 simple poles @xmath167 and @xmath103 simple zeros @xmath130 . introduce the canonical cover @xmath38 by the equation @xmath168 . the zeros of the abelian differential @xmath84 on @xmath38 all have multiplicity @xmath169 , and therefore @xmath170 is a section ( holomorphic on @xmath38 and unique up to a sign ) of a spin line bundle over @xmath38 . following @xcite we define ( ) : = ( ) v^1/2 ( ) . then the scalar equation ( [ schrint1 ] ) takes the following form in terms of @xmath171 : ( ) -uv=0 , [ inv2 ] where @xmath81 is the meromorphic function on @xmath4 given by u= --1 . and @xmath172 is the meromorphic projective connection on @xmath4 associated to the differential @xmath84 which is given by the schwarzian derivative of the flat coordinate @xmath88 : s_v()=(_x_1^x v , ) = ( ) -()^2 . choosing two linearly independent solutions @xmath173 of ( [ inv2 ] ) we construct the wronskian matrix @xmath78 which is related to the matrix @xmath174 ( [ phidef ] ) as follows = ( _ 1 & _ 2 + d_1/v & d_2/v ) v^1/2 ( ) . the matrix @xmath78 satisfies the equation = ( 0 & v + uv & 0 ) . matrix entries @xmath84 and @xmath85 of the coefficient matrix of ( [ matreq ] ) are differentials on @xmath38 ( @xmath84 is holomorphic while @xmath85 is meromorphic ) which are anti - symmetric under the involution @xmath61 . the advantage of the matrix @xmath78 over the original wronskian matrix @xmath174 is that the matrix entries of the former are functions i.e. they are independent of the choice of the local parameter @xmath19 while the matrix entries of the latter non - trivially depend on the choice of @xmath19 . the price which is paid for this independence is the appearance of the canonical covering @xmath38 in the equation satisfied by @xmath78 . therefore , @xmath78 gets additional monodromies @xmath175 around zeros @xmath145 ( @xmath174 is of course monodromy - free around @xmath145 ) . the same factors @xmath175 appear when comparing monodromies of ( [ matrixphi ] ) and ( [ matreq ] ) around poles @xmath28 . another way to rewrite equation ( [ matreq ] ) is to use the flat coordinate @xmath176 on @xmath38 ( @xmath88 can be used as a local coordinate on @xmath4 and @xmath38 outside of neighbourhoods of points @xmath145 and @xmath56 ) . then @xmath177 and the scalar equation ( [ inv2 ] ) can be written as _ zz+u=0 while the matrix equation ( [ matreq ] ) takes the form = ( 0 & 1 + u(z ) & 0 ) . there are various ways to associate a monodromy representation of the fundamental group @xmath178 to equations ( [ schrint1 ] ) , ( [ matrixphi ] ) and ( [ matreq ] ) . due to the sign ambiguity of the spinors @xmath25 the equation for @xmath174 ( [ matrixphi ] ) immediately determines only a @xmath21 monodromy representation which coincides with the monodromy representation of the schwarzian equation for function @xmath17 : ( f,)=s_b+2q it is a non - trivial problem whether or not this representation can be lifted to a @xmath23 representation ( see @xcite ) . on the other hand , if we write down the equation in terms of matrix @xmath78 ( [ matreq ] ) an @xmath23 representation can be naturally constructed ( as shown in @xcite for @xmath179 ) . let us show how an @xmath23 monodromy representation of @xmath178 can be associated to equation ( [ matreq ] ) for @xmath180 . the subtlety is in the fact that both differentials , @xmath84 and @xmath85 , forming the matrix of coefficients of the system , are differentials on @xmath38 , not on @xmath4 . however a natural @xmath23 representation of the fundamental group of @xmath181 can be defined as follows . assume that the generators of the fundamental group @xmath148 are chosen according to ( [ hab ] ) , ( [ hgg ] ) and consider the first sheet @xmath182 of the two - sheeted covering @xmath155 of the fundamental polygon @xmath156 ( fig.[chat0 ] ) . as before , the loops @xmath157 , @xmath183 are chosen to encircle the branch cuts @xmath184 $ ] ; introduce also the loop @xmath158 excircling all the remaining branch cuts @xmath185 $ ] for @xmath186 . since the coefficients of equation ( [ matreq ] ) are single - valued in the multiply - connected domain obtained by deleting all branch cuts from @xmath182 , one can define monodromy matrices along the loops @xmath153 , @xmath154 , @xmath157 and @xmath158 corresonding to initial point @xmath159 . the monodromy matrix along @xmath158 equals @xmath187 since ( see lemma 6.2 of @xcite ) the monodromy around each @xmath145 arises only due to the factor of @xmath170 relating matrices @xmath174 ( which is monodromy - free around @xmath145 ) and @xmath78 ; this monodromy equals @xmath31 . since the number of zeros encircled by @xmath158 is a multiple of 4 , the total monodromy of @xmath78 along @xmath158 equals @xmath187 . the monodromies along remaining loops @xmath157 , @xmath188 , @xmath189 satisfy therefore the required relation m__n m__1_i = g^1 m_a_i m__i m__i^-1m__i^-1=i i.e. these monodromies form an anti - representation of the fundamental group of @xmath190\}_{i=1}^n$ ] . since the monodromies of @xmath78 around @xmath145 are scalar factors of @xmath31 arising from elementary transformation between @xmath174 and @xmath78 we shall _ define _ the monodromies around poles @xmath28 to be matrices @xmath191 . these matrices obviously correspond to @xmath21 monodromies of the associate schwarzian equation ( f,)=s_b()+q ( ) . therefore , we get the following the constructed @xmath23 monodromy representation of matrix equation ( [ matreq ] ) is a lift of the @xmath21 monodromy representation of the schwarzian equation ( [ schweq ] ) . the equation ( [ sint ] ) can be written in terms of a local coordinate @xmath19 near a pole @xmath28 in the form @xmath192 the local analysis shows that the monodromy of @xmath174 around @xmath28 has two coinciding eigenvalues equal to @xmath193 . due to an additional factor of @xmath194 and @xmath31 arising from monodromies of @xmath170 around @xmath28 and @xmath145 , respectively the eigenvalues of monodromy @xmath29 of @xmath78 along @xmath195 are also coinciding and equal to @xmath196 . the variational formulas on the moduli space @xmath138 describe the dependence of the pariod matrix @xmath48 , normalized defferentials @xmath53 and the canonical bidifferential @xmath95 on moduli , which in present setting are given by periods @xmath140 . to write down these formulas in our present setting we introduce a set of generators @xmath197 in @xmath67 ( for example one can choose this set to coincide with the set @xmath137 ) and introduce the periods @xmath198 . denote by @xmath199 the set of generators dual to @xmath200 with intersection index @xmath201 ; the set of dual periods corresponding to @xmath137 is given by @xmath202 . the variational formulas on the space @xmath138 given below can be obtained by reduction of the variational formulas on spaces of holomorphic abelian differentials @xcite to a subspace consisting of riemann surfaces admitting a holomorphic involution . these formulas are only a slight modification of variational formulas on spaces of holomorphic quadratic differentials @xcite . let us introduce @xmath1 functions on @xmath38 : f_i(x)= and the function of two variables on @xmath38 b(x , y)= the variational formula for he period matrix @xmath48 looks as follows : = _ s_i^ * f_j f_k v. [ varo ] for the holomorphic normalized differentials and for the canonical bidifferential one has : = _ s_i^ * f_j(t ) b(t , y)v(t ) , [ varv ] and = _ s_i^ * b(x , t ) b(t , y)v(t ) , [ varb ] where @xmath203 and all derivatives are computed keeping @xmath204 and @xmath205 constant . the fundamental polygon of @xmath38 used to define @xmath206 and @xmath207 must be invariant under the involtion @xmath61 , such that the coordinate @xmath206 satisfies the relation @xmath208 for @xmath141 lying inside of the fundamental polygon . notice that , although the integrand in the right - hand side of ( [ varb ] ) has poles of second order at the branch points @xmath145 , its residues at these points vanish since this integrand is anti - symmetric under @xmath61 . integrating ( [ varb ] ) with respect to variable @xmath121 between any two points @xmath209 and @xmath210 along a contour @xmath211 one gets a variational formula for the normalized ( integrals over all @xmath119-periods not intersecting contour @xmath211 vanish ) differential @xmath212 of third kind with poles at @xmath213 and residues @xmath30 : = _ s_i^ * w_p_1,p_2 ( t ) b(t , x)v(t ) . [ varw ] taking the limit @xmath118 in the formula ( [ varb ] ) one gets the variational formula for the potential @xmath214 ( also at @xmath206 constant ) : = - _ s_i^ * h(x , t)v(t ) [ varq ] . where h(x , t)= the differential @xmath215 has poles of order 4 on @xmath38 at @xmath216 and @xmath217 with non - trivial residues . therefore , the choice of the class of the integration path in @xmath218 in the right - hand side of ( [ varq ] ) is important ; this choice must be made in such a way that the integration goes along parts of the boundary of the fundamental polygon of @xmath38 ; the fundamental polygon should coincide with the one used to define the coordinate @xmath206 which is kept fixed in the left - hand side of ( [ varq ] ) . let @xmath219 be solution of the matrix equation ( [ psiz ] ) normalized at the point @xmath220 via @xmath221 . let us introduce the following auxiliary matrix : ( x)=^-1(x ) _ - ( x ) [ lambdam ] where @xmath222 . the matrix @xmath223 satisfies the third order equation @xcite _ zzz-4 u(z ) _ z-2 u_z = 0 . [ 3rdor ] the components of the matrix @xmath224 are given by @xmath225 and @xmath226 ; therefore ( [ 3rdor ] ) is nothing but a well - known third order equation for products of two solutions of second order equation ( [ linearz ] ) . dependence of @xmath227 ( for fixed @xmath206 ) on moduli is given by the following proposition the following variational formula holds for solution of ( [ matreq ] ) normalized by @xmath228 : ^-1|_z(x)=const= -_x_0^x ( x ) for any @xmath229 . the integration path @xmath230 in the r.h.s . of ( [ varpsi ] ) goes along the boundary of the fundamental polygon of @xmath38 used to define the abelian integral @xmath206 . the proof of this proposition is a direct application of the variation of parameters formula for the non - homogeneus linear equation using the variational formula ( [ varq ] ) for the coefficient of the equation . an immediate corollary of the formula ( [ varpsi ] ) is the following variational formula for monodromy matrices . let @xmath231 be any element of @xmath232 . denote by @xmath233 the monodromy matrix of equation ( [ matreq ] ) along @xmath231 ( as defined in section [ mondef ] ) . then the following variational formulas hold : m_^-1=- _ ( x ) . let @xmath234 be a set of holomorphic local coordinates on @xmath42 . then for any cotangent vector @xmath102 we can define its coordinates @xmath43 via @xmath235 . introduce the canonical symplectic structure @xmath70 on @xmath0 : _ can=_i=1 ^ 3g-3+n p_iq_i and the corresponding symplectic potential ( the liouville 1-form ) _ can=_i=1 ^ 3g-3+n p_i q_i define the `` homological '' symplectic structure on @xmath236 in terms of periods @xmath140 : _ hom=2_i=1 ^ 3g-3+n a_ib_i and its symplectic potential _ hom=2_i=1 ^ 3g-3+n a_i db_i notice that the form @xmath237 is independent of the choice of basis @xmath238 in @xmath67 but @xmath239 depends on this choice . on the other hand , the 1-form @xmath240 ( such that @xmath241 ) is also independent of a choice of symplectic basis in @xmath67 . consider first the case @xmath179 and exclude the hyperelliptic locus from consideration . the natural local coordinates on the moduli space @xmath242 of unpunctured riemann surfaces of genus @xmath1 can be obtained using the torelli theorem stating that the complex structure of a riemann surface @xmath4 is uniquely determined by its period matrix @xmath48 . therefore , in a neighbourhood of any point of @xmath242 there exists a subset @xmath243 of @xmath47 matrix entries of @xmath48 which can be chosen as local coordinates @xmath244 , @xmath245 on @xmath242 . the cotangent vectors @xmath49 to @xmath242 at a given point can be identified with holomorphic quadratic differentials @xmath50 : d_jk0.5 cm ~0.5 cm v_j v_k since variation of @xmath246 under an infinitesimal deformation @xmath247 of a riemann surface defined by a beltrami differential @xmath61 ( @xmath248 can be viewed as an element of the tangent space to @xmath242 at a given point ) is given by the rauch variational formula @xmath249 . the analog of rauch s formulas on the space @xmath250 is given by ( [ varo ] ) which also confirms the identification of the tangent vector @xmath49 with holomorphic quadratic differential @xmath50 . for an arbitrary @xmath2 we are going to consider separately the case @xmath251 and the low genus cases @xmath252 and @xmath253 . local holomorphic coordinates on @xmath42 can be introduced in various ways . we are going to use the following set of coordinates : * a set of @xmath47 entries @xmath246 @xmath245 of the period matrix . these coordinates determine the complex structure of @xmath4 . the cotangent vector @xmath254 is represented by the holomorphic quadratic differential @xmath50 . these holomorphic quadratic differentials span a @xmath47-dimensional subspace in @xmath255 - dimensional cotangent space @xmath256 . * the additional @xmath2 coordinates ( outside , possibly , a a subset of @xmath257 of codimension 1 ) we are going to choose as follows : w_k=(y_k ) where @xmath53 and @xmath54 is an arbitrary pair of normalized holomorphic differentials on @xmath4 such that @xmath258 . the differential @xmath259 has to be chosen differently in different neighbourhoods on the moduli space . [ the existence of a holomorphic differential non - vanishing at any given point @xmath56 can be easily proved . namely , assuming that all holomorphic differentials vanish at @xmath56 we conclude that all periods of the second kind differential @xmath260 vanish ; thus the antiderivative of @xmath260 is a meromorphic function whose only simple pole is at @xmath56 which is a contradiction ] . + the cotangent vector @xmath261 can be represented by the following ( generically meromorphic ) differential @xmath262 whose only pole of first order is at the marked point @xmath56 : q_k(t)= b(t , y_k ) where @xmath95 is the canonical bimeromorphic differential . if the numerator of ( [ qk ] ) vanishes at @xmath56 then the quadratic differential @xmath262 is holomorphic ; thus the function @xmath263 can not be used as local coordinate on the subspace ( of codimension 1 ) of @xmath42 defined by equation @xmath264 . for our purposes , however , it is sufficient to cover @xmath42 by coordinate charts outside of a subspace of codimension one since all equalities of forms we derive can be extended to the whole space by analyticity . + to verify that ( [ qk ] ) indeed corresponds to cotangent vector @xmath265 we use variational formulas ( [ varv ] ) to get = _ s_i^ * let @xmath267 . the dimension of @xmath266 equals @xmath2 . the first coordinate @xmath268 on @xmath269 is the period @xmath48 . the holomorphic quadratic differential corresponding to the cotangent vector @xmath270 is given by @xmath271 , in analogy to the higher genus case . the remaining @xmath272 coordinates @xmath273 are defined by integrating the normalized differential @xmath274 between poles of @xmath11 : q_k=_y_1^y_k v_1 for @xmath275 . the meromorphic quadratic differential with simple poles at @xmath276 and @xmath277 which corresponds to the cotangent vector @xmath278 equals q_k(t)= v_1(t ) w_y_1 y_k(t ) where @xmath279 is the normalized ( its @xmath119-period vanishes ) differential of third kind on @xmath4 with simple poles at @xmath141 and @xmath121 and residues @xmath196 and @xmath280 , respectively . the correspondence between the cotangent vector @xmath278 and the meromorphic quadratic differential @xmath262 can be verified by integrating variational formulas ( [ varv ] ) for the differential @xmath274 between @xmath276 and @xmath277 which gives = _ s_i^ * , 0.7 cm k=2, ,n consider the non - trivial case @xmath282 ; then @xmath283 . define coordinates @xmath284 on @xmath285 as follows : q_k= _ y_3^y_k+3 w_y_1 y_2 for @xmath286 , where as before w_y_1 y_2 = ( -)dx is the meromorphic differential of third kind on the riemann sphere . according to general variational formulas for the third kind differential @xmath287 ( [ varw ] ) the quadratic differential corresponding to cotangent vector @xmath55 is given by q_k= w_y_1 y_2 w_y_3 y_k+3 computing the coordinates @xmath284 ( [ qk0 ] ) and the corresponding meromorphic differentials @xmath262 ( [ qk0 ] ) explicitly , we get q_k= and q_k= ( dx)^2 the following theorem generalizes theorem 4.1 of @xcite to the case @xmath180 . being restricted to the moduli space @xmath288 , the canonical symplectic form ( [ omcan ] ) coincides with the homological symplectic form ( [ omhom ] ) : _ can=_hom . the canonical symplectic potential ( [ lican ] ) is expressed in terms of homological coordinates as follows : _ can= _ i=1 ^ 3g-3+n ( a_i db_i - b_i da_i ) , and , therefore , the function generating the transition between canonical and homological darboux coordinates is given by g=_i=1 ^ 3g-3+na_i b_i . for @xmath179 the proof is given in @xcite , th.4.1 . let @xmath75 . consider the case @xmath251 . at generic point of the moduli space @xmath288 we can represent the quadratic differential @xmath124 as a linear combination of differentials @xmath262 ( [ qk ] ) and holomorphic differentials @xmath50 where @xmath245 for some subset @xmath243 of matrix entries of the period matrix @xmath48 : q=_(jk)d p_jk v_j v_k + _ l=1^n p_l q_l , or , since @xmath289 , v=_(jk)d p_jk + _ l=1^n p_l . integrating this relation over cycles @xmath290 and @xmath291 we get a_i=_(jk)d p_jk_a_i^-+_l=1^n p_l _ a_i^- and b_i=_(jk)d p_jk_b_i^-+_l=1^n p_l _ b_i^- . taking into account variational formulas ( [ varwk ] ) and ( [ varo ] ) we rewrite ( [ aipp ] ) and ( [ bipp ] ) as follows : a_i=_(jk)d p_jk+_l=1^n p_l and b_i=-_(jk)d p_jk-_l=1^n p_l . therefore , @xmath292 = _ ( jk)dp_jkq_jk+_l=1^n p_l q_l = _ can . this completes the proof of ( [ thetaab ] ) for @xmath251 ; applying @xmath293-operator to this relation we get ( [ omom ] ) . for @xmath294 the proof is analogous . say , in @xmath252 case the sum of residues of abelian differential @xmath295 equals @xmath34 ; therefore , @xmath11 can be represented as a linear combination of @xmath271 and quadratic differentials ( [ qkgen1 ] ) : q = p_1 v_1 ^ 2 + _ k=2^n p_k q_k . dividing ( [ qqk ] ) by @xmath84 , integrating the result over cycles @xmath290 and @xmath291 and using variational formulas for @xmath48 and abelian differentials @xmath296 we get a_i = p_1 + _ k=2^n p_k and b_i =- p_1 -_k=2^n p_k which again implies ( [ thetaab ] ) . consider now the case @xmath253 . in genus 0 any quadratic differential with at most simple poles at @xmath35 can be represented as a linear combination of @xmath297 quadratic differentials @xmath262 ( [ qk0 ] ) ; thus v=_k=1^g-3 p_k for some coefficients @xmath298 . furthermore , variational formulas ( [ varw ] ) for @xmath299 imply @xmath300 which again leads to ( [ thetaab ] ) . the statement ( [ omom ] ) about coincidence of the homological and canonical symplectic structures was contained in the paper @xcite written in 2012 ( section 7.3.2 ) in the context of meromorphic quadratic differentials with second or higher order poles . however , the supporting argument in @xcite was given on the `` physics '' level of rigour . on the other hand , theorem 4.1 of @xcite for @xmath179 and theorem [ theo1 ] of this paper ( for any @xmath2 ) are proved rigorously . moreover , in our theorem [ theo1 ] and corollary 4.1 of @xcite we get a stronger result : we do nt only prove the coincidence of the symplectic forms but also explicitly compute the generating function ( [ genfun ] ) between two systems of darboux coordinates . the correspondence of notations used here ( as well as in @xcite ) and notations used in @xcite is as follows : the homological coordinates ( the `` abelian periods '' @xmath301 ) of @xmath11 are called in @xcite the `` seibeg - witten central charge functions '' and denoted by @xmath302 . the generating function ( [ genfun ] ) of this paper coincides with function @xmath303 of @xcite called there the `` seiberg - witten prepotential '' . the terminology of this paper follows classical works on teichmller theory ; in particular , periods of @xmath304 were already used in 1975 paper @xcite devoted to the theory of strebel differentials . the canonical symplectic structure on @xmath0 ( or , equivalently , the symplectic structure ( [ omhom ] ) on the space @xmath288 ) induces a symplectic and poisson structure on the space of coefficients @xmath81 of the equation ( [ matreq ] ) . the poisson bracket between @xmath86 and @xmath87(for constant @xmath88 and @xmath89 ) can be computed using the variational formula ( [ varq ] ) and theorem [ canonhomol ] : @xmath305 which gives , in analogy to the proof of proposition 4.4 of @xcite : \{u(z ) , u ( ) } = _z- _ where @xmath306 is called the `` lenard s operator '' in the theory of kdv equation @xcite . in the computation of the bracket ( [ poissonu ] ) it is assumed that the arguments @xmath88 and @xmath89 of @xmath81 are independent of moduli . the bracket ( [ poissonu ] ) does not imply poisson brackets between monodromy matrices of the equation ( [ matreq ] ) themselves since these monodromies depend also on the choice of the basepoint @xmath159 . however , the bracket ( [ poissonu ] ) defines a poisson structure on the character variety @xmath33 which consists of equivalence classes of monodromy representations ( moduli simultaneous conjugation by some matrix ) . we add the index @xmath34 to the notation @xmath33 to indicate that each of the monodromy matrices around poles @xmath28 have coinciding eigenvalues ( @xmath193 or @xmath280 ) . the coordinates on @xmath33 can be chosen to be traces of monodromy matrices @xmath233 for a sufficiently large set of loops @xmath231 . the following proposition is rather technically tedious ; it is proved in complete analogy to the case @xmath179 given in theorems 7.2 and 7.3 of @xcite : let @xmath97 and @xmath98 be two closed contours on @xmath307 . if @xmath97 and @xmath98 do not intersect then the traces of the corresponding monodromy matrices of equation ( [ matreq ] ) poisson - commute , \{m_,m_}=0 . if @xmath97 and @xmath98 intersect transversally at one point @xmath159 with @xmath308 then \{m_,m_}=(m_m_-m_m^-1 _ ) . [ gold ] from this proposition one can deduce the validity of goldman s bracket for any two contours @xmath97 and @xmath98 by finding a sufficiently large supply of loops on @xmath4 which intersect at no more than one point . the canonical poisson structure on @xmath0 implies the goldman s bracket for traces of monodromies of equation ( [ matreq ] ) , i.e. for any two loops @xmath309 one has \{m_,m_}= _ p ( m__p - m__p^-1 ) [ goldman ] where @xmath310 and @xmath311 are two ways to resolve the intersection point @xmath102 to get two new contours @xmath310 and @xmath311 for each @xmath102 . the proof is parallel to the proof of theorem 7.4 of @xcite . namely , any two loops from the set consisting of the following @xmath312 elements of @xmath96 : = \{_i , _ i , 1 i g ; _ i_j , i < j ; _ i_j , i j ; _ k , _ k_i , _ k_i , k=1, ,n , i , j=1, ,g } [ test ] either do nt intersect or intersect at only one point ; thus ( [ goldman ] ) holds for any pair of loops @xmath313 . notice that the number of functions @xmath314 for @xmath315 is always greater than the dimension @xmath316 of the character variety @xmath33 ; however , it remains to show that at generic point of @xmath33 the differentials of @xmath314 generate the whole cotangent space to @xmath33 . by simultaneous conjugation of all monodromies we can transform the matrix @xmath317 to a diagonal form ( generically ) , and the matrix @xmath318 to a matrix whose fixed point ( of associated mbius transformation ) equals 1 . then from the triple of numbers @xmath319 we find matrices @xmath317 and @xmath318 . then to find any other matrix @xmath320 one can use the triple of loops @xmath321 also contained in in the set @xmath322 . in turn , knowing all matrices @xmath320 as well as @xmath318 one can reproduce the matrices @xmath323 since for each @xmath324 the set @xmath322 contains @xmath325 and @xmath326 for @xmath327 , which is sufficient to determine @xmath328 for @xmath329 ( up to a binary choice ) . now we have enough information to determine @xmath191 from knowing @xmath330 for all @xmath331 ( actually , each of @xmath191 has only one unknown entry . the only remaining unknown matrix is @xmath332 ; this matrix can be determined knowing @xmath333 , @xmath334 and using the relation ( [ relmon ] ) as discussed in @xcite . the map from the space of meromorphic quadratic differentials to the character variety @xmath33 via the monodromy map of the schrdinger equation is highly non - canonical : it essentially depends on the choice of the base holomorphic section of the affine bundle of projective connections over an appropriate covering of the moduli space . for example , the bergman projective connection used in @xcite ( as well as in this paper ) depends on the choice of canonical basis in homology group of @xmath4 , and therefore is a holomorphic section of the affine bundle of projective connections over the torelli space . in general , to be able to compare different holomorphic sections of the affine bundle of projective connections it s convenient to consider the affine bundle @xmath335 of holomorphic projective connections over the teichmller space @xmath336 . choosing any other holomorphic section @xmath10 of @xmath335 we could also write the schrdinger equation in the form + ( s_0+q)=0 in @xcite we have discussed alternative ways of fixing @xmath10 : the schottky or quasi - fuchsian ( bers ) projective connections ( the most common fuchsian projective connection is not suitable in this framework since it depends non - holomorphically on moduli of @xmath4 ) . in particular , we have shown that if @xmath10 is the schottky projective connection then the canonical symplectic structure on @xmath337 also implies the goldman poisson structure on the character variety @xmath338 . the same statement for the case when @xmath10 is given by bers projective connection follows from the paper by s.kawai @xcite . [ we recall that the schottky projective connection is given by schwarzian derivative @xmath339 with respect to any local coordinate @xmath340 , where @xmath341 is the schottky uniformization coordinate . the bers projective is given by the schwarzian derivative @xmath342 where @xmath343 is the coordinate in the fundamental domain of the kleinian group defining simultaneous uniformization of @xmath4 and some `` fixed '' riemann surface @xmath156 with anti - holomorphic complex structure , i.e. there exists in fact infinitely many bers projective connections ; these projective connections are labeled by points of teichmller space ] . therefore , all of the choices of the base projective connection @xmath10 listed above are equivalent from symplectic point of view , which inspires the following definition : a holomorphic section @xmath10 of the affine line bundle @xmath335 over the teichmller space @xmath344 is called _ admissible _ if the canonical symplectic structure on @xmath0 implies goldman s bracket on the character variety @xmath36 under the monodromy map of equation ( [ schalt ] ) . according to discussion of @xcite , two holomorphic sections , @xmath10 and @xmath345 , of @xmath335 are equivalent iff there exists a holomorphic function @xmath346 on @xmath336 such that _ g_01= _ ( s_0-s_1 ) , where @xmath61 is an arbitrary beltrami differential . the function @xmath346 is the generating function of the corresponding symplectomorphizm @xmath347 of @xmath348 @xcite . clearly , the equivalence of @xmath10 and @xmath345 does not depend on the number of punctures @xmath2 . since the equivalence of bergman , schottky and bers projective connections follows from the analysis of @xmath179 case contained in @xcite , we can formulate the following corollary of theorem [ mainth ] . if the holomorphic section of @xmath335 in equation ( [ schalt ] ) is chosen to be either bergman ( corresponding to any torelli marking ) , schottky ( corresponding to any choice of generators ) or bers ( corresponding to any `` base '' riemann surface @xmath156 ) projective connections then the canonical symplectic structure on @xmath71 induces the goldman bracket on the character variety @xmath33 under the monodromy map of equation ( [ schalt ] ) . the definition [ defad ] allows to formulate our theorem [ theo1 ] simply as follows : 0.3 cm _ `` the bergman projective connection is admissible for any @xmath2 and @xmath1''_. 0.3 cm actually , the set of admissible holomorphic sections of @xmath335 is rather tiny ; one of examples of non - admissible holomorphic section of @xmath335 is given in remark 5.1 of @xcite . suppose now that @xmath10 and @xmath345 are two admissible holomorphic sections of @xmath335 and write down the same equation in two ways : - ( s_0 + q_0)=0 and - ( s_1 + q_1)=0 where @xmath349 and @xmath350 are two meromorphic quadratic differentails with simple poles related by q_0-q_1=(s_1-s_0 ) then the canonical coverings @xmath155 defined by @xmath351 and @xmath352 defined by @xmath353 have different conformal structure . moreover , the periods @xmath354 of @xmath355 and @xmath356 of @xmath274 are not related to each other in any simple way ( even if @xmath10 and @xmath345 are bergman projective connections corresponding to two different torelli markings ) . however , since both @xmath10 and @xmath345 are admissible , the goldman bracket on the character variety of equation ( [ eq1 ] ) , ( [ eq2 ] ) implies the coincidence of the homological symplectic forms defined by @xmath349 and @xmath350 : _ i=1 ^ 3g-3+n da^0_id b^0_i = _ i=1 ^ 3g-3+n da^1_id b^1_i i.e. each admissible projective connection defines its own set of darboux coordinates for goldman s bracket . we do nt know how the relation ( [ homhom ] ) can be verified directly , without using the link with goldman s brackets . in the theory of isomonodromic deformations of a linear system @xmath357 the main role is played by the isomonodromic jimbo - miwa tau - function defined by equations = |_x_i a straightforward analog of the definition ( [ iso ] ) in the context of equation ( [ lsi ] ) : = ( 0 & v + uv & 0 ) where uv= --v , looks as follows : : = _ s_i^ * ( + 2 v)=_s_i^*(u+1)v where the cycles @xmath358 form a symplectic basis in the odd part @xmath67 of @xmath359 ; the term @xmath360 is added to provide compatibility of equations ( [ deftau ] ) . we notice that the addition of the analogous term proportional to @xmath361 to the formula ( [ iso ] ) does not change the right - hand side since all residues of @xmath361 vanish . in the case of ( [ deftau ] ) the addition of this counter - term is crucial . using the form ( [ potfor2 ] ) of the potential @xmath81 the equations ( [ deftau ] ) can be equivalently rewritten as follows in terms of the bergman projective connection : = -_s_i^ * . the tau - function defined by ( [ defberg ] ) equals the 1/8th power of the bergman tau - function @xmath362 on the moduli space of quadratic differentials studied in @xcite : = _ b^1/8 alternatively the definition ( [ defberg ] ) of the function @xmath363 can be written as follows : _ b= _ i=1 ^ 3g-3+n ( _ ib_i-_ia_i ) where _ i=_a_i^- ( u+1)v0.7 cm _ i=_b_i^- ( u+1)v and @xmath140 are , as before , the periods of the differential @xmath84 over cycles @xmath290 and @xmath291 .. the function @xmath363 is holomorphic and non - singular if zeros and poles of @xmath11 do nt merge . it satisfies the following two main properties : * transformation under change of torelli marking ( used to define the reference projective connection @xmath12 ) by an @xmath364 matrix @xmath365 : ^= ^6 ( c+d)where @xmath158 is an 8th root of unity * homogeneity property : ( q , ) = ^(5g-5+n ) ( q , ) which implies in particular the following relation between periods @xmath140 of the differential @xmath84 and the periods @xmath366 of the differential @xmath367 : _ i=1 ^ 3g-3+n ( _ i b_i-_i a_i)= ( 5g-5+n ) . variational formulas on spaces of quadratic differentials which are the main analytical tool for computation of poisson brackets in this paper and in @xcite were developed in @xcite with the goal of studying the properties of the bergman tau - function @xmath362 ; the function @xmath362 plays an important role in various areas - from the theory of isomonodromic deformations , random matrices and the theory of frobenius manifolds to spectral theory . we expect the function @xmath363 defined in present context to be closely related to the so - called `` yang - yang '' function @xmath368 of @xcite which is the generating ufnction between two systems of darboux coordinates - the first is the natural system of darboux coordinates on @xmath0 , and the second is the system of complex fenchel - nielsen coordinates on the character variety . in particular , as it was noticed in @xcite , the function @xmath368 transforms under a change of torelli marking of @xmath307 in the same was as @xmath369 . a complete elucidation of the link between @xmath363 and @xmath368 remains a challenging problem . assume that the poles of @xmath11 are placed to @xmath34 , @xmath193 , @xmath370 and @xmath340 in coordinate @xmath141 . then the differential @xmath11 can be written as q=(dx)^2 . as we see , the moduli space @xmath371 has dimension 2 ; @xmath372 can be chosen as coordinate on @xmath373 . to find the corresponding momentum we choose in ( [ qk0 ] ) @xmath374 , @xmath375 , @xmath376 and @xmath377 ; then the corresponding quadratic differential @xmath350 ( [ qk0 ] ) equals @xmath378 comparison with ( [ genus0 ] ) shows that the momentum @xmath209 equals to p_1=4i , and , therefore , the canonical poisson bracket @xmath379 implies @xmath380 the canonical bidifferential in terms of coordinate @xmath141 is written as b(x , y)= therefore , in the @xmath141-coordinate the bergman projective connection @xmath12 is identically vanishing and the equation ( [ sint ] ) takes the form + = 0 which is a special case of heun s equation @xcite . choosing two basic cycles @xmath119 and @xmath112 on @xmath38 ( say , @xmath119 goes around @xmath34 and @xmath193 and @xmath112 goes around @xmath193 and @xmath340 ) we get homological coordinates as elliptic integrals of first kind : a=_a,0.7 cm b=_b . the equation ( [ 4poles ] ) has 4 monodromy matrices @xmath381 and @xmath382 ; all of their eigenvalues equal to 1 . we can choose the generators @xmath383 and @xmath384 of the fundamental group of the 4-punctured sphere such that the matrices @xmath385 satisfy the relation @xmath386 it is also easy to derive directly the formulas for derivatives of any monodromy matrix @xmath389 corresponding to a basepoint @xmath159 with respect to the natural coordinates @xmath390 on the space of differentials ( [ genus0 ] ) . namely , introduce a wronskian matrix @xmath174 of two linearly independent solutions @xmath25 of ( [ 4poles ] ) satisfying the initial condition @xmath391 . this matrix solves the equation =( 0 & 1 + -q & 0 ) introduce cuts connecting @xmath159 with the singular points @xmath392 . differentiating ( [ phieq ] ) with respect to @xmath340 and @xmath61 and using the variation of parameters formula for non - homogeneous equations for @xmath393 and @xmath394 we get the following result : = -(x)_x_0^x and = - ( x)_x_0^x where the integration contour is not supposed to intersect the branch cuts and , as before , @xmath395 . earle , c. j. _ on variation of projective structures_. riemann surfaces and related topics : proceedings of the 1978 stony brook conference ( state univ . new york , stony brook , n.y . , 1978 ) , pp . 87 - 99 , ann . of math . , 97 , princeton univ . press , princeton , n.j . , eskin , a. , kontsevich , m. , zorich , a. , _ sum of lyapunov exponents of the hodge bundle with respect to the teichmuller geodesic flow _ hautes tudes sci . * 120 * 207 - 333 ( 2014 ) gallo , d , kapovich , m. , marden , a. _ the monodromy groups of schwarzian equations on closed riemann surfaces _ , ann . of math . , * 151 * 625 - 704 ( 2000 ) goldman , w. _ the symplectic nature of fundamental groups of surfaces _ , adv . in math . * 54 * , 200 - 225 ( 1984 ) goldman , w. _ invariant functions on lie groups and hamiltonian flows of surface group representations _ , invent . math . * 85 * , 263 - 302 ( 1986 ) goldman , w. _ trace coordinates on fricke spaces of some simple hyperbolic surfaces _ , handbook of teichmller theory . ii , 611 - 684 , irma lect . phys . , 13 , eur . soc . , zorich , ( 2009 ) gunning , r. _ affine and projective structures on riemann surfaces _ , in `` riemann surfaces and related topics '' ( i. kra and b. maskit , eds . ) , ann . of math . studies 97 , princeton univ . press , princeton , nj , ( 1981 ) kokotov , a. , korotkin , d. , `` tau - functions on spaces of abelian and quadratic differentials and determinants of laplacians in strebel metrics of finite volume '' math.sp/0405042 , preprint no . 46 of max - planck institut for mathematics in science , leipzig ( 2004 )

resolving quantum many - body problems represents one of the greatest challenges in physics and physical chemistry , due to the prohibitively large computational resources that would be required by using classical computers . a solution has been foreseen by directly simulating the time evolution through sequences of quantum gates applied to arrays of qubits , i.e. by implementing a _ digital quantum simulator_. superconducting circuits and resonators are emerging as an extremely - promising platform for quantum computation architectures , but a digital quantum simulator proposal that is straightforwardly scalable , universal , and realizable with state - of - the - art technology is presently lacking . here we propose a viable scheme to implement a universal quantum simulator with hybrid spin - photon qubits in an array of superconducting resonators , which is intrinsically scalable and allows for local control . as representative examples we consider the transverse - field ising model , a spin-1 hamiltonian , and the two - dimensional hubbard model ; for these , we numerically simulate the scheme by including the main sources of decoherence . in addition , we show how to circumvent the potentially harmful effects of inhomogeneous broadening of the spin systems . 3 cm

1504.05667

there is a large number of problems that are well known to be hardly tractable with standard computational approaches and resources , mainly due to the many - body nature of strongly correlated many particle systems . to overcome this limitation , the idea of a quantum simulator was originally proposed by feynman @xcite : any arbitrary complex quantum systems could in fact be simulated by another quantum system mimicking its dynamical evolution , but under the experimenter control . this idea was later refined and mathematically formalized in quantum information perspectives by lloyd @xcite . + over the past twenty years , different approaches have been proposed to realize quantum simulators of the most relevant models in condensed matter physics , quantum field theories , and quantum chemistry @xcite . most efficient protocols have been proposed and experimentally realized with trapped ions @xcite . generally speaking , quantum simulators can be broadly classified into two main categories : in _ digital simulators _ the state of the target system is encoded in qubits and its trotter - decomposed time evolution is implemented by a sequence of elementary quantum gates @xcite , whereas in _ analog simulators _ a certain quantum system directly emulates another one . digital architectures are usually able to simulate broad classes of hamiltonians , whereas analog ones are restricted to specific target hamiltonians . for a recent review on these different approaches , we refer to @xcite and references therein . + lately , superconducting circuits and resonators have emerged as an extremely promising platform for quantum information and quantum simulation architectures @xcite . the first and unique proposal for a general - purpose digital simulator has been put forward only very recently @xcite . in this proposal qubits encoded in transmons are dispersively coupled through a photon mode of a single resonator , and such coupling is externally tuned by controlling the transmon energies . however , the reported fidelities and the intrinsic serial nature of this setup ( i.e. , the need of addressing each pair of qubits sequentially ) , may hinder the scalability to a sizeable number of qubits . in addition , superconducting units are not ideal for encoding qubits owing to their relatively short coherence times . indeed , spin - ensembles @xcite or even photons @xcite have been proposed as memories to temporarily store the state of superconducting computational qubits . + here we consider an array of superconducting resonators as the main technological platform , on which hybrid spin - photon qubits are defined by introducing strongly coupled spin ensembles ( ses ) in each resonator @xcite . one- and two - qubit quantum gates can be implemented by individually and independently tuning the resonators modes through external magnetic fields . this setup can realize a universal digital quantum simulator , whose scalability to an arbitrary large array is naturally fulfilled by the inherent definition of the single qubits , represented by each coupled se - resonator device . the possibility to perform a large number of two - qubit gates in parallel makes the manipulation of such large arrays much faster than in a serial implementation , thus making the simulation of complex target hamiltonians possible in practice . a key novelty of the present proposal is that ensembles of effective @xmath0 spins are used in the hybrid encoding , which allows to exploit the mobility of photons across different resonators to perform two - qubit gates between physically distant qubits . this is done much more efficiently than by the straightforward approach of moving the states of the two qubits close to each other by sequences of swap gates , and makes the class of hamiltonians which can be realistically addressed much larger . long - distance operations arise whenever mapping the target system of the simulation onto the register implies two - body terms between distant qubits . besides the obvious case of hamiltonians with long - range interactions , this occurs with any two - dimensional model mapped onto a linear register , or with models containing @xmath1-body terms , including the many - spin terms which implement the antisymmetric nature of fermion wavefunctions . + the time evolution of a generic hamiltonian is decomposed into a sequence of local unitary operators , which can be implemented by means of elementary single- and two - qubits gates . then we combine the elementary gates of our setup in order to mimic the dynamics of spin and hubbard - like hamiltonians for fermions . we explicitly report our results for the digital quantum simulation of the transverse - field ising model on 3 qubits , the tunneling dynamics of a spin one in a rhombic crystal field and the hubbard hamiltonian . the robustness of the scheme is demonstrated by including the effects of decoherence in a master equation formalism . finally , we discuss the main sources of errors in the present simulations and the possibility to overcome them . in particular , we show how potentially harmful effects of inhomogeneous broadening of the spin ensemble are circumvented by operating the scheme in a cavity - protected regime . spins , placed at the antinodes of the magnetic field ( rotational lines ) of the cavity mode . the _ auxiliary _ resonator contains a nonlinear element ( transmon ) coupled to the electric field of the fundamental mode . ( b ) detailed sequence of time steps required to produce controlled-@xmath2 two - qubit gate between qubits @xmath3 and @xmath4 ( see appendix b for details ) . _ logical _ cavities are represented by square boxes , whereas _ auxiliary _ resonators are depicted as circular boxes . blue lines represent photon frequencies in the idle configuration ( @xmath5 in the _ logical _ and @xmath6 in the _ auxiliary _ cavities ) . the transmon ( @xmath7 and @xmath8 ) and spin ( @xmath9 , continuous , and @xmath10 , dashed ) transition energies are indicated by red lines . ( i ) qubits are initially into state @xmath11 , with the excitations ( red arrows ) stored into the photonic degrees of freedom ( blue lines ) ; ( ii ) _ logical _ cavity 3 is brought into resonance with the _ auxiliary _ resonator @xmath12 , thus ( iii ) bringing the photon to the auxiliary cavity . in the meantime _ auxiliary _ resonator 3 is detuned from the others to avoid unwanted photon hoppings . in ( iv ) the photon is absorbed by the transmon ( @xmath13 transition ) . the same hopping process ( v ) is repeated for the photon originally in cavity 2 , which is brought to the _ auxiliary _ resonator ( vi ) and then absorbed and emitted by the transmon ( @xmath14 transition ) in a semi - resonant rabi process ( vii ) . the procedure is then repeated to bring photons back to _ logical _ cavities 2 and 3 , leading the state back to @xmath11 with an additional phase @xmath2 acquired during the semi - resonant process.,width=302 ] the proposed quantum simulator is schematically shown in fig . [ setup ] . it consists of a one- or two - dimensional ( 1d or 2d ) lattice of superconducting resonators where hybrid spin - photon qubits are defined . we notice that large arrays of such resonators have already been shown experimentally @xcite . in this schematic implementation , qubits are encoded within square boxes . each box represents a coplanar resonator containing an ensemble of ( effective ) @xmath0 spins , whose collective excitations correspond to the transitions from the @xmath15 single - spin ground state to the @xmath16 excited states , and can be modeled by two independent harmonic oscillators . red lines represent the transition energies ( continuous @xmath17 , dashed @xmath18 transitions , respectively ) , while the blue line indicates the resonator frequency in the idle configuration . this can be varied within a nanosecond time - scale by means of squid devices properly connected to the resonator @xcite , in order to match the spin transition frequencies . in the hybrid qubit encoding , a dual - rail representation of the logical units is introduced where the @xmath19 and @xmath20 states of qubit @xmath21 are defined in the single - excitation subspace of each resonator . the logical state @xmath19 ( @xmath20 ) corresponds to zero ( one ) photons and a single ( zero ) quantum in the @xmath17 oscillator in cavity @xmath21 . this encoding has been introduced in previous works @xcite , and it is detailed in app . a for completeness . the @xmath18 oscillator represents an auxiliary degree of freedom that is exploited to store the photonic component of the qubit , if needed ( e.g. , to perform two - qubit gates between distant qubits , see app . b ) . + the basic unit of the scalable array is represented by a pair of qubits connected by an interposed auxiliary resonator containing a superconducting transmon device ( circular box ) , which is employed to perform two - qubit gates . it should be emphasized that this nonlinear superconducting element is not used to encode any information , and _ it is left in its ground state always except during the implementation of the two - qubit gates_. consequently , its possibly short coherence times do not significantly affect the quantum simulation . + in the following , we shall refer to the square boxes as the _ logical _ cavities labelled with greek letters , while the circular ones are the _ auxiliary _ cavities labeled by latin letters . photon hopping between neighboring resonators is allowed by capacitive coupling . formally , such a complex system can be described by the total hamiltonian @xmath22 the first term describes the ses as independent harmonic oscillators @xcite ( @xmath23 ) : @xmath24 where @xmath25 creates a spin excitation in level @xmath26 of resonator @xmath21 . the transmons are treated as effective three - level systems , with transition energies @xmath7 and @xmath8 , and described by @xmath27 the time - dependent photonic term is entirely responsible for the manipulation of the qubits . it can be expressed as : @xmath28 where @xmath29 and a similar expression holds for @xmath30 . @xmath31 ( @xmath32 ) creates ( destroys ) a single photon in the _ logical _ resonator @xmath21 , while @xmath33 ( @xmath34 ) creates ( destroys ) a single photon in the _ auxiliary _ cavity @xmath35 . hereafter , we will use the interaction picture , with @xmath36 . hence , within the rotating - wave approximation the spin - photon and transmon - photon coupling hamiltonian takes the form : @xmath37 \\ & + & \nonumber \!\!g_{12}\ ! \sum_j \left [ \hat{\tilde{a}}^{\dagger}_j |\psi_{1,j } \rangle \langle \psi_{2,j } | e^{i(\tilde{\omega}_c^j-\omega_{12 } ) t } + \text{h.c . } \right ] \\ & + & \!\!\ ! \sum_{m=1,-1 } \sum_\mu \bar{g}_{m}\ ! \left [ \hat{a}^{\dagger}_\mu \hat{b}_{m,\mu } e^{i(\omega_c^\mu-\omega_m ) t } + \text{h.c . } \right ] . \label{hamil4}\end{aligned}\ ] ] here , the coupling constants @xmath38 for the se are enhanced with respect to their single - spin counterparts by a factor @xmath39 , @xmath40 being the number of spins in the se @xcite . + finally , the last term in eq . ( [ hamil ] ) describes the photon - hopping processes induced by the capacitive coupling between the modes in neighboring cavities @xcite : @xmath41 single- and two - qubit gates are efficiently implemented by tuning individual resonator modes , as shown in previous works @xcite . arbitrary single - qubit rotations within the bloch sphere as well as controlled - phase ( c@xmath2 ) gates can be realized ( see app . b for a summary ) . + the present setup offers two remarkable benefits : the first is that using the hybrid encoding with an ensemble of effective @xmath0 spins ensures the possibility of implementing controlled - phase gates between distant qubits , with no need of performing highly demanding and error - prone sequences of swap gates . this is done by bringing the photon components of the two qubits into neighboring _ logical _ resonators by a series of hopping processes ( see app . b for details ) . transferring the photons with no corruption and without perturbing the qubits encoded in the interposed _ logical _ cavities is made possible by temporarily storing the photon component of these interposed qubits into the @xmath18 spin oscillator . + in addition , quantum simulations can be performed in parallel to a large degree , with resulting reduction of simulation times . this is made possible by the definitions of the single qubits , represented by each coupled se - resonator device , and by the local control of each _ logical _ or _ auxiliary _ resonator . non - overlapping parts of the register can then be manipulated in parallel . for instance , in simulating a heisenberg chain of @xmath1 spins @xmath42 , the @xmath1 two - qubits evolutions which appear at each time - step in the trotter decomposition are performed simultaneously first on all @xmath43 `` even '' bonds and then simultaneously on the remaining @xmath43 `` odd '' bonds . thus the simulation time of each trotter step does not increase with @xmath1 . while it is obvious that a universal quantum computer can be used in principle to simulate any hamiltonian , the actual feasibility of such simulations needs to be quantitatively assessed by testing whether the complex sequences of gates needed are robust with respect to errors due to decoherence . here we numerically solve the density matrix master equation for the model in eq . ( [ hamil ] ) with the inclusion of the main decoherence processes , i.e. , photon loss and dephasing of the transmons @xcite ( see app . c for details ) . the potentially harmful effects of inhomogeneous broadening in the se will be addressed in the next section . + in the following , we will consider the _ fidelity _ @xmath44 as a valuable figure of merit for the target hamiltonians to be simulated , where @xmath45 is the final density matrix and @xmath46 the target state . for the simulations shown in the following , we have chosen these operational parameters : @xmath47 ghz , @xmath48 ghz , @xmath49 ghz , @xmath50 ghz and @xmath51 ghz , @xmath52 ghz ( see the level scheme inside each cavity in fig . [ setup ] ) . we also assume realistic values of the se - resonator @xmath53 mhz , transmon - resonator @xmath54 mhz , @xmath55 mhz and photon - photon @xmath56 mhz couplings , respectively @xcite . the transmon parameters correspond to a ratio between josephson and charge energies @xmath57 @xcite . in this regime the dephasing time @xmath58 exceeds several @xmath59 while keeping a @xmath60 anharmonicity . the two chosen spin gaps can easily be achieved with several diluted magnetic ions possessing a @xmath61 ground multiplet , just by applying a small magnetic field along a properly chosen direction . we have chosen resonator frequencies @xmath62 and @xmath63 larger than usual experiments ( e.g. , twice the typical frequencies reported in ref . ) , since this helps improving the maximal fidelity of gates . however , we emphasize that the results do not qualitatively depend on these specific numbers . for instance , in the next section we show that remarkably good fidelities can be obtained by using realistic and state - of - art experimental frequencies @xcite or smaller . since most hamiltonians of physical interest can be written as the sum of @xmath64 local terms , our quantum computing architecture can be employed to efficiently simulate the time - evolution induced by any target hamiltonian of the type @xmath65 . the system dynamics can be approximated by a sequence of unitary operators according to the trotter - suzuki formula ( @xmath66 ) : @xmath67 where @xmath68 and the total _ digital _ error of the present approximation can be made as small as desired by choosing @xmath69 sufficiently large @xcite . in this way the simulation reduces to the sequential implementation of local unitaries , each one corresponding to a small time interval @xmath70 . the set of local unitary operators can be implemented by a proper sequence of single- and two - qubit gates . + the mapping of @xmath42 models onto an array of qubits is straightforward . we consider here two kinds of significant local terms in the target hamiltonian , namely one- ( @xmath71 ) and two - body ( @xmath72 ) terms , with @xmath73 . the unitary time evolution corresponding to one - body terms @xmath74 is directly implemented by single - qubit rotations @xmath75 . conversely , two - body terms describe a generic spin - spin interaction of the form @xmath76 , for any choice of @xmath73 . the evolution operator , @xmath77 , can be decomposed as @xcite @xmath78\ : e^{-i \hat{\lambda } \tau}\ : [ \hat{u}_{1 \alpha } \otimes \hat{u}_{2 \beta}]^\dagger \ , , \label{decomp}\ ] ] with @xmath79 , @xmath80 , @xmath81 , @xmath82 . the ising evolution operator , @xmath83 , can be obtained starting from the two - qubit c@xmath2 gate and exploiting the identity ( apart from an overall phase ) @xmath84\ : \hat{u}_{c{\varphi } } \:[\hat{i}_1 \otimes \hat{\phi}_2(-\varphi/2 ) ] , \label{decompzz}\ ] ] where @xmath85 . here @xmath86 is a phase gate ( see app . b ) . the time required and the fidelity for the simulation of each term of a generic spin hamiltonian are calculated by using a lindblad master equation formalism and are listed in table [ hamunits ] . we notice that the predicted fidelities are very high , even after the inclusion of realistic values for the main decoherence channels , especially for the photon loss rate @xmath87 , which is related to the resonators quality factor ( @xmath88 ) by @xmath89 . these elementary steps can be used to simulate non - trivial multi - spin models . + .*simulation of the elementary terms of the hamiltonian . * fidelity and time required to simulate the elementary terms of the spin hamiltonian . the fidelity has been calculated by assuming a random initial state . the second and third column show a comparison between the ideal fidelity ( calculated in the absence of decoherence ) and the real one ( calculated assuming a lindblad dynamics , with @xmath90 and @xmath91 ) . the implemented evolution is @xmath92 $ ] , with @xmath93 . the last column reports the fidelities with a setup operating in a cavity - protected regime , with @xmath90 and @xmath91 ( see section iv ) . [ cols="^,^,^,^,^",options="header " , ] as a prototypical example we report the digital quantum simulation of the transverse field ising model ( tim ) on a chain of 3 qubits : @xmath94 where @xmath95 are spin-1/2 operators . figure [ ising ] shows the oscillations of the magnetization , tr@xmath96 $ ] , for a spin system initialized in a ferromagnetic configuration . here @xmath45 is the three - qubit density matrix obtained at the end of the @xmath97 trotter steps of the simulation . the exact trotter evolution ( continuous line ) is compared to the simulated one ( points ) . in particular , red circles represent the ideal evolution , without including any source of decoherence . errors are , in that case , only due to a non - ideal implementation of the quantum gates ( see discussion below ) . conversely , green and black circles are calculated including the most important decoherence channels , namely photon loss ( timescale @xmath98 ) and pure dephasing of the transmon ( timescale @xmath58 ) . it turns out that photon loss is the most important environmental source of errors @xcite , while @xmath99 @xcite is sufficient to obtain high fidelities at the end of the simulation . indeed , the transmon is only excited during the implementation of two - qubit gates . the simulation has been performed for different values of the resonators quality factor . by decreasing @xmath88 the average fidelity decreases from @xmath100 ( infinite @xmath88 ) to @xmath101 ( @xmath102 ) and @xmath103 ( @xmath90 ) . for high but realistic @xcite values of @xmath104 the calculated points are close to the ones obtained in the ideal case ( with infinite @xmath88 ) : in that case the gating errors still dominate the dynamics . finally , by exploiting the auxiliary @xmath18 oscillator to store the photon component of the hybrid qubits when these are idle , the effects of photon loss are reduced and the fidelity significantly increases . the improvement is apparent in fig . [ ising ] , by comparing black circular and square points ; the final fidelity raises from @xmath105 to @xmath106 thanks to this storage . we stress again that the simulation time of each trotter step does not increase for larger systems containing more than 3 spins . indeed , even if more gates are needed , these can be applied in parallel to the whole array , independently of the system size . . the plot reports the expectation value of the total magnetization @xmath107 tr@xmath96 $ ] on the final state of the system , simulated for different values of the resonator quality factor ( points ) and compared with the exact evolution ( line).,width=321 ] the simulation of hamiltonians involving @xmath108 spin ensembles can be performed by encoding the state of each spin-@xmath109 onto that of @xmath110 qubits . as an explicit example , we consider a chain of @xmath0 spins , labelled @xmath111 , with nearest - neighbor exchange interactions and single - spin crystal - field anisotropy , described by the hamiltonian @xmath112 \ , , \nonumber\ ] ] which reduces to the paradigmatic haldane case for @xmath113 . by rewriting each spin-1 operator as the sum of two spin-1/2 ones ( @xmath114 ) , @xmath115 can be mapped onto a @xmath42 hamiltonian , @xmath116 , with twice the number of spins . indeed , if each a - b pair of qubits is initialized into a state with total spin equal to one , the dynamics of @xmath116 coincides with that of @xmath115 and can be simulated along the lines traced above . a proof - of - principle experiment , which could be implemented even by the non - scalable single - resonator setup described in ref . , would be the simulation of a single spin @xmath0 experiencing tunneling of the magnetization . in this simple case we find ( apart from a constant term ) : @xmath117 figure [ spin1 ] reports the comparison between the exact and the simulated evolution of the magnetization , assuming @xmath118 and @xmath119 , for different values of @xmath88 and @xmath58 . interestingly , quantum oscillations of @xmath120 are well captured by the simulation even for @xmath121 , and the fidelity is practically unaffected by a reduction of transmon coherence time to @xmath122 @xmath21s . + the simulation of many - spin models with @xmath123 typically requires two - qubit gates involving non - nearest - neighbor qubits . these can be handled with no need of swap gates as outlined in app . b. for a single @xmath0 spin with @xmath124 , after eq . ( [ eqspin1tun ] ) . as it is well known , the system oscillates between states with opposite magnetization due to quantum tunneling across the anisotropy barrier . points : digital quantum simulation obtained by the time evolution of two hybrid qubits for different values of the resonator quality factor , @xmath88 , and of the transmon coherence time , @xmath58 , respectively.,width=321 ] the numerical simulation of many - body fermionic systems is a notoriously difficult problem in theoretical condensed matter . in particular , quantum monte carlo algorithms usually fail due to the so - called sign - problem @xcite . our digital quantum simulator setup enables to efficiently compute the quantum dynamics of interacting fermions , even on an arbitrary two - dimensional lattice . although we focus on the paradigmatic fermi - hubbard hamiltonian , the proposed scheme can be generalized to the quantum simulation of several other fermionic models , such as the anderson impurity model . the target hamiltonian describing a two - dimensional @xmath125 lattice of wannier orbitals is @xmath126 where @xmath127 are nearest neighbors ( @xmath128 , @xmath129 ) and @xmath130 are fermionic operators . in order to simulate this hamiltonian with our setup , we exploit the jordan - wigner transformation to map fermion operators @xmath131 onto spin ones @xmath132 @xcite . however if such a transformation is applied to the hubbard model ( [ hubbard ] ) in more than one dimension , the hopping ( first ) term results into xy spin couplings whose sign depends on the parity of the number of occupied states that are between @xmath21 and @xmath133 in the chosen ordering of the wannier orbitals @xcite . this aspect makes the simulation of a fermionic system much more demanding than any typical spin system , because the resulting effective spin hamiltonian contains many - spin terms . to illustrate how we address this key issue , here we consider the simpler case of the hopping of spinless fermions on a lattice ( the general case of interacting spin fermions is discussed in app . the target hamiltonian can be mapped into the following spin model : @xmath134 where @xmath135 . we simulate this @xmath69-body interaction by taking care of the state - dependent phase , similarly to refs . . the sign factor in ( [ ht ] ) is obtained by performing a conditional evolution of the qubits interposed between the specifically addressed sites , @xmath21 and @xmath133 , depending on the state of @xmath21 . this corresponds to a series of controlled - z ( cz ) gates between qubit @xmath21 and each of the qubits @xmath136 interposed between @xmath21 and @xmath133 . hence , the sequence of gates to be implemented at each trotter step is the following : @xmath137 for instance , in fig . [ gateline ] we show the quantum circuit for the implementation of @xmath138 : controlled - phase gates ( with @xmath139 ) between qubit @xmath140 and each of the qubits interposed between @xmath140 and @xmath141 , namely @xmath142 , @xmath143 and @xmath144 , are sequentially performed before and after the central block ( dashed boxes ) , which implements the xy evolution : @xmath145 . the latter consists of two controlled-@xmath2 gates ( with @xmath146 ) , preceded and followed by proper single - qubit rotations , implementing respectively @xmath147 and @xmath148 terms of the interaction , as schematically explained in fig . [ gateline ] . by exploiting the high mobility of the photons entering into the hybrid encoding , hamiltonian terms involving distant qubits can be simulated straightforwardly . in fact , this is a remarkable advantage with respect to alternative solid - state platforms for quantum information processing . we stress that , in spite of the increment in the number of gates required to address the sign issue , a large number of hopping terms can still be implemented in parallel . in this section we show how the present scheme can be made robust against inhomogenous broadening ( ib ) of the se , which is probably the major shortcoming in its use to encode quantum information . a certain degree spin inhomogeneity is unavoidable in real ses , and may result from slightly disordered spin environments or by random magnetic fields produced by surrounding nuclear magnetic moments . due to ib , the collective ( super - radiant `` ) spin - excitation that couples to the photon field spontaneously decays into a quasi - continuum of decoupled , `` dark '' spin modes within a timescale of order @xmath149 , @xmath150 being the width of the distribution of gaps in the se . a possible way to deal with ib is to revert the associated hamiltonian evolution by echo techniques , e.g. by using external magnetic field pulses resonant with the spin gaps @xcite . while possible in principle , implementing such a solution within our simulation scheme would be very demanding . pulses should act with very high fidelity , and should be controlled independently for each logical resonator with the proper timing to restore qubits before they undergo gates . here we prefer to adopt a different strategy , which does not require additional resources and works well with the present scheme . the idea is to run the simulation by keeping the se in a `` cavity protection '' regime @xcite . indeed , a strong spin - resonator coupling provides a protection mechanism , by inducing an energy gap between the computational ( super - radiant ) and the non - computational ( dark ) modes @xcite , thus effectively decoupling these two . this mechanism has been experimentally demonstrated in ref . . + in the non - resonant ( dispersive ) regime , the energy shift of the super - radiant mode is of order @xmath151 , where @xmath152 is the collective se - cavity coupling and @xmath153 is the detuning between the resonator frequency and the spin gap . by assuming a spin ensemble with gaussian broadening and standard deviation @xmath150 , the cavity protection condition is fulfilled if @xmath154 . however , reducing the detuning ( or , equivalently , increasing the coupling strength ) leads to unwanted oscillations of a significant fraction ( @xmath155 ) of the wave - function between logical states @xmath156 and @xmath157 . a trade - off can be found in the limit of very large @xmath152 ( 200 - 300 mhz ) and @xmath158 ( @xmath159 ghz ) , but this is not within reach of present technology . nevertheless , we show that these oscillations do not prevent to implement the proposed digital simulation scheme , thus making it possible to employ experimentally available conditions . in the following we use small detunings , i.e. @xmath160 , with @xmath161 mhz , and a se with gaussian broadening and fwhm @xmath162 mhz , as assumed in @xcite . we numerically determine the time evolution of the qubit wave - function , coupled to a bath of dark modes , by exploiting the formalism developed in @xcite . we obtain a wave - function leakage to dark states at long times of only @xmath163 . as expected , this is lower than the upper bound ( @xmath164 ) obtained in @xcite . as we explicitly show below , the wave - function oscillations associated with the relatively large value of @xmath165 can be easily adjusted in our scheme , since they are coherent and correspond to single - qubit rotations . + here we proceed to illustrate the use of our simulation scheme in a cavity - protected regime and specific se parameters . the best spin systems are the so - called @xmath109-ions ( such as fe@xmath166 or gd@xmath166 ) whose orbital angular momentum vanishes because of hund s rules . this makes the ion practically insensitive to disorder in the environment . in addition , the number of nuclear spins should be minimized , since these produce random quasi - static magnetic fields causing ib . linewidths as small as a fraction of mhz are indeed observed in diluted magnetic semiconductors , such as fe@xmath166 in zns @xcite , whose nuclei are mostly spinless . even if fe@xmath166 and gd@xmath166 are @xmath123 spins , these behave as effective @xmath0 ses if all magnetic - dipole transitions from the ground stated are far detuned from resonator frequencies , apart from two . + to keep the experimental demonstration of the proposed scheme as easy as possible , we choose resonator frequencies smaller than in section ii . in particular , we assume 14 ghz for the _ logical _ , and 10.2 ghz for the _ auxiliary _ resonators , respectively , which are comparable with the resonator frequencies already employed in ref . . spin ensembles fitting our scheme with a 14 ghz superconducting resonator can be easily found . for instance , fe@xmath166 impurities in the same al@xmath167o@xmath168 matrix employed in @xcite display suitable gaps with an applied magnetic field of @xmath169 mt forming an angle of @xmath170 with the anisotropy axis ( given an easy plane axial anisotropy , with @xmath171 ghz @xcite ) . as mentioned above , the relatively small detuning @xmath160 needed for cavity protection induces an unwanted one - qubit oscillation with frequency @xmath172 ( @xmath173 mhz ) , which must be taken into account in our simulation scheme . this can be done in the implementation of one - qubit gates by choosing a starting time of the gate @xmath174 and by making the detuning @xmath175 for the short time ( @xmath176 ns ) needed to implement the phase gate , eq . we stress that this time is much shorter than that characterizing the damping due to ib . as far as two - qubit gates are concerned , the only part which is affected by the unwanted oscillations is the photon hopping process between _ logical _ and _ auxiliary _ cavities , whose starting time needs to be chosen again as @xmath174 ( note that the _ auxiliary _ resonator does not contain a se ) . in some situations ( such as the simulation of @xmath177 terms ) , when a pair of qubits is subject to a different or differently ordered sequence of operations , we need to increase the detuning for one of the two qubits to freeze its evolution , while waiting the other to complete an oscillation ( ' ' rephasing " ) . + to test the performance of this setup we first numerically determine the fidelity of simulations of the elementary terms in a one- and two - qubit hamiltonian , including also decoherence effects ( by solving the liouville - von neumann equations of motion ) . table [ hamunits ] shows that these fidelities remain high , thus demonstrating the effectiveness of our scheme even in a cavity - protected regime . an interesting proof - of - principle experiment that could be readily performed with existing setups is the simulation of the dynamics resulting from an @xmath178 interaction between two spins @xmath42 . this is also the central step in the simulation of hopping processes in fermion hamiltonians ( dashed boxes in fig . [ gateline ] ) . $ ] ) . lines represent the exact evolution , whereas points are calculated with lindblad formalism assuming @xmath179 and for two different values of @xmath58 . we use the parameters @xmath180 ghz , @xmath181 ghz , @xmath182 ghz , @xmath183 ghz , @xmath184 ghz , @xmath185 ghz . @xmath186 mhz , @xmath187 mhz , @xmath160.,width=321 ] fig . [ evoxy ] shows that the time evolution is well reproduced even with a short @xmath188s . we have proposed a digital quantum simulator based on hybrid spin - photon qubits , encoded in an array of superconducting resonators strongly coupled to spin ensembles . within this quantum computing architecture , quantum gates are implemented by a single operational tool , namely by tuning the resonators frequencies . we have shown the feasibility of the scheme with state - of the - art superconducting arrays technology , which allows the high fidelity simulation of a large class of multi - qubits spin and fermionic models . to test our predictions , we have performed numerical simulations of the master equation for the system density matrix , including the most important decoherence channels such as photon loss and pure - dephasing of the transmon involved in two - qubit entangling gates . coherence times of single spins are so long that their effect on quantum simulations can be disregarded . however , inhomogeneous broadening of the se might result in an irreversible population leakage out of the computational basis . nevertheless , we have shown in the previous section that this problem can be circumvented by implementing the scheme in a cavity - protected regime . + _ sources of errors_. we analyze here the sources of error that affect the quantum simulation , and point out possible solutions . three main simulation errors can be found : digital errors ( arising from the trotter - suzuki approximation ) , gating errors ( due to imperfect implementation of the desired unitaries ) , and decoherence errors ( due to the interaction of the quantum simulator with the environment ) . while digital errors can obviously be reduced by increasing the number of trotter steps , gating errors are accumulated by repeating a large number of quantum operations . in the same way the effects of the interaction of the system with the environment become much more pronounced if the number of gates , and thus the overall simulation time , increases . first , we notice that the present setup _ limits the role of the transmon _ , which is not involved in the definition of the qubits . all transmons are kept in their ground states apart for the specific transmons involved in two - qubit gates , which are excited only for a short time . thus , typical state - of - the - art technology , which ensures transmon dephasing times of the order of tens of microseconds , is sufficient to obtain high fidelity quantum simulations of relatively large systems . indeed , the three - spins tim model reported here can immediately be extended to simulate spin hamiltonians involving many spins , by addressing the different cavities in parallel . photon loss represents the main source of decoherence in our hybrid dual - rail encoding . finally , gating errors are mainly due to the small difference between the photon frequency and transmonic gaps in the auxiliary cavities , which induces a residual interaction that is never completely switched off . here we use the tunability of the resonator frequency as the only tool to process quantum information , but the flux control of the josephson energy of the transmons can also be exploited to increase the detuning , thus leading to even larger fidelities . + _ two - dimensional arrays_. while any model can be implemented onto a one - dimensional register ( e.g. , the one schematically illustrated in fig . [ setup ] ) at the cost of requiring long - range two - qubit gates , it is clear that a register topology directly mimicking the target hamiltonian would greatly reduce the simulation effort . in particular , there are several important hamiltonians defined on two - dimensional lattices whose simulation would greatly benefit from a two - dimensional register . here , we point out that our scheme is straightforwardly usable on such a register , but its experimental realization necessarily requires the implementation of two sub - lattices of cavities , alternatively coupled to spin and transmon qubits , respectively . fortunately , resonator arrays with complex network topologies are realistically possible , already , as each cavity can easily couple to multiple other resonators . [ scheme2d ] displays the schematic drawing of a potential two - dimensional layout showing how such sub - lattices could feasibly realize a two - dimensional simulator . > from a technological point of view , we notice that similar lattices with transmon qubits have been fabricated with more than 200 coupled cavities . while local tuning in such a lattice would require local flux bias on a separate layer , this need for local control lines applies to any adjustable quantum simulator . on the other hand , we notice that a recent technology has shown promising results to bring flux lines to the interior part of a lattice made of a small number of nodes , e.g. by using aluminum airbridge crossovers to route microwave signals into a target resonator @xcite . + _ summary_. in conclusion , the proposed setup exploits the best characteristics of distinct physical systems : the long coherence times of the spins , which can encode quantum information and protect it from decoherence , and the mobility of photons entering this hybrid encoding of qubits . in the end , this allows to realize long - range two - body interactions between distant qubits without the need for much more demanding swap gates . moreover , on - site tunability and scalability make this architecture extremely appealing and competitive with respect to alternative proposals , either based on superconducting arrays or on different technologies . . we consider a coplanar waveguide resonator containing a single photon in a mode of frequency @xmath62 , and an ensemble of @xmath40 non - interacting and equally oriented @xmath189 spins . in the low - excitation regime , the se can be modeled by two independent harmonic oscillators , related to two different magnetic - dipole transitions from the @xmath15 ground state of the single spin , to the @xmath190 and @xmath191 states , with excitation frequencies @xmath9 and @xmath10 . this can be achieved by properly choosing a system with easy - plane magnetic anisotropy , which provides a zero - field splitting between the @xmath15 ground state and the excited @xmath16 doublet , and in the presence of a small static magnetic field . we suppose to initialize the system by preparing each spin in its ground state : @xmath192 . + if the resonator frequency is tuned to match the spin gap @xmath10 , the se can absorb the photon and collectively evolve into the state @xmath193 . transitions between @xmath194 and @xmath195 are described ( in the limit of low number of excitations ) by the bosonic operators @xmath196 and @xmath197 , where @xmath198 and @xmath199 = 1 $ ] @xcite . conversely , if the resonator frequency is tuned to @xmath9 , the se can evolve into the state @xmath200 , the transition being described by the operators @xmath201 and @xmath202 , where @xmath203 . + within the single - excitation subspace of the system formed by the cavity mode and the se , we introduce the hybrid dual - rail encoding of the qubit @xmath21 : @xmath204 is the photon creation operator and @xmath205 is the vacuum state . resonant processes involving the absorption ( emission ) of the photons entering the hybrid encoding in ( eq . [ encoding ] ) are exploited to perform one- and two - qubit gates . these processes are induced by `` shift pulses '' , in which the frequency of cavity @xmath21 is varied by a quantity @xmath206 for a suitable amount of time . in the idle mode , the photon frequencies are largely detuned from the spin energy gaps , and @xmath207 is ineffective . in addition , the modes @xmath208 and @xmath209 of neighboring cavities are far - detuned and the effect of @xmath210 is negligible . single - qubit gates can thus be performed independently on each qubit , which can be individually addressed . + off - resonance pulses are employed to obtain a rotation by an arbitrary angle about the @xmath211 axis of the bloch sphere . these induce a phase difference between the @xmath156 and @xmath157 states of the hybrid qubits ( eq . [ encoding ] ) and performs the well - known phase gate : @xmath212 where we have assumed step - like pulses of amplitude @xmath206 and duration @xmath213 . + conversely , resonant pulses are employed to transfer the excitation between ses and resonators . this produces a generic rotation in the @xmath214-@xmath215 plane of the bloch sphere : @xmath216 with @xmath217 . by properly tuning the initial time we can obtain rotations about @xmath214 ( @xmath218 ) or @xmath215 ( @xmath219 ) axis , while the pulse duration controls the rotation angle . @xcite for a detailed derivation . + the controlled - phase ( c@xmath2 ) two - qubit gate is represented by the matrix : @xmath220 it can be implemented by means of two - step semi - resonant rabi oscillations of the transmon state between @xmath221 and @xmath222 . we describe here the c@xmath2 multi - step pulse sequence on two qubits initialized in the state @xmath223 , as schematically shown in fig . [ setup]-(b ) for @xmath224 , @xmath225 and @xmath12 : 1 . the first step corresponds to the hopping of the photon from _ logical _ cavity 3 to the _ auxiliary _ resonator 2 ( interposed between qubits 2 and 3 ) , by means of a @xmath226-pulse that brings the two cavities into resonance . 2 . as a second step , the frequency of resonator @xmath224 ( @xmath227 ) is tuned to @xmath7 by means of a @xmath226-pulse , which transfers the excitation to the intermediate level @xmath228 of the transmon . a @xmath226-pulse is exploited to induce the hopping of a second photon from _ logical _ cavity 2 to the _ auxiliary _ resonator . then , a semi - resonant process ( during which the resonator is detuned from the transmon gap by a small amount @xmath229 ) is exploited to induce an arbitrary phase on the @xmath230 component of the wavefunction @xcite . a pulse of duration @xmath231 , where @xmath232 is the detuning between the resonator mode and the @xmath14 transition of the transmon , adds a phase @xmath233 to the system wavefunction . finally , the repetition of the first three steps brings the state back to @xmath234 , with an overall phase @xmath2 . by properly setting the delay between the two @xmath226 pulses corresponding to the previous steps ( or by performing single - qubit phase shifts ) , the associated absorption and emission processes yield a zero additional phase . conversely , the other basis states do not acquire any phase , as required for the c@xmath235 gate , due to the absence of at least one of the two photons ( see ref . @xcite ) . for @xmath236 we obtain the usual full rabi process , which implements a controlled - z ( cz ) . + the setup is simplified with respect to our previous proposal @xcite , as each resonator contains a single photonic mode . + it is also important to note that here we are using an ensemble of effective spins @xmath0 as this ensures the possibility of implementing controlled - phase gates between distant qubits , with no need of performing highly demanding and error - prone sequences of two - qubit swap gates . long - distance two - qubit interactions are a key - resource for the digital simulation of many interesting physical hamiltonians . they appear each time that a multi - dimensional target system is mapped onto a linear chain of qubits or in models with @xmath1-body terms . among these , as discussed in section iii.b , a particular interest is assumed by problems involving interacting fermions in two or higher spatial dimensions , which are often intractable for classical computers . for instance , solving the two - dimensional hubbard model is considered by many as the ultimate goal of the theory of strongly correlated systems . in these cases the jordan - wigner mapping induces many - spin interactions @xcite which can be handled as outlined in fig . [ gateline ] , provided the ability to efficiently implementing long - range two - qubit couplings . these are obtained by bringing the photon components of the two qubits into neighboring _ logical _ resonators by a series of hoppings . the operations outlined in fig . [ setup]-(b ) are then performed to implement a c@xmath2 gate between neighboring qubits , and the photon components are finally brought back to the starting position by reverting the series of hoppings . the photons can be transferred with negligible leakage and without perturbing the interposed qubits by temporarily storing the photon component of these qubits into the @xmath18 spin oscillator . we stress that a large number of these long - range two - qubit gates can be implemented in parallel in the actual setup . the time evolution of the system density matrix @xmath45 is described within a markovian approximation and a lindblad - type dynamics , with the liouville - von neumann equation of motion @xcite : @xmath237 + \sum_{i } \gamma_i \mathcal{l}_{\hat{x}_i}[\hat{\rho } ] + \sum_{i } \gamma_i \mathcal{l}_{\hat{x}_i^\dagger \hat{x}_i}[\hat{\rho } ] , \label{evorho}\ ] ] being @xmath238 and @xmath239 respectively the damping and pure - dephasing rates of the field @xmath240 . the lindblad term for an arbitrary operator , @xmath241 , is given by @xmath242 = -\frac{1}{2 } \left ( \hat{x}^\dagger \hat{x } \hat{\rho } + \hat{\rho } \hat{x}^\dagger \hat{x } \right ) + \hat{x } \hat{\rho } \hat{x}^\dagger . \label{lindblad}\ ] ] if the operator @xmath243 destroys an excitation in the system , terms like @xmath244 $ ] account for energy losses , while pure dephasing processes are described by @xmath245 $ ] . we note @xcite that the former ones provide the most important contribution for photons @xcite ( with @xmath246 , @xmath247 ) , while the latter are very important for the transmons ( @xmath248 , @xmath249 ) . we represent each field as a matrix in the fock - states basis , and truncate it at a number of total excitations previously checked for convergence . the total hamiltonian , eq . ( [ hamil ] ) , and the density matrix master equation of the whole system , eq . ( [ evorho ] ) , are built by tensor products of these operators . then , the equation of motion for @xmath45 is numerically integrated , in the interaction picture , by using a standard runge - kutta approximation . to extend the quantum simulation of two - dimensional hubbard models to the case of fermionic systems with spin , we need to encode each fermion operator into a pair of qubits , corresponding to spin up and spin down . to achieve this , we exploit a generalization of the jordan - wigner transformation @xcite . for this mapping we need to introduce two different spin 1/2 operators @xmath250 and @xmath251 , with @xmath252 and @xmath253 , describing respectively odd and even qubits ( ordered by rows in the two - dimensional lattice ) . @xmath254 it can be shown that these operators satisfy the usual angular momentum commutator algebra , and that @xmath255 = 0 $ ] . we assume that the fermion variables are ordered by rows in the hamiltonian . the efficiency of the scheme would be increased by using a 2-dimensional setup consisting of @xmath1 rows and @xmath256 columns . we can write the hubbard hamiltonian in terms of the spin variables introduced above @xmath257 where @xmath21 and @xmath133 are nearest neighbors on the two - dimensional fermionic lattice , such that @xmath258 ( horizontal neighbors ) or @xmath259 ( vertical neighbors ) with the present labeling . odd ( even ) qubits encode spin up ( spin down ) variables . since the hopping term does not act if @xmath260 ( i.e. @xmath261 ) , we can start directly with @xmath262 , and the exponential in expressions like @xmath263 can be factorized . we note that in the case of horizontal neighbors the phase factor cancels out and that in @xmath264 do not appear terms @xmath265 , as we are not considering spin - flip processes . + to simulate such evolution we can proceed in a way analogous to the spinless case . here , however , two different series of @xmath266 should be carried out , depending if we are considering the hopping of spin @xmath267 or spin @xmath268 fermions . the former involves only odd values of @xmath136 , the second only even . notice that , in a 2-dimensional register , we need to transfer photons to implement @xmath269 or @xmath270 each time we have to couple a pair of fermions belonging to the same row ( due to the alternating @xmath267-@xmath268 mapping ) , but in that case @xmath271 is not required . the term @xmath271 , needed to correct the _ sign problem _ , is necessary only if @xmath272 ( no photon transfer in that case is needed ) . r. p. feynman , _ simulating physics with computers _ * 21 * , 467 ( 1982 ) . s. lloyd , _ universal quantum simulators _ , science * 273 * , 1073 ( 1996 ) . i. m. georgescu , s. ashab and f. nori , _ quantum simulation _ , rev . phys . * 86 * , 153 ( 2014 ) . l. lamata , a. mezzacapo , j. casanova and e. solano , _ efficient quantum simulation of fermionic and bosonic models in trapped ions _ , epj quantum technology * 1 * , 9 ( 2014 ) . lanyon , c. hempel , d. nigg , m. mller , r. gerritsma , f. zhringer , p. schindler , j. t. barreiro , m. rambach , g. kirchmair , m. hennrich , p. zoller , r. blatt and c. f. roos , _ universal digital quantum simulation with trapped ions _ , science * 334 * , 57 ( 2011 ) . s. schmidt and j. koch , _ circuit qed lattices _ , ann . phys . * 6 * , 395 ( 2013 ) . a. a. houck , h. e. treci and j. koch , _ on - chip quantum simulation with superconducting circuits _ physics * 8 * , 292 ( 2012 ) . u. las heras , a. mezzacapo , l. lamata , s. filipp , a. wallraff and e. solano , _ digital quantum simulation of spin systems in superconducting circuits _ , lett . * 112 * , 200501 ( 2014 ) . a. mezzacapo , u. las heras , j. s. pedernales , l. dicarlo , e. solano and l. lamata , _ digital quantum rabi and dicke models in superconducting circuits _ , * 4 * , 7482 ( 2014 ) . y. kubo , c. grezes , a. dewes , t. umeda , j. isoya , h. sumiya , n. morishita , h. abe , s. onoda , t. ohshima , v. jacques , a. drau , j .- f . roch , i. diniz , a. auffeves , d. vion , d. esteve , and p. bertet , _ hybrid quantum circuit with a superconducting qubit coupled to a spin ensemble _ , phys . 107 * , 220501 ( 2011 ) . s. probst , a. tkalec , h. rotzinger , d. rieger , j - m . le floch , m. goryachev , m. e. tobar , a. v. ustinov , and p. a. bushev , _ three - dimensional cavity quantum electrodynamics with a rare - earth spin ensemble _ , phys . b. , * 90 * , 100404(r ) ( 2014 ) . c. grezes , b. julsgaard , y. kubo , m. stern , t. umeda , j. isoya , h. sumiya , h. abe , s. onoda , t. ohshima , v. jacques , j. esteve , d. vion , d. esteve , k. mlmer and p. bertet , _ multimode storage and retrieval of microwave fields in a spin ensemble _ , phys . rev . x * 4 * , 021049 ( 2014 ) . m. mariantoni , h. wang , t. yamamoto , m. neeley , r. c. bialczak , y. chen , m. lenander , e. lucero , a. d. oconnell , d. sank , m. weides , j. wenner , y. yin , j. zhao , a. n. korotkov , a. n. cleland and j. m. martinis , _ implementing the quantum von neumann architecture with superconducting circuits _ , science * 34 * , 61 ( 2011 ) . j. ghosh , a. galiautdinov , z. zhou , a. n. korotkov , j. m. martinis , and m. r. geller , _ high - fidelity controlled-@xmath273 gate for resonator - based superconducting quantum computers _ , phys . a * 87 * , 022309 ( 2013 ) . s. carretta , a. chiesa , f. troiani , d. gerace , g. amoretti and p. santini , _ quantum information processing with hybrid spin - photon qubits _ , * 111 * , 110501 ( 2013 ) . a. chiesa , d. gerace , f. troiani , g. amoretti , p. santini and s. carretta , _ robustness of quantum gates with hybrid spin - photon qubits in superconducting resonators _ , phys . a * 89 * , 052308 ( 2014 ) . d. l. underwood , w. e. shanks , j. koch and a. a. houck , _ low - disorder microwave cavity lattices for quantum simulation with photons _ , phys . a * 86 * , 023837 ( 2012 ) . a. palacios - laloy , f. nguyen , f. mallet , p. bertet , d. vion and d. esteve , _ tunable resonators for quantum circuits _ , j. low temp . phys . * 151 * , 1034 ( 2008 ) . m. sandberg , f. persson , i. c. hoi , c. m. wilson and p. delsing , _ exploring circuit quantum electrodynamics using a widely tunable superconducting resonator _ scr . * t137 * , 014018 ( 2009 ) . z. l. wang , y. p. zhong , l. j. he , h. wang , j. m. martinis , a. n. cleland and q. w. xie , _ quantum state characterization of a fast tunable superconducting resonator _ , applied physics letters * 102 * , 163503 ( 2013 ) . y. kubo , i. diniz , a. dewes , v. jacques , a. drau , j .- f . roch , a. auffves , d. vion , d. esteve and p. bertet , _ storage and retrieval of a microwave field in a spin ensemble _ , phys . a * 85 * , 012333 ( 2012 ) . j. h. wesenberg , a. ardavan , g. a. d. briggs , j. j. l. morton , r. j. schoelkopf , d. i. schuster and k. mlmer , _ quantum computing with an electron spin ensemble _ , phys . lett . * 103 * , 070502 ( 2009 ) . d. i. schuster , a. p. sears , e. ginossar , l. dicarlo , l. frunzio , j. j. l. morton , h. wu , g. a. d. briggs , b. b. buckley , d. d. awschalom and r. j. schoelkopf , _ high - cooperativity coupling of electron - spin ensembles to superconducting cavities _ , phys . lett . * 105 * , 140501 ( 2010 ) . p. santini , s. carretta , f. troiani and g. amoretti , _ molecular nanomagnets as quantum simulators _ , phys . rev . lett . * 107 * , 230502 ( 2011 ) . j. clarke and f. k. wilhelm , _ superconducting quantum bits _ , nature * 453 * , 1031 ( 2008 ) . a. megrant , c. neill , r. barends , b. chiaro , y. chen , l. feigl , j. kelly , e. lucero , m. mariantoni , p. j. j. omalley , d. sank , a. vainsencher , j. wenner , t. c. white , y. yin , j. zhao , c. j. palmstrom , j. m. martinis and a. n. cleland , _ planar superconducting resonators with internal quality factors above one million _ . lett . * 100 * , 113510 ( 2012 ) . m. troyer and u. j. wiese , _ computational complexity and fundamental limitations to fermionic quantum monte carlo simulations _ , phys . * 94 * , 170201 ( 2005 ) . e. fradkin , _ jordan - wigner transformation for quantum - spin systems in two dimensions and fractional statistics _ * 63 * , 322 ( 1989 ) . w. shaofeng , _ jordan - wigner transformation in a higher - dimensional lattice _ , phys . e * 51 * , 1004 ( 1995 ) . b. bock and m. azzouz , _ generalization of the jordan - wigner transformation in three dimensions and its application to the heisenberg bilayer antiferromagnet _ b * 64 * , 054410 ( 2001 ) . d. s. abrams and s. lloyd , _ simulation of many - body fermi systems on a universal quantum computer _ . lett . * 79 * , 2586 ( 1997 ) . g. ortiz , j. e. gubernatis , e. knill and r. laflamme , _ quantum algorithms for fermionic simulations _ a * 64 * , 022319 ( 2001 ) . r. somma , g. ortiz , j. e. gubernatis , e. knill and r. laflamme , _ simulating physical phenomena by quantum networks _ a * 65 * , 042323 ( 2002 ) . b. julsgaard , c. grezes , p. bertet and k. mlmer , _ quantum memory for microwave photons in an inhomogeneously broadened spin ensemble _ , phys . lett . * 110 * , 250503 ( 2013 ) . s. putz , d. o. krimer , r. amsss , a. valookaran , t. nbauer , j. schmiedmayer , s. rotter and j. majer , _ protecting a spin ensemble against decoherence in the strong - coupling regime of cavity qed _ , nat . physics * 10 * , 720 ( 2014 ) . i. diniz , s. portolan , r. ferreira , j. m. gerard , p. bertet and a. auffves , _ strongly coupling a cavity to inhomogeneous ensembles of emitters : potential for long - lived solid - state quantum memories _ , a * 84 * , 063810 ( 2011 ) . z. kurucz , j. h. wesenberg and k. mlmer , _ spectroscopic properties of inhomogeneously broadened spin ensembles in a cavity _ , phys . rev . a * 83 * , 053852 ( 2011 ) . w. c. holton , j. schneider , and t. l. estle , _ electron paramagnetic resonance of photosensitive iron transition group impurities in zns and zno _ , phys . rev . * 133 * , a1638 ( 1964 ) . + a. rauber and j. schneider , z. naturforsch . * 17a * , 266 ( 1962 ) . w. g. farr , d. l. creedon , m. goryachev , k. benmessai , and m. e. tobar , _ ultrasensitive microwave spectroscopy of paramagnetic impurities in sapphire crystals at millikelvin temperatures _ phys . rev . b * 88 * , 224426 ( 2013 ) . z. chen , a. megrant , j. kelly , r. barends , j. bochmann , y. chen , b. chiaro , a. dunsworth , e. jeffrey , j. y. mutus , p. j. j. omalley , c. neill , p. roushan , d. sank , a. vainsencher , j. wenner , t. c. white , a. n. cleland , and j. m. martinis , _ fabrication and characterization of aluminum airbridges for superconducting microwave circuits _ appl . . lett . * 104 * , 052602 ( 2014 ) . m. o. scully and m. s. zubairy , _ quantum optics _ ( cambridge university press , cambridge , uk , 1997 ) . h. wang , m. hofheinz , m. ansmann , r. c. bialczak , e. lucero , m. neeley , a. d. oconnell , d. sank , j. wenner , a. n. cleland and j. m. martinis , _ measurement of the decay of fock states in a superconducting quantum circuit _ * 101 * , 240401 ( 2008 ) . r. a. bari , _ classical linear - chain hubbard model : metal - insulator transition _ , phys . b * 7 * , 4318 ( 1973 ) .

percolation clusters are random fractals whose geometrical and transport properties can be characterized with the help of probability distribution functions . using renormalized field theory , we determine the asymptotic form of various of such distribution functions in the limits where certain scaling variables become small or large . our study includes the pair - connection probability , the distributions of the fractal masses of the backbone , the red bonds and the shortest , the longest and the average self - avoiding walk between any two points on a cluster , as well as the distribution of the total resistance in the random resistor network . our analysis draws solely on general , structural features of the underlying diagrammatic perturbation theory , and hence our main results are valid to arbitrary loop order .

0809.2342

percolation @xcite clusters are random fractals . due to their randomness , it is natural to characterize their properties , such as geometrical and transport properties , by means of probability distribution functions . the perhaps most basic of these distribution functions is the the percolation correlation function @xmath0 which measures the pair connection probability , i.e. , the probability that two sites located at space coordinates @xmath1 and @xmath2 are connected . for many applications , it is useful to consider percolation models that allow for additional processes to unfold on the clusters . the most prominent model of this kind is the random resistor network ( rrn ) where occupied bonds are viewed as resistors so that the clusters can transport electric currents . one of the typical questions that are addressed when studying the rrn is that of the total resistance @xmath3 between 2 leads located at points @xmath1 and @xmath2 _ on the same cluster_. because the underlying cluster is random , the @xmath3 is a random number governed by a distribution function , @xmath4 . experimentally , this distribution might be determined as follows : measure the resistance between all pairs of sites with separation @xmath5 using an ohmmeter . discard all measurements yielding an infinite value as these measurements probe sites belonging to different clusters . prepare a histogram that collects the measured finite values . then , @xmath6 corresponds to the number of points in the histogram between @xmath7 and @xmath8 divided by the number of measurements with finite outcome . a similar experiment has been performed , e.g. , some 20 years ago by rammal , lemieux and tremblay @xcite . other interesting measurable quantities that are intimately related to the transport properties of percolation clusters include the masses ( the number of bonds ) of various fractal substructures of the clusters . the most prominent substructures of a cluster are perhaps the backbone ( the set of bonds that carry current ) , the red bonds ( the bonds that are singly connected and hence carry the full current ) as well as the shortest , the longest and the average self - avoiding walk ( saw ) on the cluster . their masses are random numbers , and the corresponding distribution functions can be measured in much the same way as described above for the total resistance . in this paper , we study the distribution functions of the pair connection probability , the total resistance and the masses of fractal substructures by using renormalized field theory . our work is based on the harris - lubensky ( hl ) model @xcite for the rrn and its nonlinear generalization by harris @xcite . in this generalized model , the bonds are assumed to have a nonlinear current - voltage characteristic of the type @xmath9 @xcite . the main benefit of this generalization is that it allows to map out the aforementioned fractal substructures by taking the appropriate limits with respect to @xmath10 : the limit @xmath11 leads to the backbone , whereas @xmath12 , @xmath13 and @xmath14 lead to the red bonds , the shortest and the longest self - avoiding path , respectively . of course , one retrieves the linear rrn for @xmath15 . via these limits , the distribution functions of the total resistance and the fractal masses are all related to one single distribution function , namely that of the total nonlinear resistance . regarding the average saw , the story is somewhat more complex . obviously , the length of the average saw lies between the lengths of the shortest and the longest saw . in a certain sense , the average saw is concealed somewhere in the discontinuity at @xmath16 , and it not clear how to extract its length by taking a limit of the nonlinear rrn . recently @xcite , we proposed a method to calculate the scaling form of this length via using our real - world interpretation of the feynman diagrams of the rrn , which we will explain below . in this paper , we use this method ( which is intimately related to a correct interpretation of the meir - harris model @xcite ) to determine the distribution of this length . in their seminal paper @xcite , hl studied among many other interesting things the asymptotic form at criticality of the distribution @xmath17 of the linear total resistance for large @xmath7 . their calculation is based on the assumption that the _ unconditional _ probability @xmath18 for measuring @xmath7 , which differs from @xmath17 in that infinite resistances are not discarded , becomes independent of @xmath1 for large @xmath7 . this assumption is problematic , however , because large @xmath7 corresponds due to scaling to short distances , and in critical phenomena , short distances generically lead to extra singularities which has qualitative consequences @xcite . our careful analysis , based on the so - called short - distance expansion as a field - theoretic tool , reveals that this assumption breaks down beyond mean - field theory , and hence the long - standing result by hl turns out being incorrect for dimensions @xmath19 . before moving into the depths of field theory , we would like to summarize for the convenience of the reader the resulting picture of the various distribution functions . it is well established , that the pair connection probability obeys the scaling form @xmath20 where @xmath21 is the control parameter that measures the deviation from the percolation point . in bond percolation , e.g. , @xmath21 is proportional to @xmath22 , where @xmath23 is the bond occupation probability , and @xmath24 is its critical value . @xmath25 and @xmath26 are the order parameter and correlation length exponents for isotropic percolation , respectively . @xmath27 is a scaling function , whose asymptotic behavior we calculate for small and large @xmath28 . for small @xmath28 , we find @xmath29 where @xmath30 and so on are expansion coefficients . for large @xmath28 , we obtain @xmath31 with constants @xmath32 and @xmath33 . the distribution @xmath34 of the total nonlinear resistance obeys at criticality , @xmath35 , the scaling form @xmath36 where @xmath37 is the resistance exponent of the nonlinear rrn . as discussed above , the distribution function for the total linear resistance is implied in this result for @xmath15 , that of the backbone mass for @xmath11 and so on . for large @xmath38 , we find that the scaling function @xmath39 behaves as @xmath40 where @xmath41 and so on are constants , and where @xmath42 is the so - called wegner exponent . for small @xmath38 , we get @xmath43 with constants @xmath44 and @xmath45 . equations ( [ scalefromprcrit ] ) to ( [ omegatsmalls ] ) also describe the distribution of the length of the average saw ; we merely have to replace @xmath37 by @xmath46 , where @xmath47 is the usual scaling exponent of the average saw on a percolation cluster , and we also have to replace nonuniversal constants @xmath48 and so on by other nonuniversal constants , say @xmath49 and so on . we note that similar asymptotic laws were derived a long time go by descloiseaux @xcite , fisher @xcite , and mckennzie and moore @xcite for saws on isotropic non - disordered substrates . the remainder of this paper has the following outline : section [ reviewpottsrrn ] briefly reviews the field theoretic formulation of the potts model and the hl model and its nonlinear generalization to provide background information and to establish notation . section [ representations ] treats the renormalization of composite fields that play a role in the short distance expansion . section [ shortdistanceexpansion ] contains the core of our short distance expansion . section [ shapedistfuncts ] presents the derivation of our final results for the distribution functions . section [ concludingremarks ] gives a few concluding remarks . there are two main field - theoretic approaches for percolation problems : one is based on the so - called generalized epidemic process , and the other is based on the @xmath50-state potts model in the limit @xmath51 . the latter is intimately related to the rrn and thus it is particularly useful for studying transport properties of percolation clusters . in this paper we focus on the potts model and the rrn along with its nonlinear generalization . here , in this section we review some aspects of the field theory of the potts model and the rrn that are essential for our present work , in order to provide the reader with background information and to establish notation . in the literature , the minimal model or field theoretic hamiltonian for the @xmath52-state potts model has been written in several forms . the perhaps simplest form @xmath53 is achieved if the @xmath54-component order parameter field @xmath55 satisfies the constraint @xmath56 because , in this case , @xmath55 shows most directly the underlying symmetry of the potts model that arises from the equivalence of the @xmath57 states . the order parameter with constraint ( [ pottsconstraintoriginal ] ) constitutes an irreducible representation of the group of permutations of @xmath57 elements , i.e. , the symmetric group @xmath58 . in group theoretical language @xcite this is the representation @xmath59 @xcite . the parameter @xmath21 measures the distance from the critical point in the phase diagram , which is given in mean field theory by @xmath60 . note that there is only a single expansion coefficient or coupling constant @xmath61 decorating the cubic term since the third order direct product @xmath62 contains the identity representation only once . hence , there is just one cubic invariant of the order parameter . for the model to describe ( purely geometric ) percolation , one has to take the limit @xmath63 . for addressing certain questions , it can be useful to switch from @xmath55 to its fourier transform @xmath64 and vice versa via the relations @xmath65 with @xmath66 given by @xmath67 the prime in eq . ( [ pottsfouriertrafo1 ] ) indicates that @xmath68 is excluded from the sum over @xmath66 due to the constraint ( [ pottsconstraintoriginal ] ) . alternatively , one can admit @xmath68 if one demands that @xmath64 satisfies the constraint @xmath69 that is a consequence of eq . ( [ pottsconstraintoriginal ] ) . in terms of @xmath64 , the field theoretic hamiltonian for the potts model reads @xmath70 \tilde{\varphi}_{\lambda } \nonumber \\ & + \frac{g}{6}\sum_{\lambda,\lambda^{\prime } } \tilde{\varphi}_{\lambda}\tilde{\varphi}_{\lambda^{\prime}}\tilde{\varphi}_{-\lambda-\lambda^{\prime}}\bigg\ } \,.\end{aligned}\ ] ] throughout the rest of this paper , we will draw heavily on the well established methods of renormalized field theory . we will outline the steps that we take as much as possible . given the space - constraints of this type of paper , however , we have to refer to the textbooks on field theory @xcite for general background and technical details . at and below the upper critical dimension @xmath71 , mean field theory , i.e. , saddlepoint - integration of functional integrals followed by naive perturbation expansions ( with a physical finite cutoff ) , breaks down . naive perturbation expansions produce ir - divergencies near the critical point , and renormalization group methods must be employed instead to sum up the leading divergencies of the complete expansion . in diagrammatic perturbation calculations using dimensional regularization instead of a finite cutoff , poles in @xmath72 arise . these @xmath73-poles can be eliminated from the theory by employing the renormalization scheme [ reno ] @xmath74 where the ring @xmath75 denotes bare , unrenormalized quantities . @xmath76 is an effective coupling constant that appears naturally in any diagrammatic calculation based on the potts hamiltonian ( [ pottshamiltonian ] ) . @xmath77 is an inverse length - scale , and the factor @xmath78 makes @xmath76 dimensionless in the sense that its engineering dimension vanishes . @xmath79 is a convenient numerical factor depending on space dimension . the renormalization factors @xmath80 , @xmath81 and @xmath82 are known to three - loop order @xcite . as far as explicit diagrammatic calculations are concerned , we will not go in the present paper beyond two - loop order , to which the renormalization factors are given by @xmath83 if one uses minimal subtraction , as we do . the critical value of the control parameter @xmath84 , where the symanzik - function @xmath85 contains simple ir - poles at all values @xmath86 with @xmath87 , becomes formally zero in the @xmath73-expansion . note that the renormalization factors are at any order in perturbation theory polynomials in @xmath88 , and hence it is guaranteed that they are well behaved in the limit @xmath63 . in their seminal work on linear rrn , hl showed , based on earlier ideas by stephen @xcite , that the field theoretic hamiltonian for the potts model can be extended to describe the rrn . we refer to their field theoretic model as the hl model . shortly after the hl model was introduced , harris @xcite took up ideas by kenkel and straley @xcite and generalized the hl model so that it captures nonlinear voltage - current characteristics of the type @xmath9 . more precisely speaking , this nonlinear rrn features bonds between nearest neighboring sites @xmath89 and @xmath90 that obey a generalized ohm s law @xmath91 , where @xmath92 is the nonlinear bond - resistance . as is customary , we use here and in the following the symbol @xmath10 for the nonlinearity parameter . this parameter must not be confused with the magnitude of the space position @xmath1 . the nonlinearity parameter always appears as a power or a subscript , respectively , and when read in its context , the meaning of @xmath10 should be clear . the main benefit of the nonlinear rrn is that it allows to map out the important fractal substructures of percolation clusters by taking the appropriate limits with respect to @xmath10 . of course , one retrieves the linear rrn for @xmath15 . we will refer to the model introduced in ref . @xcite as the nonlinear hl model . to capture the rrn , one distributes the @xmath93 states of the @xmath54-state potts model on a @xmath94-dimensional periodic lattice ( torus ) with site - coordinates @xmath95 , where @xmath96 . the combination @xmath97 is a free parameter of the theory which plays a subtle role . we will comment on this role as we move along . the order parameter field of the ( nonlinear ) hl model is given by the generalized potts spin @xmath98 subject to the constraint @xmath99 it is related to its fourier transformed counterpart via @xmath100 where @xmath101 with @xmath102 and @xmath103 . the prime in eq . ( [ voltagecurrenttrafo ] ) indicates that @xmath104 is excluded from the sum . in the following , we will drop this prime for notational simplicity unless stated otherwise . through this fourier transformation , the constraint ( [ constraint ] ) translates into @xmath105 . expressed in terms of @xmath106 , the nonlinear hl model takes on the form @xmath107 \tilde{\varphi}_{{\vec{\lambda } } } \nonumber \\ & \qquad\qquad + \frac{g}{6}\left . \tilde{\varphi}_{{\vec{\lambda}}}\tilde{\varphi}_{\vec{\lambda } ^{\prime}}\tilde{\varphi}_{-\vec{\lambda}-\vec{\lambda}^{\prime}}\bigg\ } , \end{aligned}\ ] ] where @xmath108 the control parameter @xmath109 is proportional to the resistance of the occupied bonds . the parameter @xmath10 specifies the non - linearity of the network with @xmath110 corresponding to the linear rrn . the hl model constitutes a potts model with a specific quadratic symmetry breaking . purely geometrical percolation is retrieved for @xmath111 provided one takes the limit @xmath112 with @xmath113 finite . in contrast , for the model to describe the ( nonlinear ) rrn , one has to take the limits @xmath114 and @xmath115 ( keeping @xmath116 fixed ) with the order of the limits being of crucial importance . in the following , we will refer to this double limit as the rrn - limit . there are some subtle , however important , features of the theory associated with the rrn - limit . as long as @xmath113 is finite , we have a model for percolation with quadratic symmetry breaking which can be handled by simple insertions of the operator @xmath117 . however , in the continuum - limit of the @xmath118-variables , @xmath119 , sums of the form @xmath120 diverge like @xmath121 even if @xmath112 , and a more careful treatment is in order @xcite . in the rrn - limit , the free parameter @xmath122 becomes obsolete , or in the language of field theory , it becomes a redundant parameter . as such , it is possible to eliminate it via rescaling of @xmath109 . that this elimination can indeed be achieved hinges on the fact that in perturbation theory in conjunction with the rrn - limit , @xmath109 and @xmath123 always appear in the combination @xmath124 . this in turn implies that only this combination , as opposed to @xmath109 or @xmath123 alone , has a well defined engineering dimension , namely @xmath125 like @xmath21 or @xmath126 . the success of the hl model and its nonlinear generalization stems , at least in part , from the fact that the two - point correlation function @xmath127 provides for an elegant and efficient route to calculate the average resistance between two points on the same cluster . in the following , we will use the notation @xmath128 for the two - point function , as we can because @xmath129 and @xmath109 always appear in the combination @xmath124 , see above . that the two - point function ( [ defcorr ] ) is helpful in calculating the average resistance becomes transparent if one notices that it constitutes a generating function : @xmath130_{av } \nonumber \\ & = \chi({\bm{{\rm r } } } ) \left[{\exp } \big ( -w \lambda_r ( \vec{\lambda}){r}_r ( { \bm{{\rm r } } } | c ) \big)\right]_{av}^\prime \ , .\end{aligned}\ ] ] where @xmath131 is proportional to the total nonlinear resistance between an arbitrary point @xmath132 and @xmath133 for a given random percolation configuration @xmath134 . here , we have absorbed the resistance @xmath135 of the conducting microscopic bonds into the control parameter @xmath109 such that @xmath136 . @xmath137 is an indicator function which is one if @xmath138 is connected to @xmath133 in configuration @xmath134 , and zero otherwise . @xmath139_{av}$ ] denotes the disorder average over all configurations of the diluted lattice . @xmath139_{av}^\prime$ ] stands for disorder averaging conditional to the constraint that @xmath138 and @xmath133 are connected . @xmath140_{av}\end{aligned}\ ] ] is the usual percolation correlation function , i.e. , the probability for any two sites a distance @xmath141 apart being connected . from eq . ( [ genfkt ] ) it follows that one can extract the average resistance @xmath142_{av}^\prime\end{aligned}\ ] ] and its moments simply via taking the derivatives of @xmath143 with respect to @xmath144 . as mentioned above , the great value of the nonlinear rrn lies in the fact that it can be used to map out different fractal substructures of percolation clusters . inspection of the overall dissipated electric power shows readily that @xmath145 where @xmath146 is proportional to the average mass ( number of bonds ) of the backbone ( between any two site a distance @xmath147 apart , of course ) . moreover , it has been shown by blumenfeld and aharony @xcite that @xmath148 where @xmath149 stands for the mass of the red bonds and @xmath150 where @xmath151 is the mass ( length ) the shortest self - avoiding walk ( saw ) . likewise , on can show that @xmath152 where @xmath153 is the mass ( length ) of the longest saw . as alluded to above , it is not known how to extract the mass ( length ) @xmath154 of the average saw from @xmath155 . however , we recently proposed @xcite a method for calculating @xmath154 based on the rrn that does not invoke @xmath155 directly . the idea behind this method is exploit the real - world interpretation of the rrn feynman diagrams , which will be explained in the next subsection , by placing saws onto these diagrams ( for details see ref . @xcite ) . using this approach , one can calculate @xmath154 in much the same way as @xmath151 and @xmath153 . from the above , it should be clear that the renormalization factors @xmath80 , @xmath81 and @xmath82 for the rrn are identical to those for the potts model with @xmath93 states . the control parameter @xmath109 , however , needs an independent renormalization @xmath156 with the renormalization factor @xmath157 having no counterpart in the usual potts model . this renormalization can be calculated in an elegant and efficient way by using our real world interpretation of the hl model s feynman diagrams @xcite . this interpretation is based on the observation that the diagrammatic perturbation theory of the hl model in the rrn - limit can be formulated in such a way that the feynman diagrams resemble real rrns . in this approach , the diagrams feature conducting propagators corresponding to occupied , conducting bonds and insulating propagators corresponding to open bonds . the conducting bonds carry replica currents @xmath158 . the resistance of a conducting bond is given by its schwinger parameter in a parametric representation of the propagators . because of the interpretation of @xmath129 ( apart from the factor @xmath159 ) as a replicated current , we refer to the space in which @xmath129 takes on its values as the replicated current space . likewise , we refer to the space in which the conjugate variable @xmath160 lives as the replicated voltage space . when this real - world interpretation is used , the task of calculating @xmath157 is by and large reduced to calculating the total nonlinear resistance of the conducting feynman diagrams . to one - loop order , one obtains @xcite @xmath161 the function @xmath162 is given by @xmath163 ^{r}}\,\end{aligned}\ ] ] note that it vanishes for @xmath164 und @xmath165 if @xmath10 is positive . for negative @xmath10 , @xmath162 it takes the form @xmath166 ^{\left\vert r\right\vert } \,,\ ] ] which is equal to @xmath167 for @xmath164 und @xmath165 . here , we encounter another manifestation of the discontinuity at @xmath16 which , physically , is related to the different types of saws on the percolation clusters . following standard procedures of renormalized field theory , one can extract from the above renormalizations a gell - mann low renormalization group equation ( rge ) that governs the scaling behavior of the correlation functions . for the 2-point functions , in particular , this rge leads to the well known result @xmath168 were @xmath169 is , up to scaling factors , a universal scaling function . the critical exponents @xmath25 and @xmath26 are known to 3-loop order @xcite . to 2-loop order , they are given by @xmath170 the resistance exponent @xmath37 is know for general @xmath10 to 1-loop order up to a remaining integral @xcite , @xmath171 with @xmath162 given by eq . ( [ defcrgreaterzero ] ) . for specific values of @xmath10 , @xmath37 is known to higher order . for the linear rrn , the resistance exponent @xmath172 is known to 2-loop order @xcite . in the limit @xmath11 relevant for the backbone , we calculated @xmath37 to 3-loop order @xcite . in the limit @xmath12 relevant for the red bonds , we showed explicitly to 3-loop order that @xmath173 @xcite as had to be expected from rigorous results by coniglio @xcite . furthermore , we calculated @xmath37 in the limit @xmath13 describing the shortest saw @xcite as well as the exponents governing the longest and the average saw to 2-loop order @xcite . as discussed in the introduction , it is useful for a comprehensive characterization of the properties of the rrn and the various fractal substructures of percolation clusters to go beyond just studying the average nonlinear resistance or its higher moments and to address the more general problem of the form of the distribution function of the nonlinear resistance . the distribution @xmath174 of the conditional probability that the total nonlinear resistance between @xmath1 and @xmath133 is @xmath175 given that @xmath1 and @xmath133 are on the same cluster can be defined as @xmath176_{av}^\prime \ , .\ ] ] comparison with eq . ( [ genfkt ] ) reveals that @xmath177 is the laplace transform of @xmath174 . thus , one can calculate @xmath174 via calculating the two point correlation function @xmath178 and the applying inverse laplace transformation to @xmath179 , @xmath180 this is the main route that we will follow in the remainder of this paper . the distributions of the various fractal masses can be defined by modifying eq . ( [ defresdistr ] ) in an evident manner . these distributions can be extracted from @xmath174 simply by taking the appropriate limit with respect to @xmath10 except for the distribution @xmath181 of the length @xmath182 of the average saw . nevertheless , by combining the route described above with the method proposed in ref . @xcite , we can calculate @xmath181 in much the same way as the distributions of the other fractal masses . before embarking on field theory , we conclude this review section by considering mean field theory . at 0-loop order , the 2-point correlation function is given simply by the fourier transform of the gaussian propagator , @xmath183 where @xmath184 is an abbreviation for @xmath185 , and where @xmath186 is the modified bessel function ( basset function ) with index @xmath187 . using this 2-point function in conjunction with eq . ( [ defpi ] ) in eq . ( [ inversetrafoyieldingp ] ) and introducing the quantity @xmath188 which becomes unity at criticality , @xmath189 , we obtain @xcite @xmath190 for the resistance distribution in mean field theory . note that , from this expression , one can read off the mean field results for the correlation length and resistance exponents , @xmath191 and @xmath173 . the mean - field result for @xmath181 is of the same form as eq . ( [ eq : meanfielddist ] ) with @xmath7 replaced by @xmath182 . this implies the well known mean - field result @xmath192 . at criticality , the distribution function takes on the scaling form ( [ scalefromprcrit ] ) . the corresponding mean field scaling function is plotted in fig . [ fig : meanfielddist ] . for large and small arguments , it behaves asymptotically as the leading term in eq . ( [ omegatsmalls ] ) and eq . ( [ omegatlarges ] ) , respectively , with @xmath193 and @xmath194= 2 $ ] . ) with @xmath195 and @xmath196.,width=294 ] as mentioned above , hl based their calculation on the assumption that the unconditional distribution function @xmath197 becomes independent of @xmath1 for @xmath198 . noting from eq . ( [ scaleform2point ] ) that @xmath199 and recalling from eq . ( [ omegatsmalls ] ) that @xmath200 with @xmath201 in this limit , we find @xmath202 with @xmath203 . in mean field theory , where @xmath204 and @xmath205 , @xmath206 vanishes so that @xmath207 is indeed independent of @xmath1 . beyond mean field theory , however , @xmath206 is non - zero , and the assumption by hl is incorrect . because of this shortcoming , the hl result @xmath208 for @xmath209 is not correct for @xmath210 , although it produces the correct mean - field value . now we turn to renormalized field theory to study the connection probability and the nonlinear resistance distribution beyond mean field theory . in particular , we will calculate their asymptotic forms in the limits @xmath211 and @xmath212 , respectively . these calculations are quite non - trivial because we need to determine the short - distance behavior of the 2-point correlation function . this behavior can be analyzed systematically with help of the well established short - distance expansion @xcite . we will explain the application of this technique to the current problem in some detail in sec . [ shortdistanceexpansion ] . it turns out , that the composite field or composite operator @xmath213 plays a pivotal role in the short - distance expansion . in this section we discuss the renormalization of this and related operators employing group theory @xcite . our approach is guided by the work of wallace and young @xcite , and we review parts of their work in a language that is adapted to the problem at hand . we are interested in the direct product of the order parameter with itself . as discussed below eq . ( [ pottsconstraintoriginal ] ) , the order parameter @xmath98 is an object which transforms as the fundamental irreducible representation @xmath59 of the symmetric group @xmath58 , if the invariant identity representation @xmath54 is ruled out through the constraint @xmath214 . the product of the fundamental representation with itself is reducible and it can be decomposed into a sum of irreducible representations , @xmath215 comprising the identity representation @xmath54 , the fundamental representation @xmath59 , the symmetric tensor representation @xmath216 and antisymmetric tensor representation @xmath217 . the latter is absent for a product of two identical fields . applied to the product @xmath218 , the decomposition leads to the identity representation @xmath219 the fundamental representation @xmath220 and the symmetric tensor representation @xmath221 to simplify the discussion to follow , it is useful to switch from replicated voltage space to replicated current space via the fourier transformation ( [ voltagecurrenttrafo ] ) . then , the above operators take on the form [ fouriertransirredops ] @xmath222 if @xmath129 and @xmath223 are non - zero . otherwise @xmath224 . hence , there does not exist any component with zero current . as mentioned above , it will be the operator @xmath225 which plays a central role in the sde . evidently , this operator satisfies @xmath226 . as we will discuss in the following , the renormalization of @xmath227 is intimately related to that of the operators @xmath228 and @xmath229 . equation ( [ fouriertransirredops ] ) implies that the operators @xmath227 , @xmath228 and @xmath230 are related for @xmath231 via @xmath232 note that the decomposition of @xmath227 in irreducible parts contains a factor @xmath233 which could be potentially problematic in the limit @xmath63 . this factor , however , cancels in perturbation theory . form the group theoretic considerations above , we draw the three important conclusions . ( i ) the operator @xmath234 can mix additively with the order parameter field @xmath106 and its derivatives because both transform according to the same irreducible representation @xmath59 . ( ii ) the operators @xmath228 and @xmath229 are renormalizable individually by multiplicative renormalization , because they belong to representations which are different from one another and @xmath59 . ( iii ) the operator @xmath227 , which will play a central role in the sde , contains both the independent renormalizations of @xmath228 and @xmath230 and mixes these . equipped with this information , we will now determine the renormalizations of the various composite operators . with the renormalization scheme ( [ reno ] ) , the renormalized version of the generalized potts hamiltonian is given by @xmath235 \tilde{\varphi } _ { _ { \vec{\lambda } } } \nonumber\\ & + z_{u}^{1/2}\ , \frac{g}{6}\sum_{\vec{\lambda},\vec{\lambda}^{\prime}}\tilde{\varphi } _ { _ { \vec{\lambda}}}\tilde{\varphi } _ { \vec{\lambda}^{\prime}}\tilde{\varphi } _ { -\vec{\lambda}-\vec{\lambda}^{\prime}}\bigg\ } \,.\end{aligned}\ ] ] from this renormalized hamiltonian , we can infer without much effort the renormalizations of @xmath228 and @xmath234 . as far as @xmath228 is concerned , we note that if @xmath21 is viewed as a function of @xmath1 one can generate insertions of the operator @xmath236 into renormalized correlation functions via functional differentiation with respect to @xmath237 . thus , the renormalized version @xmath238 of @xmath228 is given by @xmath239 turning to @xmath234 , we note that the functional derivative with respect to @xmath240 leads to the equation of motion @xmath241 where we have suppressed the current - index of most of the fields for notational simplicity . as a consequence of this equation of motion , insertions of the operator @xmath242 into renormalized correlation functions are already completely renormalized . this implies that the renormalized version @xmath243 of @xmath234 is given by @xmath244 as noted above , @xmath234 mixes under renormalization additively with @xmath106 and its derivatives . next , we move on to @xmath227 and @xmath230 . our strategy is to determine the renormalization of @xmath227 with the help of that of @xmath230 and @xmath228 via relation ( [ relabphi ] ) , and we will switch back and forth between considering @xmath227 and @xmath230 to achieve this goal . as we know from our group theoretic considerations above , the renormalization of @xmath230 is independent of that of @xmath228 and @xmath234 , and thus it is renormalizable by multiplicative renormalization . also because of this independence , we can not simply extract the renormalization of @xmath230 from the above . as the typical quadratic symmetry - breaking term in the hamiltonian , the tensor @xmath245 determines the so - called crossover exponent . thus , we denote its renormalization factor by @xmath246 so that @xmath247 to approach the renormalization of @xmath227 through perturbation theory , we consider its insertions into vertex functions @xmath248 , i.e. , @xmath249 . according to the above , @xmath227 itself is not renormalizable multiplicatively as it requires additive contributions from @xmath228 , i.e. , the renormalized version of @xmath227 will be of the type @xmath250 any result for @xmath251 produced by diagrammatic perturbation theory has to be of the form @xmath252 where @xmath253 and @xmath254 are polynomials in @xmath88 which at each order in @xmath88 have a perturbation expansion in powers of @xmath76 containing @xmath73-poles . summing both sides of eq . ( [ formgamma2ains ] ) over @xmath129 produces @xmath255 we already know that @xmath256 is renormalized by multiplication with @xmath257 , i.e. , @xmath258 is free of @xmath73-poles , and thus @xmath259 from which @xmath257 can be constructed order for order in @xmath76 with polynomial factors in @xmath88 . equipped with this information , we now insert the renormalized version of @xmath227 as given in eq . ( [ formrena ] ) into the renormalized two - point vertex function and determine @xmath260 and @xmath261 such that @xmath262 is cured at any order in @xmath76 from @xmath73-poles , i.e. , such that @xmath263 is finite . it follows from summation over the @xmath88 nonzero values of @xmath129 that @xmath264 comparison of eqs . ( [ finite1 ] ) and ( [ finite2 ] ) now reveals that @xmath265 hence , the remaining task in determining the renormalization of @xmath227 is to determine @xmath266 . to this end , we substitute the result ( [ resy ] ) into eq . ( [ formrena ] ) and we then note that @xmath267 which reveals that @xmath268 because @xmath269 , @xmath257 and @xmath261 are at each order in @xmath76 polynomials in @xmath88 , it follows that in particular for @xmath270 that @xmath271 whereby we now have completely determined the renormalization of @xmath227 in the percolation limit . note that these results are valid to arbitrary order in perturbation theory . to double - check our reasoning , we also calculated @xmath246 and @xmath261 explicitly in a two - loop calculation yielding @xmath272 \frac{u^{2}}{4\varepsilon}\,,\\ \label{resy2loop } y & = -\frac{u}{\varepsilon}+\left [ \frac{5n-11}{\varepsilon}-\frac { 23n-65}{12}\right ] \frac{u^{2}}{4\varepsilon } \,.\end{aligned}\ ] ] in conjunction with eq . ( [ 2looprestau ] ) , the results clearly confirm our above general results . it is worth pointing out that the equality of temperature renormalization and crossover renormalization in the replica limit is a generic feature of all replicated field theories which therefore have a crossover exponent which is strictly equal to 1 . for example , this is also true for the @xmath273 limit of the @xmath88-component @xmath274-theory describing linear polymers . surprisingly , this important general feature has not been much appreciated in the literature , at least as far as we know . as discussed in sec . [ probdistdefandmf ] , the distribution function for the nonlinear resistance can be extracted from the two - point correlation function @xmath275 , whose scaling form is well known to be given by eq . ( [ scaleform2point ] ) . for treating the limit @xmath276 , however , we need to know how the scaling function @xmath277 in eq . ( [ scaleform2point ] ) behaves for small arguments . naively , one might attempt a taylor expansion of @xmath277 for small arguments which is erroneous , however , because the scaling function contains infra - red ( ir ) singularities . to treat these ir singularities properly , one has to resort to a short - distance expansion ( sde ) . for the @xmath278-model , this was done a long time ago by brezin , amit , and zinn - justin @xcite , and brezin , dedominicis , and zinn - justin @xcite . to follow their route , we will briefly review some essentials of the sde and its basis , the operator product expansion ( ope ) @xcite . then we will investigate the short distance behavior of @xmath275 by applying the sde which will involve the operators @xmath227 and @xmath228 whose renormalization we discussed in sec . [ representations ] . in the following , we will suppress the subscript @xmath7 on renormalized operators for notational simplicity . the ope is based on the fact that a product of two renormalized operators @xmath279 and @xmath280 ( which may be plain or composite fields ) at two different but close points can be expanded as @xcite @xmath281 where the @xmath282 are pure functions of @xmath283 . of course , any decomposition into symmetry representations has to be done such that both sides of this equation remain consistent . the statement ( [ generalexp ] ) is meaningful only when inserted in correlation or vertex functions and only in the critical domain . the same holds true for many of the equations in the remainder of this section , and we ask the reader to keep this in mind when going through this section . if the @xmath279 are eigenoperators of the rg , i.e. , if @xmath284 where @xmath285 is the scaling dimension of @xmath279 , then the @xmath282 are given by @xmath286 the higher the scaling dimension of an operator @xmath287 in the expansion ( [ generalexp ] ) , the less it contributes if @xmath5 is small . thus one naturally obtains an expansion for short distances , or sde , if one orders the operators in the ope according to their scaling dimension . in general , the unit operator will be present on the right hand side of eq . ( [ generalexp ] ) . however , if one considers connected correlation functions , as we do , it does not contribute . the leading contributing operators are then those which have at @xmath288 the naive dimension @xmath289 , viz . @xmath290 and @xmath291 , and thus the leading contributing terms in the sde of @xmath292 read @xmath293 all other composite fields such as @xmath294(\mathbf{r})$ ] , @xmath295(\mathbf{r})$ ] , @xmath296(\mathbf{r})$ ] , and their sum over @xmath129 do at least have naive dimension @xmath297 at @xmath288 . the same holds true for the operator @xmath298 ( \mathbf{r } ) = : w \ , \mathcal{c}^\prime(\mathbf{r } ) \,.\end{aligned}\ ] ] note , that this composite operator must be renormalized as a whole ; its renormalization does not simply follow from the renormalization of its part @xmath299\sim \mathcal{a}_{\vec{\lambda}}$ ] in the rrn - limit . this was overlooked in earlier work @xcite and led the wrong result @xmath173 for the crossover exponents . however , the operator @xmath300 plays a special role due to the double limit @xmath114 and @xmath115 . when the other dimension @xmath297 operators such as @xmath301(\mathbf{r})$ ] are inserted into feynman diagrams , they leave the superficial degree of divergence unchanged . when @xmath302 is inserted into feynman diagrams , it reduces the superficial degree of divergence of any of these diagrams by @xmath125 just as @xmath291 does . thus , @xmath302 should be included on the same footing as @xmath291 as far as making insertions to avoid extra uv singularities if the limit @xmath303 is concerned . all other operators whose naive dimension is @xmath297 or higher , can be neglected in the following . having established the specific form of the sde for the problem at hand , we will now turn to various rges that will eventually reveal the short distance behavior of the 2-point correlation function . these rges can be set up using standard procedures exploiting the fact that the unrenormalized theory must be independent of the inverse length - scale @xmath77 introduced by renormalization @xcite , cf . the renormalization scheme ( [ reno ] ) . in the rges for the current problem we encounter on various occasions the rg differential operator @xmath304 which contains the rg - functions @xmath305 and the rg - function @xmath306 where the @xmath307 indicates that bare quantities are kept fixed while taking the derivatives . all these rg - functions are well known and explicit results for them can be straightforwardly reconstructed from the renormalizations and @xmath80-factors reviewed in sec . [ reviewpottsrrn ] . first , let us state the rges for the operators @xmath227 , @xmath228 and @xmath308 , which we will need as input as we move along : [ rgeabc ] @xmath309 \ , \mathcal{a}_{\vec{\lambda}}(\mathbf{r})+y\ , \mathcal{b}(\mathbf{r})=0\ , , \\ & \left [ d_\mu + \kappa\right ] \ , \mathcal{b}(\mathbf{r})=0\ , , \\ & \left [ d_\mu + \zeta_r\right ] \ , \mathcal{c}^{\prime}(\mathbf{r})=0\,,\end{aligned}\ ] ] with the rg - function @xmath310 given by @xmath311 where we used on the right hand side our 2-loop result ( [ resy2loop ] ) for @xmath261 . next , we set up and solve rges for the sde coefficients @xmath312 and @xmath313 . from eq . ( [ specificformsde ] ) we learn by using the distributivity of @xmath314 and by exploiting the rges ( [ rgeabc ] ) that @xmath315 \left ( \tilde{\varphi}_{\vec{\lambda } } ( \mathbf{r}+{\bm{{\rm r}}}/2 ) \tilde{\varphi}_{-\vec{\lambda}}(\mathbf{r}-{\bm{{\rm r}}}/2)\right ) \nonumber\\ & \approx \left [ d_\mu + \gamma\right ] \left ( c_{\mathcal{a}}({\bm{{\rm r}}})\mathcal{a}_{\vec{\lambda}}(\mathbf{r})+c_{\mathcal{b}}({\bm{{\rm r}}})\mathcal{b}(\mathbf{r})\right ) \nonumber\\ & = \mathcal{a}_{\vec{\lambda}}(\mathbf{r})\left [ d_\mu + \gamma\right ] c_{\mathcal{a}}({\bm{{\rm r}}})+c_{\mathcal{a}}({\bm{{\rm r}}})d_\mu \mathcal{a}_{\vec{\lambda}}(\mathbf{r } ) \nonumber\\ & + \mathcal{b}(\mathbf{r})\left [ d_\mu + \gamma\right ] c_{\mathcal{b}}({\bm{{\rm r}}})+c_{\mathcal{b}}({\bm{{\rm r}}})d_\mu \mathcal{b}(\mathbf{r})\nonumber\\ & = \mathcal{a}_{\vec{\lambda}}(\mathbf{r})\left [ d_\mu + \gamma-\kappa\right ] c_{\mathcal{a}}({\bm{{\rm r } } } ) \nonumber\\ & + \mathcal{b}(\mathbf{r } ) \left ( \left [ d_\mu + \gamma-\kappa\right ] c_{\mathcal{b}}({\bm{{\rm r}}})-yc_{\mathcal{a}}({\bm{{\rm r}}})\right ) \,.\end{aligned}\ ] ] because the operators @xmath227 and @xmath228 are linearly independent , this implies that @xmath316 c_{\mathcal{a}}(\mathbf{r})=0\,,\\ & \left [ d_\mu + \gamma-\kappa\right ] c_{\mathcal{b}}(\mathbf{r})-y\ , c_{\mathcal{a}}(\mathbf{r})=0\,.\end{aligned}\ ] ] at the rg fixed - point @xmath317 , determined by @xmath318 , these coupled rges for @xmath312 and @xmath313 are readily solved with the result @xmath319 \,.\end{aligned}\ ] ] the additional logarithmic contribution @xmath320 appears because the operators @xmath227 and @xmath228 have the same scaling dimension , @xmath321 , in the percolation limit @xmath63 . @xmath322 and @xmath323 , where @xmath324 and so on , are the usual critical percolation - exponents , cf . [ renscalingint ] . for completeness , and because we will use this later on , we also note that the resistance exponent @xmath37 is determined by the fixed point values of the rg - functions as @xmath325 . now we are in the position to address the short distance behavior of @xmath275 . we must not naively simply plug the expansion ( [ specificformsde ] ) into the correlation function ( [ defcorr ] ) because this would generate correlation functions such as @xmath326 which require additional additive renormalizations . this problem can be avoided by taking derivatives with respect to @xmath21 and @xmath109 which generate insertions of @xmath228 and @xmath308 that reduce the superficial degree of divergence . it turns out that 3-fold differentiation is sufficient to suppress additional @xmath327-singularities . in the following , we denote derivatives with respect to @xmath328 and @xmath329 with superscripts @xmath330 and @xmath113 , e.g. , @xmath331 where @xmath332 and likewise for @xmath333 . till noted otherwise , we assume that @xmath334 . then it is safe to use the sde ( [ specificformsde ] ) in the connected correlation function with insertions ( [ corrwithder ] ) , which leads to @xmath335 this equation makes it transparent that the expansion has the benefit of factorizing the spatial dependence from the dependence on the control parameters @xmath21 and @xmath109 . this feature will be important when it comes to solving the rge for @xmath336 . as pointed out above , this expansion is valid only in the critical domain , i.e. , for @xmath337 and @xmath338 . next , we discuss the 2 contributions to eq . ( [ g2exp ] ) . the second term on its right hand side vanishes in the limit @xmath63 . this can be seen as follows . let us abbreviate @xmath339 , where we have reinstated the prime on the sums over replicated currents to emphasize that @xmath104 is excluded from these sums , c.f . the comment below eq . ( [ deflambdaalpha ] ) . switching to an unrestricted sum , we can rewrite @xmath340 as @xmath341 . by expressing @xmath342 in terms of its laplace transform , switching from discrete to continuous replica currents and then performing the integration over @xmath129 with help of the saddle - point method , we find that @xmath343 in the limit @xmath112 . thus , the second term on the right hand side of eq . ( [ g2exp ] ) vanishes in this limit . for discussing the first term on the right hand side of eq . ( [ g2exp ] ) , we denote @xmath344 an rge for this connected correlation function is readily derived : @xmath345 f_{\mathcal{a}}^{(n , m)}(\tau , w\lambda_r(\vec{\lambda}),\mu)=0\,.\ ] ] at the fixed point , it is straightforward to solve this rge . the result is @xmath346 this solution needs to be augmented with a dimensional analysis , @xmath347 to account for the non - vanishing naive dimensions of its various ingredients . by combining eqs . ( [ solrgefa ] ) and ( [ dimanargefa ] ) and by exploiting freedom of choice regarding the flow parameter @xmath348 by setting @xmath349 , we obtain the scaling form @xmath350 where @xmath351 is a scaling function . for @xmath352 around @xmath353 , this scaling function is certainly analytic because the model is for @xmath354 away from its critical point , and hence there can be no ir - singularities . the situation is more subtle for @xmath355 because then the question arises whether a nonzero @xmath109 makes the system noncritical or not . the most economical way to address this question is to calculate @xmath356 in dimensional regularization to lowest order beyond mean - field theory . diagrammatically , this calculation involves the attaching of external legs ( propagators ) to the usual 1-loop self - energy diagrams followed by an integration over the external momentum ( which amounts essentially to a 2-loop calculation ) . note that it is sufficient to work for @xmath357 in dimensional regularization , at least formally , because this regularization formally sets the otherwise required additive renormalizations equal to zero . we perform this calculation for the physically relevant limits of @xmath10 discussed above . for @xmath358 , we find that @xmath359 is sufficient to make the system noncritical for @xmath35 . thus , we expect that [ assformsfnm ] @xmath360 for @xmath361 , where @xmath362 is some constant . for @xmath363 it appears as if @xmath359 is not sufficient to make the system noncritical for @xmath35 . however , we encounter the problem that the calculation produces a contribution proportional to @xmath364 that has no counterpart in mean - field theory , i.e. , @xmath356 is not renormalizable with the usual renormalization factors for @xmath363 . currently , we do not know how to overcome this problem or what its physical significance might be . nevertheless , we expect that @xmath365 for @xmath361 if @xmath363 . until the problem with the renormalizability is resolved , this formula should be understood as a proposal , as do our final results for the distribution functions pertaining to @xmath363 . we plan to return to the unresolved difficulties for @xmath363 in a future publication . now , we move on to the rge for @xmath336 . in its generic form , it reads @xmath366 g^{(n , m)}({\bm{{\rm r}}},\tau , w\lambda_r(\vec{\lambda}),\mu)=0\,.\end{aligned}\ ] ] for solving this equation , it is useful to split it up into 2 parts such that one of the parts collects the derivatives with respect to @xmath21 and @xmath109 contained in @xmath314 . at the fixed point @xmath317 , this procedure leads to @xmath367g^{(n , m ) } \nonumber\\ & = -[\kappa^{\ast}\tau\partial_{\tau}+\zeta_r^{\ast}w\partial_{w}]g^{(n , m)}\nonumber\\ & = \kappa^{\ast}\tau g^{(n+1,m)}+\zeta_r^{\ast}w\lambda_r(\vec{\lambda})g^{(n , m+1)}\nonumber\\ & \approx c_{\mathcal{a}}\big[\kappa^{\ast}\tau f_{\mathcal{a}}^{(n+1,m)}+\zeta_r^{\ast } w\lambda_r(\vec{\lambda } ) f_{\mathcal{a}}^{(n , m+1)}\big]\,,\end{aligned}\ ] ] where we have dropped the arguments of @xmath336 and so on for notational simplicity . note that the part of the equation that has been moved over to the right hand side constitutes the leading terms in an expansion of @xmath336 in powers of @xmath21 and @xmath368 . thus , after the rge is written in the form of eq . ( [ rgeg2sep ] ) , the sde is safe for @xmath369 . for the right hand of the rge ( [ rgeg2sep ] ) , we can draw on the results we established above for @xmath312 and @xmath370 . we obtain @xmath371g^{(n , m ) } ( \mathbf{r},\tau , w\lambda_r(\vec{\lambda}),\mu)\nonumber\\ & \approx \mu^{d-2(n+m+1)}\left ( \mu\left\vert \mathbf{r}\right\vert \right ) ^{2-\eta-1/\nu}\left ( \tau/\mu^{2}\right ) ^{(d\nu-1)-(n+m\phi_r ) } \nonumber\\ & \times \phi^{(n , m)}\left(\frac{w\lambda_r(\vec{\lambda})/\mu^{2}}{(\tau/\mu^{2})^{\phi_r } } \right ) , \end{aligned}\ ] ] where @xmath372 is a scaling function which sums up the contributions of the different @xmath373 , and whose details are unimportant for our main quest . any solution of the inhomogenous rge consists , of course , of the general solution of the underlying homogenous equation and a particular solution of the inhomogenous equation . the homogenous equation is independent of @xmath21 and @xmath109 , and its solution is found to be @xmath374 as a particular solution of the inhomogenous equation , we find @xmath375 where @xmath376 is some scaling function determined by a linear first order differential equation containing @xmath377 . now , we are finally in the position to write down the short distance scaling form of the 2-point correlation function via adding eqs . ( [ solhomrge ] ) and ( [ solinhomrge ] ) and subsequent integration with respect to @xmath21 and @xmath109 . introducing the scaling variables @xmath378 we arrive at @xmath379 where @xmath30 and so on are constants , and where @xmath380 is yet another scaling function that , in principle , can be calculated from @xmath376 . based on eq . ( [ assformsfnm ] ) , we expect that @xmath380 is analytic near @xmath381 , and for @xmath382 @xmath383 we conclude this section by asserting what additional contributions we get if we proceed to next order in the sde @xcite , i.e. , if we include operators that have at @xmath288 a naive dimension of @xmath297 . these operators are @xmath384 , @xmath385 , @xmath386 $ ] , @xmath387 $ ] and @xmath388 , and their respective scaling exponents are @xmath389 , @xmath390 , @xmath389 , @xmath391 and @xmath392 , where @xmath393 is the so - called wegner exponent . the operator @xmath388 does not contribute to the sde , because when inserted into correlation functions , it produces zero due to translational invariance : @xmath394 . therefore , the upshot here is that the dimension @xmath297 operators lead to additional analytic contributions in the brackets in eq . ( [ finresg2 ] ) which are of third order in @xmath28 and @xmath395 , such as @xmath396 , @xmath397 and so on , as well as additional non - analytic contributions of the form @xmath398 , @xmath399 and @xmath400 . in this section , we exploit our renormalization group results for 2-point correlation function derived in the previous section to determine the shape of connection probability and the probability distribution for the nonlinear resistance . in the remainder , we will suppress the inverse length scale @xmath77 for notational simplicity . as reviewed in sec . [ sec : conprobandavnonlres ] , the connection probability , i.e. , the probability for any two sites a distance @xmath147 apart being on the same cluster , is proportional to the 2-point correlation function of the rrn evaluated at @xmath401 . hence , the scaling form ( [ scaleform2point ] ) of the 2-point correlation function implies that @xmath402 with @xmath403 being a scaling function that depends on the scaling variable @xmath28 defined in eq . ( [ defscalevarx ] ) . in the following we will discuss the asymptotic forms of this scaling function for small and large @xmath28 . for small @xmath28 , our sde result ( [ finresg2 ] ) implies that @xmath29 where @xmath30 and so on are expansion coefficients . when we go to next order in the sde , we get additional contributions proportional to @xmath404 , @xmath405 und @xmath406 . now we turn to large x , i.e. , to distances @xmath407 that are large compared to the correlation length @xmath408 . these distances correspond to small wave vectors @xmath409 , for which we can expand the 2-point vertex function @xmath410 , i.e. , the inverse of the 2-point correlation function , as @xmath411 thus , we obtain by applying the inverse fourier transformation @xmath412 where we have omitted higher order terms . using @xmath413 , where @xmath414 , we get @xmath415 where @xmath33 is some constant . this reveals that the large-@xmath28 behavior of the scaling function of the connection probability is given by @xmath31 with constants @xmath32 and @xmath33 . to establish the general scaling form of the resistance distribution , we recall from sec . [ probdistdefandmf ] that this distribution can be extracted from the 2-point correlation function via inverse laplace transformation . inserting eq . ( [ scaleform2point ] ) into eq . ( [ inversetrafoyieldingp ] ) , we readily obtain @xmath416 with a scaling function @xmath417 . in the remainder , we focus on criticality , @xmath35 , where @xmath418 with @xmath419 , and we discuss the asymptotic forms of the distribution when the scaling variable @xmath420 is either small or large . for large @xmath38 , we draw on our sde results yielding @xmath421 this expansion is now inserted into eq . ( [ genformpres ] ) . in performing the inverse laplace transformation , the analytic contributions in @xmath395 merely produce shorts , @xmath422 , in the form of delta functions @xmath423 and their derivatives . what remains reads @xmath424 where @xmath425 . at the critical point , we obtain in particular @xmath426 where @xmath48 and so on are constants . here , we included the next higher order terms in the sde as discussed above . finally , we turn to small @xmath38 . to this end , we expand the 2-point vertex function for small wave vectors , @xmath427 where we denote @xmath428 in conformity with our conventions of sec . [ probdistdefandmf ] and where @xmath429 and @xmath430 are scaling functions . at the critical point , we can replace these functions by constants . by dropping terms that are subdominant for @xmath431 , we obtain @xmath432 where @xmath433 in accordance with eq . ( [ defscalevary ] ) . feeding this intermediate result into eq . ( [ genformpres ] ) , we obtain @xmath434 the remaining integration can be simplified by substituting @xmath435 which yields @xmath436 where @xmath437 is some constant . for large @xmath438 , we may use the saddle - point approximation provided that @xmath439 . however , @xmath440 for @xmath441 and @xmath110 as well as @xmath442 since @xmath443 ! we finally arrive at @xmath444 where @xmath44 and @xmath45 are constants . as discussed above , it is straightforward to extract from these general results the distributions of the linear total resistance as well ass the distributions of the fractal masses of the red bonds , the backbone and the shortest and longest saws by taking the appropriate limit with respect to @xmath10 . for @xmath181 , the analysis just presented in conjunction with the method of ref . @xcite leads at criticality to the same form as given in eqs . ( [ tauzeroform ] ) , ( [ largesform ] ) and ( [ smallsform ] ) . to transcribe the above results to the average saw , we just need to replace on the right hand side of these equations @xmath7 by @xmath182 , @xmath37 by @xmath445 and the constants @xmath48 and so on by other constants , say @xmath49 and so on . in summary , we calculated the asymptotic forms of the pair - connection probability , the distributions of the total resistance and fractal masses of the backbone , the red bonds and the shortest , the longest and the average self - avoiding walk between any two points on a cluster . our analysis drew solely on general , structural features of the underlying diagrammatic perturbation theory , and hence our results for the form of the distributions are valid to arbitrary loop order , although the critical exponents featured in these distributions are known only to finite order , of course . the distributions of the transport quantities such as the total resistance or the backbone mass etc . are essentially all of the same form . in this sense , these distributions share a high degree of universality . as far as future directions are concerned , we think it is worthwhile to study the renormalizability - problem for @xmath363 discussed in sec . [ shortdistanceexpansion ] and its implications in more detail . it would be interesting to have field theoretic results for distribution functions in directed and dynamic percolation . also , it would be interesting to see experimental or simulation results for distribution functions in any kind of percolation that might be compared to our results presented here or to future results . we hope that our work stimulates further interest in this subject . see , e.g. , d.j . amit , _ field theory , the renormalization group , and critical phenomena _ ( world scientific , singapore , 1984 ) ; j. zinn - justin , _ quantum field theory and critical phenomena _ ( clarendon , oxford , fourth edition 2002 ) . k.g . wilson , phys . rev . * 179 * , 1499 ( 1969 ) ; l. kadanoff , phys . * 23 * , 1430 ( 1969 ) ; w. zimmermann : in _ lectures on elementary particles and quantum field theory _ , edited by deser , grisaru and pendleton ( mit press , cambridge , mass . , 1970 )

boundary actions for three - dimensional quantum gravity in the discretized formalism of ponzano - regge are studied with a view towards understanding the boundary degrees of freedom . these degrees of freedom postulated in the holography hypothesis are supposed to be characteristic of quantum gravity theories . in particular it is expected that some of these degrees of freedom reside on black hole horizons . this paper is a study of these ideas in the context of a theory of quantum gravity that requires no additional structure such as supersymmetry or special gravitational backgrounds . lorentzian as well as euclidean regimes are examined . some surprising relationships to liouville theory and string theory in are found . gr - qc/0002092

gr-qc0002092

this article presents the calculation , using continuum and lattice methods , of boundary terms in 3-dimensional gravity . the gravity theory is presented in first order palatini form , this being a particular example of the general class of bf models @xcite as this is the most convenient presentation for deriving the discretization . we find a variety of boundary conditions , and discuss the significance of these for different types of boundaries in space - time . the bulk theory of three - dimensional gravity is well known to be a topological field theory , however it is also well known that three - dimensional topological field theories can give rise to non- topological boundary degrees of freedom , the classic example being the cs theory giving rise to a wzw model on the boundary @xcite . in the case of three dimensional gravity with cosmological constant , one can utilize a trick that relates the action to the difference of two cs actions , and then use the standard cs - wzw relationship , however the actual boundary conditions are a little more subtle . in three dimensions this is relevant to the @xmath0 space , or more generally to btz black hole solutions . in this paper we wish to understand in the context of discretization of quantum gravity the boundary degrees of freedom that correspond to black hole entropy . this paper is directed towards a longer study of boundary terms in gravity theories , ultimately in @xmath1 dimensions , with the hope of understanding directly in a theory of quantum gravity , the possibile origin of holographic phenomena , and of the microscopic details of black hole entropy , in particular well out of the supersymmetric and extremal limits which have been very well studied in the framework of string theory . the actual type of discretization that we consider here is maybe at first sight a bit unusual . the approach is originally due to ponzano and regge @xcite where they considered a simplicial decomposition of a three - manifold and the path - integral is then defined as a summation over the possible sets of lengths of the edges of the dual lattice . the alternative of course is to fix the size of the simplices and to form the path integral by summation over possible simplicial decompositions . for the major part of this paper , we will be discussing three dimensional models that have a topological invariance in the bulk and thus the fixed decomposition is somewhat innocuous but again the use of this simplicial decomposition also for the boundary where in general we believe there are physical degrees of freedom needs to be considered more cautiously . in addition we eventually need to extend our results to the realistic case of four - dimensional gravity where we do not even have topological invariance in the bulk making things more intricate though hopefully still manageable . we begin however , in the context of euclidean three - dimensional gravity where already we find some interesting results concerning the boundary theories . we will start off with a discussion of a discretization of the bf theory that corresponds to three - dimensional euclidean gravity in the framework of the ponzano - regge discretization , that is a discretization into tetrahedra with edges labelled by so(3 ) spins , and each tetrahedron then weighted in the path integral ( sum ) by the corresponding @xmath2 symbol . from this discretization we can then derive a boundary action and will compare this to what we may expect from the corresponding bf theory . we in fact find that there are two simple types of boundary conditions , one leads to a topological boundary theory and the other to a dynamical boundary theory . in addition we discuss mixed boundary conditions which are relevant for the boundary at infinity in @xmath0 for example . we discuss modifications to these boundary actions that arise when one replaces the group @xmath3 with @xmath4 which would correspond to gravity with lorentzian signature . we also discuss the regularization via quantum groups and find some interesting relationships to work on string theory and @xmath0/cft duality . finally we make some suggestions for understanding black hole entropy in this context and we discuss briefly the extension of these methods to four - dimensional quantum gravity . we will now turn to a discretization of the bf representation of three - dimensional gravity and show how it leads to the ponzano - regge action . the bf action is a generic action for a certain class of topological field theories , @xcite . for three - dimensional gravity it actually corresponds to the palatini first order action . we will mostly use the @xmath5 variables which are related to the gravity variables via the dictionary ; @xmath6 is the dreibein and @xmath7 is the curvature of the spin - connection @xmath8 . the basic action for three - dimensional gravity in this first order formulation is then @xmath9 where @xmath10 is the curvature two form of a potential one - form @xmath11 , and @xmath12 is the dreibein . these fields transform under the action of an @xmath3 gauge group . the invariance of the action consists of a gauge transformation in @xmath11 , @xmath13 , coupled with a local gauge rotation @xmath14 , and an additional invariance only acting on the dreibein ( reparametrization ) under which @xmath15 $ ] . the parameters of these transformations are in @xmath3 and it s lie algebra respectively . the first transformation expresses the local lorentz invariance , and the second the diffeomorphism invariance . the theory as formulated is diffeomorphism invariant with no explicit appearance of the metric in the action and thus topological . the constraint that the metric is torsion free , @xmath16 , in this first order form , arises from the @xmath11 equation of motion . the total group of local symmetry is @xmath17 @xcite . we will proceed now to a discrete formulation of three - dimensional gravity . we will carry out the discretization as a means of studying the continuum theory , however we would like to point out that in @xcite some arguments are given indicating that in three - dimensional gravity the space - time is necessarily discrete . our study in fact also indicates another possibile method to prove that three - dimensional gravity is discrete . one of the original motivations leading us to consider a discrete space - time approach to quantum gravity is the following . we will throughout this paper take the view that black holes in quantum gravity behave like quantum mechanical objects , and that this leads to unitarity in quantum gravity via some type of holographic mechanism @xcite . if one considers the black hole horizon to be a quantum object capable of storing and retransmitting information , then one would imagine that this horizon follows a null or even time - like path in space - time and that the region inside the global horizon is not something that an outside observer can ever see or discuss . this is the view of black - hole complementarity developed to reconcile the apparent contradiction that unitary black hole evaporation implies that observers outside the black hole view the physics of the horizon in a very different way to freely falling observers who fall into the horizon of a large black hole @xcite . as such one may view the formation of a black - hole as the expansion of a planckian bubble in space - time to become macroscopic . inside such a bubble there is nothing . thus it seems necessary to think of the microscopic structure of space - time to be a collection of bubbles . as such there is a discretization of space - time into units of size the planck length . there are a variety of ways to approach the discretization of the bf theory in three - dimensions , although all constructions give the same final result . for other discussions of the approach that we present here see @xcite . the simplest approach to discretization is to formally carry out the path integral over the b - field , as it is simply a lagrange mutiplier for the einstein equation . the result is , @xmath18 = \int { \cal d}a \prod_x \delta(e^{f(x)})\ ] ] where @xmath19 are the co - ordinates on the closed manifold @xmath20 , and the delta function is in the group manifold of @xmath3 . the delta function can be rewritten using the identity @xmath21 where @xmath22 and the sum is over all representations of the group g. using this identity we can write , @xmath18 = \int { \cal d}a \prod_x \sum_j ( 2j + 1 ) \chi_j(e^{f(x)}).\ ] ] to make this expression tractable we now discretize the manifold @xmath20 by dividing it into tetrahedra . from this tetrahedral decomposition , we construct a dual discretization for which the vertices are at the centre of the tetrahedra @xcite , the edges pass between the centres of adjacent tetrahedra , and the faces are then bound by these edges and each dual face will be pierced by precisely one edge of the original tetrahedral decomposition . in figure 1 we show the part of the dual lattice that will live inside one of the original tetrahedra . we now assign to every face of the dual lattice ( that is every edge of the original lattice ) a representation and to every edge of the dual lattice a group element as shown in figure 2 . the product of the group elements around a dual face is then the holonomy of that cycle and thus represents a discretization of the curvature . denoting the discretization of @xmath20 as @xmath23 we can finally write , @xmath24 = { \cal n}\int \prod_{e\in\delta } \sum_{j_e } du_e ( 2j_e + 1 ) \chi_{j_e}(\prod_{\partial\tilde{e}}u)\ ] ] where @xmath25 is a normalisation factor . in this expression , @xmath12 is an edge of the tetrahedral decomposition @xmath23 and @xmath26 is the face dual to the edge @xmath12 . to actually evaluate this expression we notice that the character can be written as a sum of products of the wigner function @xmath27 where @xmath28 is the group element corresponding to an edge of the dual graph , @xmath29 where @xmath30 . then for each edge of the dual graph there will appear in the integral over the corresponding group elements three wigner functions which can be evaluated immediately using , @xmath31 ( for these and other angular momentum identities that we use below and for the definitions of the various symbols that we use , we recommend that the reader refer to the very complete monograph @xcite ) . each tetrahedron thus will contribute two @xmath32 symbols for every face , thus eight @xmath32 symbols for every tetrahedron . half of these are summed over angular momentum projections in pairs , one of the pair coming from each of two tetrahedron with a common face and the orthogonality of the @xmath32 symbols ensures that this term becomes the identity . the remaining expression is such that summing over the projection quantum number of the angular momentum the four @xmath32 of a given tetrahedron gives a single @xmath2 symbol using the identity , @xmath33 the final result for a closed manifold @xmath34 and simplicial decomposition @xmath23 is ( see @xcite or @xcite for more details ) , @xmath35 this answer is the path sum proposed by ponzano and regge to be a discretization of three dimensional quantum gravity . in fact for a tetrahedron with edge lengths @xmath36 , the corresponding weight is the 6j symbol with angular momenta @xmath37 . in the semi - classical limit @xcite ( large angular momenta and vanishing planck length such that the combination @xmath38 remains constant ) , a 6j symbol actually becomes the cosine of the regge action for the tetrahedron as a direct discretization of three dimensional gravity ( the cosine arises as the bf theory path integral sums indiscriminately over positive , negative and degenerate values for @xmath39 ) . the regge action is the direct discretization of the einstein - hilbert action @xcite . @xmath40 @xmath41 is the angle between the normals to adjoining faces and @xmath42 is the length of the edge common to the two faces labeled by @xmath43 and @xmath44 . the extra factor of half between the edge length and corresponding angular momentum is for consistency in this semi - classical limit and we can intuitively justify it by noting that the length of the angular momentum vector for the representation of spin @xmath45 is actually @xmath46 which becomes @xmath47 in the limit of large angular momentum . the above expression for the discretized path sum is not quite complete . we have ignored the fact that there could in principle be some normalization factor in front of the sum , and in fact one would hope that there is such a factor simply because the sum itself is divergent . one can simply choose a normalization factor to subtract the divergence , however it is interesting to see how the divergence arises . this was already analyzed in the original paper of ponzano and regge , and the reader should look there for the details . in short , one takes the biedenharn - elliot ( be ) identity ( appendix a. ) , which relates a product of three @xmath2 symbols summed over one angular momentum , to a product of two @xmath2 symbols without summation . geometrically this corresponds to taking three tetrahedra joined together along a common edge , and each with a face in common with two of the others . removing the common edge ( the sum in the be identity ) leaves one with two tetrahedra sharing one common face . using the orthogonality of the @xmath2 symbols , one can change this identity to one that relates a single tetrahedron to four tetrahedra formed by introducing an additional vertex at the centre of the original tetrahedron . the identity is in appendix a for the interested reader . the important point is that there is an infinite factor @xmath48 for every vertex of the simplicial decomposition . therefore we see that an infinite factor of this form must be added to the denominator of the path sum to regularize it , and that there is one such factor for every vertex in the triangulation . the actual normalization factor is then @xmath49 where @xmath50 is the number of vertices in the discretization . in addition this discussion has shown us that the path sum is actually invariant under the two transformations derived from the be identity . these two transformations are known as pachner moves @xcite and these are the discretized version of diffeomorphisms . we have therefore learnt that the path sum thus defined ( in particular with the regularization discussed ) is diffeomorphism invariant in discretized form just as the @xmath5 theory was before the discretization . the full path sum is then , @xmath51 where @xmath52 is the number of vertices in @xmath23 . in this form however it is still not very practical for calculating . there exists a different regularization that involves a q - deformation of @xmath53 due to turaev and viro @xcite . the path sum is @xmath54_q \prod_{t\in\delta } ( -)^{\sum_{i=1}^6 j_t^i } \left[\begin{array}{ccc } j_t^1 & j_t^2 & j_t^3\\ j_t^4 & j_t^5 & j_t^6 \end{array}\right]_q\notag\end{aligned}\ ] ] where @xmath55_q & = \frac{q^n - q^{-n}}{q - q^{-1}}\end{aligned}\ ] ] the parameter of the quantum deformation is a root of unity @xmath56 and the sum is regularized as the representations of @xmath57 involve angular momenta only in the range @xmath58 so the path sum now involves all finite sums and @xmath59 has been replaced by @xmath60 which is clearly finite . the semi - classical limit of the @xmath61 symbol indicates that the q - deformed path sum is related to quantum gravity in three dimensions with a positive cosmological constant . the limit must be carried out in a way that as the angular momentum become large , correspondingly also @xmath44 must go to infinity . the limit is @xcite , @xmath62 and we see in particular that the limit which makes contact with the semi - classical physics is the limit in which the cosmological constant goes to zero . note that this can not be derived directly from the action @xmath63 by any simple generalization of the discretization carried out above , as the non - linearity in @xmath39 does not allow us to easily integrate over @xmath39 to get a simple expression involving the curvature @xmath64 . it would be very interesting to find a derivation of the tv path sum from the discretization of the path integral for @xmath65 . for simple manifolds this sum can actually be evaluated giving the turaev - viro invariants that are important for the understanding of the topology of three - manifolds . the restriction on angular momentum in the quantum group representations is the same as that which must be imposed on string states in @xmath0 . we will take another look at the q - deformed action and limits thereof after we have discussed the boundary discretization and will find that in the context of gravity in @xmath0 there may indeed be a deeper meaning to this regularization . it is also interesting to consider the relationship between this construction of three - dimensional gravity using a quantum deformation and studies of quantum doubles of groups @xcite . in this article one has a different type of quantum group that does not have a fixed deformation parameter . it is used for the discussion of multi - particle states in three - dimensional gravity . each particle , creates a localized source of curvature , and in general the space is conical at infinity . it is amusing to notice that for the chern - simons description of three - dimensional gravity , at zero cosmological constant one uses the group @xmath17 @xcite , but at non - zero cosmological constant , one finds instead the group @xmath66 with the level of the chern - simons theory related to the curvature . going to the multi - particle fock space in three - dimensions means that we are allowing variable localized curvature depending upon the location and mass of the particle sources . we find that the group @xmath17 is replaced by @xmath67 @xcite but now with no additional parameter , indicating perhaps that all values of curvature are possibile depending on the number and mass of particles present . this relationship deserves to be studied in more detail as it indicates a possible second quantization that involves also the cosmological constant . let us consider the general variation of the @xmath5 action for a manifold with boundary , ( other work on this subject can be found in the papers @xcite ) . @xmath68 we see then that the field equations are not effected by the presence of the boundary provided that the variation of @xmath69 is zero on the boundary . the path integral in the presence of the boundary will then be a function of the boundary value of the spin connection . we have another choice , which corresponds to the bf theory with a boundary term @xmath70 the variation of @xmath71 is now @xmath72 and therefore the boundary condition must be that the variation of @xmath39 is zero on the boundary , and the path integral will now be a function of the boundary metric . the first boundary condition of fixed spin connection on the boundary actually gives rise to a topological field theory on the manifold plus boundary . the second boundary condition is dirichlet on the metric , and this does not give rise to a topologically invariant boundary action . in a study of asymptotic symmetries in three - dimensional gravity @xcite , it was shown that with appropriate boundary conditions one can also find a liouville theory on the boundary at infinity of @xmath0 space . such boundary conditions formulated in terms of the metric and connection are actually mixed boundary conditions , and we will give more details of how these work below . the continuum boundary action can be easily derived by following a construction similar to that used in @xcite where the wzw - cs relationship was discussed in some detail . first we consider the boundary condition @xmath73 for which there is no additional boundary term . to examine the boundary theory we will insert the solutions to the bulk equations and then examine the action of the gauge symmetries of the theory in the presence of the boundary . for the wzw - cs relationship the boundary degrees of freedom arise precisely because the bulk gauge symmetry is only a global symmetry on the boundary thus the breaking of the gauge symmetry by the presence of the boundary gives rise to new degrees of freedom . the bulk equations of motion are solved by @xmath74 @xmath75 substituting these solutions into the action we find of course that it vanishes identically as @xmath76 . the gauge variation of the action is identically zero for the local lorentz invariance , but the diffeomorphism transformation has in principle an additional boundary term equal to @xmath77 which we see also vanishes as @xmath76 from the equations of motion ( modulo some topological issues regarding the extension of a flat connection to a boundary of given topology ) . this boundary action is that of an obviously topological two - dimensional field theory in agreement with the proof by ooguri and sasakura @xcite that with the @xmath73 boundary condition the path sum is a topological invariant not just of the bulk but of the bulk plus boundary theory . for the @xmath78 boundary condition we must add to this result the boundary term @xmath79 . the boundary condition now seems to indicate that the boundary metric is important in the path sum , in fact the path sum will now be a function of the boundary triangulation . again the solutions to the bulk equations will be inserted into the action , the bulk again giving zero contribution but the boundary now gives a non zero contribution equal to @xmath80 furthermore the gauge transformations now give rise to non - trivial boundary terms , @xmath81 @xmath82 we can see from these variations that the symmetry of the boundary theory is significantly smaller than that of the bulk theory . in fact we must have @xmath83 constant for the gauge transformation to vanish and also @xmath84 . therefore the boundary theory has no diffeomporphism invariance , and is invariant only under global lorentz transformations . finally we can consider the boundary conditions used in @xcite which are related to three dimensional gravity with cosmological constant . to do this we make a small deviation into the chern - simons representation of three - dimensional gravity with cosmological constant . our action is then , @xmath85 we make the change of variables , @xmath86 and we then find that @xmath87 becomes the difference of two chern - simons theories plus an additional boundary term . @xmath88 - cs[a^- ] ) + \frac{1}{\sqrt{-3\lambda}}\int_{\partial { \cal m}}tr(a^+a^-)\ ] ] we see here that the level of the chern - simons theory is inversely proportional to the square root of the cosmological constant , and also that if we started with a @xmath5 action with no boundary term , then after the change of variables we have a boundary term that is of a mixed form , rather than of the form @xmath89 . this is due to the fact that using the variables @xmath90 we can consider boundary conditions that would be mixed boundary conditions when expressed in terms of the variables @xmath69 and @xmath39 . indeed , if we add @xmath91 to the @xmath5 action then following the construction of @xcite one finds that in the chern - simons variables the action factorizes into two pieces that represent a pair of chiral wzw theories . the boundary conditions now imply restrictions on a combination of the metric and connection . it is precisely this setup that was shown to arise for the boundary at infinity of @xmath0 in the work of brown and henneaux and afterwards coussaert , henneaux and van driel @xcite . the boundary theory is actually a liouville theory . note that to discuss this case in the discretized framework we really need to use the quantum group representations as it is only then that ones sees a cosmological constant in the semi - classical limit . the discretized boundary theory will turn out to be very similar to a discretization of liouville theory . we will show how this relationship arises in more detail once we have set up the formalism for the quantum discrete boundaries . one may already worry here that we are trying to construct some triangulation of liouville theory in the strongly coupled phase and it is well known that for euclidean surfaces such theories have very non - continuum like phases . a discussion of these problems and arguments for better behaviour in the lorentzian case are in @xcite . from the bulk calculation of the discretized path sum , we saw that for every face of the simplicial decomposition , there are two @xmath32 symbols . indeed if we consider a single tetrahedron as a discretization of the three - dimensional ball then it has a weight , @xmath92 joining now an additional tetrahedron to one of the faces of this tetrahedron , we get another decomposition of the three - ball . on the internal face there is now a @xmath32 symbol coming from each of the tetrahedra , but we must now sum over the angular momentum projections assigned to the internal faces ( now identified of course ) . using the orthogonality identity for a pair of @xmath32 symbols @xmath93 we see that the internal @xmath32 symbols vanish and we are left in the path sum with a @xmath2 symbol for every bulk tetrahedron , and a @xmath32 symbol for every boundary face . in general the integral that gave rise to this pair of @xmath32 symbols was along the link of the dual lattice that passes from the centre of one tetrahedra to the centre of an adjacent one piercing one and only one face . in the presence of a boundary only half of this integral is carried out , from the centre of the tetrahedron to the face and this integral gives rise to a @xmath32 symbol for the bulk @xmath2 symbol and additional single @xmath32 symbol for the boundary face . the new feature that has given rise to the boundary weights for the boundary faces is that now we do not have an entire dual face , but rather the dual face is cut in half by the presence of the boundary as shown in figure 3 . we therefore need to also consider the group elements that live on the edge of the dual face that is exposed by the boundary . for the bulk path - sum described in the previous section the connection was integrated away . now due to the exposed dual faces , we have a boundary dependence on the connection that we may or may not integrate over depending upon the boundary conditions chosen . in figure 4 we have labelled one such edge from @xmath94 to @xmath95 with its weight @xmath96 . for the boundary conditions that correspond to the action with no boundary term , that is the @xmath97 conditions , we are instructed to keep the connection fixed on the boundary , and thus we must not integrate over the boundary values of @xmath28 . we thus find a network with trivalent vertices , each vertex is weighted by a @xmath32 symbol , and the vertices are tied together by the matrix elements of the corresponding group elements . the one and two tetrahedra path sums above easily generalize by gluing faces of tetrahedra together and using the orthogonality condition giving one the general expression for a simplicial decomposition with boundary . @xmath98 in this expression the normalization factor is the usual one mentioned above and @xmath99 is the dual lattice . the summation is over the angular momenta assigned to edges in the bulk and the boundary , and over the angular momentum projections assigned to each triangular face of the boundary . for the situation where the group representations summed over are those of the quantum group this is precisely the bulk plus boundary action derived by ooguri and sasakura @xcite , where they show that the hilbert space of the @xmath100 theory is equivalent to that of a pair of chern - simons theories for which the boundary state is described by wilson lines joined by trivalent vertices with an identical structure to that derived above . we would also like to note that this path sum ( for all boundary group elements @xmath28 equal to the identity element ) is the same as that derived in @xcite . in contrast to our present approach , in that paper the boundary action was derived purely on the grounds of topological invariance . the boundary term @xmath101 required for the @xmath102 boundary conditions when discretized becomes , @xmath103 where @xmath104 refers to the boundary value of @xmath39 . in this expression the dreibein @xmath104 is replaced by its discretized representation that being the length of the corresponding edge of the boundary of the original lattice , and the connection is represented by @xmath28 which is the gauge field assigned to the link of the boundary of the dual lattice that is dual to the edge where @xmath104 resides . the partition function is in this case a function only of @xmath104 . we must multiply the path sum derived above for the @xmath97 boundary conditions by this additional term , remove the sum over the boundary values of the spins @xmath105 , and integrate over @xmath28 to derive the final path sum for dirichlet boundary conditions in the metric . the integral of importance is that over @xmath28 and is @xmath106 inserting this into the path sum gives the final result for fixed metric boundary conditions , @xmath107 where @xmath108 signifies the lattice without boundary components . the sum is now only over the angular momentum in the interior edges . the integral over @xmath28 on the boundary has now fixed the angular momentum projections to be associated to edges of the boundary triangulations , rather than with faces as for @xmath109 . thus we see that the action is quite similar to that for the @xmath97 boundary conditions except that now the boundary values of the angular momenta are fixed corresponding to the fixed boundary metric . note that in this path sum the factors of @xmath110 are absent for edges that lie in the boundary due to the restriction in the product over edges to @xmath111 . these factors are important in the angular momentum identities that one uses to prove topological invariance , thus indicating that for this choice of boundary conditions there is no topological invariance on the boundary agreeing with our continuum analysis . in the path integral for a fixed boundary metric , one would expect that in the quantum gravity there would be a need to sum over all possible boundary configurations that give a discretization of the continuum boundary metric . a construction of such a type will be seen to be necessary for a calculation of black hole entropy in this discretized setup . in general before fixing boundary conditions we have the expression for @xmath97 without summation over angular momenta and without the integral over the boundary gauge connection . we need to understand what boundary conditions will allow calculations relevant to black hole physics , and also what representations of the ( quantum ) group one must include in this summation . the representations and boundary conditions will be discussed in the final section when we consider the construction for lorentzian metrics . furthermore , we need to know how to implement the boundary conditions that give liouville theory in our path sum construction . we want to show a point of contact between our calculations and discrete tft s in two dimensions . for @xmath97 everything is topological and there is an easy way to get a two dimensional tft from this theory . in @xmath112 take a thickened wall and remove the bulk tetrahedra using the various pachner moves in the bulk and on the boundary . the final result will be just two dimensional , but in some sense a double layer as the two faces will both carry their own @xmath32 symbols . the two dimensional action that one finds by this procedure is , @xmath113 this is indeed a two - dimensional tft , invariant under two - dimensional pachner moves and similar actions have been studied in a collection of works @xcite . for @xmath102 we can not actually remove all the bulk tetrahedra , as the removal process that one uses for the totally topological situation of @xmath97 relies heavily on the topological invariance of the boundary theory and in particular on the elementary shelling operations . we can however take a limit that is inspired by the bulk boundary correspondence of the ads / cft conjecture @xcite . to do this we imagine that we take a semi - classical limit of the bulk action leaving the boundary angular momentum fixed . the relevant limit of the bulk @xmath2 symbols that have a face edge or vertex on the boundary were already studied in the original article of ponzano and regge . the interesting thing that we find is that the boundary answer depends crucially on the asymptotic properties of the manifold . this sort of behaviour is maybe not a surprise as it is precisely such a dependence in the ads / cft correspondence that accounts for the simplicity of the near horizon limit in the ads case . for asymptotically flat spaces however the action is not expected to be similar to the cft as it will live on a null surface rather than on a time - like surface and the asymptotic group of symmetries will be smaller . the limits of @xmath2 symbols in which only some of the angular momentum are taken to be large are of two basic types . the first involves removing one vertex to infinity , and thus the three edges connected to that vertex become large , while the three vertices that form the remaining face stay fixed , this face then is a triangle of the boundary configuration . the result is thus the @xmath32 symbol of the remaining face , where the pairwise differences between the large angular momenta make the @xmath114 quantum numbers in this @xmath32 symbol and we thus find an answer similar to that which we derived from the bf theory , a pair of @xmath32 symbols on the boundary . the answer is , @xmath115 the other possibility corresponds to holding the length of one edge fixed , this representing a tetrahedra that has only an edge in contact with the boundary . in this case one still can do one of two things with the remaining angular momenta . one can take the angular momentum on the unique edge that does not touch our chosen edge to also be fixed , and the other four go to infinity . or one can take all five to be large . this is where the dependence on the large scale asymptotics of the space have an effect . if for instance in the euclidean case we are considering a boundary that is a sphere in @xmath112 , then clearly we must take the limit where all five other angular momenta become large . on the other hand , if the boundary is a plane in @xmath112 then one need take only four angular momenta to infinity , the other two corresponding to opposite edges of the tetrahedra remain fixed . the expressions for these limits contain additional dependence on parameters of the limiting process and can be found in appendix b. the expressions for the path sums in these limits are relatively complicated . it is interesting to note that the answer for this `` near - boundary '' limit , is basically the two - dimensional double @xmath32 symbol action derived above for purely topological boundary conditions , however , with some additional structure depending upon the asymptotic behaviour of the space - time . in the next section for null surfaces in lorentzian manifolds we will find that the semi - classical limit leads to an hypothesis that simplifies the boundary discretization considerably . if we replace the @xmath3 of the euclidean construction with @xmath4 then the representation theory becomes somewhat more complicated , and all limits of the corresponding angular momentum coupling coefficients in the various representations have not been fully studied . however , the original large angular momentum limit of ponzano and regge has also been carried out for the discrete series of representations in the non - compact case @xcite . the result is basically the same as for the compact group apart from the fact that the angles are now hyperbolic given by the boost required to take the normal to a face into the normal of an adjoining face . in the limit of large angular momentum we have @xmath116 in this expression , @xmath117 is the angle between the faces @xmath43 and @xmath44 , @xmath118 and @xmath119 is the unit normal to the corresponding face . thus as one face becomes null , the corresponding normal will also become null , and the angle that this face makes with the three neighbouring faces becomes infinite . the exponential in the weight for the tetrahedron implies that the corresponding angular momentum must be zero or that two of the sides of the triangle are of equal length and the corresponding angles are of opposite sign . therefore the only configurations that can contribute have equilateral triangles and isosceles triangles where the short edge has length @xmath120 corresponding to zero angular momentum . the equilateral triangles must have all zero angular momentum labels and thus have all sides of length @xmath120 . so in a path sum involving all discretizations with a given boundary metric the path sum is dominated by discretizations with boundary triangles that have all lengths equal to @xmath120 . this is modified then by collective structures built from isosceles triangles . it is clear that the configurations involving isosceles triangles must be collective , as the presence of an isosceles triangle , implies also that neighbouring triangles are isosceles , and so on , until the structure closes again . an example of such a collective structure is shown in figure 5 . these structures are reminiscent of macroscopic loop operators in the matrix models of dynamical triangulations @xcite . we should note here that we have been a bit incautious regarding the order of limits . we took large @xmath45 and then interpreted the expression for small @xmath45 . there is indirect evidence that the result is sensible and we will discuss our reasoning below . the precise calculation that one needs to do is to take the null boundary limit of the quantum @xmath2 symbol in a similar way to the original limits studied by ponzano and regge . thus we get a picture of horizon states in discretized quantum gravity and this is a positive step towards a micrpscopic understanding of black hole entropy . in t hooft s discussions of horizon states @xcite one finds similarly a special role for the low angular momenta , @xmath121 when the horizon at fixed rindler time is represented as a collection of discretized line segments labelled by angular momenta of @xmath4 . also in the ashtekhar approach to quantum gravity , the entropy calculations indicate that entropy is derived from contributions only from the lowest spin states on the horizon @xcite and similarly in the paper @xcite . it is interesting to reflect upon the meaning of the boundary action . if we assume that the semi - classical limit of the clebsch - gordon coefficients for @xmath122 for the discrete representations are an analytic continuation of those for @xmath57 then we should find a negative cosmological constant . for the situation of @xmath123 gravity in a space of constant negative curvature , one finds as mentioned above that the boundary theory is a liouville theory . furthermore from recent work on liouville theory @xcite it is known that the representations of the virasoro algebra that arise in the n - point functions , involve the quantum group @xmath124 . in the string theory picture of @xmath0/cft duality @xcite the mass cut - off on angular momentum representations is also the same as that which arises in the discrete representations of @xmath124 . beginning as we did from the prtv ( ponzano - regge - turaev - viro ) construction , it appears that we have arrived at almost the same conclusion . note though that in the prtv construction , after changing to a lorentzian space - time signature it is not necessary that the representations are identical to those used for the euclidean geometries . maybe one should sum(integrate ) over the continuous representations that arise in the liouville approach for the boundary at infinity . on the other hand , for the null boundary at a global horizon , it is not so clear how to proceed , however some interesting insight will come from a comparison of our boundary action and string theory on @xmath0 @xcite . if we had the clebsch - gordon coefficients for the discrete series of @xmath122 we could also explicitly calculate the weight of a `` macroscopic loop '' configuration and make a direct comparison with the macroscopic loop wavefunctions calculated for example in @xcite . the clebsch - gordon coefficients are known for the continuous series @xcite and for these one should be able to directly compute the null boundary limit . the representation theory of non - compact quantum groups is still very much under development , see @xcite . also a discussion on the relationship between strings in @xmath0 and quantum groups can be found in @xcite . there are representations of @xmath122 that are discrete , and agree basically with the discrete ones for the compact group , and give a cut - off in the path sum , see also @xcite for a few more details on these . the other representations are those that arise from the quantum group representation of the virasoro algebra of the liouville cft at @xmath125 . these are similar to the continuous representations of @xmath126 . we can already make some speculative remarks derived from studies of string theory and continuum gravity in @xmath0 @xcite . one can study the various physical excitations in this space - time both from the perspective of the space - time and that of the string theory . in the space - time picture , one finds a @xmath127 liouville theory , and indeed if one considers a non - critical string theory with target equal to @xmath0 then again one will find the world - sheet theory also to be liouville with @xmath127 . the states that arise are classified by quantum group representations @xcite . however the representations that arise are not those that we are using in the turaev - viro path sum . this strongly suggests that an extension of the prtv ( ponzano - regge - turaev - viro ) construction to include the representations of @xmath122 that arise in the liouville theory corresponds to extending the quantum gravity path sum , to a string field theory path sum ( albeit with a fixed topology for the target manifold ) . liouville theory at @xmath128 also appears in the context of @xmath129 compactifications although in this case the theory appears as a consequence of @xmath130 group factors in the internal space @xcite . it would be interesting to find connections between this structure and the liouville theory that is naturally present for @xmath0 string compactifications . the liouville theory on the boundary cylinder at infinity for gravity in @xmath0 has a central charge @xmath131 where @xmath132 or @xmath133 and the correlation functions of this theory are constructed from the clebsch - gordon coefficients of the quantum group @xmath122 where @xmath134 the cosmological constant of the @xmath0 space is proportional to @xmath135 . in turn , the cosmological constant that arises in prtv is proportional to @xmath136 , where the deformation for the turaev - viro quantum group is given by , @xmath137 clearly @xmath138 and thus the groups that arise in the two approaches are indeed identical deformations of @xmath139 . thus the group and deformation parameter agree in a manner which supports the conjecture that the quantum group that arises in the regge calculus is the same as that of the liouville theory on the boundary . however , the representations that arise in the liouville theory have angular momentum in @xmath140 , while those in prtv are identical to those that arose for @xmath57 with angular momentum running from 0 to @xmath141 . of course once we changed from euclidean to lorentzian discretizations , the question already arose as to which representations one should sum over and now we see that the answer to this question may have deeper significance . actually one can make the relationship between our discrete boundary action involving the quantum group and the perturbation theory of the liouville theory on the cylinder more concrete in a very geometrical manner by examining the perturbative expansion of the liouville theory on a cylinder ( corresponding to the boundary of @xmath0 ) . write the path integral with sources and charges for all liouville vertex operators in selected representations . use bootstrap to argue that all vertices can be reduced to cubic and recall that the cubic vertex for the liouville theory is given precisely by the clebsch - gordon coefficient of the quantum group @xmath122 @xcite . furthermore the propogator of the perturbative expansion of the liouville theory is the wigner coefficient @xmath142 of the corresponding representations . such feynman diagrams correspond precisely to the dual lattice with weights as derived in the previous section and as shown in figure 3 . geometrically all genus zero amplitudes correspond to one of our quantum group boundary terms in structure but with a sum over representations different from those used by turaev - viro . if one considers the dual lattice to the boundary triangulation , one finds a trivalent graph that lives on the boundary of the manifold , being one of the feynman diagrams discussed above . at any given time - slicing this will look like a collection of particles with mass given by their spin , and as this gas evolves there are interactions coming from the trivalent graph . thus one can make a proposal for calculating the entropy using a system of particles making a gas . the @xmath143 spin chain is a possible starting point for such a calculation . this model is a chain of spins the solution to which involves the quantum group @xmath122 and which is related to liouville theory for @xmath128 and @xmath144 and also possibly for all @xmath125 @xcite . within this framework we should be able to formulate the explicit calculation that is necessary to calculate the entropy of the boundary theory and thus black hole entropy . the feynman diagrams of the boundary action describe the time - evolution of the gas ( figure 6 ) . for a null boundary the gas will be non - relativistic whereas for a time - like boundary the gas will be relativistic . in the case of a null boundary these representations will become more restricted and the three point interaction implies that during the evolution of the gas one has both creation and annihilation of particles . this may even imply some sort of dissipation in the null case . other works arguing for dissipative behaviour for a theory describing a black hole horizon have appeared in @xcite . from a deeper understanding of this gas one should be able to directly calculate the entropy and thus the black hole horizon entropy . another consideration that we have not addressed directly but that has already arisen a few times in our discussions , and also one that is intimately related to the calculation of the entropy is the following . without a boundary , it was clear that the prescription of ponzano and regge to hold fixed the simplicial decomposition was already sufficient due to the topological nature of the theory . now in the presence of a boundary it is possible that one really needs to sum over the boundary triangulations . the bulk theory is topological and is insensitive to how one describes the sum in detail , however we expect some dynamics on the boundary . this indicates the possibility of extending the path - sum to dynamical triangulations . this sounds like trouble as such triangulations give rise to the matrix model of liouville and for @xmath127 these models are badly behaved with very rough surfaces dominating the path sum . however , discretizations for lorentzian manifolds have been studied in @xcite where the authors have shown that when the simplicial decompositions are restricted by the requirement of a causal structure , the phases of the dynamical triangulations are well behaved involving smoother surfaces than in the euclidean setup . the haussdorf dimension in particular remaining @xmath145 rather than becoming fractal and equal to 4 as it does in the euclidean case . indeed , our work also implies that there is another possibly interesting type of dynamical triangulation , where the `` causal '' structure is that implied by the constraint that the surface be not lorentzian , but null . it would be interesting to study in the context of @xcite null dynamical triangulations , the results of which investigation would certainly shed light on the dynamics of black hole horizons in quantum gravity . for @xmath1 dimensions we now have some intuition for how to approach the discretization . the simplices will be labelled by @xmath146 representations . we can write the boundary path sum including boundaries following more or less the same philosophy as above . in this case from the beginning it seems that we probably need to consider dynamical triangulations as otherwise we will end up with a topological bulk theory rather than a theory containing also gravitational dynamics . various versions of discretizations of four - dimensional lorentzian manifolds have been studied as for example in @xcite . furthermore the semi - classical limit of the @xmath147 symbols that arise in these bulk path - sums , has been studied in @xcite with results agreeing with the regge discretization once more . we expect for null boundaries also in @xmath1 dimensions that some restrictions will be placed on the representations arising and that one will probably again find some sort of three - dimensional dynamical triangulation describing the behaviour of the horizon . from the work in @xcite it has been shown that also for three - dimensional lorentzian dynamical triangulations , the branched polymer and crumpled phases , can not be reached leaving hope that such a system will have a nicely behaved continuum phase transition . it would be interesting to also look at three dimensional null dynamical triangulations to see if the causality restrictions on triangulations introduce some regulator of the geometries . the way to proceed is we believe clear . one must determine the representations that are important for the theory that is being investigated , and then one must look at various limits of the @xmath45-symbols . + * appendices * the biedenharn - elliot identity relates the @xmath2 symbols associated to two different ways of combining nine angular momenta . the sum on the right hand side is replaced by a product on the left . geometrically this identity is represented by the diagram shown . @xmath148 using the orthogonality for a pair of @xmath2 symbols this identity can be rearranged as discussed in the text , up to an infinite multiplicative factor . the regularized version of this identity as first given in ponzano and regge @xcite , is @xmath149 another useful identity for understanding the relationship between bulk and boundary transformations is , @xmath150 the geometrical meaning of the left side is simply a pair of adjoining boundary faces . the right hand side involves the gluing of two faces of an additional tetrahedron to the original pair of faces resulting in a new pair of boundary triangles . this results in a @xmath151 pachner transformation in two - dimensions . using the orthogonality of the @xmath32 symbols one can rewrite this equation to give the algebraic representation of the @xmath152 transformation . here are the @xmath153 and @xmath154 limits of ponzano - regge . we will use the following labelling for the tetrahedron ( figure 7 ) . for the @xmath153 limit , we shift @xmath155 by @xmath10 and take @xmath10 to be much larger than all of @xmath156 . in the figure this limit corresponds to keeping the segments @xmath157 $ ] and @xmath158 $ ] of fixed length while all other edge lengths go to infinity . we can consider @xmath159 $ ] to be the edge of the tetrahedron that lies in the boundary . for the @xmath154 limit , we take @xmath161 and @xmath162 , where now @xmath163 are all large , though small with respect to @xmath164 . this corresponds to keeping only the segment @xmath165 of fixed length and all other edges to infinity . the one edge of small size is then @xmath165 and in the text this is the edge that lies in the boundary of the manifold , all other edges in this case being internal . the final answer is , @xmath166 } \mbox{cos}\,(t-\frac{1}{4}\pi),\ ] ] where @xmath167 and @xmath168 is the regge action for the tetrahedron . in this limit , the dependence on the asymptotic structure of the space enters as the angle that remains in the final expression is the angle between the edges @xmath169 $ ] and @xmath170 $ ] or equivalently between @xmath171 $ ] and @xmath172 $ ] . these angles enter the expression for the limit through the regge action . if for instance the boundary is on a sphere of finite volume , then as one takes this limit a tetrahedron with one edge stuck on the sphere boundary , these angles will go to infinity . if the boundary is planar in flat space then the angles will go to zero and we go back to the @xmath153 result . + the author would like to thank g.arcioni , m. blau , g. thompson , g. t hooft and k. krasnov for useful conversations during the various stages of this work . this work was commenced during the extended workshop in string theory at the abdus salam ictp in trieste in the summer of 1999 . the author is supported by the pionier programme of the netherlands organisation for scientific research ( nwo ) . 99 m. blau and g. thompson , `` a new class of topological field theories and ray - singer torsion , '' _ _ * b228 * ( 1989 ) , 64 ; `` topological gauge theories of antisymmetric tensor fields '' , _ ann . * 205 * ( 1991 ) , 130 : g. horowitz , `` exactly soluble diffeomporphism invariant theories '' , _ commun * 125 * ( 1989 ) 417 . s. elitzur , g. moore , a. schwimmer and n. seiberg , `` remarks on the canonical quantization of the chern - simons - witten theory . '' _ nucl.phys._*b326 * ( 1989 ) 108 . g. ponzano and t. regge , `` semi - classical limit of racah coefficients , '' in _ spectroscopic and group theoretical methods in physics _ f. bloch , ( north - holland , 1968 ) . e. witten , `` @xmath173-dimensional gravity as an exactly soluble system '' , _ nucl . * b311 * ( 1988 ) 46 . g. t hooft , `` quantization of point particles in @xmath173-dimensional gravity and space - time discreteness '' , _ class . * 13 * ( 1996 ) 1023 , gr - qc/9601014 . g. t hooft , `` dimensional reduction in quantum gravity . '' gr - qc/9310006 . l. susskind , `` the world as a hologram . '' hep - th/9409089 . g. t hooft , _ nucl . * 256 * ( 1985 ) , 727 ; _ nucl . _ * b335 * ( 1990 ) 138 ; _ phys . _ * t36 * ( 1991 ) 247 . l.susskind , l. thorlacius and j. uglum , _ phys . _ * d48 * ( 1993 ) 3743 . l. freidel and k. krasnov , `` spin foam models and the classical action principle . '' hep - th/9807092 , n.kawamoto , h.b . nielsen and n sato , `` lattice chern - simons gravity via ponzano - regge model . '' hep - th/9902165 . s. chung , m. fukuma and a. shapere , `` structure of topological lattice theories in three dimensions , '' hep - th/9305080 . m. reisenberger , `` a left - handed action for euclidean general relativity . '' , _ _ * 14 * ( 1997 ) , 1753 . varshalovich , a.n . moskalev and v.k . khersonskii , `` quantum theory of angular momentum '' . singapore : world scientific , 1988 . t. regge , _ nuovo cimento _ * 19 * ( 1961 ) 551 . u. pachner , `` shellings of simplicial balls and p.l . manifolds with boundary , '' _ discr.math . _ * 81 * ( 1990 ) , 37 ; `` p.l homeomorphic manifolds are equivalent by elementary shellings , '' _ europ.j.combinatorics _ * 12 * ( 1991 ) , 129 . v.g . turaev and o.ya . viro , `` state sum invariants of 3-manifolds and quantum 6j - symbols , '' _ topology_*31 * ( 1992 ) , 865 . s. mizoguchi and t. tada , `` 3-dimensional gravity from the turaev - viro invariant , '' hep - th/9110057 . f.a . bais and n.m . muller , `` topological field theory and the quantum double of su(2 ) '' , hep - th/9804130 . h. ooguri , `` partition functions and topology - changing amplitudes in the three - dimensional lattice gravity of ponzano - regge , '' _ nucl.phys._*b382 * ( 1992 ) 276 . h. ooguri and n.sasakura , `` discrete and continuum approaches to three - dimensional quantum gravity , '' _ mod.phys.lett._*a6 * ( 1991 ) 3591 . m. blau , i. jermyn and g. thompson , `` solving topological field theories on mapping tori , '' _ phys . lett . _ * b383 * ( 1996 ) 169 , hep - th/9605095 , ic/96/74 . brown and m. henneaux , `` central charges in the canonical realization of asymptotic symmetries : an example from three - dimensional gravity . '' _ comm.math.phys._*104*(1986 ) , 207 . o. coussaert , m. henneaux and p. van driel , `` the asymptotic dynamics of three - dimensional einstein gravity with a negative cosmological constant . '' gr - qc/9506019 . j. ambjrn and r. loll , `` non - perturbative lorentzian quantum gravity , causality and topology change , '' hep - th/9805108 , j. ambjrn , j. jurkiewicz and r. loll , `` a non - perturbative lorentzian path - integral for gravity '' hep - th/0002050 , j. ambjrn , j. jurkiewicz and r. loll , `` lorentzian and euclidean quantum gravity - analytical and numerical results '' hep - th/0001124 . c. bachas and p.m.s . petropoulos , _ comm.math.phys . _ * 152 * ( 1993 ) , 191 . m. fukuma , s. hosono and h. kawai , `` lattice topological field theory in two - dimensions , '' hep - th/9212154 . g. carbone , m. carfora and a. marzuoli , `` wigner symbols and combinatorial invariants of three - manifolds with boundary , '' dfnt - t 14/98 , sissa 118/98/fm , `` invariants of spin networks with boundary in quantum gravity and tqft , '' gr - qc/9905076 . j. maldacena , `` the large n limit of superconformal field theories and supergravity '' , _ adv . * 2 * ( 1998 ) 231 , hep - th/9711200 ; s. gubser , i. klebanov and a. polyakov , `` gauge theory correlators from noncritical string theory '' , _ phys . lett . _ * b428 * ( 1998 ) 105 , hep - th/9802109 . e. witten , `` anti de sitter space and holography '' , _ adv . _ * 2 * ( 1998 ) 2253 hep - th/9802150 . j.w . barrett and t.j . foxon , `` semi - classical limits of simplicial quantum gravity , '' gr - qc/9310016 . s. davids , `` semiclassical limits of extended racah coefficients , '' gr - qc/9807061 . d. gross and a. migdal , _ phys . * 64 * ( 1990 ) 127 , t. banks , m. douglas , n. seiberg and s. shenker , _ phys . _ * b238 * ( 1990 ) 279 , g. moore , n. seiberg and m. staudacher , _ nucl . _ * b362 * ( 1991 ) 665 . g. t hooft , `` transplanckian particles and the quantization of time '' , _ class . _ * 16 * ( 1999 ) , 395 ; gr - qc/9805079 . a. ashtekar , j. baez , a. corichi and k. krasnov `` quantum geometry and black hole entropy , '' _ phys . * 80 * ( 1998 ) 904 , gr - qc/9710007 . v. suneeta , r.k . kaul and t.r . govindarajan , `` btz black hole entropy from ponzano - regge gravity , '' gr - qc/9811071 . h. dorn and h .- j . otto , `` two and three - point functions in liouville theory , '' hep - th/9403141 . a.b . zamolodchikov and al.b . zamolodchikov , `` structure constants and conformal bootstrap in liouville field theory , '' hep - th/9506136 . b. ponsot and j. teschner , `` liouville bootstrap via harmonic analysis on a non - compact quantum group , '' hep - th/9911110 . a. giveon , d. kutasov and n. seiberg , `` comments on string theory on @xmath0 '' , _ adv . * 2 * ( 1998 ) 733 , hep - th/9806194 , and references herein ; j. maldacena and h. ooguri , `` strings in ads(3 ) and sl(2,r ) wzw model '' , hep - th/0001053 ; a. kato and y. satoh , `` modular invariance of string theory on ads(3 ) '' , hep - th/0001063 . g. arcioni , s. de haro and m. oloughlin , work in progress . h. steinacker , `` quantum anti - de sitter space and sphere at roots of unity . '' hep - th/9910037 . a. jevicki and s. ramgoolam , `` non - commutative gravity from the ads / cft correspondence '' _ jhep _ * 9904 * ( 1999 ) 32 , hep - th/9902059 . s. randjbar - daemi and d. jatkar , `` type iib string theory on @xmath174 '' , _ phys.lett . _ * b460 * ( 1999 ) 281 , hep - th/9904187 . l.d . fadeev and o. tirkkonen , `` connections of the liouville model and the @xmath143 spin chain , '' _ nucl . _ * b453 * ( 1995 ) 647 , hep - th/9506023 . g. t hooft , `` quantum gravity as a dissipative deterministic system '' , _ class.quant.grav . _ * 16 * ( 1999 ) 3263 ; gr - qc/9903084 . `` black holes : the membrane paradigm '' eds . k.s . thorne , r.h . price and d.a . macdonald , ( yale university press , london , 1986 ) . parikh and f. wilczek , `` an action for black hole membranes '' , _ phys.rev . _ * d58 * ( 1998 ) , 064011 , gr - qc/9712077 . j.w . barrett and l. crane , `` relativistic spin networks and quantum gravity '' , _ j. math . * 39 * ( 1998 ) 3296 , gr - qc/9709028 . j.w . barrett and l. crane , `` a lorentzian signature model for quantum general relativity , '' gr - qc/9904025 . j.w . barrett and r.m . williams , `` the asymptotics of an amplitude for the four - simplex '' , gr - qc/9809032 . m. reisenberger and c. rovelli , `` sum over surfaces formulation of loop quantum gravity '' , _ phys.rev . _ * d56 * ( 1997 ) , 3490 . m. reisenberger , `` a lattice world - sheet sum for 4-d euclidean general relativity '' , gr - qc/9711052 . j. barrett and l. crane , `` relativistic spin networks and quantum gravity '' , _ j.math.phys . _ * 39 * ( 1998 ) 3296 .

we present a possible approach to the study of the renormalization group ( rg ) flow based entirely on the information theory . the average information loss under a single step of wilsonian rg transformation is evaluated as a conditional entropy of the fast variables , which are integrated out , when the slow ones are held fixed . its positivity results in the monotonic decrease of the informational entropy under renormalization . this , however , does not necessarily imply the irreversibility of the rg flow , because entropy is an extensive quantity and explicitly depends on the total number of degrees of freedom , which is reduced . only some size - independent additive part of the entropy could possibly provide the required lyapunov function . we also introduce a mutual information of fast and slow variables as probably a more adequate quantity to represent the changes in the system under renormalization and evaluate it for some simple systems . it is shown that for certain real space decimation transformations the positivity of the mutual information directly leads to the monotonic growth of the entropy per lattice site along the rg flow and hence to its irreversibility .

0910.2097

renormalization group ( rg ) , which is a powerful instrument for analyzing different strongly coupled systems @xcite(see also @xcite for more recent reviews ) , is usually based upon an appropriate division of the whole set of variables into two subsets : so called `` fast '' and `` slow '' variables with subsequent elimination of the fast ones . the resulting slow system is usually expected to behave essentially in the same way as the initial one but with coupling constants changed . successive application of this transformation results in a flow in the space of couplings . after each step of the rg transformation the information about exact values of fast variables is lost for the observer to whom only slow variables are available . it seems natural then to expect , that this information loss should lead to the irreversible character of the corresponding rg flow . this irreversibility is a subject of an intensive study for several decades now , starting from the famous work of zamolodchikov @xcite , who showed under certain assumptions that in two dimensions ( 2d ) there exists a function on a space of coupling constants ( a kind of lyapunov function ) which always decreases along the rg flow and at fixed points coincides with the central charge @xmath0 of corresponding conformal theories . `` irreversibility '' here means that the flow in the space of couplings resembles that of a simple dissipative system with a given trajectory never returning back to its starting point , thus excluding limit cycles or more complex strange attractors . in 2d theories the @xmath0-theorem of ref . @xcite states that the rg evolution is exactly of this type and looks like a simple monotonic flow downhill in the couplings space from fixed points with large @xmath0 to those with smaller values of the central charge . much work have been done in this direction and the @xmath0-theorem was studied in detail with its possible generalizations to higher dimensions ( see e.g. @xcite and references therein and also @xcite for quite recent development ) . still the interesting possibility that in general the flow may be more complicated , and may even exhibit chaotic behavior , like in case of many other nonlinear mappings , attracts some attention @xcite ( probably the first example of a chaotic rg flow was found for spin systems on hierarchical lattices @xcite , see also @xcite ) . the models where such peculiarities are observed are however rather artificial , so that the question whether they could take place in more realistic cases remains somewhat unclear . though this problem is rather complicated mathematically it looks much simpler from the physical point of view . the wilsonian rg transformation is designed in such a way that all finite dimensionless correlation lengths in a translationally invariant system always decrease @xcite . to say this another way , all masses ( measured in units of an ultra - violet cut - off ) grow and it seems quite natural that the effective number of massless modes , ( in 2d measured by the central charge @xmath0 ) can not increase . the argument based on the growth of masses , though not completely rigorous , is quite general , hence something unusual , like limit cycles in the flow , may be expected only in some special systems with only infinite correlation lengths or with infinite number of correlation lengths which can be arbitrarily large ( when we have e.g. a self - similar spectrum of masses converging to zero @xcite ) . but in systems with finite number of well defined correlation lengths the rg flow should probably always be irreversible . this argument however seems to be unrelated to the aforementioned information loss and information aspects probably play no role in the irreversibility , which may seem somewhat strange . probably the most known attempt to study rg irreversibility using information theory tools is the approach of ref . @xcite where the relative entropy was introduced in quantum field theory and some monotonicity theorems for it , as a function of relevant couplings , were proved and its relation to @xmath0-theorem was discussed . in this paper we present a somewhat different , entirely information theory based approach to the problem of the rg flow , oriented more on discrete real space rg transformations in different lattice spin systems . clearly , information - theoretic considerations are too general to lead to some non - trivial concrete results , but still , as we shall see below they may impose some restrictions on the possible character of the flow . first , in section 2 we evaluate an exact amount of information loss for a single step of the renormalization as a kind of conditional entropy of fast variables ( compare with @xcite ) . as one might expect , this information loss is equal to the decrease of entropy under the rg transformation . thus it is possible to establish a general theorem about the monotonic decrease of the total entropy ( and relative entropy as well ) under general coarse graining as a result of information loss . the monotonic decrease of entropy along the rg flow is neither something unusual ( see e.g. @xcite ) nor very interesting , because the entropy is proportional to the total number of degrees of freedom and should certainly decrease when some of them are integrated out . hence this monotonic behavior does not directly lead to the irreversibility of the flow . an interesting thing may happen however , if the entropy of a finite system possesses an additive size - independent part . if this part of the entropy also decreases , as the total entropy does , then it may provide the proper lyapunov function , because it does not explicitly depend on the number of variables and its change is entirely due to the flow of coupling constants . we will show that this takes place in 1d ising model , where this subextensive part of the entropy is known as the `` excess entropy '' and was studied in detail as a measure of statistical complexity of spatial structures @xcite . this excess entropy equals to the mutual entropy of two halves of the system which is actually the reason why it decreases under coarse graining . unfortunately we are unable to present a general theorem concerning the monotonicity of such excess entropy in higher dimensions . it appears possible though to prove quite a different monotonicity theorem , but only for a certain class of real space lattice rg transformations , namely for decimations , when exactly half of variables are eliminated on each step . the proof is based on the analysis of a new quantity introduced in section 3,mutual information of slow and fast variables . contrary to the information loss this mutual information shows how much information about eliminated fast variables is still present after a single step of the wilsonian renormalization . the nonnegativity of the mutual information may alone impose some restrictions on the character of the flow leading to its irreversibility . for the decimations of the aforementioned type in lattice models we will show that the entropy per lattice site monotonically grows as a result of the positivity of the mutual information of spins on two identical sublattices , one of which is eliminated in the course of the rg transformation . therefore such decimation transformations , though they results in highly non - linear mappings , always lead to irreversible flows in the space of couplings . simple examples of such a behavior are discussed in section 4 , where one- and two - dimensional classical spin models and also the continuum limit gaussian model are considered . note , however , that the theorem holds true only for systems with real actions , when information theory can be applied , while e.g. it is not valid for the ising model with complex external magnetic field where the decimation rg flow is known to be chaotic @xcite . the theorem is also invalid for spin models on hierarchical lattices @xcite , because decimations are not associated there with decomposition of these finite and inhomogeneous lattices into two identical sublattices . the entropy per site growth may seem somewhat unexpected , because it takes place not only in the trivial case , when the flow goes to the high - temperature fixed point , but even when we start from points located in the ordered phase . rg trajectories that start from points below the phase transition can never end in the fully disordered trivial fixed point , since for decimation the mean magnetization is conserved along the flow , but nevertheless the entropy per site ( and hence the amount of disorder ) will always grow . this is a peculiar feature of the decimation and does not take place for other possible renormalization schemes , like e.g. the majority rule block spin transformation also discussed in section 4 . suppose that we somehow manage to divide our variables into `` fast '' @xmath1 and `` slow '' ones @xmath2 . for the wilsonian momentum space rg the fast variables are high fourier harmonics of the fields , but for real space renormalization in lattice systems @xmath1 s are just spins on some sublattice , so that they are not actually `` fast '' in any sense . still we will use these names to distinguish degrees of freedom to be integrated out and those that remain . the partition function is given by the sum over all fields , which symbolically may be written as @xmath3 where @xmath4 is the action ( or energy divided by temperature for classical lattice systems ) . the fields normally depend on all space - time coordinates , so that the sums in eq . ( [ z ] ) are actually path integrals , properly regularized . now , the first step of the rg transformation is to integrate out the fast variables @xmath1 to obtain an effective action for the slow ones , @xmath5 , defined by @xmath6 this transformation is then repeated many times resulting in the rg flow of the corresponding actions ( or coupling constants if the form of the action is fixed ) . certainly , after the fast variables are eliminated , information about their exact values is lost . one can ask then , what is the precise amount of the information loss . for a moment let us fix the slow variables @xmath2 , then the conditional probability for @xmath1 ( normalized to unity ) is @xmath7 it seems natural then to define the information ( measured in ` nats ' ) related to some field configuration @xmath1 in the fixed background @xmath2 by the usual shannon formula @xmath8 and the information loss after the fields @xmath1 are integrated out as an average of ( [ i ] ) over the distribution ( [ pcond ] ) @xmath9 this quantity still depends on the chosen configuration of the slow fields @xmath2 . actually @xmath10 is similar , in a sense , to the boltzmann entropy of a ` macrostate ' specified by the coarse - grained variables @xmath2 . indeed , in case of ` equipartition ' , when @xmath11 does not depend on @xmath1 in some domain of fast variables the information loss eq.([ia ] ) is just the logarithm of the number of ` microstates ' corresponding to a given macrostate . we propose then that the total information loss @xmath12 is an average of @xmath13 over all configurations of @xmath2 with the corresponding weight @xmath14 , i.e. @xmath15 this is obviously the _ conditional entropy _ @xmath16 of two sets of random variables @xmath1 and @xmath2 well known from the information theory @xcite . this is the average entropy of the fast variables given the state of the slow subsystem . the conditional entropy of this kind usually measures the amount of information loss in a noisy channel when @xmath1 is an input signal and @xmath2 is an output . one can also view the rg transformation as an action of a smoothing filter , then @xmath12 is the information loss due to smoothing . the conditional entropy as a measure of information loss also have been introduced in ref . @xcite where a coarse graining of polymer configurations was considered . the above estimate of the information loss introduces a somewhat new paradigm of renormalization . we see now that it is possible to view the fast fields @xmath1 as an input signal , not directly available to a receiver , while the slow field @xmath2 is an output . after renormalization only the output signal is available and one can only try to restore the input signal more or less accurately . this is a typical situation of a signal transmission through a noisy channel where information loss may be estimated as a conditional entropy ( [ it ] ) . but this analogy suggests that it may probably be of more use to discuss not only the information loss but also the information that is preserved , i.e. information about fast variables that is stored in the action for the slow ones ( see section 3 ) . substituting the probability distributions into eq . ( [ it ] ) we easily obtain @xmath17 where the brackets @xmath18 mean the average over the fields with the corresponding actions ( clearly , for continuous theory some regularization is required for these formulas to make sense ) , the initial informational entropy @xmath19 of the distribution @xmath20 is given by @xmath21 where @xmath22 and the entropy of the slow field @xmath23 is given by the same formula with @xmath24 and @xmath25 . recall that our rg transformation is defined in such a way that the partition function @xmath26 remains unchanged . also note that the new action @xmath27 , defined by ( [ eff ] ) , always contains an additive constant term , but it does not contribute to the entropy and may be ignored in the evaluation of @xmath28 . the shannon entropy @xmath19 used here is the entropy of the fluctuating fields and it measures in fact how large these fluctuations are . for a classical system @xmath19 coincides with its thermodynamic entropy @xmath29 , but in quantum systems ( in the continuum limit ) they are different , because @xmath29 is related to fluctuations in occupation of energy levels and not to that of fields as functions of space - time coordinates . while @xmath19 is dominated by an ultra - violet divergent `` bulk '' contribution , proportional to the space - time volume , the quantum thermodynamic entropy @xmath29 looks , in a sense , as a finite - size correction to a leading term in @xmath19 . indeed , for example , for a 2d system on a large cylinder of length @xmath30 and circumference @xmath31 which may be viewed as a 1d quantum system at a temperature @xmath32 we have @xmath33 and the same is true for @xmath19 , but this bulk term obviously does not contribute to the thermodynamic entropy @xmath34 . hence non - zero @xmath29 arises only from finite size corrections to @xmath35 ( see section 4.5 for another example ) . thus the information loss @xmath12 is merely the difference of the initial entropy @xmath36 and the entropy of the remaining slow variables @xmath37 . for discrete variables @xmath2 , @xmath1 ( e.g. for classical spin systems ) both entropies and the conditional entropy are always nonnegative , @xmath38 , @xmath39 and @xmath40 and hence @xmath41 which means that the entropy of the remaining variables decreases along the rg flow . at first glance this decrease of entropy may seem strange , since if the flow is e.g. directed toward the high temperature fixed point the system obviously gets more disordered after renormalization . recall , however , that it is the total entropy , an extensive quantity proportional to the system size , that decreases according to eq . ( [ eg ] ) and this decrease is mainly due to the fact that the total number of variables is reduced after some of them are integrated out . for example , in large lattice systems @xmath42 , where @xmath43 is the number of sites which decreases under real space rg transformation . the entropy per lattice site @xmath44 which is a more adequate measure of disorder may well increase along the rg flow . therefore the information loss @xmath12 is not very informative about the change in physics under renormalization and the monotonic decrease of @xmath19 does not in general give rise to a proper lyapunov function defined on the space of coupling constants . this is not very surprising because the total information loss is a rather crude characteristic of the rg transformation . an interesting possibility still remains , that some part of this information loss may provide the required monotonic function . the most natural choice probably would be some volume - independent additive part in @xmath19 for a finite system , something that may be called e.g. information about boundary conditions . an example of such a function will be explicitly presented in section 4.1 for the one - dimensional ising model . how to obtain possible extension of this result to higher dimensions is however not clear . to conclude this subsection it should be mentioned , that for continuous field variables when we deal not with true probabilities but probability distributions and sums are replaced by integrals , direct application of the above formulas may lead to negative values of entropies @xcite and to negative @xmath12 which certainly is not acceptable . to avoid this one should either somehow make the variables @xmath2 , @xmath1 discrete ( `` digitize '' them ) or better use some other quantities , which do not suffer from such a drawback , like e.g. the _ relative entropy_. relative entropy is the basic concept of the approach to renormalization developed in a series of papers @xcite . for the sake of completeness we present here a short discussion of its properties with regard to the wilsonian rg transformation . consider some statistical system with variables @xmath45 , and two probability distributions @xmath46 and @xmath47 . then we may introduce the relative entropy @xmath48 as the so called kullback - leibler distance ( divergence ) @xmath49 between these probability distributions @xmath50 this distance , though not symmetric , is nonnegative even for continuous @xmath45 and @xmath51 for @xmath52 . now , the main idea of the approach @xcite is to consider probability distributions corresponding to two actions @xmath53 and @xmath54 which differ by the values of coupling constants , and to study the relative entropy @xmath55 where @xmath56 and @xmath57 , and then choose @xmath58 to correspond to a rg fixed point distribution . thus physically this @xmath48 measures how far our system with the action @xmath53 is from a given fixed point . then we expect the relative entropy to behave monotonically under the change of the coupling constant , corresponding e.g. to some relevant perturbation of the fixed point . in fact , some rather general results are known about the behavior of @xmath48 under a coarse - graining . let us proceed as usual and decompose the whole set of fields @xmath45 into fast @xmath1 and slow ones @xmath2 . if we introduce conditional probability @xmath11 ( normalized to unity @xmath59 ) and make use of @xmath60 we can write the following chain of equalities @xmath61 the first term in the resulting expression is obviously the relative entropy after renormalization , @xmath62 , while the last term is known in information theory as a _ conditional relative entropy _ which is always nonnegative @xcite . hence we have @xmath63 this result , as well as the derivation of eq . ( [ ca ] ) , is in fact widely known and usually referred to as a decrease of relative entropy under any kind of coarse - graining @xcite . the decrease of relative entropy for continuum gaussian model with lowering the ultra - violet cutoff @xcite is just one example of this general phenomenon . one can also view eq . ( [ ca ] ) as a generalization of the expression ( [ i ] ) for the information loss . then the last term in eq . ( [ ca ] ) may probably be called `` conditional information loss '' . since this is nonnegative also for continuous variables it looks more preferable for quantum field theory , than the information loss from the previous subsection . unfortunately inequality ( [ ha ] ) for the relative entropy is also not very useful because normally @xmath48 is an extensive quantity and its monotonic decrease again may be attributed to the reduction in the number of degrees of freedom ( effective shrinking of the system under renormalization ) and it is a specific quantity @xmath64 , where @xmath43 is e.g. a number of lattice sites , which is physically relevant . a more careful analysis shows however that @xmath48 is much better that entropy @xmath19 or similar quantities and possesses interesting monotonicity properties with respect to couplings ( see refs . @xcite for details ) . for example , close to a critical point with a diverging dimensionless correlation length @xmath65 the relative entropy taken with respect to the fixed point distribution in many cases behaves as @xmath66 ( for @xmath67-dimensional lattice system ) and is universal , depending not on the total number of variables , but only on the number of blocks of size @xmath65 . this means that relative entropy does not contain redundant `` low - level '' information . when this is true @xmath68 is obviously monotonic along the wilsonian rg flow , which is a particular example of how the informational `` distance '' grows on a trajectory moving away from the unstable fixed point . if we take @xmath69 in ( [ ra ] ) to be a uniform distribution @xmath70 , then @xmath71 , where @xmath19 is the entropy of a given distribution @xmath72 . therefore , if the first term @xmath73 remains constant , the decrease of the relative entropy under a coarse - graining is usually related to the growth of the entropy @xmath19 @xcite . in our case , however , the first term also changes , because @xmath74 due to normalization condition for @xmath69 and therefore @xmath48 and @xmath19 both decrease . we now introduce another information - related quantity , the so called _ mutual information _ of the fast and slow variables , which seems to be free of the aforementioned drawbacks . we shall demonstrate now that the nonnegativity of this mutual information may itself impose some restriction on the rg flow . to define this mutual information , we need to perform a transformation opposite , in a sense , to the usual rg transform of eq . ( [ eff ] ) , namely to integrate out the slow variables . this will result in an effective action @xmath75 for the fast variables @xmath1 , their probability distribution being @xmath76 , and the corresponding entropy will be denoted by @xmath77 . then the mutual information of two sets of variables @xmath1 and @xmath2 is defined as @xmath78\ ] ] where @xmath79 is the probability distribution for the initial system ( joint probability for @xmath2 and @xmath1 variables ) . the mutual information is certainly a kind of relative entropy which in our case measures the `` distance '' @xmath80 between the original system and some hypothetic one where fast and slow variables are independent , but described by their true corresponding effective actions , i.e. @xmath81 and @xmath82 in eq . ( [ ra ] ) . substituting expressions for probabilities in terms of actions one can easily derive the well known formulas from elementary information theory @xmath83-h(\psi,\eta)\end{aligned}\ ] ] here @xmath84 is the information loss after the normal rg transformation and @xmath85 is the information loss after elimination of the slow variables , or if we adopt the view that renormalization is the information transmission through the noisy channel @xmath86 , then @xmath85 is a measure of the noise @xcite . in bipartite systems the mutual information is nonnegative and is known to measure the information about one subsystem stored in the other @xcite ( for independent subsystems @xmath87 ) , i.e. in our case it shows what amount of information about the fast fields @xmath1 is still contained in the effective action @xmath5 . since usually rg transformation results in the change of coupling constants it seems natural to think that the information about the eliminated fast variables is now hidden in the new values of these constants . therefore @xmath88 should be directly related to the rg flow of the coupling constants . note also that @xmath88 is represented in the last equation of eq . ( [ mu ] ) as a difference of two quantities with roughly the same size dependence , so that it should not be so sensitive to the reduction of degrees of freedom under the rg transformation . while in general @xmath88 is more difficult to calculate than the entropy , there exists an interesting situation when @xmath89 and @xmath90 . this happens e.g. for some decimation transformations in lattice systems , when the system is divided into two identical sub - lattices ( see next section ) and exactly half of variables ( those on one sub - lattice ) are integrated out . for large enough system with @xmath43 sites we may write @xmath91 and @xmath92 where @xmath44 and @xmath93 are the entropies per lattice site before and after the renormalization . then @xmath94 and hence the condition @xmath95 directly leads to the growth of the entropy per lattice site along the rg trajectory . this means that such decimations are always irreversible , but suggests that corresponding rg flows do not have fixed points for finite generic values of couplings . indeed , if there exists a fixed point then at this point @xmath96 and hence the bulk mutual information @xmath88 is zero . but this means that spins on two sub - lattices are essentially uncorrelated and remain uncorrelated under renormalization , which seems impossible for realistic spin systems with reasonable finite interactions . note , that this does not imply that decimation does not have nontrivial fixed points at all , but such fixed points ( apart from the trivial high - temperature one ) should probably lie at the `` boundary '' of the parameter space , when some couplings are set to infinity , as one can see e.g. in the classical spin-1 model on a line @xcite ( see also section 4.2 ) . the absence of a nontrivial fixed point corresponding to the phase transition for pure decimation in 2d ising model have been noticed already in @xcite and was related to the fact that decimation does not eliminates only `` short range '' correlations but inevitably sums over some long - range behavior @xcite , i.e. it is actually not a good rg transformation . let us now take a look at some simple examples . the classical one - dimensional ferromagnetic ising model is the most known example of the rg transformation which here can be performed exactly . the model is defined on a lattice with @xmath43 sites and the initial action is given by @xmath97 , where @xmath98 . the real space rg procedure ( decimation ) is defined as follows : we decompose the lattice into two sub - lattices and sum over spins on one sublattice @xcite ( see fig . the resulting model is again of the ising type with @xmath99 , with twice the initial lattice spacing and @xmath100 sites . the entropy @xmath19 is here the true entropy of the ising model and the information loss , defined by eq . ( [ i ] ) , is @xmath101 where @xmath102 is the entropy per lattice site @xmath103 and @xmath104 the rg flow is very simple since the running coupling constant @xmath105 flows from the zero temperature fixed point @xmath106 to the high temperature one at @xmath107 . though @xmath108 decreases during the rg procedure and @xmath109 it is easy to verify that @xmath12 is always positive , has its maximum @xmath110 at @xmath107 and then monotonically decreases to zero as @xmath111 . it is interesting that for large @xmath112 the information loss can be obtained independently , so that eq . ( [ ii ] ) can be used to derive @xmath108 . indeed , if @xmath113 most probable spin configurations are that of well separated kinks ( domain walls ) . if a spin , which is eliminated , falls inside a domain , no information is lost , because its value does not fluctuate at large @xmath105 and is completely determined by adjacent spins . but a spin that lies on a domain wall is completely unconstrained and hence the information loss is @xmath114 . since there are on the average @xmath115 kinks at @xmath111 we conclude that @xmath116 where dots indicate terms of higher order in @xmath117 . at large @xmath105 the leading term in entropy is @xmath118 hence eq . ( [ ii ] ) results in @xmath119 which has the solution @xmath120 this is the correct expression for the renormalized coupling at @xmath111 and we see now that @xmath114 here is in fact of informational origin . now we discuss the relative entropy eq . ( [ ra ] ) of the ising model at some coupling @xmath121 taken with respect to the state at a different coupling @xmath122 . the general formula , derived in @xcite states that @xmath123 where @xmath124 in our case @xmath125 and hence @xmath126 we can look now at some limiting cases . first consider the entropy relative to the high temperature fixed point @xmath127 . then @xmath128 one can easily see that @xmath48 decreases under the rg transformation when @xmath99 and @xmath129 and @xmath64 also tends to zero . the entropy relative to the zero temperature fixed point should be obtained in the limit @xmath130 but the limiting expression diverges . fortunately , we can take @xmath131 which is almost the same for our purposes . then @xmath132.\ ] ] again the total relative entropy monotonically decreases along the rg flow as the general proof suggests , but now @xmath133 increases from zero to @xmath134 . close to the critical point @xmath135 where @xmath136 is the correlation length . thus , in this limit @xmath48 is not changed by the rg transformation @xmath137 , @xmath129 and actually measures the number of blocks of size @xmath65 as the general analysis suggests @xcite . this example clearly illustrates that @xmath48 always decreases , but this behavior is in part due to the effective shrinking of the system @xmath129 and implies only that @xmath138 . the relative entropy per lattice site @xmath133 is more useful because it really measures the `` distance '' from the fixed points and its evolution is related to the direction of the flow . since here the rg flow is from the zero temperature fixed point to the high temperature one , @xmath139 decreases while @xmath140 monotonically grows ( as @xmath141 close to the critical point ) . it is easy also to calculate the mutual information @xmath88 for this model . since the two sublattices are identical and similar to the original one , @xmath142 and hence @xmath143\ ] ] we see then , that indeed @xmath88 , as expected , is determined entirely by the change of the coupling constant and from @xmath95 it follows that @xmath144 i.e. the entropy per site @xmath102 always grows along the rg trajectory . this is rather trivial for the model in question since the rg flow leads us to the disordered fixed point @xmath107 with maximum entropy regardless of the initial state because there is no phase transition . nevertheless this simple example clearly shows how the irreversibility of the rg flow can be derived within the informational approach . so far we have considered only the thermodynamic limit @xmath145 . let us now take a more careful look at the finite system . the partition function for finite block of ising spins is well known @xmath146 . from this expression it is easy to obtain the entropy and its expansion in powers of @xmath147 @xmath148 where @xmath44 is given by eq . ( [ h ] ) and @xmath149 this @xmath150 monotonically decreases from @xmath134 at @xmath106 to zero at @xmath107 . it is interesting , that at fixed points @xmath150 is equal to universal values @xmath151 , where @xmath152 is the degeneracy of the largest eigenvalue of the transfer matrix @xmath153 , @xmath154 . indeed , @xmath155 at zero temperature fixed point ( due to two ground states with opposite magnetization ) while @xmath156 in the fully disordered state . this subextensive part of the entropy is also known as the _ excess entropy _ and was studied in detail as a possible measure of statistical complexity of spatial structures in 1d spin systems @xcite . it can be shown that if we divide our finite system into two halves then excess entropy @xmath150 is equal to the mutual information of these halves ( see @xcite and references therein ) . but any mutual information is a kind of relative entropy and so it should in general decrease under coarse graining ( if it preserves the partition of the block in two parts ) @xcite ( see also eq . ( [ ca ] ) ) and since it does not depend on @xmath43 this decrease is not related to the change of size @xmath129 . thus for 1d ising model the excess entropy @xmath150 has an independent meaning and its monotonic decrease under renormalization can be derived directly from the information theory . more interesting critical behavior may be obtained in one dimension if we take a spin chain with larger value of spin @xmath158 . we start with the most studied example of a spin-1 chain which is defined by the action @xmath159 where spin now takes three possible values @xmath160 . this is again a ferromagnetic model for @xmath161 , but with much richer behavior than before because we can now vary the effective length of the spin by changing the parameter @xmath162 . the action ( [ spoh ] ) retains its form under the decimation transformation and only the coupling constants are changed . it is widely believed that there are no phase transitions in one - dimensional systems with short range interactions . this is true , however , only if all the couplings are finite . non - trivial critical behavior in this model , analyzed thoroughly in ref . @xcite , takes place when both @xmath122 and @xmath162 tend to infinity while @xmath163 remains finite . the remaining two - dimensional space of couplings may be parameterized by new coordinates @xmath164 and the decimation transformation acts as @xmath165 at some fixed points.,width=288 ] the resulting quite interesting rg flow diagram obtained in @xcite is shown schematically in fig . 2 . its main features are four fixed points and a critical line ac . since this flow arises from the decimation transformation our general arguments require that the entropy per site @xmath166 should increase along the flow . though the model can be solved exactly ( see @xcite for details ) it is easy to evaluate the entropy at the fixed points without much computations . the point a corresponds to fully ordered state ( actually there are three such states with all @xmath167 either @xmath168 , or @xmath169 , or 0 ) hence @xmath170 at @xmath145 with only subextensive contribution to @xmath19 . the same is true also for the point @xmath171 with two ordered states . the point b is similar to a high temperature fixed point of the ising model , with uncorrelated @xmath98 and no zero spins , hence @xmath172 . the most interesting case is the critical point c at @xmath173 . it corresponds to @xmath174 while @xmath175 . in this case there exist the lowest energy state with all @xmath176 and zero energy . for periodic boundary condition all configurations with some finite fraction of spins changed to @xmath177 have infinite energy at @xmath178 while states with all @xmath98 are gapped with @xmath179 . but there are @xmath180 such states and their statistical weight exactly compensates the boltzmann factor @xmath181 which results in the partition function @xmath182 and mean energy @xmath183 . therefore at c we effectively have two equally probable states with highly degenerate disordered excited one and the resulting entropy @xmath184 at @xmath145 . the values of entropies are shown in fig . 2 near the corresponding fixed points . we see now that the entropy per site can in fact be used to understand the ordering of the fixed points along the flow since @xmath44 monotonically grows when we pass from a to c and finally to b. certainly this is not the only way to classify these fixed points . in fact we can use also the degeneracy @xmath152 of the largest eigenvalue of the transfer matrix @xmath185 ( actually the degeneracy of the ground state of the corresponding quantum hamiltonian @xmath186 , defined by @xmath187 ) . in 1d @xmath152 plays , in a sense , the same role as the central charge @xmath0 in two dimensions , because @xmath152 always decreases under renormalization ( this also resembles the @xmath188-theorem @xcite for 1d quantum systems ) . the global decrease of @xmath152 in 1d spin system with nearest - neighbor interactions when @xmath189 is rather trivial and follows directly from the decimation rg transformation written as a transfer matrix mapping @xmath190 where for spin with @xmath191 possible values @xmath185 is a nonnegative @xmath192 matrix . eigenvalues of @xmath185 transform as @xmath193 so that the degeneracy once reduced will never return back . for strictly positive @xmath185 frobenius theorem requires @xmath156 , but when some couplings turn to infinity , as we have seen above for the spin-1 model , some entries in @xmath185 may vanish and @xmath152 may differ from unity . for the spin-1 model all these eigenvalues were calculated in ref . @xcite and @xmath194 for point a , @xmath155 for c ( also for the point at the origin ) and @xmath156 for point b in fig . 2 . for these models it is possible also to evaluate the excess entropy at the fixed points . fixed points for decimation are defined by @xmath195 where @xmath196 is some positive number ( multiplication of @xmath185 by a constant does not change the physics ) and it can be easily shown that in these cases @xmath185 may have @xmath197 eigenvalues @xmath196 while other eigenvalues are zeros . then for the periodic boundary condition @xmath198 and since @xmath199 entropy at fixed points has the form @xmath200 with @xmath201 . in this case the subextensive part in @xmath19 coincides with that in @xmath35 and this number monotonically decreases when we pass from one fixed point to another along the rg flow ( this is somewhat similar in spirit to the recently proposed @xmath202-theorem for field theories in odd dimensions @xcite ) . unfortunately , for periodic boundary conditions @xmath150 equals @xmath151 also away from fixed points ( at @xmath145 ) and it is discontinuous , because @xmath152 changes abruptly , while excess entropy for a finite block in an infinite chain , which is related to mutual information of two halves and is expected to be continuous and monotonic along the rg flow , is much harder to evaluate at @xmath157 and in general is no more related to @xmath152 in a simple way even at fixed points . consider now a two - dimensional classical spin system , with spins placed on sites of a square lattice . various spin interactions , including e.g. nearest - neighbor , second - nearest - neighbor and similar interactions , are parameterized by a set of coupling constants @xmath203 . to define a `` checkerboard '' decimation transformation we first divide the lattice into two interpenetrating square sub - lattices and then sum over all spins on one sublattice @xcite ( see fig . just as in one dimension we obtain then a new spin system again on a square lattice ( with the lattice spacing @xmath204 times larger ) but with a new energy @xmath27 with couplings @xmath205 . this is the first step of the rg transformation , which is then repeated further . since two sub - lattices are identical the mutual information between them is again given by the eq . ( [ imu ] ) , i.e. @xmath206 for a large lattice with @xmath43 sites , where @xmath207 is the entropy per site . hence we conclude that in this case the entropy per site should also increase in the course of the rg transformation and the rg flow is irreversible . the entropy growth in two dimensions may seem a rather counterintuitive result , since in systems with phase transition one may navely expect that at low temperatures ( below the transition ) the rg transformation should lead us deeply into the ordered phase with @xmath208 decreasing to zero . but for the decimation this intuition obviously does nt work , since in this case the average spin remains unchanged so that the spontaneous magnetization is constant along the rg flow . for better understanding of why the entropy per site can grow under renormalization at low temperatures let us now consider the ferromagnetic ising model with @xmath209 , where the sum is over nearest - neighbors and @xmath98 . after the first step of the rg we have ( up to an unimportant constant ) @xcite @xmath210 where the second term is the next - nearest - neighbor interaction and the last one is the four spin interaction ( @xmath167 , @xmath211 are spins on the corners of a plaquette ) with the sum over all plaquettes . the new coupling constants are known to be @xmath212 we see now that @xmath213 ( though @xmath214 ) , the next - nearest - neighbor interaction is also ferromagnetic , @xmath215 , but @xmath216 , i.e. the four - spin interaction tries to decrease the ferromagnetic order . next , we estimate the entropy at low temperatures . at @xmath217 all spins point in one direction and the entropy is different from zero only due to rare spin flips . the partition function may be represented as a low temperature expansion @xmath218 where @xmath219 is the energy of all spins aligned and the energy needed to reverse one spin is @xmath220 in original system since after the reversal four links have `` wrong '' alignment of adjacent spins . hence at low temperatures the entropy per site is @xmath221 after the decimation the system is still ferromagnetic but the energy of spin reversal is now @xmath222 or @xmath223 at large @xmath121 . hence the new excitation energy is smaller than @xmath224 and it becomes easier to reverse a spin . however , since the difference is exponentially small one has to include also the next term in @xmath207 due to pairs of reversed spins @xmath225 ( the corresponding term in @xmath208 can be easily evaluated as @xmath226 and is unimportant ) . but even with this contribution taken into account the entropy per site after the decimation @xmath227 is larger , @xmath228 thus , it is possible to check at least for the first iteration at low temperatures that indeed the entropy per site grows after the decimation . physically the result ( [ e ] ) for the mutual information or @xmath229 is due to the reversed pairs of adjacent spins in the original lattice . only through these pairs the two sub - lattices have non - trivial information about each other at @xmath230 ( there should be also a term @xmath134 in @xmath88 since the orientations of the overall magnetization in sub - lattices are correlated , but we are interested only in terms @xmath231 in the mutual information ) . now it is possible to explain , what this increase in entropy at low temperatures actually means . since the average magnetization conserves along the decimation rg flow , the average number of reversed spins per unit volume remains the same , but now more of them are isolated ones ( with less energy cost of reversal ) and the number of reversed close pairs or larger clusters is relatively smaller than in the initial system . surely this picture may also be interpreted as a decrease in the correlation length for spin fluctuations . unfortunately the next iteration of the decimation can not be performed exactly and is known to lead to non - gibbsian measures at low temperatures @xcite , so that it is not possible to describe renormalization as a flow in usual space of couplings . still there should be a flow of probability measures , probably with limiting points corresponding to product measures describing independent spins with fixed total magnetization . consider now a classical lattice system of ising spins @xmath167 with the action @xmath232 $ ] and first divide the lattice into a set of nonoverlapping identical blocks with , say , @xmath233 adjacent spins in each block @xmath234 , assuming that blocks form a lattice , similar to the original one ( see fig . 3 for an example ) . then , following kadanoff prescription @xcite associate a new spin variable @xmath235 with @xmath236-th block and define the rg transformation @xmath232\rightarrow s'[t]$ ] according to @xmath237)=\sum_{s}t(t , s)\exp(-s_0[s])\ ] ] with @xmath238 obeying the condition @xmath239 which guarantees that the partition function @xmath26 is the same for both @xmath158 and @xmath240 variables . . blocks @xmath234 are shaded , spins @xmath167 are located at black sites while block variables @xmath235 may be placed in the centers of blocks.,width=240 ] this transformation is of the general type of eq . ( [ eff ] ) but now all @xmath167 are `` fast '' variables and @xmath235 are `` slow '' ones . hence the total `` action '' @xmath241=-\ln t(t , s)+s_0[s]$ ] now defines the joint probability for initial and `` renormalized '' spins . the entropy @xmath19 now is not equal to the physical entropy @xmath242 of the initial spin system but is given by @xmath243 where we have used the condition ( [ t ] ) in the calculation of @xmath244 . the most popular form of @xmath245 is @xmath246 where @xmath247 is a free parameter and @xmath248 is the sum of initial spins in the block @xmath234 . note that @xmath249 enters in the partition function as a spatially varying magnetic field constant over a block . hence at @xmath250 all block spins @xmath251 will be aligned strictly along @xmath235 . now making use of the identity @xmath252 we easily obtain @xmath253 where the sum is over all blocks . next , we can easily sum @xmath254)$ ] over @xmath240 because of the identity ( [ t ] ) to obtain @xmath255=s_0[s]$ ] and hence @xmath256 now we can use the definition of the mutual information ( [ mu ] ) and eqs . ( [ hk ] ) , ( [ ktild ] ) to derive the final expression @xmath257 note that @xmath242 cancels away from the resulting expression for the mutual information . next we introduce the entropy per lattice site for renormalized spins according to @xmath258 and instead of @xmath259 which was valid for decimation , we have at large @xmath43 quite a different expression @xmath260\geq 0,\ ] ] where the sum of spins @xmath167 is over one block , @xmath233 is the number of spins in a block and the average is over the initial configurations with the weight @xmath261)$ ] . the second term in eq . ( [ mubs ] ) looks like the entropy of the total spin of a block at temperature @xmath262 averaged over different values of its magnitude . thus the entropy after the renormalization @xmath208 is always bounded from below , provided the second term in eq . ( [ mubs ] ) is nonzero . this is due to the noise induced by the renormalization at finite @xmath247 . the most interesting case however corresponds to @xmath250 which is known to lead to the so called majority - rule renormalization @xcite . this means that we define @xmath235 as @xmath177 depending on the sign of the total block spin @xmath251 . we assume also that @xmath233 is odd to avoid ambiguities . then @xmath263 and from eq . ( [ tav ] ) it follows that @xmath264 at @xmath250 leading to @xmath265 this result is rather obvious however , because from eq . ( [ mu ] ) we have @xmath266 but for majority - rule rg the initial configuration of spins @xmath167 completely determines that of renormalized spins @xmath235 so that @xmath267 ( there is no noise in the transmission channel ) and hence their mutual information coincides with the final entropy @xmath268 . therefore in this case the positivity of mutual information gives nothing new and @xmath208 can be either greater or smaller than @xmath269 . contrary to decimation now @xmath95 does not prohibit fixed points and the entropy per site can even go to zero along the rg flow . now we proceed to quantum field theory , using ordinary quantum mechanics as the simplest example . consider a quantum particle of mass @xmath270 , moving along a line in a harmonic potential well with a frequency @xmath271 . its partition function at a temperature @xmath185 is given by a path integral , which in a lattice regularization , when the imaginary time interval @xmath272 $ ] ( @xmath273 ) is divided into @xmath43 intervals of length @xmath274 , has the form @xmath275 with @xmath276 . shannon entropy , as mentioned above , equals @xmath277 where brackets mean the average with weight @xmath278 . @xmath19 coincides with the entropy of the statistical system of variables @xmath279 placed on sites of a one - dimensional lattice of length @xmath31 with nearest - neighbor interaction and periodic boundary condition . this system is actually the one - dimensional gaussian model near its critical point @xmath280 @xcite . however , at @xmath281 this model is actually ( 0 + 1 ) euclidean free field theory and its thermodynamic entropy is given by a different formula @xmath282 where @xmath283 is the mean energy of the particle and @xmath284 is its free energy . let us now briefly discuss the connection between these two entropies . the mean velocity squared @xmath285 which enters in @xmath286 diverges as @xmath281 @xcite and its direct evaluation is not so easy , but the average action can be evaluated simply by differentiating @xmath26 with respect to mass @xmath287 where the first term in the r.h.s . comes from the differentiation of @xmath270 in the integration measure . but for the harmonic oscillator @xmath26 does not depend on mass , and therefore @xmath288 . in the continuum limit @xmath289 and at a given @xmath271 we have the following asymptotics for the entropies at @xmath290 @xmath291 we see that the leading `` bulk '' term in @xmath19 is divergent in the continuum limit @xmath281 and @xmath29 resembles , in a sense , the finite size correction in @xmath19 . the different behavior of @xmath19 and @xmath29 comes from the fact that @xmath29 describes the information related to energy levels occupation and hence tends to zero at zero temperature when the system is in its ground state , while @xmath19 describes fluctuations of `` paths '' in configuration space which obviously grow as @xmath292 . the entropy per lattice site here is @xmath293 at @xmath281 . it may seem strange that @xmath44 decreases with correlation length @xmath294 , because for the model in question we may also apply the decimation transformation , similar to that in the ising model , integrating away @xmath279 s on alternating sites ( see e.g. @xcite ) and our general result states that in this case @xmath44 should increase along the rg flow . under such a decimation @xmath137 and @xmath44 from eq . ( [ hf ] ) does not behave in the required way . the solution to this apparent contradiction is rather simple . note first that summation over variables @xmath295 is defined now as @xmath296 and the integration measure itself depends on mass and on lattice spacing . this integration measure is fine - tuned to the action so that @xmath26 has a well defined correct continuum limit ( [ hoha ] ) . but after the rg transformation this fine - tuning is destroyed . the effective action @xmath27 now contains renormalized mass @xmath297 , frequency @xmath298 and a new lattice spacing @xmath299 , but the measure ( [ sum ] ) remains the same . this means that in this case _ the entropy after renormalization is no longer given by the same formula ( [ ha ] ) as before_. we may transform the measure to the correct form but this leads to an additional factor @xmath300 in @xmath26 , since the lattice now has @xmath100 sites . this additional factor does not affect the mean action @xmath301 but certainly change the @xmath35 part of the entropy . simple gaussian integration like that in ref . @xcite shows that @xmath302 up to the terms of order @xmath303 at @xmath304 and the additional factor is simply @xmath305 . then the new entropy per site equals @xmath306 at @xmath281 . this quantity is larger than @xmath307 and the mutual information @xmath88 of two sublattices is given by @xmath308 at @xmath281 . note also that the total entropy is reduced after the decimation , because the information loss @xmath309 is positive . let us now analyze this system more carefully . after the decimation is repeated many times we get away from the continuum limit because the lattice spacing grows . in this region it is more convenient to rewrite the renormalized action in the form @xmath310 where @xmath105 , @xmath311 are new running coupling constants with the initial values @xmath312 , @xmath313 . the entropy per lattice site for this action can be derived from the exact solution of the gaussian model @xcite @xmath314 where @xmath312 comes from the integration measure in eq . ( [ za ] ) . note that while all correlations depend only on @xmath315 , the entropy depends on both @xmath105 and @xmath311 because the integration measure contains the unrenormalized coupling @xmath316 . decimation acts as @xmath317 or @xmath318 . this rg equation has a fixed point @xmath319 which corresponds to @xmath280 and under renormalization @xmath320 flows to @xmath321 . close to the fixed point @xmath322 one can define a continuous theory by choosing @xmath323 and @xmath324 with @xmath281 and rg equations turn into @xmath325 , @xmath326 . it is now easy to verify that in the vicinity of @xmath322 the initial value of the entropy is @xmath327 and that it indeed acquires an additional @xmath328 contribution at each decimation step provided we are still close to the fixed point . note that @xmath44 grows even for @xmath322 because in the two dimensional space of @xmath105 and @xmath311 there is no critical fixed point and the point @xmath322 corresponds to a line @xmath329 with the rg flow along it . after many decimations @xmath330 , i.e. @xmath105 decreases much faster than @xmath311 and the entropy starts to behave as @xmath331 with logarithmic accuracy . since @xmath332 means the loss of correlations between @xmath295 s on different sites , this expression for the entropy @xmath44 is in fact the entropy production rate for the gaussian white noise @xmath333 with dispersion @xmath334 , well known in the information theory . in this paper we have tried to understand what limitations on the rg flow may follow from the information theory , because wilsonian rg transformation , when some fast variables are integrated out , is obviously related to the information loss . this transformation is similar in a sense to a signal transmission through a noisy channel , the fast variables being the input , while the slow ones play the role of the output available to a receiver . the information loss is defined here as an average conditional entropy of the eliminated fast variables when the remaining slow variables are fixed . from the nonnegativity of this quantity ( which holds at least for systems with discrete variables , like e.g. classical spins ) it follows immediately that the informational entropy @xmath19 of the system can not increase under renormalization . unfortunately this result is not very interesting , and does not lead to the irreversibility of the rg flow , because it applies to the total extensive entropy , proportional to the size of the system , which effectively shrinks after renormalization . this shrinking is evident for real space rg schemes , where some lattice sites are eliminated and the system size is smaller if measured in new lattice constant , but equally applies to momentum space wilsonian renormalization due to final rescaling of momenta . only if @xmath19 possesses some volume independent additive part @xmath150 , then this part , if it also decreases under renormalization , may provide the required lyapunov function , that depends only on couplings . physically its decrease may be attributed to a loss of information about massless modes , but this does not follow directly from that of @xmath19 and should be deduced independently . then , to be monotonous @xmath150 must have also some independent definition . for 1d ising model this quantity is just the subextensive excess entropy and we explicitly check its monotonicity under rg transformation . this behavior for a finite block of spins in an infinite chain follows directly from information theory , because excess entropy in this case is just the mutual information of two halves of the system @xcite , which decreases under coarse graining . note also that for 1d systems with periodic boundary conditions the excess entropy is @xmath151 where @xmath152 is the degeneracy of the largest eigenvalue of the transfer matrix which always decreases along the rg flow . this is somewhat similar to what is known for critical 2d systems defined on a long cylinder of size @xmath335 with , say , @xmath336 and periodic boundary condition in the @xmath295 direction . in this case the first finite size correction in the expansion of entropy in @xmath337 is @xmath338 ) @xcite , where @xmath0 is the central charge . therefore the subextensive part @xmath150 in entropy in 2d should decrease as a consequence of the @xmath0-theorem . however in both these examples the subleading term in entropy is discontinuous and does not directly lead to a proper lyapunov function . this means that the extraction of the required part from the entropy of finite size system is somewhat ambiguous and depends on the boundary condition . one may suggest , however , that in general there might exist some size - independent information - based quantity defined on the whole space of couplings ( maybe some kind of relative entropy or higher order mutual information @xcite ) that is monotonic along the flow and at fixed points coincides with some kind of excess entropy , but this point is not clear . the main aim of the present paper was , however , apart from demonstrating some examples of informational approach , to introduce a different quantity , namely the mutual information of fast and slow variables @xmath88 , which shows how much information about the eliminated variables is still stored in the renormalized effective action . this quantity does not suffer from the `` shrinking size '' trouble and is less sensitive to the overall decrease in the number of degrees of freedom , though it is generally more difficult to evaluate . there exist however some real space decimation transformations for which this mutual information can be easily calculated . this happens e.g. if the original lattice may be decomposed in two identical sublattices and rg transformation eliminates spins on one sublattice . then the nonnegativity of @xmath88 results in the increase of entropy per lattice cite and rg flow resembles a relaxation to equilibrium . this result does not depend on interactions and may be true also for a larger set of models . what is actually needed is that the rg transformation must be represented as a product of such decimations . the approach to irreversibility based on the mutual information is different from those trying to generalize zamolodchikov s @xmath0-theorem and possibly may provide some additional information about the rg flows . it may appear interesting also to apply the present approach to the momentum space renormalization in field theory or to study more complicated correlations along the flow and some higher - order mutual information . i am very grateful to a.c.d . van enter , j. gaite , v. losyakov , a. marshakov , a. morozov for valuable discussions and comments , and to l. shchur for some useful advices . the work was supported in part by rfbr grants 09 - 02 - 00886 , 10 - 02 - 00509 , rfbr - ukraine 09 - 01 - 90493 and by federal agency for science and innovations of russian federation under contract 14.740.11.0081 . wilson and j. kogut , phys . rep . * 12 * 75 ( 1974 ) ; k.g . wilson , rev . phys . * 47 * 773 ( 1975 ) c. bagnuls and c bervillier , phys . rep . * 348 * 91 ( 2001 ) ; b. delamotte , arxiv : cond - mat/0702365 a.b . zamolodchikov , pisma zh . teor . fiz . * 43 * 565 ( 1986 ) [ jetp lett . * 43 * 730 ( 1986 ) ] j.l . cardy , phys . lett . * b215 * 749 ( 1988 ) ; i. jack and h. osborn , nucl . phys . * b343 * 647 ( 1990 ) ; a. cappelli , d. friedan and j.i . latorre , nucl . * b352 * 616 ( 1991 ) ; a.h . castro neto and e. fradkin , nucl . * b400 * 525 ( 1993 ) ; s. forte and j.i . latorre , nucl . * b535 * 709 ( 1998 ) ; t. appelquist , a.g . cohen and m. schmaltz , phys . d * 60 * 045003 ( 1999 ) ; a. cappelli , g. dappollonio , r. guida and n. magnoli , arxiv : hep - th/0009119 ; d. anselmi , acta phys . * 52 * 573 ( 2002 ) ; e. barnes , k.a . intriligator , b. wecht and j. wright , nucl . phys . * b702 * , 131 ( 2004 ) d.l . jafferis , i.r . klebanov , s.s . pufu and b.r . safdi , arxiv:1103.1181 ; a. amariti and m. siani , arxiv:1105.3979 ; i.r . klebanov , s.s . pufu and b.r . safdi , arxiv:1105.4598 . a. morozov and a.j . niemi , nucl . phys . * b666 * 311 ( 2003 ) s.r . mckay , a.n . berker and s. kirkpatrick , phys . rev , lett . * 48 * 767 ( 1982 ) ; n.m . vraki , j. kertsz and w. selke , j. phys . a * 15 * l427 ( 1982 ) ; b. derrida , j .- eckmann and a. erzan , j. phys . a * 16 * 893 ( 1983 ) p.h . damgaard and g. thorleifsson , phys . a * 44 * 2738 ( 1991 ) ; p.h . damgaard , int . j. of mod . a * 7 * 6933 ( 1992 ) b.p . dolan , phys . e * 52 * 4512 ( 1995 ) s.d . gazek and k.g . wilson , phys . lett . * 89 * 230401 ( 2002 ) ; e. braaten and h .- w . hammer , phys . lett . * 91 * 102002 ( 2003 ) ; s. d. gazek and k.g . wilson , arxiv : cond - mat/0303297 j. gaite and d. oconnor , phys . d * 54 * 5163 ( 1996 ) j. gaite , phys . lett . * 81 * 3587 ( 1998 ) ; j. gaite , phys . rev . d * 61 * 045006 ( 2000 ) ; phys . d * 62 * 125023 ( 2000 ) p.l . ferrari and j.l . lebowitz , j. phys . a * 36 * 5719 ( 2003 ) a. robledo , j. stat . phys . * 100 * 475 ( 2000 ) j.p . crutchfield and d.p . feldman , phys . e * 52 * 1239r ( 1997 ) ; d.p . feldman and j.p . crutchfield , phys . e * 67 * 92 ( 2003 ) t.m . cover and j.a . thomas , elements of information theory , 2nd ed . ( wiley , hoboken , ny ) 2006 s. krinsky and d. furman , phys . lett . * 32 * 731 ( 1974 ) ; phys . b * 11 * 2602 ( 1975 ) l.p . kadanoff and a. houghton , phys . b * 11 * 377 ( 1975 ) l. sneddon and m.n . barber , j. phys . c : solid state phys . * 10 * 2563 ( 1977 ) i. affleck and a.w.w . ludwig , phys . lett . * 67 * 161 ( 1991 ) ; d. friedan and a. konechny , phys . lett . * 93 * 030402 ( 2004 ) a.c.d . van enter , r. fernndez and a.d . sokal , j. stat . * 72 * 879 ( 1993 ) ; a.c.d . van enter , f. redig and e. verbitskiy , arxiv:0804.4060 l.p . kadanoff , rev . * 49 * 267 ( 1977 ) th . niemeijer and j.m.j . van leeuwen , in phase transition and critical phenomena , vol . 6 , eds . c. domb and m.s . green ( academic press , new york , 1976 ) t.h . berlin and m. kac , phys . 86 * 821 ( 1952 ) r.p . feynman and a.r . hibbs , quantum mechanics and path integrals ( mcgraw hill inc . , new york , 1965 ) kadanoff , statistical physics . statics , dynamics and renormalization ( world scientific , singapore , 2000 ) h. matsuda , phys . e * 62 * 3096 ( 2000 ) h.w.j . blte , j.l . cardy and m.p . nightingale , phys . lett . * 56 * 742 ( 1986 ) ; i affleck , phys . rev . lett . * 56 * 746 ( 1986 )

within the massive field theoretical renormalization group approach the expressions for the- and- functions of the anisotropic-vector model are obtained for general space dimension in three - loop approximation . resumming corresponding asymptotic series , critical exponents for the case of the weakly diluted quenched ising model ( , ) , as well as estimates for the marginal order parameter component number of the weakly diluted quenched-vector model are calculated as functions of in the region . conclusions concerning the effectiveness of different resummation techniques are drawn . * key words : * critical phenomena , diluted spin systems , ising model , renormalization group . * pacs numbers : * 64.60.ak , 61.43.-j , 11.10.gh _ to appear in j.stat.phys . vol . 92 , nos 5/6 _

cond-mat9807401

study of the critical behaviour of the ising model has several attractions . on the one hand , the ising - like models are simple enough , which is of a special advantage in the statistical physics . on the other hand , in spite of their simplicity , such models show rich and interesting behaviour at the critical point . also , the existence of the exact solution for the two - dimensional ising model often makes it an object for verifying different approximation schemes . all the stated above yielded the high interest devoted to the problem . in particular a great deal of generalization of the model appeared . among different ways of generalization , much attention has been devoted to the affect of the impurities on the critical behaviour of the ising - like models as well as to the investigation of critical regimes of the models on the lattices of a non - integer dimension ( @xmath3 ) . there have been devised different realizations of the last stated generalization . for example , one can approach the concept of non - integer dimensionality either by explicit construction of the non - integer dimensional object , which leads to the concept of a fractal @xcite , or by formal carrying out an analytic continuation of the function , which by definition depends on a natural value of dimension . within the theory of critical phenomena the latter ambiguity was reflected in examining the critical behaviour of the many - particle systems on fractal @xcite or on abstract hypercubic lattices of the non - integer @xmath3 . there arosed a question whether a model on a fractal lattice ( being scale invariant ) possesses universality as well as a system on a hypercubic lattice ( having translation invariance ) . the problem has been widely studied but still remains open @xcite . today s point of view states that the usual demand for strong universality ( in sense of critical properties depending only on symmetry of the order parameter , interaction range and space dimension ) seems not to be obeyed by fractal lattice systems , and for them the concept of universality itself should be revised @xcite . speaking about the study of ising - like models on analytically continued hypercubic lattices of non - integer @xmath3 , one should note a great variety of theoretical approaches devised for these problems . these include : the wilson - fisher @xmath9-expansion @xcite improved by the summation method @xcite ; kadanoff lower - bound renormalization applied to some special non - integer dimensions @xcite ; high - temperature expansion improved by a variation technique @xcite ; finite - size scaling method applied to numerical transfer - matrices data @xcite ; new perturbation theory based on the physical branch of the solution of the renormalization group equation @xcite ; fixed dimension renormalization group technique @xcite applied directly to arbitrary non - integer @xmath3 @xcite . perhaps the first paper devoted to the study of the ising model in different , however not non - integer dimension , was @xcite where non - universal properties of the model were discussed . all these approaches , as well as the computer simulations , confirm the correctness of the universality hypothesis also for non - integer @xmath3 hypercubic lattices and allow to obtain the critical exponents as functions of @xmath3 with high accuracy . in spite of the variety of approaches to treat the general non - integer @xmath3 case @xcite the results for the ising model critical exponents obtained on their basis lay close to each other ; the above mentioned analytic continuations may appear actually equivalent . note , however , that the analytic continuation in @xmath10 at general non - integer dimension leads to the fail of the yang - lee theorem @xcite . on the other hand , the study of ising - like spin systems on non - integer dimension hypercubic lattices can not be reduced to the fractal lattice systems @xcite except for the case of vanishing lacunarity limit @xcite and thus is the task of individual interest . returning to the study of the critical behaviour at integer @xmath3 , one should note that the problem becomes more complicated when studying spin systems with a structural disorder . whereas the case of the annealed disorder is of less interest from the point of view of determining asymptotical values of critical exponents @xcite , the weak * quenched * disorder has been a subject of intensive study . here the harris criterion @xcite has been devised . it states that if the heat capacity exponent @xmath11 of a pure model is negative , that is the heat capacity has no divergence at the critical point , impurities do not affect the critical behaviour of the model in the sense that critical exponents remain unchanged under dilution . only in the case @xmath12 , the critical behaviour of the disordered model is governed by a _ new set of critical exponents_. as far as for a @xmath13 @xmath7-vector spin model only the @xmath13 ising model ( @xmath4 ) is characterized by @xmath14 , it is the ising model which is of special interest . and because of the triviality of the annealed disorder in the sense mentioned above , the most interesting object for study is just the quenched ising model . the appearance of a set of new critical exponents for that model at @xmath15 is confirmed by the experiments @xcite , renormalization group @xmath16 calculations @xcite , monte - carlo @xmath17 @xcite and @xmath18 @xcite simulations . the situation is not so simple for the @xmath19 ising model . onsager exact solution of the pure model proves the logarithmic divergence of heat capacity , which yields @xmath20 , and allows one , in accordance with the harris criterion , to clasify this case as a marginal one . most of the theoretical works suggest that the @xmath19 ising model with a quenched disorder has the same critical behaviour as the @xmath19 pure ising model ( except for logarithmic corrections ) @xcite ( see also review @xcite ) . this result is corroborated by @xmath21-simulations on two - dimensional lattices @xcite and experiments @xcite . deviations from the expected critical exponents , which sometimes are observed during such computations , are explained by a system being not in the asymptotic region ( see @xcite for recent study ) . nevertheless , some authors assert that for the @xmath19 ising model with a quenched disorder a new critical behaviour appears @xcite while the undiluted ising model at non - integer @xmath3 was a subject of intensive study @xcite , it is not the case for the diluted ising model . only the work @xcite can be mentioned here , where the model was studied within the golner - riedel scaling field @xcite approach . it is worthwhile to note that the @xmath22-expansion technique applied to this model , due to the fact that @xmath23-equations appear to be degenerated on the one loop level , results in @xmath24-expansion for the critical exponents @xcite . the latter is known up to the three - loop order @xcite . the equations of the massive field theory at fixed integer @xmath3 @xcite first applied to the diluted ising model at @xmath25 in @xcite were found to be the most effective method for investigating this problem . in order to consider an arbitrary non - integer @xmath3 the parisi approach @xcite was generalized in @xcite where critical behaviour of the model was studied in a two - loop approximation . the aim of the present work , based on the massive field theoretical approach , is to make a more detailed investigation of the critical behaviour of the diluted @xmath26-vector model at arbitrary @xmath3 . though it is the case @xmath4 in which we are interested most of all , we consider the @xmath23-equations for any @xmath7 , which also allow us to study the crossover in the model at any @xmath3 . we will obtain the @xmath23-equations within the 3-loop approximation and apply to their analysis different resummation procedures in order to find the most reliable one . the set - up of the article is as follows . in the next section we introduce the model and the notation . then we describe the @xmath23-procedure adopted here and give the series for the @xmath23-functions of the weakly diluted quenched @xmath7-vector model in the three - loop approximation . being asymptotic , these series are to be resummed . this is done in section 2 where different ways of resummation are used . section 3 concludes our study giving results for the quantitative characteristics of the critical behaviour and discussing them . in the conclusions we give some general comments to the present work . in the appendix we list some lengthy expressions for the coefficients of the @xmath23-functions in the three - loop approximation as it is well known , the critical behaviour of the quenched weakly - diluted @xmath7 - vector model is governed by a lagrangian with two coupling constants @xcite : @xmath27 + { v_{0}\over 4 ! } \left(\sum_{\alpha=1}^{n}|\vec{\phi}^\alpha|^2 \right)^2 + \nonumber \\ \label{2.1 } & & { u_{0}\over 4 ! } \sum_{\alpha=1}^{n}\left(|\vec{\phi}^\alpha|^2 \right)^2 \big\},\end{aligned}\ ] ] in replica limit @xmath28 . here any @xmath29 is a @xmath7-component vector + @xmath30 ; @xmath31 are bare coupling constants ; @xmath32 is bare mass . as it was already stated above , we adopt here the massive field theory renormalization scheme @xcite in order to extract the critical behaviour governed by ( [ lagrangian ] ) . we start from the defined by ( [ lagrangian ] ) unrenormalized one - particle irreducible vertex functions @xmath33 depending on the wave vectors @xmath34 , bare parameters @xmath35 and the ultraviolet momentum cutoff @xmath36 . the vertex functions dependence on the space dimension @xmath3 is explicitly noted here as well . we impose the renormalization conditions at zero external momenta and non - zero mass ( see @xcite for instance ) at the limit @xmath37 for the renormalized functions @xcite @xmath38 : @xmath39 with @xmath40 being the renormalized mass @xmath41 and couplings @xmath42 . from these conditions there follow expansions for the renormalized constants for field @xmath43 , vertices @xmath44 @xmath45 , @xmath46 @xmath47 and @xmath48 insertion @xmath49 . subsequently , these define the coefficients @xmath50 entering the corresponding callan - symanzik equation : @xmath51 to be defined by simultaneous zero of both @xmath0-functions : @xmath52 the @xmath53-function gives the critical exponent @xmath54 of the pair correlation function : @xmath55 the correlation length critical exponent @xmath56 is defined in the stable fixed point by : @xmath57 using familiar scaling relations , one can easily calculate any other critical exponents on the base of @xmath54 and @xmath56 . applying the described above procedure , one obtains in the three - loop approximation @xcite @xmath0- and @xmath1-functions in the form @xcite : @xmath58 u^2 + \frac{96}{(m+8)(mn+8 ) } \times \nonumber\\ & & \big [ ( m+5)(i_1-\frac{1}{2})+ \frac{m+2}{6 } i_2 \big ] uv + \frac{24}{(mn+8)^2 } \times \nonumber\\ & & \big [ ( mn+14)(i_1-\frac{1}{2 } ) + \frac{mn+2}{3 } i_2 \big ] v^2 + \beta_u^{(3la ) } + \dots \big \ } \\ \label{beta_v } \beta_v(u , v ) & = & -(4-d ) v \big \ { 1 - v - \frac{2(m+2)}{m+8 } u + \frac{8}{(mn+8)^2 } \times \\ & & \big [ ( 5mn+22)(i_1-\frac{1}{2})+(mn+2 ) i_2 \big ] v^2 + \frac{96(m+2)}{(m+8)(mn+8 ) } \times \nonumber\\ & & \big [ i_1-\frac{1}{2 } + \frac{i_2}{6}\big ] uv + \frac{24(m+2)}{(m+8)^2 } \big [ i_1-\frac{1}{2 } + \frac{i_2}{3 } \big ] u^2 + \beta_v^{(3la ) } + \dots \big \ } \nonumber\\ \gamma_{\phi}(u , v ) & = & - 2 ( 4-d ) \big \ { \big [ \frac{2(m+2)}{(m+8)^{2 } } u^{2 } + \frac{4(m+2)}{(m+8)(mn+8 ) } uv + \nonumber\\ & & \frac{2(mn+2)}{(mn+8)^{2 } } v^{2 } \big ] i_2 + \gamma_{\phi}^{(3la ) } + \dots \big \ } \\ \label{gamma } \bar{\gamma}_{\phi^2}(u , v ) & = & ( 4-d ) \big \ { \frac{m+2}{m+8 } u + \frac{mn+2}{mn+8 } v - 12 \big [ \frac{m+2}{(m+8)^2 } u^2 + \\ & & \frac{2(m+2)}{(m+8)(mn+8 ) } uv + \frac{mn+2}{(mn+8)^2 } v^2 \big ] ( i_1-\frac{1}{2 } ) + \bar{\gamma}_{\phi^2}^{(3la ) } + \dots \big \ } \nonumber\end{aligned}\ ] ] here @xmath3 is the space dimension , @xmath7 is the order parameter component number , @xmath59 is the replica index , @xmath60 and @xmath61 are dimensionally dependent two - loop integrals . the corresponding coefficients for three - loop parts are listed in the appendix . the values for the three - loop integrals @xmath62 which appear in three - loop coefficients for integer @xmath63 are listed in @xcite . in particular , substituting loop integrals @xmath64 as well as @xmath65 in ( [ beta_u])-([gamma ] ) by their values at @xmath15 we get at @xmath66 the corresponding functions of the @xmath13 weakly diluted ising model , which in the 3-loop approximation were obtained in @xcite . at @xmath15 , @xmath67- arbitrary corresponding expressions coincide with those , obtained for the @xmath13 anisotropic @xmath2-vector model in @xcite . our idea is to keep the dimensional dependence of the loop integrals and , being based on their numerical values for arbitrary @xmath3 @xcite , to study the @xmath68-model at arbitrary ( non - integer ) @xmath3 as well . but for the reason explained above , the point of main interest here will be the replica limit @xmath5 of the anisotropic @xmath2-vector model , especially the case @xmath4 . expressions for @xmath0- and @xmath1-functions will be the starting point for the qualitative study of the main features of the critical behaviour which will be done in the next section . as we have already mentioned , the values of the @xmath1-functions in a fixed point @xmath69 lead to the values of the critical exponents @xmath54 and @xmath56 . however , it is well known now that the series for @xmath23-functions are of asymptotic nature @xcite and imply the corresponding resummation procedure to extract reliable data on their basis . let us note , however , that as to our knowledge the asymptotic nature of the series for @xmath23-functions have been proved only for the case of the model with one coupling @xcite , and the application of a resummation procedure to the case of several coupling constants is based rather on general belief than on a proved fact . one of the resummation procedures , which in different modifications is most commonly used in the studies of asymptotic series , is known as the integral borel transformation @xcite . however , this technique implies explicit knowledge of the general term of a series and thus can not be applied here , where only truncated sums of the series are known . to get over this obstacle one represents the so - called borel - leroy image of the initial sum in the form of a rational approximant and in such a way reconstitutes the general term of the series . the technique which involves a rational approximation and the borel transformation together , is known as the pad - borel resummation technique ( in the field - theoretical @xmath23 content see @xcite as an example of its application ) . note here that the resummation technique , based on the conformal mapping , which is widely used in the theory of critical phenomena @xcite , can not be applied in our case because its application postulates information on the high order behaviour of the series for @xmath0- and @xmath1-functions . the latter is still unknown for the theory with the lagrangian ( [ lagrangian ] ) . to summarize up the stated let us write that the pad - borel resummation is performed as follows : * constructing the borel - leroy image of the initial sum @xmath70 of @xmath59 terms : @xmath71 where @xmath72 is the euler s gamma function and @xmath73 is an arbitrary non - negative number . the special cases @xmath74 and @xmath75 correspond to resumming @xmath0-functions without or with prefactors @xmath44 and @xmath46 in accordance with the structure of the functions ( [ beta_u])-([beta_v ] ) ; * the borel - leroy image ( [ leroy ] ) is extrapolated by a rational approximant @xmath76 ( xt)$ ] , where by @xmath76 $ ] one means the quotient of two polynomials ; @xmath77 is the order of the numerator and @xmath78 is that of the denominator ; * the resummed function @xmath79 is obtained in the form : @xmath80 ( xt).\ ] ] in the two variables case only the first step is changed ; namely , here we define the borel - leroy image as @xmath81 generalization to the many - variable case is trivial . now one can take into account that the second step of the stated scheme can be done in different ways . one can write down various pad approximants in the variable @xmath82 to obtain within the three - loop approximation the expressions of the structure @xmath83 $ ] , @xmath84 $ ] and @xmath85 $ ] . it is also possible to use chisholm approximants @xcite in the variables @xmath44 and @xmath46 , which , generally speaking , in the same number of loops can be of type @xmath86 $ ] , @xmath87 $ ] , @xmath88 $ ] and @xmath89 $ ] , but the explicit definition of any approximant needs some additional equations now @xcite . the technique , which involves chisholm approximation together with the integral borel transformation is referred to as the chisholm - borel resummation technique . to be consistent , one would have to apply the all different resummation frameworks in order to obtain reliable results on their basis and find which of the methods is the most effective . however , strong restriction on the number of choices can be imposed . first of all , an approximant should be chosen in the form reconstituting the sign - alternating high - order behaviour of the general term of @xmath0- and @xmath1-functions , which was confirmed in the particular case @xmath4 , @xmath90 and @xmath91 @xcite . the approximant generating a sign - alternating series might be chosen in a form @xmath92 $ ] with the positive coefficients at the variable @xmath82 ( or @xmath44 and @xmath46 ) . choosing an approximant with a non - linear denominator , generally speaking , one does not ensure the desired properties . direct calculations affirm the argumentation : @xmath0-functions , resummed with the pad - borel and the chisholm - borel methods with approximants @xmath76 , n>1 $ ] , for @xmath93 give the roots which lie far from the expected values which for @xmath15 are known up to the order of four loops @xcite and for general @xmath3 were calculated from the two - loop @xmath0-functions @xcite . this is true for any @xmath73 . the stated results permit us to eliminate from the consideration approximants with a non - linear denominator . note as well that choosing representation of the @xmath23-functions ( [ beta_u])-([gamma ] ) in the form of pad or chisholm approximant of type @xmath94 $ ] might result in the appearance of a pole in the expression . here we use an analytical continuation of the resulting expressions by evaluating the principal value of the integral . treating the task in this way one notes that the topological structure of the lines of zeros for the resummed by the pad - borel technique @xmath0-functions is very different in the region near the solution for the mixed fixed point and strongly irregular when passing through the number of loops . in particular this yields that in the three - loop approximation there exist two solutions close to the expected value of the mixed fixed point . to compare , the results obtained within the frames of the chisholm - borel method do not have these faults and are more stable from the point of view of proceeding in number of loops . so , the results given below are obtained by the chisholm - borel method applied to the approximant of type @xmath86 $ ] . in order to determine the form of this approximant completely one must define two additional conditions . the approximants are expected to be symmetric in variables @xmath44 and @xmath46 , otherwise the properties of the symmetry related to these variables would depend , except for the properties of the lagrangian , on the method of calculation . by the substitution @xmath95 all the equations which describe the critical behaviour of the diluted model are converted into appropriate equations of the pure model . however , if pure model is solved independently , the resummation technique with the application of pad approximant is used . thus , chisholm approximant is to be chosen in such a way that , by putting any of @xmath44 or @xmath46 equal to zero , one obtains pad approximant for a one - variable case . this also implies a special choice of additional conditions . in the present study amidst all the possible expressions which satisfy the stated demand we choose chisholm approximant @xmath86 $ ] by putting coefficients at @xmath96 and @xmath97 to be equal to zero . now we are going to apply the mathematical framework which was discussed in previous sections in order to obtain numerical characteristics of the critical behaviour of the weakly - diluted ising model in general dimensions . it was noted in the section 1 that the critical behaviour of the quenched weakly - diluted ising model is described by the effective lagrangian ( [ lagrangian ] ) in the case @xmath4 and zero replica limit @xmath5 . namely , the task in the end comes to obtaining fixed points which are defined by simultaneous zero of the both @xmath0-functions . among all the possible fixed points one is interested only in those in the ranges @xmath98 and only in stable ones where the stability means that two eigenvalues @xmath99 of the stability matrix @xmath100 , @xmath101 are positive or possess positive real parts . the structure of the @xmath0-functions ( [ beta_u])-([beta_v ] ) yields the possibility of four solutions for the fixed points . the first two @xmath102 and @xmath103 in our case at @xmath104 are out of physical interest , while the second pair which consists of pure @xmath105 and mixed @xmath106 points , are responsible for two possible critical regimes . the critical behaviour of the diluted model coincides with that of the pure model when the pure fixed point appears to be stable . if the mixed point is stable , the _ new _ ( diluted ) critical behaviour of the system takes place . the type of the critical behaviour depends on the number @xmath7 of the order parameter components and on the dimensionality @xmath3 : at any @xmath107 a system with large enough @xmath7 is not sensitive to the weak dilution in the sense that asymptotic values of critical exponents do not change ; only starting from some marginal value @xmath6 , at @xmath108 a mixed fixed point becomes stable and the crossover to the random critical behaviour occurs . the problem of determining @xmath6 as a function of @xmath3 will be discussed later . now we would like to state that @xmath109 for any @xmath110 , and thus just the mixed fixed point governs the asymptotic critical behaviour of the diluted ising model . if one attempts to find the fixed points from the @xmath0-functions ( [ beta_u])-([beta_v ] ) without resummation , there always appears only the gaussian @xmath102 trivial solution ; the existence of the rest possible three fixed points depends on the concrete details of the @xmath0-functions portions in the braces in expressions ( [ beta_u])-([beta_v ] ) . in a @xmath13 case it appears that without a resummation the non - trivial mixed fixed point does not exist in one- , two- and four - loop approximations @xcite . it is only the three - loop approximation where all the four solutions of the set of equations ( [ fixedpoint ] ) exist @xcite . in figure [ fig1 ] we show the behaviour of the non - resummed @xmath0-functions of the three - dimensional weakly diluted ising model in the three - loop approximation . resummed functions are shown in the same approximation in figure [ fig2 ] . note that in this approximation the shape of the functions remains alike in the region of small couplings @xmath44 and @xmath46 . fixed points correspond to the crossing of the lines @xmath111 as it is demonstrated in figures [ fig3 ] , [ fig4 ] . the left - hand column in figures [ fig3 ] , [ fig4 ] shows the lines of zeros of non - resummed @xmath0-functions in three - dimensions in one- , two- , three- and four - loop ( results of @xcite ) approximations . one can see in the figures that without resummation all non - trivial solutions are obtained only within the three - loop level of the perturbation theory . in the next order all fixed points disappear which is a strong evidence of their accidental origin . at any arbitrary @xmath3 , @xmath8 the qualitative behaviour of the functions is very similar to that shown in figures [ fig3 ] and [ fig4 ] . as it has already been mentioned , in order to reestablish the lost pure and mixed points one applies the resummation procedure to @xmath0-functions . in the three - dimensional space the result of resummation is illustrated by the right - hand column in figures [ fig3 ] and [ fig4 ] . here we have used the chisholm - borel resummation technique choosing chisholm approximant in the form discussed in the previous section with @xmath75 in successive approximation in the number of loops . the icons in the figures which correspond to a one - loop level are the visual proof of the degeneracy of the @xmath0-functions in this order of the perturbation theory : the plots of root - lines are parallel independently of resummation . the rest three images in the right - hand columns are a good graphic demonstration of the reliability of the chisholm - borel resummational method : two- , three- and four - loop pictures are quantitatively similar , the coordinates of the pure and mixed point are close . the numerical results of our study are given in table [ table1 ] . here , the coordinates of the stable mixed fixed point and the values of the critical exponents of the quenched weakly diluted ising model are listed as functions of @xmath3 between @xmath112 and @xmath113 . the eigenvalues @xmath114 and @xmath115 of the stability matrix are given as well . it was already noted that the values of @xmath1-functions in a stable point yield the numerical characteristics of the critical behaviour of the model . for example , given the resummed functions @xmath116 and @xmath117 , the pair of equations @xmath118 allows us to find the exponents @xmath54 and @xmath56 . all other exponents can be obtained from the familiar scaling laws . however , one can proceed in a different way . that is , by means of the scaling laws it is possible to reconstitute the expansion in coupling constants of any exponent of interest or of any combination of exponents , and only after that to apply the resummation procedure . if exact calculation were performed the answer would not depend on the sequence of operations . however , this is not the case for the present approximate calculations . we have chosen the scheme of computing where the resummation procedure was applied to the combination @xmath119 and @xmath120 . the exponents @xmath121 , @xmath0 and @xmath54 have been calculated on the basis of numerical values of the exponents @xmath1 and @xmath56 . the resummation scheme appears to be quite insensitive to the choice of the parameter @xmath73 given by ( [ leroy ] ) , ( [ p ] ) . however note , that computations have been performed here , as well as in @xcite , with @xmath75 . comparing our data from table [ table1 ] for the critical exponents at @xmath112 with the results for the pure ising model one can see that the exponent @xmath1 differs from the exact value @xmath122 by the order of @xmath123 , the exponent @xmath56 is smaller from the exact value @xmath124 less than by @xmath125 . this confirm the conjecture that the critical behaviour of the weakly diluted quenched ising model at @xmath112 within logarithmic correction coincide with that of the pure model ( see @xcite for review ) . it is also interesting to compare numbers given in table [ table1 ] with those obtained for general @xmath3 within the 2-loop approximation @xcite : all the exponents of the three - loop level lie slightly farther from the expected exact values of onsager than those of the two - loop approximation . this may be explained by the oscillatory nature of approaching to the exact values depending on the order of the perturbation theory . it is also interesting to note that the two - loop approximation yields better estimates for the heat capacity critical exponent @xmath121 for all @xmath3 in the range under consideration . namely , in accordance with the harris criterion , the exponent @xmath121 for the diluted ising system should remain negative . this picture is confirmed much better by the two - loop approximation where @xmath121 is negative in the whole range of @xmath3 , unlike the three - loop level of the perturbation theory , the results of which yield @xmath126 for @xmath127 . however , table [ table1 ] shows that the next ( third ) order does improve our understanding of the critical behaviour of the model in general dimensions . the results of the two - loop calculations @xcite show that starting from some marginal space dimension the approach to the stable point becomes oscillatory : the eigenvalues @xmath114 and @xmath115 turn to be complex possessing positive real parts . this is an artifact of the calculation scheme and therefore it was expected @xcite that by increasing the accuracy of calculations one decreases the region of @xmath3 which corresponds to the complex eigenvalues . it is really the case . in the three - loop approximation the region of complex @xmath128 is bounded from below by @xmath129 , whereas in the two - loop approximation @xcite the corresponding value is lower and is equal to @xmath130 . thus , the region of @xmath3 characterized by the oscillatory approach to the stable fixed point shrinks with the increase of the order of the perturbation theory . the comparison of the three - dimensional value of @xmath56 with the four - loop result @xcite @xmath131 gives the accuracy of @xmath132 for our computations ( compare with @xmath133 for two - loops , where the value @xmath134 was obtained ) . thus , it may be stated that the general accuracy of calculations decreases when passing from @xmath135 to @xmath112 which , in particular , results from the fact that our approach is asymptotically exact at upper critical dimension @xmath135 . the comparison of the present results with the other data available is provided by figure [ fig5 ] . here , the behaviour of the correlation length critical exponent @xmath56 obtained by different methods is demonstrated in general dimensions . the results of the massive field - theoretical scheme are plotted by solid ( three - loop approximation ; the present paper ) and dashed ( two - loop approximation ; ref . @xcite ) lines . one can see that the two lines practically coincide far enough from @xmath112 , in particular , both lie very close to the most accurate result for @xmath15 @xcite which is shown by the box . the application of the scaling - field method @xcite yields numbers shown in figure [ fig5 ] by stars . the limit from below ( @xmath136 ) of the method applicability is caused by the truncation of the set of scaling - field equations , which was considered in @xcite . one can also attempt to obtain some results by resumming the @xmath24-expansion which is known for the diluted ising model up to three - loop order @xcite and for the exponents @xmath56 and @xmath54 reads : @xmath137 where @xmath138 is rieman s zeta function . the corresponding results are shown by open diamonds . they were obtained by applying the pad - borel resummation scheme to the series of @xmath24- expansion ( [ nu_sqrteps ] ) @xcite . the value of @xmath56 obtained in such a way is of physical interest only very close to @xmath135 . even in the next orders of the expansions the values of critical exponents are not improved @xcite ; this is an evidence of the @xmath24-expansion unreliability in tasks like the one under consideration . to compare , one can state that the situation with the applied in the present paper theoretical scheme is contrary to the @xmath24-expansion . while the two - loop approximation is valid in ranges @xmath139 , the next order of the perturbation theory enlarges the upper bound up to @xmath113 . one can expect that the next steps within the perturbation theory will allow one to obtain the description of the critical behaviour of the model with enough accuracy for any @xmath3 , @xmath8 . let us recall now that expressions ( [ beta_u])-([gamma ] ) for the @xmath23-functions , as well as their three - loop parts listed in the appendix , allow us to study asymptotic critical properties of the @xmath2-vector model with arbitrary @xmath7 and @xmath59 in arbitrary @xmath3 not only for the case @xmath4 , @xmath5 . in particular , by keeping @xmath7 as an arbitrary number and putting @xmath5 one can obtain the numerical estimates for the marginal order parameter component number @xmath6 which divides the diluted ( governed by the mixed fixed point ) asymptotic critical behaviour from the pure one , when the @xmath26-symmetric fixed point remains stable . in accordance with the harris criterion the case @xmath140 corresponds to zero of the heat capacity critical exponent @xmath121 of the model . one may extract the value of @xmath6 from this condition . however , the above discussed results of the three - loop approximation do not yield enough accuracy for @xmath121 . alternatively , the fixed mixed point should coincide with the pure fixed point at @xmath140 , which in particular means that @xmath141 . the last condition was chosen as a basis of our calculation . the appropriate numbers of the present three - loop approximation ( thick solid line ) together with the data of the two - loop approximation ( dashed line ) @xcite are shown in figure [ fig6 ] . the result of @xmath22-expansion @xmath142 is depicted by the thin solid line . in the three - loop approximation we obtain @xmath143 and @xmath144 . these values are to be compared with the exact results of onsager which yield @xmath145 at @xmath112 , and the theoretical estimate @xmath146 @xcite . one can see that the two - loop results are closer to the expected values for both @xmath112 and @xmath15 . for a two - dimensional case the two - loop value @xmath147 @xcite differs from the exact one by @xmath148 , while the three - loop number decreases the accuracy to @xmath149 . the case @xmath150 contradicts the suggestion that the @xmath151-model asymptotic critical behaviour should not change under dilution in three - dimensions . the reason for decreasing the calculation accuracy with increasing the order of the perturbation theory may lie in oscillatory approach to the exact result . one can expect that already the four - loop case will improve the estimates for @xmath6 for all @xmath152 . let us also note that the determination of @xmath6 may serve as a test for improving the resummation scheme . the goal of this paper is to study the critical behaviour of the weakly diluted quenched ising model in the case when the space dimension @xmath3 continuously changes from @xmath112 to @xmath135 . as it was mentioned in the introduction , the study of the pure ising model at arbitrary @xmath3 , which corresponds to a scalar field - theoretical model with one coupling constant , is the subject of a great deal of papers . it is not the case for the model with a more complicated symmetry . in particular , here we study a model with two couplings corresponding to terms of different symmetry in the lagrangian ( [ lagrangian ] ) . such a problem was studied previously on the basis of the scaling - field method @xcite , and field - theoretical fixed dimension renormalization group calculations within a two - loop level of the perturbation theory are available @xcite . our calculations hold within the theoretical scheme of @xcite . this approach appears to be one amidst other possible calculation schemes for many tasks ; however , in our case it seems to have no alternatives within the field - theoretical approach . being asymptotic , the resulting series for the @xmath23-functions are to be resummed . in the present study we have chosen the pad - borel and the chisholm - borel resummation techniques . restricting ourselves to analytic expressions for the resummed functions , we present numerical data mainly obtained on the basis of the chisholm - borel resummation technique . note that the absence of any information on the high - order behaviour of the obtained series for the @xmath23-functions does not allow one to apply other resummation schemes , e.g. those based on the conformal mapping technique @xcite . the quantitative description of the critical behaviour of the model is steady from the point of view of passing from the two- to the three - loop approximation . smaller agreement between the two- and the three - loop approximations at @xmath3 far away from @xmath135 may be explained in a way that the precision of computing falls down with the increase of the expansion parameter which takes place at decrease of @xmath3 . the real parts of eigenvalues corresponding to the mixed point seem to remain positive up to @xmath135 , which testifies that at arbitrary @xmath3 the weakly diluted quenched ising model is described by the mixed fixed point . the work was supported in part by the ukrainian foundation of fundamental studies ( grant no 24/173 ) . 99 mandelbrot b. b , fractals : form , chance and dimension . freeman , san francisco , cal . , 1977 . gefen y. , mandelbrot b. b. , aharony a. , phys . lett . , 1980 , vol . 45 , no 11 , p. 855 gefen y. , aharony a. , mandelbrot b. b. , kirkpatrick s. , phys . lett . , 1981 , 47 , no 25 , p. 1771 - 1774 . bonnier b. , leroyer y. , meyers c. , phys . b , 1988 , vol . 37 , no 10 , p. 5205 - 5210 . jeewski w. , tomczak p. , physica a , 1991 , vol . 171 , p. 209 - 222 . jeewski w. , tomczak p. , phys . lett . a , 1991 , vol . 157 , no 8 - 9 , p. 507 jeewski w. , physica a , 1994 , vol . 210 , p. 73 - 86 . hu b. , phys . b. , 1986 , vol . 33 , no 9 , p. 6503 - 6504 . wu y. , hu b. , phys . a. , 1987 , vol . 35 , no 3 , p. 1404 - 1411 . wilson k. g. , fisher m. e. , phys . , 1972 , vol . 28 , no 4 , p. 240 le guillou j. g. , zinn - justin j. , j. phys . ( paris ) , 1987 , vol . 48 , no 1 , p. 19 katz s. l. , droz m. , gunton j. d. , phys . b. , 1977 , vol . 15 , no 3 , p. 1597 - 1599 . bonnier b. , hontebeurie m. , j. phys . i ( paris ) , 1991 , vol . 1 , p. 331 - 338 . novotny m. a. , europhys . lett , 1992 , vol . 17 , no 4 , p. 297 novotny m. a. , phys . b , 1992 , vol . 46 , no 5 , p. 2939 - 2950 . filippov a. e. , radievskii a. v. , zh . eksp . fiz . , 1992 , 102 , no 6 , p. 1899 - 1920 filippov a. e. , teor . fiz , 1992 , vol . 91 , no 2 , p. 320 filippov a. e. , radievskii a.v . fiz , 1992 , vol . 56 , no 2 , p. 87 breus s. a. , filippov a. e. , physica a , 1993 , vol . 192 , no 1 , p. 486 - 515 . parisi g. , 1973 , _ in proceedings of the cargrse summer school _ , unpublished parisi g. , j. stat . , 1980 , vol . 23 , no 1 , p. 49 - 82 . holovatch yu . , theor . phys . , 1993 , vol . 96 , no 3 , p. 482 holovatch yu . , krokhmalskii t. , j. math . , 1994 , vol . 35 , no 8 , p. 3866 - 3880 . fisher m. e. , gaunt d.s . , 1964 , vol . 133 , no 1a , p. a224-a239 . baker g. a.,jr . , benofy l. p. , j. stat . , 1982 , vol . 29 , no 4 , p. 699 - 716 . here so called fisher renormalization @xcite holds . it states that the critical exponents of an annealed system are determined by those of the corresponding pure one . fisher m. e. , phys . , 1968 , vol . 176 , no 1 , p. 257 - 272 . harris a. b. , j. phys . c , 1974 , vol . 7 , p. 1671 - 1692 . birgeneau r. j. , cowley r. a. , shirane g. , joshizawa h. , phys . b , 1983 , vol . 27 , no 11 , p. 6747 - 6753 . mitchell p. w. , cowley r. a. , yoshizawa h. , bni p. , uemura y. j. , birgeneau r. j. , phys . b , 1986 , vol . 34 , no 7 , p. 4719 - 4725 . thurston t. r. , peters c. j. , birgeneau r. j. , phys . b , 1988 , vol . 37 , no 16 , p. 9559 - 9563 . lubensky t. c. , phys . b , 1975 , vol . 11 , no 9 , p. 3573 - 3580 . khmelnitskii d. e. , zh . , 1975 , vol . 68 , no 5 , p. 1960 - 1968 . grinstein g. , luther a. , phys . b , 1976 , vol . 13 , no 3 , p. 1329 - 1343 . sokolov a. i. , shalaev , b. n. fiz . tiel . , 1981 , vol . 23 , no 7 , p. 2058 - 2063 . newman k. e. , riedel e. k. , phys . b , 1982 , vol . 25 , no 1 , p. 264 - 280 . jug g. , phys . b , 1983 , vol . 27 , p. 609 - 612 . mayer i. o. , sokolov a. i. , fiz . , 1984 , vol . 26 , no 11 , p. 3454 - 3456 . mayer i. o. , sokolov a. i. , shalaev b.n . , ferroelectrics , 1989 , vol . 95 , p. 93 . mayer i. o. , j. phys . a , 1989 , vol . 22 , p. 2815 - 2823 . janssen h. k. , oerding k. , sengespeick e. , j. phys . a , 1995 , vol . 28 , no 21 , p. 6073 - 6085 . chowdhury d. , staufer d. , j. stat . phys . , 1986 , vol . 44 , no 1/2 , p. 203 - 210 . marro j. , labarta a. , tejada j. , phys . b , 1986 , vol . 34 , no 1 , p. 347 - 349 . wang j .- s . , whlert m. , mhlenbein h. , chowdhury d. , physica a , 1990 , vol . 166 , p. 173 - 179 . heuer h .- o . b , 1990 , vol . 42 , no 10 , p. 6476 - 6484 ; + europhys . lett . , 1990 , vol . 12 , no 6 , p. 551 heuer h .- o . , j. phys . a , 1993 , vol . 26 , p. l333-l339 . holey t. , fhnle m. , phys . b , 1990 , vol . 41 , p. 11709 - 11712 . dotsenko vik . s. , dotsenko vl . s. , j. phys . c , 1982 , vol . 15 , p. 495 - 507 . shankar r. , phys . lett . , 1987 , vol . 58 , no 23 , p. 2466 - 2469 . shalaev b. n. , fiz . tiel . , 1988 , vol . 30 , p. 895 . ludwig a. w. , phys . lett . , 1988 , vol . 61 , p. 2388 . shalaev b. n. , fiz . , 1989 , vol . 31 , p. 93 . ludwig a. w. w. , nucl . b , 1990 , vol . 330 , p. 639 - 680 . shalaev b. n. , phys . reports , 1994 , vol . 237 , no 3 , p. 129 wang j .- s . , selke w. , dotsenko vl . s. , andreichenko v. b. , europhys . lett . , 1990 , vol . 11 , p. 301 - 305 ; + physica a , 1990 , vol . 164 , p. 221 - 239 . andreichenko v. b. , dotsenko vl . s. , selke w. , wang j .- s . , nuclear . b , 1990 , vol . 344 , p. 531 - 556 . heuer h .- o . rev b , 1992 , vol . 45 , no 10 , p. 5691 - 5694 . talapov a. l. , shchur l. n , preprints hep - lat-9404002 and hep - lat-9404011 selke w. , shchur l. n. , talapov a. l. monte carlo simulations of dilute ising models , in : _ annual reviews of computational physics _ world scientific , singapore cowley r. a. , birgeneau r. j. , shirane g. , guggenstein h. j. , ikada h. , phys . rev . b , 1980 , vol . 21 , no 9 , p. 4038 - 4048 . hagen m. , cowley r. a. , nicklow r. m. , ikeda h. , phys . b , 1987 , vol . 36 , no 1 , p. 401 - 408 . dotsenko vik . s. , dotsenko vl . , 1983 , vol . 32 , no 2 , p. 129 - 172 . kim j .- k . , patrascioiu a. , phys . rev . lett . , 1994 , vol . 72 , no 17 , p. 2785 - 2788 . golner g. , riedel e. , phys . rev . lett . , 1975 , 34 , p. 856 ; + phys . . a , 1976 , vol . 58 , p. 11 . shalaev b. , zh . eksp . fiz . , 1977 , vol . 73 , p. 3201 . jayaprakash c. , katz h. j. , phys . b , 1977 , vol . 16 , no 9 , p. 3987 - 3990 . holovatch yu . , shpot m. , j. stat . phys . , 1992 , vol . 66 , no 3/4 , p. 867 - 883 . brzin e. , le guillou j. c. , zinn - justin j. field theoretical approach to critical phenomena , in _ phase transitions and critical phenomena * 6*. academic press , newyork , 1976 d.j.amit . field theory , the renormalization group , and critical phenomena . singapore , world scientific , 1984 . the subscripts @xmath153 at @xmath154 denote the @xmath68-symmetric and hypercubic parts of the four - point vertex function . holovatch yu . , unpublished we have used the change of the notations @xmath155 , @xmath156 and @xmath157 , @xmath158 with d being one loop integral @xmath159 in order to define convenient numerical scale in which the first two coefficients of the functions @xmath160 and @xmath161 are @xmath162 and @xmath163 . nickel b. g. , meiron d. i. , baker g. a. compilation of 2pt and 4pt graphs for continuous spin model . university of guelph report , 1977 . shpot n. a. , phys . lett . a , 1989 , vol . 142 , no 8,9 , p. 474 - 478 . lipatov l. , sov . jetp . , 1977 , vol . 45 , no 2 , p. 216 brzin e. , le guillou j. , zinn - justin j. , phys . d , 1977 , vol . 15 , no 6 , p. 1544 - 1557 . brzin e. , parisi g. , j. stat . phys . , 1978 , vol . 19 , no 3 , p. 269 eckmann j .- p . , magnen j. , seneor r. , commun . phys . , 1975 , vol . 39 , no 4 , p. 251 hardy g. h. divergent series . oxford , 1948 . note that for _ finite _ number of terms changing of order of integration and summation can always be performed baker g. a. , nickel b. g. , green m. s. , meiron d. i. , phys . lett . , 1976 , 36 , p. 1351 . baker g. a. , nickel b. g. , meiron d. i. , phys . b , 1978 , vol . 17 , no 3 , p. 1365 - 1374 . le guillou j. c , zinn - justin j. , phys . b , 1980 , vol . 21 , no 9 , p. 3976 - 3988 . chisholm j. s. r. , math . , 1973 , vol . 27 , p. 841 - 848 . kleinert h. , thoms s. , phys . d , 1995 , vol . 52 , no 10 , p. 5926 - 5943 . holovatch yu . , yavorskii t. , unpublished bervillier c. , phys . b , 1986 , vol . 34 , no 11 , p. 8141 - 8143 here we have collected the most lengthy expressions for the three - loop contributions to the @xmath23-functions . the three - loop part of the @xmath160-function reads : @xmath164 where @xmath165 ; \nonumber\\ \beta_u^{2,1 } & = & -\frac{2}{(m+8)^2(mn+8 ) } \big [ -12(17m^2 + 256m+780)i_1 + \nonumber\\ & & ( m+2)\big(-(3dm+42d+16m+80)i_2 + 12(m+14)i_3 + \nonumber\\ & & 18(m+8)i_8 \big)+ 24(3m^2 + 70m+224)i_4 + \nonumber\\ & & 6(15m^2 + 158m+448)i_5 + 6(3m^2 + 32m+100)i_6 + \nonumber\\ & & 48(5m+22)i_7 + 6(9m^2 + 146m+448 ) \big ] ; \nonumber\\ \beta_u^{1,2 } & = & -\frac{1}{(m+8)(mn+8)^2 } \big [ -12(19m^2n+80mn+470m+2032)i_1 - \nonumber\\ & & \big ( 8(mn+8)(3d+4 ) + m(3dmn+40mn+78d+176 ) \big ) i_2 + \nonumber\\ & & 12(m^2n+8mn+26m+64)i_3 + 48(m^2n+8mn+68 m + \nonumber\\ & & 292)i_4 + 12(11m^2n+34mn+136m+584)i_5 + \nonumber\\ & & 6(m^2n+8mn+50m+256)i_6 + 576(m+5)i_7 + \nonumber\\ & & 36(m+2)(mn+8)i_8 + 12(5m^2n+22mn+136m+584 ) \big ] ; \nonumber\\ \beta_u^{0,3 } & = & -\frac{4}{(mn+8)^3 } \big [ 6(mn+10)(-2(mn+23)i_1 + 3i_6)+ \nonumber\\ & & ( mn+2 ) \big(-(4mn+9d+8)i_2 + 36i_3 + 3(mn+8)i_8\big)+ \nonumber\\ & & 72(3mn+22)i_4 + 9(m^2n^2 + 14mn+88)i_5 + \nonumber\\ & & 24(mn+14)i_7 + 3(m^2n^2 + 38mn+264 ) \big ] . \nonumber\end{aligned}\ ] ] @xmath167 ; \nonumber\\ \beta_v^{1,2 } & = & -\frac{4(m+2)}{(mn+8)^2(m+8 ) } \big [ -4(28mn+275)i_1 - \nonumber\\ & & ( 3dmn+4mn+15d+56)i_2 + 12(mn+5)i_3 + \nonumber\\ & & 24(2mn+27)i_4 + 3(13mn+100)i_5 + 6(3mn+13)i_6 + \nonumber\\ & & 96i_7 + 9(mn+8)i_8+(29mn+316 ) \big ] ; \nonumber\\ \beta_v^{2,1 } & = & -\frac{m+2}{(mn+8)(m+8)^2 } \big [ -12(mn+42m+224)i_1- \nonumber\\ & & ( 3dmn+12dm-8mn+48d+16m+256)i_2 + \nonumber\\ & & 12(mn+4m+16)i_3 + 48(5m+34)i_4 + \nonumber\\ & & 12(13m+56)i_5 + 6(3mn+14m+40)i_6 + \nonumber\\ & & 144i_7 + 36(m+8)i_8 + 12(11m+64 ) \big ] ; \nonumber\\ \beta_v^{3,0 } & = & -\frac{2(m+2)}{(m+8)^3 } \big [ -4(11m+70)i_1 - 3(dm+2d+16)i_2 + \nonumber\\ & & 6(m+2)(2i_3 + 3i_6)+ 2(m+8)(12i_4 + 3i_5 + 3i_8 + 5 ) \big ] . \nonumber\end{aligned}\ ] ] the three - loop part of the @xmath53-function reads : @xmath168(3i_8 - 4i_2 ) . \nonumber\end{aligned}\ ] ] the three - loop part of the @xmath169-function reads : @xmath170 where @xmath171 ; \nonumber\\ \bar{\gamma}_{\phi^2}^{2,1 } & = & \frac{m+2}{(m+8)^2(mn+8 ) } \big [ -12(mn+10m+70)i_1 + \nonumber\\ & & ( mn+2m+6)\big(-(3d-8)i_2 + 12i_3 + 18i_6\big ) + \nonumber\\ & & 6(m+8)(12i_4 + 3i_5 + 5 ) \big ] ; \nonumber\\ \bar{\gamma}_{\phi^2}^{1,2 } & = & \frac{3(m+2)}{(m+8)(mn+8)^2 } \big [ -4(11mn+70)i_1 + \nonumber\\ & & ( mn+2)\big(-(3d-8)i_2 + 12i_3 + 18i_6\big ) + \nonumber\\ & & 2(mn+8)(12i_4 + 3i_5 + 5 ) \big ] ; \nonumber\\ \bar{\gamma}_{\phi^2}^{0,3 } & = & \frac{mn+2}{(mn+8)^3 } \big [ -4(11mn+70)i_1 + \nonumber\\ & & ( mn+2)\big(-(3d-8)i_2 + 12i_3 + 18i_6\big ) + \nonumber\\ & & 2(mn+8)(12i_4 + 3i_5 + 5 ) \big ] . \nonumber\end{aligned}\ ] ] figure [ fig3 ] the lines of zeros of non - resummed ( left - hand column ) and resummed by the chisholm - borel method ( right - hand column ) @xmath0-functions for @xmath173 in different orders of the perturbation theory : one- and two - loop approximations . circles correspond to @xmath174 , thick lines depict @xmath175 . thin solid and dashed lines show the roots of the analytically continued functions @xmath160 and @xmath161 respectively . one can see the appearance of the mixed fixed point @xmath176 in the two - loop approximation for the resummed @xmath0-functions . figure [ fig4 ] the lines of zeros of non - resummed ( left - hand column ) and resummed by the chisholm - borel method ( right - hand column ) @xmath0-functions for @xmath173 in three- and four - loop approximations . the notations are the same as in figure [ fig3 ] . close to the mixed fixed point the behaviour of the resummed functions remains alike with the increase of the order of approximation . this is not the case for non - resummed functions . figure [ fig5 ] . the correlation length critical exponent @xmath56 of the weakly diluted ising model as a function of the space dimension @xmath3 . the results of two- @xcite and three - loop ( the present paper ) approximations are shown by the dashed and the solid lines respectively , the square reflects the number of the four - loop approximation @xcite at @xmath15 , stars correspond to work @xcite and open diamonds refer to the resummed @xmath177-expansion . figure [ fig6 ] . the dependence of the marginal order parameter component number @xmath6 on the space dimension @xmath3 . two- and three - loop results are shown by the dashed and thick solid lines respectively , the @xmath22-expansion data @xmath142 are depicted by the thin solid line . lllllllll @xmath3 & @xmath178 & @xmath179 & @xmath1 & @xmath56 & @xmath121 & @xmath54 & @xmath114 & @xmath115 + + 2.0 & 2.0268 & -0.2802&1.840&0.966 & 0.067&0.097 & 0.2176 & 1.5189 + 2.1 & 2.0327 & -0.3156&1.768&0.923 & 0.062&0.084 & 0.2373 & 1.4608 + 2.2 & 2.0412 & -0.3523&1.703&0.884 & 0.056&0.073 & 0.2562 & 1.4011 + 2.3 & 2.0525 & -0.3908&1.643&0.848 & 0.049&0.064 & 0.2742 & 1.3395 + 2.4 & 2.0671 & -0.4312&1.588&0.816 & 0.041&0.055 & 0.2913 & 1.2759 + 2.5 & 2.0854 & -0.4740&1.536&0.787 & 0.033&0.047 & 0.3074 & 1.2100 + 2.6 & 2.1081 & -0.5196&1.489&0.760 & 0.025&0.040 & 0.3226 & 1.1418 + 2.7 & 2.1359 & -0.5687&1.445&0.735 & 0.016&0.034 & 0.3370 & 1.0709 + 2.8 & 2.1698 & -0.6219&1.404&0.712 & 0.007&0.028 & 0.3505 & 0.9971 + 2.9 & 2.2113 & -0.6803&1.365&0.691&-0.002&0.023 & 0.3635 & 0.9197 + 3.0 & 2.2621 & -0.7454&1.328&0.671&-0.016&0.019 & 0.3764 & 0.8380 + 3.1 & 2.3250 & -0.8190&1.294&0.652&-0.021&0.015 & 0.3905 & 0.7504 + 3.2 & 2.4039 & -0.9038&1.261&0.634&-0.030&0.012 & 0.4095 & 0.6528 + 3.3 & 2.5044 & -1.0040&1.230&0.618&-0.038&0.009 & 0.4653 & 0.5127 + 3.4 & 2.6359 & -1.1259&1.200&0.602&-0.046&0.006 & 0.4436@xmath180&0.4436@xmath180 + 3.5 & 2.8140 & -1.2804&1.171&0.587&-0.054&0.004 & 0.3946@xmath180&0.3946@xmath180 + 3.6 & 3.0678 & -1.4869&1.144&0.572&-0.061&0.002 & 0.3411@xmath180&0.3411@xmath180 + 3.7 & 3.4570 & -1.7849&1.116&0.558&-0.066&0.001 & 0.2822@xmath180&0.2822@xmath180 + 3.8 & 4.0852 & -2.2303&1.087&0.544&-0.066&0.000 & 0.2136@xmath180&0.2136@xmath180 +

we investigate the bose - einstein condensation patterns , the critical and multicritical behaviors of three - dimensional mixtures of bosonic gases with short - range density - density interactions . these systems have a global u(1)u(1 ) symmetry , as the system hamiltonian is invariant under independent u(1 ) transformations acting on each species . in particular , we consider the three - dimensional bose - hubbard model for two lattice bosonic gases coupled by an on - site inter - species density - density interaction . we study the phase diagram and the critical behaviors along the transition lines characterized by the bose - einstein condensation of one or both species . we present mean - field calculations and numerical finite - size scaling analyses of quantum monte carlo data . we also consider multicritical points , close to which it is possible to observe the condensation of both gas components . we determine the possible multicritical behaviors by using field - theoretical perturbative methods . we consider the u(1)u(1)-symmetric landau - ginzburg - wilson theory and determine the corresponding stable fixed points of the renormalization - group flow . the analysis predicts that , in all cases , the multicritical behavior is analogous to the one that would be observed in systems of two identical gases , with an additional exchange symmetry .

1601.07675

the complex behavior of mixtures of bosonic gases has been extensively investigated experimentally in particular , in cold - atom systems @xcite and theoretically @xcite . these systems exhibit a rich behavior , at zero and finite temperature , with several different phases separated by transition lines , along which one or more components of the system undergo bose - einstein condensation ( bec ) . in this paper we consider three - dimensional ( 3d ) mixtures of two different bosonic gases with short - range interactions that only depend on the local densities of the two gases . the hamiltonian of these systems is invariant under u(1 ) transformations acting independently on each species , so that the model is u(1)@xmath0u(1 ) symmetric . in particular , we consider the 3d two - component bose - hubbard model with an on - site inter - species density - density interaction . this is a realistic model for two bosonic species in optical lattices @xcite . we determine the finite - temperature phase diagram by using a variety of techniques . first , we consider the mean - field ( mf ) approximation , determining the qualitative phase behavior of the system as a function of the model parameters , such as the chemical potentials and the on - site inter- and intra - species couplings . we find several different phases , in which each species may be in the normal or superfluid state , and identify critical lines and multicritical points ( mcps ) where some transition lines meet . the 3d phase diagram is investigated in the hard - core ( hc ) limit of each species by a finite - size scaling ( fss ) analysis of quantum monte carlo ( qmc ) simulations . the numerical data allows us to identify the universality class of the transition lines that correspond to the bec of one of the two species . we show that , independently whether the noncritical component is in the normal or superfluid phase , the critical behavior of the condensing species belongs to the 3d xy universality class characterized by the breaking of a global u(1 ) symmetry and by short - range effective interactions . this is the same universality class associated with the bec of a single bosonic gas @xcite ( and also with the superfluid transition in @xmath3he @xcite , with transitions in some liquid crystals characterized by density or spin waves and in magnetic systems with easy - plane anisotropy , etc . this result implies an effective decoupling of the critical modes of the condensing species from those of the noncritical component , independently whether the latter is in the normal or superfluid phase . the phase diagram of mixtures of bosonic gases also presents particular points where some transition lines meet . multicritical behaviors develop at these mcps , arising from the competition of the two u(1 ) order parameters associated with the bec of the two species . to identify the possible universality classes of the multicritical behaviors , we use the field - theoretical approach , considering the effective landau - ginzburg - wilson ( lgw ) @xmath1 theory for two complex fields with global u(1)@xmath0u(1 ) symmetry . we study the renormalization - group ( rg ) flow in the quartic - parameter space , identifying the stable fixed points ( fps ) , which control the critical behavior , and their attraction domain . the paper is organized as follows . in sec . [ model ] we define the bose - hubbard model for two lattice bosonic gases . in sec . [ phmf ] we determine the phase diagram of the model in the mf approximation , showing that , by changing the model parameters , one can obtain qualitatively different behaviors . in sec . [ critbeh ] we present our numerical results and determine numerically the critical behavior along the transition lines where one species undergoes bec . in sec . [ multi ] we study the multicritical behaviors at mcps where some transition lines meet in the phase diagram . finally , in sec . [ conclu ] we draw our conclusions . [ 5loop ] reports the five - loop perturbative series of the minimal - subtraction scheme , which are used in the rg study of the multicritical behavior . the two - species bose - hubbard ( 2bh ) model is a lattice model appropriate to describe mixtures of bosonic gases with local density - density interactions . its hamiltonian reads @xmath4 where @xmath5 is a site of a cubic lattice , @xmath6 labels a lattice link connecting two nearest - neighbor sites , the subscript @xmath7 labels the two species , and @xmath8 is the density operator of the @xmath7-species . the 2bh model is symmetric under the u(1 ) transformations @xmath9 acting independently on the two species . therefore , the global symmetry group is @xmath10 . for @xmath11 , @xmath12 , and @xmath13 , the 2bh model ( [ sbh ] ) describes the behavior of a mixture of two identical bosonic gases and it has been extensively studied in ref . @xcite . in this case , the model has an additional @xmath14 exchange symmetry , so that the symmetry group becomes @xmath15 $ ] . in the hc limit , or , more generally , when @xmath16 , both components condense at the transition . thus , the global symmetry breaks to @xmath17 $ ] . the critical behavior is controlled by a decoupled 3d xy fixed point @xcite : the critical behaviors of the two gases are effectively decoupled and belong to the 3d xy universality class associated with the symmetry breaking @xmath18 . although the inter - species density - density interaction does not change the leading critical behavior , it plays an important role close to criticality , as it gives rise to very - slowly - decaying scaling corrections , which are not present at the bec transition of a single bosonic species . in the opposite case , i.e. , for @xmath19 , only one component condenses , so that the global symmetry is broken to @xmath20 . the associated critical behavior belongs to a different 3d universality class @xcite . in the following we consider mixtures of nonidentical gases described by the 2bh model ( [ sbh ] ) . as we shall see , their finite - temperature phase diagrams present several phases where the two species are in the normal or superfluid state , separated by transition lines along which only one species condenses . moreover , we will discuss mcps , which are points in the phase diagram where some transition lines meet . the hard - core ( hc ) limit for the @xmath7 component is obtained by taking @xmath21 . in this limit , using the particle - hole transformation , we can relate the spectrum of the hamiltonian for two different sets of parameters . for instance , assume that @xmath22 , so that @xmath23 can only assume the values 0 and 1 . under a particle - hole transformation the kinetic term is unchanged , while @xmath24 . thus , the spectrum of the model with chemical potentials @xmath25 and @xmath26 and interaction @xmath27 is related to that of the model with chemical potentials @xmath28 and @xmath29 and interaction @xmath30 with @xmath31 indeed , the energy levels of the two models differ by an irrelevant constant term proportional to @xmath25 . an analogous relation holds for the second species if @xmath32 . these relations imply that , if one of the two components has a hard core , one can limit oneself to study the phase diagram for @xmath33 . , as a function of @xmath25 and and @xmath27 . the symbol @xmath34 marks the superfluid phases : @xmath35 and @xmath36 indicate the phases in which only components 1 and 2 are superfluid , respectively ; in phase @xmath37 both components are superfluid . the symbols @xmath38 and @xmath39 mark mott phases for species @xmath7 with filling 0 and 1 , respectively . in the leftmost phase marked @xmath40 , we have @xmath41 for the superfluid density and @xmath42 , which are the values corresponding to a single gas with vanishing chemical potential . ] and two values of @xmath27 , @xmath43 ( top ) and @xmath44 ( bottom ) , as a function of @xmath25 and @xmath45 . in phase @xmath35 component 1 is in the superfluid state , while component 2 is in the normal state . in phase @xmath36 the opposite occurs . in phase @xmath37 both gases are superfluid . in the upper ( and , for @xmath43 , rightmost ) region with no label both gases are in the normal state . the full and dashed lines indicate the normal - to - superfluid transition lines of gas 1 and 2 , respectively . , title="fig : " ] and two values of @xmath27 , @xmath43 ( top ) and @xmath44 ( bottom ) , as a function of @xmath25 and @xmath45 . in phase @xmath35 component 1 is in the superfluid state , while component 2 is in the normal state . in phase @xmath36 the opposite occurs . in phase @xmath37 both gases are superfluid . in the upper ( and , for @xmath43 , rightmost ) region with no label both gases are in the normal state . the full and dashed lines indicate the normal - to - superfluid transition lines of gas 1 and 2 , respectively . , title="fig : " ] some qualitative or semi - quantitative aspects of the phase diagram can be inferred by mf calculations . for this purpose we make the approximation @xmath46 \left [ ( b_{s\bm{y } } - \phi_s ) + \phi_s \right ] \nonumber \\ & \approx & \phi_s b^\dag_{s\bm{x } } + \phi_s^ * b_{s{\bm y } } - |\phi_s|^2 , \label{approx - meanfield}\end{aligned}\ ] ] where @xmath47 are two complex space - independent variables , which play the role of order parameters at the bec transitions . the parameters @xmath48 are related to the superfluid densities by @xmath49 . ( [ approx - meanfield ] ) allows us to rewrite hamiltonian ( [ sbh ] ) as a sum of decoupled one - site hamiltonians @xmath50 where @xmath51 . since the spectrum of the theory is invariant under @xmath52 , where @xmath53 are two independent phases , the two parameters @xmath48 can be assumed to be real without loss of generality . they are determined by minimizing the single - site free energy @xmath54 where @xmath55 are the eigenvalues of @xmath56 @xcite . at zero temperature the minimization of the free energy corresponds to the minimization of the ground - state energy @xmath57 . in the following , we restrict ourselves to the case @xmath58 and @xmath59 . moreover , we set @xmath60 , so that all energies are expressed in units of @xmath61 . the model shows a complex phase diagram , with transition lines ( surfaces ) along which one component undergoes a transition from the normal state to a superfluid one . note that , in the limit of zero temperature , the normal phases become mott insulating phases or simply correspond to the vacuum . in the following we present mf results for some selected values of the model parameters , which should be representative of the different finite - temperature behaviors that can be observed by varying the parameters . to begin with , we consider the hc limit @xmath62 . [ fig : t0_hc ] shows the zero - temperature phase diagram as a function of @xmath25 and @xmath27 , when component 2 has zero chemical potential , i.e. , for @xmath63 . we observe several mott and superfluid phases , separated by continuous transition lines . the mott phases appearing in fig . [ fig : t0_hc ] are somewhat trivial , as they correspond to the vacuum or to unit filling ( one particle per site ) . if one chooses more general values for @xmath64 and @xmath65 , one may obtain more complex mott phases at @xmath66 . for instance , for @xmath12 and @xmath67 , the phase diagram also shows a mott phase with global unit filling ( @xmath68 but @xmath69 ) and , therefore a degenerate ground state @xcite . the large degeneracy of the ground state may be described in terms of _ isospin _ degrees of freedom per site , interacting by means of an effective low - energy spin hamiltonian @xcite . at finite temperature the vacuum and mott phases are replaced by normal phases . as suggested by the @xmath66 phase diagram of fig . [ fig : t0_hc ] , we may have different behaviors depending on the strength of the inter - species interaction @xmath27 . for example , fig . [ fig : tpos_hc ] shows the phase diagram for @xmath63 as a function of @xmath25 and @xmath45 , for two values of @xmath27 , @xmath43 and @xmath44 . for @xmath43 there is a single multicritical point ( mcp ) where four transition lines meet . for @xmath44 , instead , two different mcps are present . the change of behavior occurs for @xmath70 . it can be related to the phase boundary @xmath71 appearing in the phase diagram of a single hc bosonic gas . in the mf approximation all phase boundaries correspond to continuous transitions . and @xmath44 , as a function of @xmath25 and @xmath72 . the different phases are labelled as in fig . [ fig : t0_hc ] . for @xmath73 , a first - order transition line ( indicated by a thick line ) runs along @xmath74 . the phase @xmath75 , @xmath36 occurs only for @xmath76 . ] and @xmath44 as a function of @xmath45 and @xmath25 . we report results for several values of the intra - species @xmath72 repulsion , for @xmath72 close to @xmath77 ( at @xmath66 , the phase @xmath75 , @xmath36 occurs only for @xmath76 , see fig . [ fig : t03_sc ] ) . the different phases and the transition lines are labelled as in fig . [ fig : tpos_hc ] . ] in order to understand the role of finite intra - species couplings , we consider a finite @xmath72 . in this case , mott phases with higher integer fillings are possible . moreover , when @xmath78 one may have first - order transition lines between the superfluid phases of the two components . for example , fig . [ fig : t03_sc ] shows the zero - temperature behavior for @xmath63 and @xmath44 as a function of @xmath72 and @xmath25 . in this case , the phase @xmath35 , in which component 1 is superfluid and component 2 is depleted , is separated from phase @xmath36 , in which component 2 is superfluid and component 1 is depleted , by a first - order transition line with @xmath74 and @xmath79 . finite intra - species interactions lead to significant changes of the phase diagram also at finite temperature . for example , fig . [ fig : tpos1_sc ] shows the phase behavior for @xmath44 , @xmath80 , and some finite values of @xmath72 , as a function of @xmath25 and @xmath45 . it should be compared with fig . [ fig : tpos_hc ] , where we report the phase diagram in the hc limit for the same values of @xmath27 and @xmath26 . as @xmath72 is increased from @xmath81 to @xmath82 , the phase diagram changes qualitatively . for @xmath83 only one mcp is present , while for @xmath84 three mcps occur . moreover , for @xmath85 there are two different @xmath37 phases . in particular , both components condense for any @xmath86 , if @xmath45 is not too large . such a large-@xmath25 @xmath37 phase occurs for any finite @xmath72 : for @xmath25 larger than a @xmath72 dependent value @xmath87 , i.e. , for @xmath88 , both components always condense for small @xmath45 . for @xmath89 we have @xmath90 , so that such phase does not exist in the hc case , as shown in the lower panel of fig . [ fig : tpos_hc ] . when @xmath78 the finite - temperature phase diagram changes significantly from that observed in the regime @xmath91 . indeed , as shown in fig . [ fig : fst_order ] , in the phase diagram for @xmath63 , @xmath92 , and @xmath43 three transition lines meet at a mcp : two continuous normal - to - superfluid transition lines and a first - order transition line separating the superfluid phases of gases 1 and 2 along the line @xmath93 . this is of course consistent with what observed at zero temperature , see fig . [ fig : t03_sc ] . , @xmath43 , and @xmath92 , as function of @xmath25 and @xmath45 . species 1 and 2 are superfluid for @xmath94 and @xmath95 , respectively . these phases are separated by a first - order transition line along @xmath93 , ending at the point where the continuous normal - to - superfluid transition lines of the two gases meet . in the uppermost phase with no labels , both components are in the normal state . ] mf calculations can be straightforwardly extended to more general cases , such as @xmath96 and/or @xmath67 . however , the main features of the possible finite - temperature behaviors should be already present in the results shown above . in this section we investigate numerically the nature of the transitions occuring in two - component bosonic systems described by the hamiltonian ( [ sbh ] ) . as already discussed in the mf approximation , in most of the cases the transition lines are associated with normal - to - superfluid transitions of one of the two species . however , it is also possible to have first - order transition lines between two phases in which only one component is superfluid ( this behavior is expected in the soft - core regime , see fig . [ fig : fst_order ] ) and mcps where three or four transition lines meet . since the normal - to - superfluid transition of a single species is related to the spontaneous breaking of the u(1 ) symmetry of the condensing species , it seems natural that the critical behavior belongs to the standard 3d xy universality class @xcite . this conjecture , however , requires that the second ( spectator ) component plays no role at the transition , i.e. , an effective decoupling of the critical modes of the condensing species from the modes of the spectator one . this hypothesis is quite natural when the spectator species is in the normal phase , which is characterized by short - range correlations . however , if the second component is in the superfluid phase , in which long - range spin - wave ( goldstone ) modes develop , the asymptotic decoupling of the two species is no longer obvious . indeed , the goldstone modes may give rise to effective long - range interactions among the condensing particles . as a consequence , one might observe a different critical behavior ( we recall that the standard 3d xy universality is observed only if the interactions decay sufficiently fast with the distance ) , or xy behavior with peculiar slowly - decaying scaling corrections , as it occurs in the case of mixtures of identical bosonic gases @xcite . to investigate these issues , we perform quantum monte carlo ( qmc ) simulations of the 2bh model in the hc limit for both species ( @xmath97 ) . our results provide a robust evidence that the critical behaviors belong to the 3d xy universality class along all continuous transition lines , including those where one species condenses in the superfluid background of the other one . in other words , the local inter - species density - density interaction is an irrelevant rg perturbation at all bec transition lines . there is also no evidence of slowly - decaying scaling corrections . apparently , the leading scaling corrections are always controlled by the leading irrelevant rg operator which appears in the standard xy model or at the bec transition of a one - component bosonic gas . as we shall see in sec . [ multi ] , these features change when we approach a mcp , where several transition lines meet . in that case the competition of the two condensing order parameters gives rise to more important effects . we consider the 2bh model ( [ sbh ] ) with @xmath98 in the hc limit @xmath99 , for cubic @xmath100 lattices with periodic boundary conditions . we perform qmc simulations @xcite for @xmath63 and two values of @xmath27 , @xmath43 and @xmath44 , using the same algorithm employed in ref . @xcite ( we refer to this reference for technical details ) . in the mf approximation , the phase behavior as a function of @xmath45 and @xmath25 is reported in fig . [ fig : tpos_hc ] . here , we verify that the mf diagram is qualititavely correct . moreover , we determine the nature of the critical behavior at a few selected points , at which the spectator species is both in the normal and in the superfluid phase . for this purpose , we focus on the finite - size scaling ( fss ) behavior of the helicity modulus , which generally provides the most precise numerical results to characterize the critical behavior . the helicity modulus @xmath101 of species @xmath102 is obtained from the response of species @xmath7 to a twist in the boundary conditions by an angle @xmath103 , i.e. , @xmath104 where @xmath105 is the partition function for twisted boundary conditions in one direction and periodic boundary conditions in the two orthogonal directions . in qmc simulations @xmath101 is simply related to the linear winding number @xmath106 of species @xmath7 , through the relation @xmath107 . we also computed expectation values of other observables , such as the two - point function @xmath108 , its spatial integral , and the second - moment correlation length . in the following we do not report the corresponding results . we only mention that they substantially confirm the conclusions drawn from the analysis of the helicity modulus . the helicity modulus at the bec transition of the @xmath7-species is expected to behave as @xmath109 where @xmath110 is the correlation - length exponent , and the linear scaling field @xmath111 is a linear combination of @xmath45 and of the model parameters , which vanishes at the critical point . assuming for simplicity that @xmath26 is fixed , at a generic critical point @xmath112 , the scaling field can be written as @xmath113 where @xmath114 and @xmath115 are nonuniversal coefficients . thus , if we fix @xmath25 to its critical value , that is we set @xmath116 , and we investigate the transition by varying @xmath45 , we have simply @xmath117 . if instead @xmath45 is fixed to its critical value , we have @xmath118 . the scaling function @xmath119 is universal , provided that coefficients @xmath114 and @xmath115 appearing in eq . ( [ usca ] ) are properly defined . however , it depends on the shape and boundary conditions of the system . a straightforward consequence of eq . ( [ eq : rscaling ] ) is that the curves @xmath120 at fixed @xmath121 cross each other at the critical point , where their slopes are controlled by the correlation - length exponent @xmath110 . we wish now to verify our conjecture that the critical behavior of the condensing component always belongs to the 3d xy universality class , irrespective of the ( normal or superfluid ) state of the spectator component . if true , the exponent @xmath110 appearing in eq . ( [ eq : rscaling ] ) equals that of the 3d xy universality class @xcite @xmath122 . moreover , also the scaling function @xmath119 must be equal to that of the 3d xy universality class apart from a trivial multiplicative rescaling of the argument . in particular , we should find @xcite @xmath123 at the transition . we should note that an accurate determination of the critical parameters requires us to take into account the corrections to the asymptotic scaling behavior ( [ eq : rscaling ] ) . including the leading corrections we have @xmath124 where @xmath125 is a scaling function and @xmath126 a universal exponent . for the standard 3d xy universality class numerical simulations give @xcite @xmath127 . at @xmath128 , @xmath63 , @xmath43 , as a function of @xmath25 , close to the normal - to - superfluid transition . ] for the two gases at the normal - to - superfluid transition of gas 1 , for @xmath128 , @xmath44 , @xmath63 , and @xmath129 . for @xmath130 , @xmath131 approaches the xy critical value @xmath132 , while @xmath133 increases linearly with @xmath121 as it is appropriate for a gas in the superfluid phase . ] ( we plot the same data reported in fig . [ u10a ] ) versus @xmath134 , @xmath135 , taking @xmath110 equal to the xy value @xmath136 . we use @xmath137 , obtained by fitting the data to eq . ( [ eq : rscaling_taylor ] ) . ] in the mf approximation , for @xmath63 , @xmath43 , and sufficiently small temperature values , the system undergoes three different phase transitions as @xmath25 is decreased at fixed @xmath45 , see fig . [ fig : tpos_hc ] . ( i ) first , starting from large values of @xmath25 , component 1 undergoes a normal - to - superfluid transition , while component 2 remains in the normal phase . ( ii ) as @xmath25 is further decreased , also component 2 condenses , while component 1 is in the superfluid phase . ( iii ) finally , a second normal - to - superfluid transition of component 1 occurs , but in this case the second component is superfluid . figs . [ u10a ] , [ u10b ] and [ u10c ] show qmc results at @xmath128 for the transition ( iii ) at @xmath95 , up to @xmath138 . the estimates of @xmath131 show a crossing point at @xmath139 , see fig . [ u10a ] , indicating that component 1 undergoes a phase transition , while @xmath133 increases linearly with @xmath121 , see fig . [ u10b ] , which is the appropriate behavior expected for a gas in the superfluid phase . in order to check that the critical behavior belongs to the 3d xy universality class , we verify that the data are consistent with eq . ( [ eq : rscaling2 ] ) , taking the xy values for the critical exponents . in practice , we fit the data to @xmath140 with @xmath141 . we set @xmath136 and @xmath142 , which are the best available estimates of the two exponents for the 3d xy universality class . for our data @xmath143 is small , so that we have replaced the scaling functions @xmath119 and @xmath125 with their expansions ( to order @xmath144 and @xmath145 , respectively ) around @xmath146 . the values of @xmath144 and @xmath145 have been chosen by checking the quality of the fit and the stability of the results with respect to the order of the expansions . around @xmath147 , good fits are obtained by taking @xmath148 or 2 and @xmath149 . correspondingly , we estimate @xmath150 . the quality of this xy - biased fit can be checked by plotting @xmath151 versus @xmath152 with @xmath135 , see fig . we observe a good collapse of the data , confirming the xy nature of the transition . this fss analysis confirms the conjecture that the critical behaviors along the normal - to - superfluid transition lines of a single species belong to the 3d xy universality class , even when the other species is in the superfluid phase . moreover , corrections to scaling always decay as @xmath153 , where @xmath154 is the leading irrelevant exponent for the xy universality class . therefore , the interactions between the critical and the noncritical component give rise to corrections that are quite suppressed , decaying at least as fast as @xmath153 . ) in the hc limit , for @xmath63 and @xmath44 , as a function of @xmath25 and @xmath45 . transitions where component 1 condenses are labelled with squares , those where component 2 condenses with circles . the interpolating lines are only meant to guide the eye . the same phase diagram , computed in the mf approximation , is shown in fig . [ fig : tpos_hc ] . ] and @xmath133 versus @xmath155 at @xmath63 , @xmath44 and @xmath156 . two different crossing points are visible , providing evidence for two distinct transitions . ] and the corresponding critical values in the @xmath45-@xmath25 plane are indicated in the labels . we show data at @xmath156 ( at the two transitions considered in fig . [ fig : mua2_decoupling ] ) , @xmath157 ( at the normal - to - superfluid transitions of both gases ) , and at @xmath128 at the two transitions of gas 1 driven by @xmath25 . for comparison , we also report results for the bec transition of the single - species bose - hubbard model at @xmath158 and in the hc limit . the data are plotted versus @xmath134 , where @xmath159 , @xmath160 for the transitions at fixed @xmath25 , and @xmath161 for those at fixed @xmath45 , see eq . ( [ usca ] ) . the constants @xmath114 or @xmath115 ( they assume different values at each transition ) are optimized to obtain the best collapse of the data . ] for @xmath63 and @xmath44 we have repeated the fss analysis of sec . [ nrcb ] for other values of the model parameters , with the purpose of determining an approximate phase diagram , to be compared with that obtained in the mf approximation , see fig . [ fig : tpos_hc ] . our numerical results are in qualitative agreement with the mf predictions , confirming the presence of two mcps where four transition lines meet , see fig . [ 3dphdiau4 ] . to verify the predicted behavior we have performed simulations at different values of @xmath25 , @xmath162 , 2 , and 6 , varying the temperature @xmath45 . in fig . [ fig : mua2_decoupling ] we show the helicity modulus of the two gases at @xmath163 as a function of @xmath164 . we observe that @xmath131 and @xmath133 cross at two different values of @xmath45 , indicating the presence of two separate ( but close ) normal - to - superfluid transitions . if we move down from the high - temperature phase decreasing @xmath45 , we first observe the condensation of gas 1 at @xmath165 and then that of gas 2 at @xmath166 . two different transitions are also observed at @xmath167 . here , however , the order is reversed . decreasing the temperature , first gas 2 condenses at @xmath168 , then gas 1 condenses at @xmath169 . at @xmath162 we have observed only one transition , related to component 2 . we have also considered a different line in the phase diagram , keeping the temperature fixed , @xmath170 , and varying @xmath25 . in this case we observe two normal - to - superfluid transitions of the same component , gas 1 , at the boundaries of the superfluid phase @xmath37 , at @xmath171 and @xmath172 . fss analyses analogous to those described in sec . [ nrcb ] confirm that all transitions belong to the 3d xy universality class . as a further check , in fig . [ fig : collapse_all_transitions ] we show the helicity modulus close to the transitions we have investigated , as a function of @xmath173 . by tuning appropriately the constants @xmath114 or @xmath115 , cf . ( [ usca ] ) , at each transition ( if our data are obtained at fixed @xmath174 we optimize the constant @xmath114 , while , if data are obtained at fixed @xmath175 , we optimize @xmath115 ) we obtain a perfect collapse of the data , confirming the universality of the scaling function @xmath119 defined in eq . ( [ eq : rscaling ] ) . we also report the helicity modulus at the bec transition of a single hc bose - hubbard gas . results fall on top of those obtained for the mixture , confirming the xy nature of the transition . as already anticipated by the mf computations , the phase diagram reported in fig . [ 3dphdiau4 ] has two mcps , where four transition lines meet . at each mcp , both gases simultaneously condense . their locations can be inferred from the numerical results of ref . @xcite , where the finite - temperature bec transitions of the 2bh model for two equal bosonic gases were studied . in particular , when @xmath176 , the two identical gases condense at @xmath177 for @xmath44 , and at @xmath178 for @xmath179 . using the particle - hole relation ( [ transf ] ) , the latter transition implies an analogous transition at @xmath180 , @xmath63 , @xmath44 , and @xmath178 . clearly , the two transitions at @xmath44 , @xmath93 , @xmath63 , @xmath181 and @xmath44 , @xmath180 , @xmath63 , @xmath182 , must correspond to the mcps of the phase diagram reported in fig . [ 3dphdiau4 ] , since they are characterized by the simultaneous bec of both gases . as shown in ref . @xcite , the critical behavior of the transition of two equal hc bosonic gases with on - site inter - species density - density interaction is controlled by a decoupled xy fp . however , in this case the competition of the two u(1 ) order parameters leads to unusual slowly - decaying scaling corrections . indeed , the inter - species density - density interaction gives rise to scaling corrections that decay very slowly , as @xmath183 , where @xmath184 is the diverging length scale at the transition . such scaling corrections are not present in standard transitions belonging to the xy universality class , where they decay as @xmath185 with @xmath186 . the competition of distinct types of order gives generally rise to multicritical phenomena . more specifically , a multicritical point ( mcp ) is observed at the intersection of two critical lines characterized by different order parameters . multicritical behaviors occur in several physical contexts : in anisotropic antiferromagnets , high-@xmath187 superconductors , multicomponent polymer solutions , disordered systems , etc . , see , e.g. , refs . the scaling behavior at a mcp is controlled by the stable fixed point ( fp ) of the rg flow , which can be studied by field - theoretical approaches based on the appropriate lgw @xmath1 theory . the lgw theory describing the competition of the two different u(1 ) order parameters is obtained by constructing the most general @xmath1 theory of two complex fields @xmath188 , which is invariant under independent u(1 ) transformations acting on each of them . its hamiltonian is @xmath189 , \nonumber\end{aligned}\ ] ] with two quadratic parameters @xmath190 and @xmath191 , and three quartic parameters @xmath192 , @xmath193 , and @xmath111 . a multicritical behavior is obtained by tuning the quadratic parameters @xmath190 and @xmath191 simultaneously to their critical values , keeping the quartic parameters fixed . note that the theory is well defined ( the quartic potential is bounded from below ) for @xmath194 , @xmath195 , and @xmath196 . mean - field calculations show that the u(1)@xmath0u(1 ) lgw theory ( [ mcphi4s ] ) leads to two different phase diagrams @xcite , see fig . [ mcpd ] , depending on the sign of @xmath197 . if @xmath198 , four critical lines meet at the mcp ( tetracritical behavior ) , as in the left panel of fig . [ mcpd ] , while , if @xmath199 , two critical lines and one first - order line ( bicritical behavior ) are present , see the right panel of fig . note that , in the hc limit , the 2bh model should correspond to the lgw theory with @xmath198 , because of the correspondence @xmath200 and @xmath201 . therefore , we expect a tetracritical behavior . a bicritical behavior is expected instead in the opposite limit @xmath202 . the mf results presented in sec . [ phmf ] are completely consistent with this prediction . in the lgw theory the transition lines appearing in fig . [ mcpd ] are obtained by tuning one of the two quadratic parameters @xmath190 and @xmath191 to its critical value . multicritical behaviors arise when both of them are tuned to criticality . therefore , generic multicritical behaviors are associated with two relevant scaling fields @xmath203 and @xmath204 ( analytic functions of the model parameters such that @xmath205 at the mcp ) with positive rg dimensions @xmath206 and @xmath207 . for example , in the case of the 2bh model ( [ sbh ] ) @xmath203 and @xmath204 may be taken as linear combinations of the temperature and of the chemical potentials of the two gases . in the absence of external fields , the singular part of the free - energy density is expected to obey the scaling law @xmath208 where @xmath115 is an arbitrary _ blocking _ variable . here , we have introduced additional irrelevant scaling fields @xmath209 , @xmath210 with rg dimensions @xmath211 , that give rise to scaling corrections at the critical point . neglecting their contribution and appropriately fixing the arbitrary variable @xmath115 as @xmath212 , we obtain the asymptotic scaling expression @xmath213 where @xmath214 are universal scaling functions , which depend on the sign of @xmath203 : @xmath215 should be considered for @xmath216 , @xmath217 in the opposite case . close to the mcp , all transition lines correspond to constant values of the product @xmath218 . within the lgw theory , a standard multicritical behavior can only be observed if there exists a stable fp for the corresponding rg flow and the system is in its attraction domain . in the opposite case , the flow generically runs to infinity and the transition is discontinuous . note that this can also occur if a stable fp exists , but the system is outside its attraction domain . we should note that the unstable fps of the theory are also associated to multicritical behaviors . however , in this case one should perform additional tunings of the parameters and correspondingly introduce additional relevant scaling fields . for example , consider a fp that is unstable with respect to one of the rg quartic perturbations , i.e. , such that the flow ends at the fp only if one performs one additional tuning of the initial parameters . this means that there is an additional relevant scaling field . ( [ freeenb ] ) still holds , but now @xmath219 . therefore , the contribution of @xmath220 can no longer be neglected approaching the critical point . from a more phenomenological point of view , this higher - order multicritical behavior can be observed by varying three model parameters . in the corresponding parameter space , one has surfaces of standard critical transitions . these surfaces intersect along lines that correspond to standard multicritical behavior . the transition may be continuous , controlled by the stable fp , or of first - order , if the rg flow goes to infinity . the higher - order multicritical behavior is observed at the intersection of the multicritical lines . of course , such points are quite difficult to observe in practice . the critical behavior at a continuous transition is controlled by the fps of the rg flow , which are determined by the common zeroes of the @xmath221 functions associated with the parameters appearing in the quartic potential . the presence of a stable fp controls the universal features of the critical behavior if the transition is continuous . if no stable fp exists , the generic transition is expected to be of first order . the @xmath221 functions of the theory can be computed using perturbation theory . in the calculation one should be careful to tune @xmath190 and @xmath191 to their critical value to obtain the critical theory . this requirement is automatically satisfied if one considers the @xmath222 expansion , which is based on dimensional regularization around four dimensions @xcite . indeed , in this regularization scheme , one considers directly the massless critical theory . the same is true in the related 3d scheme of ref . @xcite , the so - called @xmath223 scheme without @xmath222 expansion . here , one also considers the @xmath223 perturbative series , but does not expand in powers of @xmath222 , setting @xmath224 . the hamiltonian fields and parameters are renormalized @xcite by setting @xmath225 , @xmath226 , @xmath227 , where @xmath228 are the @xmath223 renormalized quartic couplings . the five renormalization functions @xmath229 and @xmath230 are normalized so that @xmath231 , @xmath232 and @xmath233 at tree level . here @xmath234 is a @xmath235-dependent constant given by @xmath236 . the @xmath223 @xmath221 functions are obtained by differentiating the renormalized couplings with respect to the scale @xmath237 , keeping the bare couplings @xmath238 , @xmath193 , and @xmath111 fixed . the two - loop @xmath221 functions associated with the quartic couplings are @xmath239 the complete five - loop series are reported in app . [ 5loop ] . the zeroes of the @xmath221 functions provide the location of the fps of the rg flow . their stability is controlled by the matrix @xmath240 [ the indices correspond to the three quartic couplings @xmath241 evaluated at the given fp . the fp is stable , if all eigenvalues @xmath242 of the stability matrix have positive real part . we first determine the fps and their stability properties close to four dimensions , using the first few terms of the standard @xmath222 expansion @xcite . we find six different fps , of which the only stable one is located at @xmath243 in the general o(@xmath244)@xmath0o(@xmath245 ) model this fp is named biconical fp ( bfp ) @xcite , it generally satisfies @xmath246 . in the u(1)@xmath0u(1 ) case , however , @xmath247 and therefore this fp also appears in the theory with @xmath248 and @xmath249 , @xmath250 , \nonumber\end{aligned}\ ] ] which is symmetric under the larger symmetry group @xmath251 $ ] . note that this fp is degenerate with the o(4 ) fp at leading order in @xmath222 . as a consequence , @xmath145-loop calculations at this fp provide results to @xmath252 only . thus , the available five - loop series reported in app . [ 5loop ] allow us to determine the location of the fp only to order @xmath253 . for the same reason the smallest eigenvalue of the stability matrix is of order @xmath254 : @xmath255 . the u(1)@xmath0u(1 ) lgw theory reduces itself to the o(4)-symmetric @xmath1 theory when @xmath256 and @xmath257 . correspondingly , the rg flow has an o(4)-symmetric fp at @xmath258 this fp is unstable in the full theory ( [ mcphi4s ] ) , since one eigenvalue of the stability matrix is negative , @xmath259 . therefore , it corresponds to a higher - order multicritical behavior with three relevant scaling fields , of rg scaling dimensions @xmath260 , @xmath261 , and @xmath262 . the lgw theory decouples into two identical u(1 ) @xmath1 theories when @xmath263 . we can therefore identify a decoupled fp ( dfp ) with @xmath264 . at the dfp , the two u(1 ) order parameters are decoupled , with a critical behavior belonging to the xy universality class . the dfp is located at @xmath265 the dfp is stable within each u(1 ) theory . therefore , its stability properties in model ( [ mcphi4s ] ) depend only on the rg dimension @xmath266 of the coupling @xmath111 associated with the quartic term @xmath267 that couples the two fields . the rg dimension @xmath266 can be evaluated using general scaling arguments @xcite . at the dfp , the operator @xmath267 scales as the product of two energy - like operators of the @xmath235-dimensional xy universality class . therefore , the rg dimension @xmath266 is given by @xmath268 using @xmath269 , we obtain @xmath270 . therefore , the dfp is unstable close to four dimensions . the other three fps also have @xmath264 . their stability matrix has two or three negative eigenvalues , and hence they can only be observed by tuning four or five different system parameters . they are of little relevance for interacting bose gases . the above calculations can be straightforwardly extended to @xmath271 [ @xmath272 in the case of the stable fp ] using the complete series reported in app . [ 5loop ] . however , methods based on the @xmath222 expansion allow us to find only those 3d fps which can be related , by analytic continuation , to those present close to four dimensions . but new fps may emerge in three dimensions , which can not be detected by the @xmath222 expansion , because they do not have a 4d counterpart . this means that the @xmath222 expansion may not provide the correct description of the 3d rg flow . for example , this occurs for the ginzburg - landau model , in which a complex scalar field is coupled to a gauge field , which is appropriate to describe superconductors and the nematic smectic - a transition in liquid crystals @xcite . although @xmath222-expansion calculations do not find a stable fp @xcite therefore , they predict a first - order transition numerical analyses show that these systems can also undergo continuous transitions in three dimensions , see , e.g. , refs . this implies the presence of a stable fp in the 3d ginzburg - landau theory , in agreement with experiments in liquid crystals @xcite . other examples are provided by the o(2)@xmath273o(@xmath274 ) lgw @xmath1 theories describing frustrated spin models with noncollinear order @xcite , the @xmath275he superfluid transition from the normal to the planar phase @xcite , etc ... therefore , a more conclusive analysis of the rg flow in three dimensions requires a direct 3d study . we now extend the analysis to the 3d case . since the @xmath222 expansion suggests that the relevant fps belong to the plane @xmath276 , we first consider the 3d fps that appear in the lgw theory ( [ phi4s ] ) , and discuss their stability in the multicritical theory in which the exchange symmetry is broken . the analysis of the fps for model ( [ phi4s ] ) is reported in ref . two stable fps are identified : the dfp and a second fp , named asymmetric fp ( afp ) . the dfp controls the transitions at which two identical gases condense simultaneously . the afp , instead , is the relevant fp for transitions at which only one gas undergoes bec , breaking the exchange symmetry of model ( [ phi4s ] ) . let us now discuss the stability of these two fp within the multicritical theory ( [ mcphi4s ] ) , starting with the dfp . for @xmath277 the model corresponds to two noninteracting u(1 ) systems . we thus obtain for the rg dimensions of the quadratic operators @xmath278 . the two quartic perturbations that are present for @xmath279 can be identified with the quartic perturbation of the standard u(1 ) theory , so that @xmath280 . the rg dimension of the perturbation coupling the two xy models can be computed as in the 4d case , using eq . ( [ yug ] ) and @xmath122 @xcite . we obtain @xmath281 which is also negative , confirming the stability of the dfp in three dimensions , at variance with the behavior close to four dimensions . note , however , as already discussed in ref . @xcite , that @xmath266 is quite small . thus , it gives rise to very slowly decaying corrections , that are quite difficult to detect . since @xmath282 , we have @xmath283 . thus , the dfp should be relevant for systems that have a tetracritical mcp , see fig . the asymptotic decoupling of the critical modes allows us to simplify eq . ( [ freeenb ] ) . neglecting scaling corrections , i.e. setting @xmath284 for @xmath285 , we can simply rewrite @xmath286 where @xmath287 are constants . by appropriately choosing @xmath203 and @xmath204 , we can additionally set @xmath288 , so that @xmath289 . the scaling fields are defined by the requirement that the transition lines correspond to @xmath290 or @xmath291 . the second stable fp of the reduced theory ( [ phi4s ] ) is the afp that controls the critical behavior when only one of two bosonic species condenses @xcite . it also appears in the o(2)@xmath273o(2 ) lgw theory describing the critical modes of some frustrated spin models with noncollinear order @xcite . within the multicritical theory ( [ mcphi4s ] ) , the afp should describe bicritical points ( right panel of fig . [ mcpd ] ) , essentially because it is associated with the bec of only one species . the rg dimensions of the two relevant perturbations at the afp correspond to the dimensions of the quadratic operators . the rg dimension @xmath206 is obtained from the relation @xmath292 , where @xmath110 is the correlation - length exponent of the lgw theory ( [ phi4s ] ) . the analysis of the perturbative expansions within two different renormalization schemes @xcite ( the so - called @xmath293 mzm and @xmath223 schemes ) gives @xmath294 and @xmath295 , respectively , so that @xmath296 . the rg dimension @xmath207 can be derived from the results reported in ref . @xcite for the quadratic perturbations to the fps of the o(2)@xmath273o(2 ) theory . we obtain @xmath297 and @xmath298 in the two different perturbative schemes . the stability properties of the afp within the multicritical theory ( [ mcphi4s ] ) depend also on the rg dimension @xmath299 of the quartic coupling @xmath300 associated with the operator @xmath301 breaking the exchange symmetry . the rg dimension @xmath299 can be computed by a perturbative analysis in the @xmath223 scheme without @xmath222 expansion . we obtain the estimate @xmath302 , where the error takes into account how the estimate changes as the fp position varies within one error bar ( we use the fp estimates of ref . @xcite ) and the dependence on the resummation parameters . since @xmath303 , the quartic perturbation @xmath304 is irrelevant and therefore the afp is stable in the multicritical theory . note that the afp is not connected with the stable biconical fp found close to 4d . indeed , they describe different symmetry - breaking patterns . the stable fp close to 4d corresponds to a tetracritical mcp , in contrast with the afp , which gives rise to a bicritical behavior . apparently , the biconical fp disappears when approaching 3d , while the afp , which is absent close to four dimensions , appears only close to three dimensions @xcite . it is also interesting to discuss the o(4 ) fp , which is already unstable in the reduced theory ( [ phi4s ] ) . a complete analysis shows that there are three different relevant scaling fields at the o(4 ) fp in the full theory ( [ mcphi4s ] ) . summarizing , the rg dimensions of the relevant scaling fields are @xmath305 associated with the scalar quadratic perturbation at the o(4 ) fp ( obtained using @xmath306 @xcite ) , @xmath307 associated with the spin-2 quadratic perturbation @xcite , and @xmath308 associated with the spin-4 quartic perturbation @xcite . the analysis we have presented considers only the fps with @xmath309 . _ a priori _ other fps may be present with @xmath310 . as we shall see in the next section , the analysis of the general rg flow does not find any evidence of additional stable fps . in this section we study the rg flow . in particular , we determine the rg trajectories starting from the gaussian fp , where the quartic couplings vanish . this study allows us to determine the stable fps and their attraction domain in the space of the hamiltonian ( bare ) quartic parameters . we use here the @xmath223 scheme without @xmath222 expansion @xcite , which provides a genuine 3d critical scheme . the rg trajectories are obtained by solving the differential equations @xmath311 , \label{betaf}\\ & & - \lambda { dv_{sr } \over d\lambda } = \beta_{v_s}[u_r(\lambda),v_{sr}(\lambda ) ] , \nonumber \end{aligned}\ ] ] where @xmath102 , @xmath312 , with the initial conditions @xmath313 note that the trajectories do not depend on the hamiltonian parameters individually , but only through their dimensionless ratios . for this purpose , we rescale @xmath314 , so that the initial conditions become @xmath315 for @xmath316 , @xmath317 and @xmath318 for the derivatives of @xmath319 and @xmath320 , respectively . note that the lgw theory ( [ mcphi4s ] ) is stable for @xmath321 . one expects tetracritical or bicritical behavior if @xmath322 and @xmath323 , respectively . moreover , the symmetry of the model under @xmath324 , @xmath325 , allows us to restrict the analysis to @xmath326 . to obtain meaningful results , the perturbative series are resummed by employing the pad - borel technique , see , e.g. , refs . @xcite . typical results are shown in fig . [ fig : rgflow-3d ] , where we report a projection of the trajectories in the @xmath327 plane for @xmath328 ( results for other values of @xmath329 are qualitatively analogous ) . for @xmath330 slightly larger than @xmath331 ( see the behavior for @xmath332 in the figure ) , it is not possible to follow the flow beyond a certain value of @xmath333 , since the borel transform becomes singular on the positive real axis . these trajectories clearly correspond to systems that undergo discontinuous transitions . if we further increase @xmath330 , we observe that trajectories flow to the dfp , which is the stable fp relevant for small values of @xmath111 . if @xmath330 is increased again , the relevant fp changes and the trajectories end up at the afp . if @xmath330 is further increased , trajectories run into the non - borel summable region @xmath334 and @xmath335 ( see the behavior for @xmath336 in the figure ) . it is interesting to observe that the range of values of @xmath330 corresponding to trajectories flowing to the afp is quite small . for the approximant shown in the figure , we should have @xmath337 . for any @xmath338 the trajectories flow to infinity . this suggests that most of the bicritical mcps undergo first - order transitions . note that the numerical analysis does not provide any evidence of additional fps . apparently , all relevant fps belong to the symmetric model with @xmath339 . in this paper we study the critical and multicritical behaviors that can be observed in 3d mixtures of bosonic gases interacting by short - range density - density interactions . these systems have a global u(1)@xmath0u(1 ) symmetry , related to independent u(1 ) transformations acting on each species . as a representative of this class of systems , we consider the 3d bose - hubbard model for two lattice bosonic gases coupled by an on - site inter - species density - density interaction , whose hamiltonian is given in eq . ( [ sbh ] ) . however , the qualitative features of the finite - temperature phase diagram and the results for the universality classes associated with the critical and multicritical behaviors apply to generic bosonic mixtures . the generic features of the phase diagram of the 2bh model have been determined in the mf approximation and additionally confirmed by qmc simulations . the qualititave behavior depends on the model parameters , such as the chemical potentials and the on - site inter- and intra - species couplings . by varying them , one can observe several transition lines , along which one of the two species undergoes a normal - to - superfluid transition , and different types of multicritical behavior . the transition lines separating the different phases generally correspond to the bec condensation of one of the two species . we show that , independently whether the other species is in the normal or superfluid phase , the critical behavior of the condensing species belongs to the 3d xy universality class , characterized by the breaking of a global u(1 ) symmetry and short - ranged effective interactions , which is the same universality class associated with the bec of a single bosonic gas . therefore , the critical modes of the condensing gas effectively decouple from those of the other species , independently whether the latter is in the normal or superfluid phase . the phase diagram of mixtures of bosonic gases also presents particular points where some transition lines meet . at these points multicritical behaviors develop , due to the competition of the u(1 ) order parameters related to the two bosonic gases . we investigate them by a field - theoretical approach based on the effective lgw @xmath1 theory for two complex scalar fields with global u(1)@xmath0u(1 ) symmetry . the possible universality classes that describe the multicritical behaviors are associated with the stable fps of the rg flow . they can be determined by studying the rg trajectories in the critical theory , starting from the unstable gaussian fp in the quartic - parameter space . for this purpose , we consider the so - called @xmath223 scheme without @xmath222 expansion @xcite . we start from the five - loop @xmath340 @xmath221 functions , resum them using the pad - borel technique , and solve the flow equations . we find two stable fps , that also belong to the @xmath1 theory ( [ phi4s ] ) , which has an additional @xmath2 symmetry related to the exchange of the two order parameters . this more symmetric model has already been discussed in the context of the critical behavior of a mixture of two identical gases @xcite . if the system has a tetracritical continuous transition , see fig . [ mcpd ] , the critical behavior is controlled by a decoupled fp . each component shows an xy critical behavior correspondingly , the rg dimensions of the two relevant operators are @xmath341 but with very slowly - decaying scaling corrections ( they decay as @xmath342 , where @xmath184 is the correlation length ) due to inter - species coupling . if , instead , the system undergoes a bicritical continuous transition , the critical behavior is associated with a different asymmetric fp , with @xmath343 and @xmath344 . recent experiments on atomic gas mixtures @xcite , either using two different atomic species or the same atomic species in two different states , have already obtained several interesting results on the properties of the low - temperature condensed phase and on the interplay of the different condensates . they have also demonstrated the possibility of a robust control of the model parameters , which may allow the observation of the different phases , such as those found in the present study , and the determination of the nature of the critical and multicritical behaviors . our results should provide a complete characterization of the possible bec patterns and of the critical behaviors that these systems may develop along their transition lines . most cold - atom experiments have been performed in inhomogeneous conditions , due to the presence of space - dependent trapping forces , which effectively confine the atomic gas within a limited space region @xcite . the trapping potential is effectively coupled to the particle density , which may be taken into account by adding a further hamiltonian term to the 2bh hamiltonian ( [ sbh ] ) , i.e. , @xmath345 where @xmath65 is the space - dependent potential associated with the external force . the inhomogeneity arising from the trapping potential introduces an additional length scale @xmath346 into the problem , which drastically changes the general features of the behavior at the phase transitions . experimental data for inhomogeneous trapped cold - atom systems are usually analyzed using the local - density approximation , see , e.g. , ref . . however , this approach fails to describe the emergence of large - scale correlations @xcite . this problem may be overcome experimentally by using ( almost ) flat traps , giving rise to a finite space region where the system is effectively homogenous @xcite . otherwise , one may infer the critical behavior by studying the scaling behavior with respect to the trap size @xmath346 , which is expected to be universal and controlled by the critical exponents of the universality class of the corresponding homogenous system , in the large trap - size limit @xcite . we report here the five - loop perturbative series of the @xmath221 functions used in sec . [ 3drgflow ] to analyze the rg flow of the u(1)@xmath0u(1 ) lgw @xmath1 theory . we consider the perturbative expansions obtained in @xmath347 dimensions , using the dimensional regularization and in the modified minimal - subtraction ( @xmath340 ) scheme . they were computed in ref . @xcite for general o(@xmath244)@xmath0o(@xmath245 ) theories , but they have never been reported . apart from the first few orders , coefficients are reported with a @xmath348 numerical precision , although they are computed in terms of fractions and @xmath349 functions ( the exact series are available on request ) . to simplify the formulas , the renormalized couplings are named @xmath238 , @xmath193 and @xmath111 instead of @xmath350 , @xmath351 , and @xmath352 . the five - loop @xmath221 functions read @xmath353 k. m. mertes , j. w. merrill , r. carretero - gonzlez , d. j. frantzeskakis , p. g. kevrekidis , and d. s. hall , nonequilibrium dynamics and superfluid ring excitations in binary bose - einstein condensates , phys . 99 * , 190402 ( 2007 ) . s. tojo , y. taguchi , y. masuyama , t. hayashi , h. saito , and t. hirano , controlling phase separation of binary bose - einstein condensates via mixed - spin - channel feshbach resonance , phys . a * 82 * , 033609 ( 2010 ) . soltan - panahi , j. struck , p. hauke , a. bick , w. plenkers , g. meineke , c. becker , p. windpassinger , m. lewenstein , and k. sengstock , multi - component quantum gases in spin - dependent hexagonal lattices , nat . * 7 * , 434 ( 2011 ) . l. de forges de parny , f. hbert , v. g. rousseau , r. t. scalettar , and g. g. batrouni , ground state phase diagram of spin-1/2 bosons in a two - dimensional optical lattice , phys . rev . b * 84 * , 064529 ( 2011 ) . m. a. garca - march , b. juli - daz , g. e. astrakharchik , th . busch , j. boronat , and a. polls , quantum correlations and spatial localization in one - dimensional ultracold bosonic mixtures , new j. phys . * 16 * , 103004 ( 2014 ) m. campostrini , m. hasenbusch , a. pelissetto , and e. vicari , theoretical estimates of the critical exponents of the superfluid transition in @xmath3he by lattice methods , phys . b * 74 * , 144506 ( 2006 ) . the mf hamiltonian is in general an infinite - dimensional matrix so that , to obtain numerical results , one must fix a cutoff on the site occupancy . we fix both occupation numbers of the two species to be at most @xmath355 , so that becomes a @xmath356 real symmetric matrix . of course , the cut - off is arbitrary and may induce a bias in mf calculations . for this reason , the stability of the results must be checked a posteriori , repeating the computation for increasing values of @xmath355 . note that , in the hc limit , no truncation is needed and the mf hamiltonian is simply a @xmath357 matrix . zhang , a unified theory based on so(5 ) symmetry of superconductivity and antiferromagnetism , science * 275 * , 1089 ( 1997 ) ; s .- c . zhang , j .- hu , e. arrigoni , w. hanke , and a. auerbach , projected so(5 ) models , phys . b * 60 * , 13070 ( 1999 ) . m. hasenbusch , f. parisen toldin , a. pelissetto , and e. vicari , magnetic - glassy multicritical behavior of 3d @xmath359 ising model , phys . rev . b * 76 * , 184202 ( 2007 ) ; multicritical nishimori point in the phase diagram of the @xmath359 ising model on the square lattice , phys . rev . e * 77 * , 051115 ( 2008 ) . a. aharony , dependence of universal critical behaviour on symmetry and range of interaction . in : phase transitions and critical phenomena . vol . 6 , edited by c. domb and m.s . green , ( academic , new york , 1976 ) p. 357 . m. campostrini and e. vicari , critical behavior and scaling in trapped systems , phys . lett . * 102 * , 240601 ( 2009 ) ; ( e ) * 103 * , 269901 ( 2009 ) ; trap - size scaling in confined particle systems at quantum transitions , phys . a * 81 * , 023606 ( 2010 ) .

the hurwitz space is a compactification of the space of rational functions of a given degree . the lyashko - looijenga map assigns to a rational function the set of its critical values . it is known that the number of ramified coverings of by with prescribed ramification points and ramification types is related to the degree of the lyashko looijenga map on various strata of the hurwitz space . here we explain how the degree of the lyashko - looijenga map is related to the intersection theory on this space . we describe the cohomology algebra of the hurwitz space and prove several relations between the homology classes represented by various strata . _ _ [ cols= " > , < " , ]

math0303218

in a series of two papers , the present one and the second one by the second author @xcite , we continue the study of the hurwitz problem @xcite concerning counting ramified coverings of the @xmath1-sphere . roughly speaking , this problem can be formulated as follows : _ given a set of fixed ramification points on the target @xmath1-sphere and a set of ramification types over these points , count the number of non - isomorphic ramified coverings @xmath2 by a @xmath1-surface @xmath3 , having the prescribed ramification types over the prescribed ramification points_. two coverings are considered to be isomorphic if there is a homeomorphism of the covering surfaces taking the first covering to the second one . in fact , we count isomorphism classes with the weight equal to the inverse order of the group of automorphisms of the class ; in this form the problem is more natural , and admits a lot of applications . in a more modern setting this problem can be reformulated as the computation of the gromov witten invariants of the complex projective line . a detailed exposition of the history of the problem and a description of various approaches to its solution can be found in @xcite . hurwitz @xcite himself gave an explicit answer to the problem in the case , where the covering surface is also a sphere , and there is one ramification point with arbitrary ramification , while all others are points of simple ramification . in spite of the importance of the problem , only recently hurwitz s original results were extended to more general cases and treatable answers were obtained : * i. goulden and d. jackson @xcite solved the combinatorial problem that allows one to give the answer to hurwitz s problem for polynomials ( that is , ramified coverings of the sphere by a sphere having a point with a single preimage and arbitrary ramification over other points ) ; later their results were reestablished in @xcite and @xcite by absolutely different methods ; * t. ekedahl , s. lando , m. shapiro , and a. vainshtein @xcite gave an expression for the number of ramified coverings of the sphere by a surface of arbitrary genus with a single point of arbitrary ramification type and all other points of simple ramification , in terms of intersection indices on moduli spaces of curves ( when the covering surface is the sphere , the answer given by the formula coincides with hurwitz s one ) ; another proof of the formula was given in @xcite . here we address the special case of the hurwitz problem , where the covering surface is also a sphere , but in contrast to the hurwitz case , ramification points of arbitrary type are allowed . we are far from solving the problem in this generality , but we suggest a general approach , and show how it works in some special cases , thus producing new enumerative results . to a ramification point (= a critical value ) of multiplicity @xmath4 we assign a partition @xmath5 of @xmath4 , called the _ ramification type_. the elements of @xmath5 are the multiplicities of the critical points that correspond to our critical value . recall the hurwitz theorem . the number @xmath6 of degree @xmath7 coverings of the sphere by the sphere , having the ramification type @xmath8 over one point , @xmath9 , while all other ramification points are fixed and simple , is @xmath10 here @xmath11 is the order of the automorphism group of the partition @xmath5 , @xmath12 is the number of nondegenerate ramification points , and @xmath13 is the number of simple preimages of the multiple ramification point . as a consequence of our approach we obtain , in the second part of this paper @xcite , for example , the following result . the number @xmath14 of degree @xmath7 coverings of the sphere by the sphere , having two double critical points and @xmath15 simple critical points , all the @xmath16 critical values being fixed , is @xmath17 as far as we know , this formula , as well as similar formulas for some other ramification types , is new . first of all we reduce the problem to the calculation of the degree of some map , called the lyashko looijenga map . this step is now standard . the lyashko looijenga map ( below , the @xmath18 map ) takes a meromorphic function to the set of its critical values . its source space can be chosen in a variety of ways ; here we define the @xmath18 map on the hurwitz space @xmath19 constructed in @xcite . it is the space of stable maps from genus zero complex curves to the projective line , having trivial ramification over infinity . all spaces of functions possessing degenerate ramification are considered as subvarieties in this space . these subvarieties form a stratification of the hurwitz space . [ thmml ] the number @xmath20 of ramified coverings of the sphere , having the ramification types @xmath21 over prescribed ramification points with multiplicities @xmath22 , is given by the formula @xmath23 where @xmath24 is the degree of the @xmath18 map restricted to the stratum @xmath25 consisting of functions with these ramification types . this is an instance of a general situation , see e.g. @xcite . indeed , if we fix a ramified covering @xmath26 and choose a complex structure on the target sphere @xmath27 , then there exists a unique complex structure on the covering surface @xmath3 making the function @xmath28 into a meromorphic function . this complex structure is produced by the riemann construction . hence , there is a one - to - one correspondence between ramified coverings with fixed ramification points of given types and meromorphic functions with fixed critical values of the same types . the latter number is exactly the degree of the @xmath18 map on the corresponding moduli space . the coefficient @xmath29 results from the fact that in our construction of the space @xmath19 we choose a numbering of the @xmath7 poles of each rational function . the factor @xmath30 is due to the fact that if we permute the ramification types @xmath31 preserving the multiplicities @xmath22 , we obtain a new set of ramified coverings that lie on the same stratum @xmath25 and have the same image under the @xmath18 map . the space @xmath19 is , in fact , a vector bundle over the deligne mumford moduli space @xmath32 of stable genus zero curves . the latter space is a smooth projective variety . the multiplicative group @xmath33 of nonzero complex numbers acts on this bundle fiberwise . this action consists in just multiplying a meromorphic function by a constant . it preserves the stratification of the space @xmath19 because the multiplication does not change the type of singularities . deleting the zero section of the bundle and taking the quotient modulo this action we reduce the variety to the projectivization @xmath34 of the vector bundle @xmath19 . the projectivization @xmath34 carries the tautological line bundle @xmath35 , and a natural cohomology class @xmath36 , @xmath37 , the class of a hyperplane section . the degree of the lyashko looijenga map restricted to a subvariety in @xmath19 is related to the intersection index of the subvariety with the complementary power of the class @xmath38 . namely , the following statement is true . [ thm1.4 ] the degree @xmath24 of the @xmath18 map restricted to the stratum @xmath25 is equal to @xmath39 here @xmath4 is the dimension of the stratum @xmath40 , @xmath41 is the multiplicity of the @xmath42th ramification point ( or the sum of the elements of @xmath43 ) , and @xmath44 is the coupling between a homology and a cohomology class . this theorem is proved in section [ ssec : strathur ] . theorem [ thm1.4 ] implies that our approach requires the study of the cohomology ring @xmath45 . all cohomology groups we consider are with complex coefficients , and we do not specify the coefficients explicitly . we need * a reasonable description of this ring , say , in terms of generators and relations ; * reasonable expressions for the cohomology classes of the strata and the class @xmath38 in terms of the generators . the solution of the first problem is known since the space @xmath19 is a vector bundle over the moduli space of stable rational curves , whose cohomology is known . the second problem seems to be much more difficult . in fact , we need less than the whole cohomology ring . the symmetric group @xmath46 acts on the space @xmath19 by permuting the indices , whence it acts on the cohomology space @xmath45 . all the cohomology classes we are interested in are symmetric with respect to this action . therefore , we only need to know the @xmath46-symmetric part of the cohomology algebra @xmath45 . this is an additional problem , but its solution can lead to a simplification of the decompositions of the strata . the situation would become even easier if we restrict ourselves to the subalgebra in the cohomology algebra @xmath45 generated by the classes of the strata . consider two closed codimension 1 subvarieties @xmath47 and @xmath48 in @xmath19 ; the first of them is called the _ caustic _ and it is the closure of the space of functions having a critical point of order @xmath1 , the second one is the _ maxwell stratum _ and it is the closure of the space of functions having two distinct critical points with coinciding critical values . [ conj : kazarian ] the subalgebra of the cohomology algebra @xmath45 generated by the cohomology classes of the strata is generated by the two classes @xmath47 and @xmath48 in @xmath49 . up to now , the basis of this conjecture is not too solid . but it does not contradict our sample calculations , and it has a nonformal justification coming from kazarian s theory . kazarian s theory concerns cohomology classes of multisingulatities of a map @xmath50 of two complex manifolds , @xmath51 . its main statement claims that there is a universal way to express these cohomology classes in terms of the characteristic classes of the tangent bundles over the two manifolds @xmath52 and @xmath53 , more precisely in terms of the class @xmath54 , where @xmath55 is the total chern class . it is a development of the theory of the thom polynomial . up to now , only a preliminary version @xcite of the text describing the theory is available , and we are not going to refer directly to statements from it . however , the ideology of the theory seems to be applicable in the situation we are studying , and it leads to a number of conjectures concerning the part of the cohomology ring of the hurwitz spaces we are interested in . in our situation , the variety @xmath52 is the universal curve @xmath56 over the hurwitz space @xmath34 , while the variety @xmath53 is the quotient of @xmath57 by the natural @xmath33 action . the map @xmath28 is the universal map over @xmath34 . it is easy to see that the map @xmath28 almost identifies the tangent spaces @xmath58 and @xmath59 ; more precisely , one has @xmath60 , where @xmath61 and @xmath62 are first chern classes of some linear bundles . however , new complications arise , since we are interested in cohomology classes of @xmath34 and not of @xmath56 . kazarian s ideology leads to conjecture [ conj : kazarian ] . if the conjecture is true , then the computation of intersection numbers of the strata in the hurwitz space with the complementary powers of the class @xmath36 can be split into two independent stages : * express the cohomology class of the required stratum as a ( homogeneous ) polynomial in the classes @xmath48 and @xmath47 ; * find the intersection index of each monomial in @xmath48 and @xmath47 with the complementary degree of the class @xmath38 . another cohomology class in @xmath49 is also distinguished . this is the class @xmath63 dual to the subvariety of functions defined on singular curves . this class also can be expressed in terms of the classes @xmath47 and @xmath48 . this means that the subring in @xmath45 generated by the classes @xmath47 and @xmath48 coincides with that generated by @xmath38 and @xmath63 . the linear relations relating these four classes are as follows : @xmath64 the coefficients in these linear relations are polynomial in @xmath7 , which ( together with the informal support of kazarian s theory ) allows us to sharpen conjecture 1.5 in the following way : each cohomology class dual to a stratum in @xmath34 can be expressed as a homogeneous polynomial in @xmath65 whose coefficients that are polynomials in @xmath7 . therefore the calculations must even be simplified . in section 2 we give precise definitions of notions mentioned in this introduction . section 3 is devoted to the description of the lyashko looijenga map . in section 4 the structure of cohomology of the hurwitz spaces is analyzed . we are greatful to a. zorich who invited us both to rennes , where the main part of this work was accomplished . the work was completed during our participation in the mini - workshop on hurwitz theory and its ramifications , and we thank its organizers j. h. kwak and a. d. mednykh for inviting us . we also thank b. shapiro and m. shapiro for their interest and m. kazarian for explaining the basics of his theory . in this section we briefly recall the definitions of stable curves and stable maps ( we will restrict our attention to curves of genus 0 ) , as well as the definition of the hurwitz space @xmath19 from @xcite . consider the riemann sphere @xmath0 with @xmath66 distinct numbered marked points on it . two choices of marked points are considered equivalent if there exists an automorphism of @xmath0 sending one set of marked points to the other , preserving their numbering . the space of all nonequivalent choices of @xmath7 numbered marked points on @xmath0 can be endowed with a natural structure of a smooth irreducible ( noncompact ) complex manifold of dimension @xmath67 . we denote this moduli space by @xmath68 . [ exm3m4 ] the space @xmath69 is a point , because any triple of points on @xmath0 can be sent to any other triple by a biholomorphic map . similarly , @xmath70 is isomorphic to @xmath71 ( here we suppose that the projective line is endowed with an arbitrary complex coordinate ) . indeed , for any four marked points on @xmath0 one can send the first three to @xmath72 , @xmath73 , and @xmath74 . the fourth marked point will be sent to some uniquely determined point @xmath75 , different from @xmath72 , @xmath73 , and @xmath74 , and this value of @xmath75 is the point of @xmath70 . the spaces @xmath68 are compactified by adding to them new points corresponding to singular stable curves . all the curves under consideration are compact . the only singularities allowed are simple _ nodes _ , or _ double points _ ( as at the origin in the plane curve @xmath76 ) ; such curves are called _ nodal_. separating the two branches of a nodal curve at each node , we obtain a disjoint union of smooth compact curves , which is called the _ normalization _ of the original curve . the connected components of the normalization are the _ irreducible components _ of the curve . associate to a nodal curve a graph whose vertices correspond to the connected components of the normalization , and two vertices are connected by an edge iff the corresponding components intersect at a node ( loops appear if a component intersects itself ) . this graph is called the _ modular graph _ of the nodal curve . by definition , the genus of a nodal curve is zero if all connected components of its normalization are projective lines , and the modular graph is a tree . a _ stable genus 0 curve _ is a nodal curve of genus @xmath72 with @xmath7 distinct marked and numbered points satisfying the following conditions : ( i ) the marked points do not coincide with the nodal points ; ( ii ) the number of automorphisms of the curve preserving the marked points is finite . the second condition is equivalent to the following easily verifiable condition ( ii ) : the total number of marked and nodal points on every irreducible component of the nodal curve is greater than or equal to @xmath77 . the space of genus zero stable curves with @xmath7 marked points is endowed with a natural structure of a smooth irreducible compact complex manifold of dimension @xmath67 ; we denote this space by @xmath78 . the space @xmath68 of smooth curves form a dense subvariety in @xmath78 . apart from smooth curves , there are exactly @xmath77 stable genus 0 curves with @xmath79 marked points . they are shown in fig . [ figm4 ] . if @xmath75 from example [ exm3m4 ] tends to @xmath72 ( respectively , @xmath73 or @xmath74 ) , the corresponding smooth curve tends to the 3rd ( respectively 2nd , or 1st ) stable curve in the figure . accordingly , @xmath80 is obtained from @xmath70 by adding three initially punctured points . actually , @xmath80 is isomorphic to @xmath0 . now consider the space of meromorphic functions on @xmath0 with exactly @xmath81 numbered simple poles . this space has a structure of a smooth ( noncompact ) complex manifold of dimension @xmath82 . in the case when @xmath66 , if we forget the function itself and only remember the positions of its poles , we obtain a map from the space of functions onto @xmath68 . because of this , we will often say `` marked points '' instead of `` poles of @xmath28 '' . we are going to construct a compactification of the space of meromorphic functions following @xcite , similar to that of the moduli space @xmath68 . to do that , we will need to define meromorphic functions on nodal curves . a meromorphic function @xmath28 on a nodal curve @xmath3 is simply a meromorphic function defined on each irreducible component of the nodal curve . if two components intersect at a nodal point , the function must take the same value at this point on both components . two pairs @xmath83 and @xmath84 are equivalent if there exists a biholomorphic isomorphism @xmath85 that preserves the numbering of the poles and makes the following diagram commutative : ( 21,15)(0,0 ) ( 0,12)@xmath86 ( 16,12)@xmath87 ( 6.5,-2)@xmath0 ( 5,13.5)(1,0)9 ( 1.5,10.5)(3,-4)6 ( 17.5,10.5)(-3,-4)6 ( 8,16)@xmath88 ( .5,5)@xmath89 ( 16,5)@xmath90 a _ stable genus 0 meromorphic function _ is a function @xmath91 defined on a nodal curve @xmath3 and satisfying the following conditions . ( i ) the function @xmath28 does not have poles at nodal points . ( ii ) the number of automorphisms of the pair @xmath92 is finite . as above , a comment must be made about the second condition . an automorphism of the pair @xmath92 is a map @xmath88 from @xmath3 to itself that makes the following diagram commutative : ( 21,15)(0,0 ) ( 0,12)@xmath3 ( 16,12)@xmath3 ( 6.5,-2)@xmath0 ( 5,13.50)(1,0)9 ( 1.5,10.5)(3,-4)6 ( 17.5,10.5)(-3,-4)6 ( 8,16)@xmath88 ( .5,5)@xmath28 ( 16,5)@xmath28 condition ( ii ) is again easy to check , because it is equivalent to ( ii ) : if the function @xmath28 is constant on some irreducible component of @xmath3 , then the number of nodal points on this component is greater than or equal to @xmath77 . ( note that such a component can not contain marked points precisely because @xmath28 is constant on it . ) the _ hurwitz space _ @xmath19 ( for @xmath93 ) is the space of all stable genus 0 meromorphic functions with @xmath7 simple numbered poles . there is a natural action of @xmath33 on @xmath19 defined by simply multiplying a stable function by a constant . the invariant points of this action are stable functions such that the nodal curve has at most one pole on every component and the function vanishes on the components without poles . these points form the _ zero section _ of the hurwitz space . ( we will soon see that the hurwitz space is a vector bundle and the points in question indeed form its zero section . ) the _ projectivized hurwitz space _ @xmath34 is the quotient of @xmath19 , without the zero section , by @xmath33 . for @xmath66 , @xmath34 has the structure of a smooth compact complex manifold of dimension @xmath94 . @xmath95 is a compact smooth orbifold . @xmath95 is isomorphic ( as an orbifold ) to the weighted projective line with weights @xmath1 and @xmath73 , that is to @xmath96 indeed , if we are given a meromorphic map with 2 poles on @xmath0 , we can move its poles to @xmath72 and @xmath74 and write the function as @xmath97 . the numbers @xmath98 are considered up to ( simultaneous ) multiplication by a scalar factor . moreover , making a change of variables @xmath99 , we see that the function @xmath100 is equivalent to @xmath28 . thus @xmath101 $ ] is a set of homogeneous coordinates in @xmath95 . the point @xmath102 corresponds to the stable meromorphic function defined on a nodal curve with two irreducible components . the space @xmath103 , as will soon be proved , is isomorphic to @xmath104 . now we are going to prove that for @xmath66 , @xmath34 is the projectivization of an @xmath105-dimensional holomorphic vector bundle on @xmath78 . first let us define several line bundles on @xmath78 . consider a stable curve @xmath3 with @xmath7 marked points . for @xmath42 from @xmath73 to @xmath7 , let @xmath106 and @xmath107 be the complex line respectively cotangent and tangent to @xmath3 at the @xmath42th marked point . [ defl_i ] @xmath108 ( respectively @xmath109 ) is the line bundle over @xmath78 whose fiber over a point @xmath110 is the line @xmath106 ( respectively @xmath107 ) . denote by @xmath111 the trivial line bundle and let @xmath112 momentarily denote the following @xmath105-dimensional bundle over @xmath78 : @xmath113 [ proph_n = p(e_n ) ] for @xmath66 , we have @xmath114 . [ [ proof . ] ] proof . + + + + + + we will show that giving a stable meromorphic function on a nodal curve is equivalent to giving a point in the total space of the bundle @xmath112 . considering stable functions up to a scalar factor corresponds to taking the projectivization of @xmath112 . first of all , note that if a meromorphic function @xmath28 has a simple pole at some smooth point @xmath115 of a nodal curve @xmath3 , then it determines a tangent vector to @xmath3 at @xmath115 ; more precisely , the principal part of the pole is such a tangent vector . indeed , if we want to couple this tangent vector with a cotangent vector represented by a @xmath73-form @xmath116 , the result will be the residue of @xmath117 at @xmath115 . now , given a stable meromorphic function on a nodal curve , we can divide its poles into two parts . the poles of the first type lie on the irreducible components of the curve that contain at least @xmath77 marked and double points . the poles of the second type lie on the irreducible components that contain a unique pole and a unique nodal point . ( these are the only possible cases . ) if we forget the function itself and retain only the positions of its poles , we obtain a nodal curve with @xmath7 marked points . this curve is not necessarily stable . however , we can make it stable by contracting every component that contains only 1 nodal and 1 marked point . the former node becomes the marked point . thus we obtain a map from @xmath19 to @xmath78 . further , consider the stable curve that we have obtained . we assign to its marked points of the first type the principal parts of the corresponding poles ( they are tangent vectors at the marked points ) . we assign zero tangent vectors to the marked points of the second type . finally , we assign to the stable meromorphic function a complex number : the sum of its critical values given by the lyashko - looijenga map ( see section [ ssec : ll ] below ) . thus we have assigned to the stable meromorphic function a point in the total space of @xmath112 . conversely , given a point in the total space of @xmath112 , we can easily recover the stable meromorphic function . first we extend the stable curve by adding new irreducible components at each marked point of the second type ( those , whose tangent vector is equal to 0 ) . then , knowing the principal parts of the poles , we can recover the function @xmath28 uniquely up to an additive constant . knowing the sum of its critical values allows us to fix the constant . the vector bundle @xmath19 can be understood more conceptually . consider the space @xmath118 of all stable degree @xmath7 maps from genus @xmath72 curves @xmath3 to @xmath0 . inside this space there is a suborbifold canonically isomorphic to @xmath119 , where @xmath46 is the symmetric group acting on @xmath78 by renumbering the poles . this subvariety consists of stable maps of the following form . take a stable genus @xmath72 curve @xmath120 with @xmath7 marked but not numbered points . attach to @xmath120 a new spherical component at each marked point . instead of the former marked points , mark one point on each of the new spherical components . thus we have obtained a nodal curve @xmath3 . consider the map @xmath121 that sends the whole curve @xmath120 to @xmath72 and is of degree one on each of the new spherical components , the poles being at the new marked points . such stable maps @xmath92 form an orbifold isomorphic to @xmath119 . now , the normal bundle to @xmath119 in @xmath118 ( in the orbifold sense ) is canonically identified with the quotient @xmath122 . [ def : psi ] we denote by @xmath123 the canonical line bundle over @xmath34 and by @xmath124 the first chern class of the dual line bundle . consider the space of unordered sets of @xmath125 ( not necessarily distinct ) complex numbers . this space is isomorphic to the space of monic polynomials of degree @xmath125 : just take a set of @xmath125 numbers to the polynomial whose roots are these numbers . the _ lyashko - looijenga map _ @xmath18 is the map that assigns to a meromorphic function with @xmath7 simple poles on @xmath0 the ( unordered ) set of its @xmath82 critical values counted with multiplicities . we recall that a _ critical point _ of a function @xmath28 is a point , where @xmath126 , while a _ critical value _ of @xmath28 is its value at a critical point . the _ multiplicity of a critical point _ is the multiplicity of the zero of @xmath127 at this point . in other words , if the function @xmath28 has the form @xmath128 in some local coordinate @xmath115 , then @xmath129 is a critical point of multiplicity @xmath130 . the _ multiplicity of a critical value _ is the sum of multiplicities of the critical points at which the critical value in question is attained . if we consider the function as a ramified covering of @xmath0 , then its critical values are exactly those points in the image over which the covering is ramified . the notion of a _ cone _ ( or of a cone bundle ) generalizes that of a vector bundle to the case of singular fibers , see @xcite . the essential point is that cones carry a fiberwise action of the multiplicative group @xmath33 of nonzero complex numbers . a cone morphism preserves this action . the lyashko - looijenga map extends to a cone morphism @xmath131 the proof of the proposition is based on the extension of the notion of set of critical values to maps defined on nodal curves . first consider all the irreducible components of @xmath3 where @xmath28 is constant . consider the union of these irreducible components , and let @xmath132 be a connected component of this union . ( in general @xmath132 consists of several irreducible components intersecting at double points . ) suppose @xmath132 contains @xmath133 nodal points at which it intersects other irreducible components ( on which @xmath28 is not constant ) . then the value of @xmath28 on @xmath132 is considered a critical value of multiplicity @xmath134 . second , consider a nodal point lying on two irreducible components of @xmath3 such that @xmath28 is not constant on either of them . the value that @xmath28 takes at such a point is considered a double critical value . finally , the critical values of @xmath28 on all the irreducible components are taken into account as ordinary critical values . one can prove that in that way we have obtained a set of @xmath82 critical values , and that this set depends continuously on the point of @xmath19 . see @xcite for more details . consider a degree @xmath7 ramified covering @xmath26 of the sphere . let @xmath135 be a ramification point of @xmath28 of multiplicity @xmath4 . to this ramification point we can assign a partition @xmath5 of @xmath4 . this partition consists of the multiplicities of all critical points of @xmath28 in @xmath3 whose image under @xmath28 is equal to @xmath136 . we will write such a partition in the multiplicative form , @xmath137 , where @xmath138 , @xmath139 is the number of critical points in the preimage of @xmath136 having multiplicity @xmath42 , @xmath140 . if the covering has @xmath55 ramification points , @xmath141 , then @xmath55 partitions @xmath21 are associated to it . sets of partitions of @xmath7 that can appear as sets of ramification partitions of a ramified covering are subject to a number of constraints . if the covering surface @xmath3 also is the sphere , the riemann hurwitz formula implies that the total degeneracy of the set of ramification partitions of a meromorphic function must be @xmath82 , @xmath142 however , this property alone does not guarantee the existence of a ramified covering with the prescribed ramification type . [ defstrata ] the set of all rational functions on @xmath0 with @xmath7 simple numbered poles such that the set of partitions assigned to their critical values is @xmath143 is called an _ open stratum_. its closure in @xmath19 is called just a _ stratum _ and denoted by @xmath144 . the projectivizations of these strata ( lying in @xmath34 ) are defined in the obvious way . they will be denoted by @xmath145 . sometimes we will omit reference to simple partitions @xmath146 in the index , omitting in this case the braces as well . in the sequel we will often describe strata by specifying the set of functions that belong to their open parts and omitting the words `` closure of '' . the reader should bear in mind that we always consider closed strata , unless explicitly stated otherwise . [ excm ] there are two strata of complex codimension 1 . they are the caustic @xmath147 and the maxwell stratum @xmath148 . the _ caustic _ @xmath149 is ( the closure of ) the set of functions with a double critical point . the _ maxwell stratum _ @xmath150 is ( the closure of ) the set of functions with the same critical value taken at two distinct critical points . [ defmultiplicity ] the _ degree _ of the @xmath18 map on a stratum @xmath144 is the number of the preimages in @xmath144 of a generic point in @xmath151 . the degree is denoted by @xmath152 . since the space @xmath153 is a point , the space @xmath154 is isomorphic to @xmath155 , where the @xmath106 are the cotangent lines at the marked points on @xmath0 . it can be identified with the space of rational functions of the form @xmath156 in figure [ h3 ] we represent the stratification of the projectivized hurwitz space @xmath103 . the figure shows the real part of this space with a slight simplification : instead of a 3-dimensional projective space we have represented a 2-dimensional projective space , dropping the parameter @xmath4 . the interesting subvarieties in @xmath103 are the following ones . first , the projectivized caustic @xmath157 , which is a singular cubic with one self - intersection point . it has the homogeneous equation @xmath158 since our picture shows the real part of @xmath103 , the self - intersection point , which has the homogeneous coordinates @xmath159 , appears to lie apart from the rest of the cubic . second , the three projective lines @xmath160 , @xmath161 , @xmath162 , which correspond to reducible curves with one pole lying on one component and the other two poles on the other one . the maxwell stratum is empty . the cubic intersects each of the three lines at one point , but the intersection has multiplicity three . these three points lie on the line @xmath163 . the boxes around the picture of @xmath103 show , for each of the subvarieties , what are the corresponding stable rational functions . short traits represent poles , crosses represent critical points , while crosses with double lines represent double critical points . the numbers written near the arrows are the multiplicities of the @xmath18 map on the corresponding subvariety . now we are able to prove theorem [ thm1.4 ] from the introduction . we restate it here . consider a stratum @xmath164 ( see definitions [ defstrata ] and [ defmultiplicity ] ) determined by partitions @xmath31 . let @xmath165 be the multiplicities of the critical values . let @xmath166 be the number of automorphisms of the set @xmath167 , i.e. , the number of permutations @xmath168 of @xmath55 elements such that @xmath169 for all @xmath42 . [ thmmult ] the degree @xmath24 of the lyashko - looijenga map on @xmath25 equals @xmath170 where @xmath4 is the dimension of the stratum @xmath40 and @xmath44 is the coupling between a homology and a cohomology class . [ [ proof - of - theoremthmmult . ] ] proof of theorem [ thmmult ] . + + + + + + + + + + + + + + + + + + + + + + + + + + + to prove the theorem we consider a @xmath171-sheeted ramified covering @xmath172 of @xmath34 . by definition , a point of @xmath172 corresponds to a stable meromorphic function _ with numbered critical points_. more precisely , consider the zariski open subset @xmath173 of @xmath34 constituted by functions @xmath28 defined on @xmath0 , with @xmath82 simple critical points . over this open subset we can construct a @xmath171-sheeted non - ramified covering @xmath174 , whose points correspond to functions with numbered critical points . it is a connected complex manifold . then this covering is uniquely extended to a ramified covering of @xmath34 in the following way . for simplicity we start by describing the set of preimages of every point @xmath175 in the covering @xmath172 . consider an @xmath175 and a small ball @xmath176 surrounding @xmath177 . a point @xmath178 has @xmath171 preimages in the covering . the local monodromy group @xmath179 acts by permutations on these preimages . the preimages of @xmath177 in the covering are the orbits of this action . now let us give an algebro - geometric definition . consider an affine chart @xmath180 in @xmath34 and let @xmath181 be the algebra of algebraic functions on @xmath180 , @xmath182 . let @xmath183 , be a function whose set of zeroes is the ramification divisor . then the localization @xmath184 is the algebra of algebraic functions on the complement @xmath185 to the ramification divisor . the map @xmath186 is an injection , because @xmath181 has no zero divisors . let @xmath176 be the algebra of algebraic functions on the lifting @xmath187 of @xmath185 on @xmath174 . then @xmath188 is again an injection . the composition of the two injections gives an injection @xmath189 . we denote by @xmath190 the _ integral closure _ of @xmath181 in @xmath176 . by definition , the ramified covering over @xmath180 is then the variety @xmath191 . such ramified coverings of affine charts can then be glued into a covering of the total space @xmath34 . [ lem : normal ] the variety @xmath172 is normal . [ [ proof.-1 ] ] proof . + + + + + + recall that a normal algebraic variety is a variety such that each of its points possesses an affine neighborhood whose algebra of functions is integrally closed in its field of fractions ( see @xcite ) . on normal varieties chern classes of vector bundles and local intersection indices of subvarieties are well - defined . normality is a local property , therefore we can restrict our attention to an affine chart @xmath180 as above . let us prove that @xmath190 is integrally closed . first of all , @xmath176 is integrally closed , because it is an algebra of functions on a smooth variety . thus an element @xmath192 of the field of fractions of @xmath190 integral over @xmath190 belongs to @xmath176 . but @xmath190 is integrally closed in @xmath176 ( being the integral closure of @xmath181 ) , thus @xmath192 necessarily belongs to @xmath190 . in general , @xmath172 is not smooth . for example , the @xmath1-sheeted ramified covering of @xmath193 ramified over the axes @xmath194 and @xmath195 is the cone given by the equation @xmath196 . our description of the preimage of a point in @xmath34 under the projection @xmath197 follows from the fact that a normal complex algebraic variety is locally irreducible as a complex manifold ( @xcite , theorem 6.6 ) . we denote by @xmath198 the map @xmath199 that assigns to a meromorphic function with numbered critical points in @xmath172 the set of its numbered critical values . ( recall that both the function and the set of its critical values are considered up to a common scalar factor . ) denote by @xmath200 the tautological line bundle over the target space @xmath201 . recall that @xmath123 is the tautological line bundle over @xmath34 . denote by @xmath202 the pull - back of @xmath123 under the projection @xmath203 . then @xmath202 coincides with the pull - back of @xmath200 under @xmath198 . indeed , consider a point of the total space of the bundle @xmath202 , outside its zero section . it corresponds to a nonvanishing meromorphic function with numbered critical points ( and , this time , not considered up to a scalar factor ) . similarly , the total space of the bundle @xmath200 without the zero section is @xmath204 , and it corresponds to sets of @xmath82 numbered critical values , not all of which are equal to @xmath72 . it follows that the map @xmath198 can be lifted to the total spaces of the bundles @xmath202 and @xmath200 : it suffices to assign to a meromorphic function the set of its critical values . this lifting identifies the fibers of the two bundles . it follows that for any subvariety @xmath205 of pure dimension @xmath4 such that its image @xmath206 is irreducible and also of dimension @xmath4 , we have @xmath207 indeed , we have @xmath208 where @xmath209 is a generic projective subspace of dimension complementary to that of @xmath210 . now , @xmath211 is the intersection of @xmath4 projective hyperplanes . a hyperplane is a section of @xmath212 ( the line bundle dual to the tautological line bundle @xmath200 ) . let us count in two different ways the number of preimages in @xmath213 of the points of intersection between @xmath210 and @xmath211 . first , since each intersection point has @xmath214 preimages , the number of preimages equals @xmath215 second , the pull - back to @xmath172 of a generic hyperplane is a generic section of the line bundle @xmath216 dual to @xmath202 . taking the intersection between @xmath4 sections like that and the subvariety @xmath213 , we obtain @xmath217 hence , these two numbers are equal . the same equality holds if @xmath210 is not irreducible , but all generic points of @xmath210 have the same number of preimages . now let @xmath213 be the lifting to @xmath172 of a stratum @xmath40 in @xmath34 . denote by @xmath4 their common dimension . for a partition @xmath5 denote by @xmath218 the product of the factorials of its elements . it is easy to see that a generic point in @xmath40 has @xmath219 liftings to @xmath220 . further , the image of @xmath213 under the lifted lyashko - looijenga map @xmath198 is a union of @xmath221 projective subspaces of dimension @xmath4 . finally , denote by @xmath222 the degree of @xmath198 on @xmath223 , which is the number of preimages in @xmath223 of a generic point in the image of @xmath223 . simple combinatorial considerations show that @xmath224 the factor @xmath225 is the number of ways to number the critical points once the critical values are numbered . putting everything together we see that @xmath226 @xmath227 @xmath228 @xmath229 this proves the theorem . in this section we describe the cohomology algebra of the moduli spaces and of the hurwitz spaces . we also prove several cohomological identities which will be used in the subsequent paper @xcite to derive recurrence relations on hurwitz numbers . denote by @xmath232 an unordered partition of the set @xmath233 of marked points into two disjoint parts ( a partition @xmath234 coincides with @xmath235 ) . each part must contain at least @xmath1 points . consider all stable curves with exactly two irreducible components such that one component contains the marked points of the set @xmath181 and the other one of the set @xmath176 . such stable curves form a ( complex ) codimension 1 subvariety of @xmath78 . its closure is a smooth closed subvariety of @xmath78 of complex codimension 1 . therefore it represents a 2-cohomology class which we will denote by @xmath236 $ ] . the generators @xmath236 $ ] are not linearly independent . for example , for @xmath237 all the three generators corresponding to the partitions @xmath238 represent the same cohomology class , ( poincar dual to ) a point . more generally , let us fix four distinct numbers @xmath239 between @xmath73 and @xmath7 . denote by @xmath240 $ ] the sum of all the generators @xmath236\in h^2(\ocm_n)$ ] corresponding to partitions , where @xmath42 and @xmath241 belong to one of the two parts , while @xmath133 and @xmath242 belong to the other part . then the generators @xmath236 $ ] satisfy the relations @xmath243 = [ ik\d jl ] = [ il\d jk].\ ] ] these relations can be deduced using the forgetful map @xmath244 which forgets all the marked points except for @xmath239 and contracts all the components of the curve that have become unstable . finally , we say that two partitions @xmath246 and @xmath247 are _ compatible _ if the set @xmath248 can be divided into a disjoint union of three sets @xmath181 , @xmath176 , @xmath190 , in such a way that @xmath249 it is easy to see that if two partitions @xmath246 and @xmath247 are not compatible , then the geometric intersection of the corresponding cycles has codimension greater than @xmath1 . therefore , the intersection of two such generators in the cohomology algebra vanishes . the cohomology algebra @xmath230 is generated by the generators @xmath236 $ ] modulo two kinds of relations : ( i ) the linear relations @xmath245 , ( ii ) the multiplicative relations @xmath250 [ d_2 ] = 0 $ ] for incompatible partitions @xmath246 and @xmath247 . in what follows , we will require particular @xmath1-cohomology classes @xmath252 . recall that @xmath253 denotes the line bundle over @xmath78 whose fiber coincides with the cotangent line at the @xmath42th marked point . let @xmath257 . we denote by @xmath258 $ ] the sum of all the classes @xmath236 $ ] corresponding to partitions containing @xmath42 in one part , and @xmath259 in the other part . the star in the notation reminds that an irreducible component with a single node can not contain a single marked point . we construct a holomorphic section of the line bundle @xmath253 in the following way . consider a stable curve such that the marked points @xmath261 are on the same irreducible component . on this component there exists a unique meromorphic differential with simple poles at the points @xmath241 and @xmath133 , whose residues at these poles are equal to @xmath73 and @xmath262 respectively . the value of this differential at the point @xmath42 is an element of the line @xmath106 . thus we obtain a section of the bundle @xmath253 , and it extends in a unique way to a meromorphic section over the entire base @xmath78 . it is easy to see that this section has no poles ( thus the section is , actually , holomorphic ) . its zeroes are precisely the classes that add up to @xmath263 , and it intersects the zero section of @xmath253 transversally . since @xmath19 is a vector bundle over @xmath78 , the cohomology algebra of @xmath78 can be canonically seen as a subalgebra of the cohomology algebra of @xmath34 . we will therefore use the same notation for cohomology classes on @xmath78 and their pull - backs on @xmath34 . the assertion follows immediately from the description of @xmath19 as a vector bundle over @xmath78 ( section [ ssec : bundle ] ) and from the following well - known fact ( see , for example , @xcite ) . for any rank @xmath7 complex vector bundle @xmath265 over a complex manifold @xmath52 we have @xmath266\,\ , / \,\ , \bigl(\psi^{n } + c_1(e ) \psi^{n-1 } + \dots + c_{n}(e ) \bigr),\ ] ] that is , the cohomology algebra of @xmath267 is isomorphic to the algebra of polynomials in one variable @xmath268 with coefficients in @xmath269 , modulo the chern polynomial in @xmath268 . here the new variable @xmath268 can be identified with the first chern class of the tautological line bundle over @xmath267 . now , let @xmath232 be a partition of the set @xmath233 into a disjoint union of two nonempty subsets . the subsets are no longer required to have at least 2 elements . consider the set of stable meromorphic functions defined on nodal curves with @xmath1 irreducible components , the first component containing the marked points of the subset @xmath181 and the second one of the subset @xmath176 . this set is an open submanifold of @xmath19 of complex codimension @xmath73 , so its closure is a cycle poincar dual to a cohomology class in @xmath49 . this class will be denoted by @xmath236 $ ] . let @xmath270 . denote by @xmath271\in h^2(\p\ch_n)$ ] the sum of all the classes @xmath236 $ ] corresponding to partitions such that @xmath42 belongs to one part and @xmath241 and @xmath133 belong to the other one . ( there is no star after @xmath42 , because the singleton @xmath272 is now an allowed subset . ) recall that @xmath38 is the first chern class of the tautological line bundle on @xmath34 . a section of the vector bundle @xmath274 dual to @xmath19 will automatically provide a section of @xmath275 ( because a linear form on each fiber of @xmath19 can be restricted to any line contained in that fiber ) . but a section of the bundle @xmath274 can be obtained from a section of any of the line bundles @xmath253 by taking its direct sum with zero sections of each of the other line bundles @xmath276 for @xmath277 . thus the class @xmath38 is equal to the sum of the class @xmath278 and of the class poincar dual to the projectivization of the subbundle @xmath279 the latter @xmath1-cohomology class is exactly the class that corresponds to the partition @xmath280 using the expression for @xmath256 given in proposition [ proppsi_i ] we obtain @xmath273 $ ] . the first assertion clearly follows from the second one . let us prove the second assertion . for @xmath281 , @xmath95 is topologically a sphere , and the class @xmath236 = [ \{1\ } \sqcup \{2\}]$ ] is a point ; thus the proposition is true . for @xmath66 , the classes @xmath236 $ ] corresponding to partitions with at least two elements in each part generate the cohomology algebra of @xmath78 , and those corresponding to partitions with a single element in one part produce the class @xmath38 . therefore , they generate the cohomology algebra of @xmath34 . in this section we study various relations between the cohomology classes poincar dual to strata in @xmath34 . note that the symmetric group @xmath46 acts on @xmath19 by permuting the poles . therefore , it also acts on the cohomology algebra of @xmath34 . strata of the stratification , as well as most of other cohomology classes that we consider , are invariant under this action . therefore , their cohomology classes belong to the subalgebra of @xmath46-invariant cohomology classes . let @xmath282 and @xmath283 be two positive integers , @xmath284 , @xmath285 . we have introduced the cohomology class @xmath286 $ ] assigned to a partition @xmath234 of the set @xmath287 into two nonempty parts . denote by @xmath288 the sum of the classes @xmath286 $ ] for all partitions such that @xmath289 , @xmath290 . ( if @xmath291 the partitions @xmath234 and @xmath235 are two different terms of the sum , although they determine the same cohomology class . ) in order to prove the second identity we consider once again the covering @xmath172 of @xmath34 introduced in the proof of theorem [ thmmult ] . denote by @xmath297 the pull - back of the class @xmath38 on @xmath34 . in the proof of theorem [ thmmult ] we established that @xmath298 ( the pull - back under the lifted lyashko - looijenga map of the class of a hyperplane in @xmath201 equals @xmath299 . ) consider a particular union of hyperplanes in the image space @xmath201 of @xmath198 . namely , the union of all hyperplanes where the @xmath42th and the @xmath241th critical values of @xmath28 ( i.e. , the @xmath42th and the @xmath241th coordinates in @xmath201 ) are equal ( for @xmath300 ) . this union represents @xmath301 times the class of a hyperplane . now consider the preimage of this union under the map @xmath198 . it consists of the union of the strata @xmath302 , @xmath303 , and @xmath304 ( the liftings to @xmath172 of the caustic @xmath47 , the maxwell stratum @xmath48 , and the boundary stratum @xmath63 ) . moreover , @xmath303 , @xmath304 are simple preimages , but @xmath302 is a triple preimage . ( this means that if we take a generic point @xmath305 and its image in @xmath201 , and then take a generic point close to the image , then this point will have three preimages close to @xmath28 . ) now , the projection of @xmath172 onto @xmath34 is ramified over @xmath47 and @xmath63 , and the order of ramification is @xmath1 in both cases . on the contrary , it is not ramified over @xmath48 . this implies that the geometrical liftings of the strata @xmath47 and @xmath63 to @xmath172 represent only one half of the pull - backs of the corresponding homology classes . on the other hand , the homology class of the geometrical lifting of @xmath48 is equal to the lifting of the homology class of @xmath48 . putting everything together we obtain @xmath306 whence @xmath307 first of all , note that it suffices to prove the first of the four identities @xmath308 the other three identities follow from this one and from proposition [ propidentities ] . moreover , the @xmath1-cohomology classes in the left- and the right - hand sides of the identity are _ symmetric _ , i.e. , invariant under the action of the symmetric group @xmath46 by renumbering the poles . the symmetric part of @xmath49 is spanned by the classes @xmath288 ( see section [ ssec : h*(hn ) ] ) . therefore , we must construct a set of @xmath1-homology classes of @xmath34 such that their couplings with the classes @xmath288 allow one to distinguish all linear combinations of the classes @xmath288 . it is then enough to show that their couplings with @xmath47 and @xmath309 coincide . to construct such a set of @xmath1-homology classes , we consider several embeddings of @xmath0 into @xmath34 . for any positive integers @xmath310 such that @xmath311 we are going to consider the following embedding of @xmath0 in @xmath34 . let @xmath312 $ ] be the homogeneous coordinates on @xmath0 . consider the following family of rational functions in the variable @xmath115 : @xmath313 where the @xmath314 and the @xmath315 are generically chosen complex numbers . the rational functions of this family do not belong to @xmath19 for @xmath316 , @xmath317 or @xmath318 ( since they have less than @xmath7 poles ) ; however , there is a unique way to extend the family to a well - defined map from @xmath0 to @xmath34 , because @xmath34 is compact . we denote by @xmath319 the @xmath1-homology class of the image of this map . we are going to study its couplings with the @xmath1-cohomology classes @xmath47 , @xmath63 , and @xmath288 . note that these classes are defined as poincar dual classes to particular codimension @xmath73 subvarieties of @xmath34 . therefore we will usually speak of their _ intersection indices _ ( rather than couplings ) with the families @xmath288 . the three following lemmas describe completely the intersection indices of the family @xmath319 with the classes @xmath320 and @xmath47 . for shortness , we use the following convention . if in a lemma we give the intersection indices @xmath321 and @xmath322 , then in the particular case @xmath323 , @xmath324 , the two intersection indices must be added . for example , when we write @xmath325 , in the particular case @xmath291 the corresponding intersection index is equal to @xmath1 . before proving the lemmas , let us make use of them to compute the total intersection indices . ( for @xmath38 we use the expression from proposition [ propidentities ] . ) we have @xmath335 @xmath336 @xmath337 @xmath327 for other pairs @xmath328 . and @xmath338 this is consistent with the identity @xmath339 that we must prove . moreover , looking at the intersection indices with the classes @xmath288 we conclude immediately that the families @xmath319 for different @xmath282 and @xmath283 allow one to distinguish any linear combinations of the classes @xmath288 . therefore , the identity follows from the three lemmas , that we will now prove . such a stable function belongs to the stratum @xmath340 , but not to other strata @xmath320 , nor to the caustic @xmath47 . it is easy to check that the intersection with @xmath340 is simple ( unless @xmath291 , in which case it is double by the definition of @xmath341 ) . when @xmath330 the function @xmath28 is defined on a curve with @xmath7 irreducible components shown in figure [ fig : case2 ] . such a stable function belongs to @xmath342 different irreducible components of @xmath343 and to one irreducible component of @xmath344 . it is easy to check that the intersection with each component is simple . now we must find the index of intersection of the family @xmath319 with the caustic @xmath47 at the point that we are considering . this is more difficult , because this point does not belong to the smooth part of the caustic . in order to find the intersection index we have to use a result of the subsequent paper @xcite , theorem 3.9 . in the particular case under consideration the assertion of this theorem is as follows . consider the subvariety of @xmath34 obtained from the stable function @xmath28 by moving in all possible ways the @xmath345 points of intersection of the peripheral components with the central component of the curve in figure [ fig : case2 ] , without changing the restrictions of @xmath28 to the peripheral components . the closure of this subvariety is isomorphic to @xmath346 . now , the theorem says that the neighborhood of this subvariety in @xmath34 is isomorphic to the neighborhood of @xmath347 in @xmath348 . moreover , the intersection of the caustic @xmath47 with this neighborhood is isomorphic to @xmath349 , where @xmath350 is the ( non - projectivized ) caustic in the smaller hurwitz space . the intersection of the family @xmath319 with the neighborhood is a generic complex curve that intersects the subvariety @xmath346 transversally and is not contained in @xmath349 . now we claim that the intersection of the caustic @xmath350 with a generic fiber of @xmath351 over @xmath346 is a hypersurface of degree @xmath352 , which is therefore the index of intersection between @xmath319 and @xmath47 at the point under consideration . it remains to check that @xmath352 is indeed the degree of the intersection of @xmath350 with a generic fiber of @xmath351 over @xmath346 . to do that it suffices to find the degree of @xmath353 as a polynomial in variables @xmath354 ( discrim means discriminant , numer means numerator ) . indeed , this polynomial is precisely the equation of the caustic @xmath350 in the fiber , where the poles are fixed at the points @xmath355 . the division by @xmath356 is needed because , as one can easily check , the hyperplanes @xmath357 are simple zeroes of the discriminant , but do not belong to the caustic . the degree of the above polynomial is indeed equal to @xmath352 for homogeneity reasons . in order to calculate the intersection of @xmath47 with the family @xmath319 outside the points where some poles get glued together , we will consider the discriminant of the derivative of the function @xmath28 of our family @xmath358 the degree of this discriminant as a homogeneous polynomial in @xmath359 and @xmath360 equals @xmath361 . one can check that this polynomial has double zeroes at the points @xmath330 . moreover , it has a zero of multiplicity @xmath362 at @xmath316 and a zero of multiplicity @xmath363 at @xmath317 . subtracting the multiplicities of all these zeroes , we obtain @xmath334 zeroes outside the points @xmath364 . notes of the scientific seminar of the mathematical department of st - petersburg steklov institute ( pomi ) , vol . * 292 * , ( 2002 ) , representation theory , dynamical systems , combinatorics , and algorithmic methods , no . 7 , p. 92119 . arxiv : math.co/0209013 .

we study some properties of a class of random connected planar fractal sets induced by a poissonian scale - invariant and translation - invariant point process . using the second - moment method , we show that their hausdorff dimensions are deterministic and equal to their expectation dimension . we also estimate their low - intensity limiting behavior . this applies in particular to the `` conformal loop ensembles '' defined via poissonian clouds of brownian loops for which the expectation dimension has been computed by schramm , sheffield and wilson . msc classification : 28a80 , 82b43 , 28a78

1009.4782

in this paper , we study certain random planar fractals that are close in spirit to random cantor sets constructed via independent iterations . before describing the class of sets that we will focus on , let us first recall some features of the `` classical '' planar random cantor sets , sometimes known as mandelbrot percolation or fractal percolation : * mandelbrot percolation . * define a set @xmath0 by removing independently each dyadic square inside the unit square @xmath1 ^ 2 $ ] with probability @xmath2 . this self - similar iterative procedure defines a random fractal that has been studied extensively @xcite . in the case where one replaces dyadic by triadic , one gets a natural `` random sierpinski carpet '' . clearly , one can interpret the set @xmath0 as the limit of a galton - watson tree . this implies immediately that when @xmath3 , then @xmath0 is almost surely empty , and that when @xmath4 , then @xmath5 . it is then possible and easy to compute the hausdorff dimension of @xmath0 as a function of @xmath2 ( see for instance @xcite ) . in fact , when @xmath2 is small enough and the set @xmath0 is non - empty , @xmath0 has non - trivial connected components . it seems clear that their dimensions must be a deterministic function of @xmath2 , but in general it appears not possible to compute it explicitely in terms of @xmath2 . the set @xmath0 is statistically invariant under dyadic scaling in a rather obvious sense : for a dyadic square @xmath6 ^ 2 $ ] , the law of @xmath7 conditioned by the event that it is non - empty is equal to the law of @xmath0 itself conditioned not to be empty . a similar invariance under dyadic translations can be stated . however , this invariance is restricted to `` dyadic transformations '' and it can not be extended to more general maps . a natural way to obtain stronger scale and translation invariance is to define the set @xmath0 by removing from @xmath1 ^ 2 $ ] all squares of a statistically translation - invariant and scale - invariant poisson point process of squares . in fact , one could also replace these squares by other planar shapes , such as line segments or disks . this gives rise to self - similar poisson percolation models , as studied for instance in @xcite . in these papers , the focus is on the existence and the nature of the phase transition for the connectivity property of @xmath0 in terms of the intensity @xmath8 of the poisson point process ( that replaces the factor @xmath9 ) : for small @xmath8 ( i.e. small @xmath2 ) , the set @xmath0 can have non - trivial connected components , and one can define the `` carpet '' @xmath10 that consists of all points of @xmath11 that can be connected to the boundary of the unit square by a path that remains in @xmath0 ( a typical point in @xmath0 will in fact not satisfy this property ; see figure [ square_soup ] for an example , and section [ remaining ] for a more complete discussion ) . [ square_soup ] recently , it has been pointed out that a special and interesting case is to remove the interiors of brownian loops instead of squares . in other words , one removes the interiors of a poisson collection of brownian loops called the brownian loop - soup and introduced in @xcite . indeed , planar brownian motion is conformally invariant , so that this collection of loops is not only scale - invariant , but also conformally invariant in a rather strong sense ( related to the conformal restriction property described in @xcite ) . this allows @xcite to derive links with other conformally invariant objects such as the gaussian free field or the stochastic loewner evolutions ( sle ) , via the notion of conformal loop ensembles ( cle ) studied in @xcite . the link with sle enables the study of many properties of those random cantor sets ( see sheffield - werner @xcite ) : * we know the carpet is non - trivial if and only if @xmath12 ( the value @xmath13 depends of course on the choice of normalization for the brownian loop measure ) . * the complement of the carpet is made of disjoint `` holes '' whose boundaries are sle(@xmath14 ) loops for some explicit @xmath15 and the dimension of these loops is known ( @xcite ) . * schramm , sheffield and wilson @xcite have computed ( as a function of @xmath14 ) the `` expectation dimension '' of the carpet that measures the mean number of @xmath16-balls needed to cover it . this last result was our initial motivation for the present paper . we show here that the hausdorff dimension of the carpet is deterministic and equals the expectation dimension . in combination with the connection between cles and brownian loop soups derived in @xcite and the explicit formula for the expectation dimension in @xcite , this completes the determination of the almost sure hausdorff dimensions of cle carpets . more generally , our paper illustrates the fact that the `` brownian loop - soup '' approach to cle and sle can be helpful in the derivation of second - moment estimates that are used to determine hausdorff dimensions ( these second - moment estimates can turn out to be difficult to handle directly in the sle setting , see for instance @xcite for the dimension of the sle curve itself ) . [ brownian_soup ] * the contribution of the present paper . * we study the dimension of the random carpet obtained from a poisson point process corresponding to a self - similar and translation invariant measure @xmath17 . in particular , we show that the carpet s hausdorff dimension @xmath18 is non - random , that it is equal to its `` expectation dimension '' , and we interpret the first terms of the expansion of @xmath18 as @xmath19 in terms of @xmath17 . the techniques that we use are rather classical , and are based on second - moment estimates ( see e.g. @xcite ) . one can also note that in the brownian loop - soup case , the first term in the expansion @xmath20 when @xmath19 is related to the expected area of the filled brownian loop computed by @xcite . the paper is organized as follows . in section 2 , we define a class of invariant poisson point processes on planar curves and prove some of their elementary properties . in section 3 we show that the hausdorff dimensions of random carpets defined by a random soup are deterministic , and that first moment computations are enough to determine these dimensions . in section 4 , we study the dimension of the carpet when @xmath19 using an approximation of the carpet by a simpler set . we are interested in measures @xmath17 on the set of compact planar curves @xmath21 ( a curve @xmath22 is defined here as the image of a continuous function from @xmath1 $ ] into the complex plane ) . we will not need to use a strong topology on this set , so we can just view these curves as compact subsets of the plane and simply use the hausdorff topology . we say that the measure @xmath17 is scale - invariant ( resp . translation - invariant ) if it is invariant under the transformations @xmath23 for all @xmath24 ( resp . @xmath25 for all @xmath26 ) . we will focus on the case where the measure @xmath17 is * `` locally finite '' * in the sense that the @xmath17-mass of the set of curves @xmath22 with diameter greater than @xmath13 that are included in a @xmath27 square is finite ( note that we could replace @xmath27 by @xmath28 for any given @xmath29 ) . local finiteness is in fact necessary in our setup , otherwise the carpets that we define are almost surely empty . we define @xmath30 to be the set of all translation - invariant , scale - invariant and locally finite measures @xmath17 that are not `` one - dimensional '' . more precisely , we require that for any real @xmath31 , @xmath32 ( note that if this quantity is not @xmath33 , then it is infinite because of translation - invariance ) . we need this condition to exclude degenerate cases such as @xmath17 being supported on line segments parallel to the @xmath34-axis . throughout this paper , @xmath35 will denote the diameter of a bounded set @xmath36 . we denote by @xmath37 the ball centered at @xmath38 with radius @xmath39 , and by @xmath40 its boundary circle . we define @xmath41 to be the area of the set `` surrounded '' by @xmath36 , that is , of the complement of the unbounded connected component of the complement of @xmath36 ( if @xmath36 is a circle , then @xmath41 is the area of the corresponding disc ) . for a non - empty compact set @xmath42 , we define @xmath43 as the point in @xmath42 with the smallest @xmath34-coordinate ( if there are several of them , choose the one among them with smallest @xmath44-coordinate ) . our first result states that any measure in @xmath30 can be constructed as the product of three components that describe location , scale , and shape . [ l1 ] * suppose that @xmath45 is a measure on the set of compact planar curves such that @xmath46 . then , consider the following product measure on @xmath47 : @xmath48 if we define @xmath49 , the previous measure induces a measure @xmath50 on the set of compact planar curves defined by : @xmath51 for all measurable bounded positive @xmath0 . then , the measure @xmath50 is translation - invariant , scale - invariant and locally finite . * conversely , any measure @xmath17 in @xmath30 can be constructed in this way . in fact , it is always possible to construct @xmath17 via ( [ procedure ] ) starting from a finite measure @xmath45 that is supported on the set of curves with diameter @xmath52 that are included in the square @xmath53 ^ 2 $ ] . * finally , if a measure @xmath54 can be defined starting from two different finite measures @xmath45 and @xmath55 as before then @xmath56 and @xmath57 . we will denote this last quantity by @xmath58 . note that defining @xmath59 as above , or @xmath60 under @xmath61 is the same ( via the change of variable @xmath62 ) . that is , we can either translate first or scale first , but the distribution of the scaling factor differs in the two cases . * let us suppose that @xmath45 is a measure on the set of compact planar curves such that @xmath46 . then , the measure @xmath50 defined via ( [ procedure ] ) is clearly scale - invariant and translation invariant . we have to check that it is locally finite . let us compute the @xmath50-mass of curves that fall in the square @xmath53 ^ 2 $ ] and have diameter greater than @xmath13 . note first that @xmath63 but for each given @xmath22 , the lebesgue measure of @xmath64 ^ 2 \}$ ] is @xmath65 . hence , integrating the previous identity over @xmath38 , it follows that the @xmath50 mass of the set of curves that have diameter greater than @xmath13 and are subsets of @xmath53 ^ 2 $ ] is bounded by @xmath66 . this ensures that @xmath67 . * conversely , suppose that @xmath68 . we would like to find a corresponding measure @xmath45 . intuitively @xmath45 should be the `` projection '' of @xmath69 on the set @xmath42 of curves @xmath22 that have diameter @xmath13 and such that the point @xmath70 is at the origin . we can write any curve @xmath71 where @xmath72 , and this induces a representation of the measurable space @xmath73 as a product @xmath74 . therefore , if we define , for all translation - invariant and scale - invariant functions @xmath0 on planar curves , @xmath75 } 1 _ { z ( \gamma ) \in [ -1,1]}),\ ] ] this induces a measure @xmath45 on @xmath42 . since @xmath69 is locally finite , the measure @xmath45 has finite mass ( take @xmath76 in the previous expression ) . we then define @xmath17 from @xmath45 as in ( [ procedure ] ) . for @xmath0 translation - invariant and scale - invariant , we have @xmath77 } 1 _ { z ( \gamma ) \in [ -1,1 ] } ) & = & \int f ( z+\rho\gamma ) \ , 1_{| z+\rho\gamma | \in [ 1,2 ] } \ , 1 _ { z ( z+\rho\gamma ) \in [ -1,1 ] } \ , d^2 z \times \frac { d\rho } { \rho^3 } \times \pi(d\gamma ) \\ & = & \int f ( \gamma ) \ , 1_{\rho \in [ 1,2 ] } \ , 4 \times \frac { d\rho}{\rho^3 } \times \pi(d\gamma ) \\ & = & \frac 3 2 \pi(f ) \\ & = & \tilde \mu \bigl ( f ( \gamma ) 1_{| \gamma | \in [ 1,2 ] } 1 _ { z ( \gamma ) \in [ -1,1]}\bigr ) . \end{aligned}\ ] ] + clearly , the measure @xmath17 is also scale - invariant and translation - invariant . it follows that for any @xmath78 and @xmath79 , @xmath80 and finally , using the product representation of @xmath81 , we get @xmath82 . * to check the final statement , compute the @xmath17-mass of the set of curves @xmath22 such that @xmath83 ^ 2 $ ] and @xmath84 in terms of @xmath45 . this equals @xmath85 ^ 2 } d^2 z \times \int_0^\infty \int_{\cal u } \frac { d \rho}{\rho^3 } d\pi ( \gamma ) 1_{{\cal a } ( \rho \gamma ) \ge 1 } = \pi \left ( \int_{1 / \sqrt { { \cal a } ( \gamma ) } } ^\infty \frac { d \rho}{\rho^3 } \right ) = \pi ( { \cal a } ( \gamma ) ) / 2.\ ] ] this last quantity is therefore the same for @xmath45 and @xmath55 . exactly the same argument implies that @xmath56 and in fact works for any function of @xmath22 that scales with @xmath86 . since any locally finite @xmath17 can be constructed from a measure @xmath45 supported on curves of diameter 1 , it follows that @xmath87 is finite . it is worth emphasizing that the scale - invariance of the measures @xmath54 is with respect to the transformations @xmath88 that move the point @xmath89 away from the origin when @xmath90 is large . one could also study how @xmath17 behaves under the transformations @xmath91 that magnify @xmath22 without changing its lowest - left - most point . the lemma in fact implies that the image of @xmath17 under @xmath92 is @xmath93 . recall that a poisson point process with intensity @xmath17 where @xmath94 is a random countable collection of curves @xmath95 in the plane such that for any disjoint ( measurable ) sets of curves @xmath96 , the numbers @xmath97 of curves in @xmath98 that are respectively in @xmath99 are independent poisson random variables with respective mean @xmath100 ( when @xmath101 , then @xmath102 almost surely ) . we call this a * random soup * in the plane with intensity @xmath17 . for any domain @xmath103 , we define the random soup in @xmath104 as the set of all curves in the soup @xmath98 that are contained in @xmath104 . in other words , if @xmath105 , then the soup in @xmath104 is @xmath106 . note that @xmath107 is itself a poisson point process with intensity @xmath108 . the properties of @xmath17 ensure that @xmath98 is translation - invariant and scale - invariant . the fact that @xmath17 is locally finite implies that almost surely for each bounded domain @xmath104 and each @xmath109 , @xmath107 contains only a finite number of curves of diameter greater than @xmath39 . a few examples of natural measures @xmath45 and their corresponding random soups are the following : * the measure @xmath45 is a constant times the law of a brownian loop of time - length 1 : this defines the * brownian loop soups * introduced in @xcite . it is immediate to check that the measure @xmath17 is locally finite in our sense ( i.e. just check that @xmath110 . * we can also look at the law of the outer boundary of the brownian loop of time - length @xmath13 . this is in fact a sle(@xmath111 ) loop that also corresponds to scaling limits of critical percolation cluster outer boundaries @xcite . it defines a soup of `` outer boundaries of brownian loops '' . local finiteness is a consequence of that of the brownian loop measure . * the measure @xmath45 is a constant times the dirac mass supported on the unit circle ( so @xmath17 is also supported on circles ) : this is the * circle soup*. * the measure @xmath45 is a constant times the law of the segment with endpoints @xmath112 and @xmath113 , where @xmath112 is chosen uniformly at random on the unit circle ( so @xmath17 is supported on segments ) : this is the * stick soup*. a variant of this is the * discrete stick soup * , where @xmath112 is a uniformly chosen random vertex of a regular @xmath114-gon . we now define a condition on @xmath17 that will roughly prevents the soups from being too `` dense '' . [ mu_cross ] let @xmath115 be the set of curves that intersect the unit disk and have diameter at least @xmath116 . suppose that @xmath54 . then the following statements are equivalent : * for some @xmath117 , @xmath118 . * for all @xmath117 , @xmath119 . * the measure @xmath17 is constructed as in lemma [ l1 ] via ( [ procedure ] ) from a finite measure @xmath45 such that @xmath120 holds and that satisfies @xmath121 when the statements hold , we say that the corresponding soup is * thin*. note in particular that if a soup is thin , then the ( two - dimensional ) lebesgue measure of @xmath22 is @xmath45-almost surely ( and therefore @xmath17-almost surely ) equal to zero ( otherwise the third statement would not hold ) so that a given point in the plane belongs almost surely to no curve of the corresponding soup . suppose that @xmath54 is defined via ( [ procedure ] ) from a measure @xmath45 supported on the set of curves @xmath22 with @xmath120 and @xmath122 . the curve @xmath123 has diameter @xmath124 , and it intersects the unit disk if and only if @xmath125 is at a distance less than @xmath126 from @xmath22 . hence , for all @xmath116 , we have @xmath127 the equivalence between the three statements follows readily ( note that this integral in @xmath39 can diverge only near @xmath128 ) . [ thin_large_r ] * if a soup is thin , then @xmath129 . * a random soup is thin if and only if almost surely , for every ring in the plane , only a finite number of curves in @xmath98 do intersect both the inner circle and the outer circle . the first statement follows from the dominated convergence theorem . for the second part , let @xmath130 be the set of curves that intersect both circles of radius @xmath131 and @xmath132 . it follows easily that for @xmath133 , @xmath134 . if @xmath135 is finite , then so is @xmath136 , and lemma [ mu_cross ] implies that the soup is thin . conversely , if the soup is thin , then @xmath137 is finite , so all @xmath135 are finite , and by scale invariance so are all @xmath130 . the rings with rational radii form a countable dense set among all rings , and this completes the proof . the circle soups , the square soups and the stick soups are obviously thin . the third condition in the lemma shows that as soon as the mean area of the @xmath39-neighborhood of @xmath22 ( defined under @xmath138 ) decays for instance faster than @xmath139 as @xmath140 , the corresponding @xmath17-soup is thin . for instance , if @xmath45 is supported on curves with hausdorff dimension @xmath141 , the size of the neighborhood decays like @xmath142 and the soup should be thin , assuming the bound holds in expectation . the brownian loop - soup is not thin , but we now show that the soup of its outer boundaries is thin ( and this will be enough for our purposes since they will define the same carpets ) . the soup of outer boundaries of brownian loops is thin . since outer boundaries @xmath22 of brownian loops have dimension @xmath143 ( see @xcite ) , the area of their @xmath39-neighborhoods decays typically at least like @xmath144 when @xmath140 . however , lemma [ mu_cross ] requires looking at expectations , rather than at `` typical '' behavior , so additional arguments are needed . here is a short self - contained proof that does not rely on @xcite . consider the brownian loop - measure , defined on brownian loops @xmath145)$ ] of time - length @xmath13 that start and end at the origin . a point @xmath38 is in the @xmath39-neighborhood of the outer boundary of the loop iff it lies at distance at most @xmath39 of the loop , and the disc @xmath37 is not disconnected from @xmath146 by the loop . using circular re - rooting of the loop , the mean area of the @xmath39-neighborhood of the outer boundary is clearly bounded by four times the mean area of the set of points @xmath38 that lie at distance at most @xmath39 from @xmath147 $ ] and such that @xmath37 is not disconnected from infinity by @xmath148 $ ] . it is easy to check that the law of @xmath149 $ ] is absolutely continuous with respect to that of @xmath150 $ ] , where @xmath151 is a standard brownian motion , and that its radon - nikodym derivative is bounded . hence , it is sufficient to bound the expected area of the set of points @xmath38 that lie at distance less than @xmath39 of @xmath152 $ ] and such that @xmath37 is not disconnected from infinity by @xmath153 $ ] . using the strong markov property of @xmath151 , it follows immediately that this quantity is bounded by the mean area @xmath154 of the @xmath39-neighborhood of @xmath155 $ ] times the probability that a planar brownian motion started at distance @xmath39 from the origin does not disconnect @xmath156 before time @xmath157 . this last probability is bounded by @xmath39 to some positive power when @xmath158 ( see any introductory paper on disconnection exponents this is due to the fact that one the one hand the probability that it stays in @xmath159 during the time - interval @xmath160 $ ] is very small , and on the other hand that the probability that it does not disconnect @xmath156 before reaching the circle of radius @xmath161 decays at least like a positive power of @xmath39 when @xmath158 ) , and the former mean area @xmath154 is bounded . hence , the thinness of the soup of outer boundaries of brownian loops follows . we will be interested in the fractal carpets defined using random soups . to simplify the discussion , we assume in this section that the measure @xmath45 ( and therefore also @xmath17 ) is supported on * simple * loops , that is , injective continuous maps from the unit circle into the plane . this will exclude the discrete stick soups ( discussed briefly at the end of this section ) and the brownian loop soup ( but we can define and recover the carpet for the latter via soups of outer boundaries of brownian loops ) . the carpet corresponds to a connected component of the random cantor set obtained by removing the interiors of all the loops . there are _ a priori _ various ways to define it . consider a random soup @xmath107 in a simply connected domain @xmath104 ( such that @xmath162 ) , consisting of loops @xmath163 . each @xmath164 defines an interior @xmath165 ( the bounded connected component of @xmath166 ) . the set @xmath167 is then a random closed subset of @xmath104 that we shall sometimes refer to as the `` remaining set '' . our interest is in its connected components : the * carpet * @xmath10 is the set of points @xmath38 in @xmath104 such that there exists a continuous path from any neighborhood of @xmath38 to @xmath168 that stays in @xmath0 . loosely speaking , the carpet is the connected ( by arcs ) component of @xmath0 that has @xmath168 as part of its boundary . note that , for technical reasons , we allow the connecting paths to intersect the loops in the soup ( but not their interiors ) . we will comment in a moment on whether this definition is equivalent to saying that there exists a continuous path from @xmath38 itself to @xmath168 that stays in @xmath0 . another almost equivalent approach is to look at * clusters of loops*. we say that two loops @xmath22 and @xmath169 of a random soup are connected if one can find a finite sequence @xmath170 in the soup such that for all @xmath171 , @xmath172 clearly , connection between loops forms an equivalence relation , and one can then define the clusters of the soup as the union of all @xmath164 s for the loops in the same equivalence class . note that for technical reasons , we require here not only that the two curves @xmath164 and @xmath173 intersect but also that their interiors do . in many cases , this definition can be relaxed , as translation invariance can be easily be used to show that almost surely ( for a sample of the soup ) , if two curves intersect then so do their interiors . the clusters may be nested , this may occur for example when one loop lies inside another one . given a loop - soup in the entire plane , one can wonder whether there exist clusters of infinite diameter . @xmath174 law implies in the standard way that this event has probability either 0 or 1 , depending on the law of the soup . furthermore , if the measure @xmath17 is supported on loops with positive inner area , it follows immediately that any disk @xmath175 is almost surely contained in the interior of some loop @xmath164 . since this holds for arbitrarily large @xmath116 , it follows that if there exists a cluster of infinite diameter , then it is unique . in this case , let @xmath176 denote the distance between the unbounded cluster and the origin . clearly , @xmath176 is a scale - invariant finite real random variable , so that @xmath177 almost surely . it follows that for a loop soup in the entire plane : * either all clusters are bounded almost surely * or there almost surely exists exactly one cluster , and this cluster is dense and unbounded . since our main interest in the present paper is the geometry of the clusters rather then the phase transition , we will focus on the case when all clusters are bounded almost surely and the carpets are not empty . more precisely , our assumption on @xmath17 goes as follows : * subcriticality assumption 1 . * _ all clusters in the full - plane loop soup are almost surely bounded . _ in fact , we will assume an _ a priori _ slightly stronger condition : * subcriticality assumption 2 . * _ with positive probability , there exists a ( random ) closed loop @xmath178 in the plane that surrounds the origin and does not `` cross '' any loop of the loop - soup . _ in this definition , and throughout the rest of this paper , we say that two loops `` cross '' if each one of the two does intersect the interior of the other one . let us stress that the loop @xmath178 is not a loop of the loop - soup . note that a simple @xmath174 argument then implies that any given point is almost surely surrounded by infinitely many such loops ( of arbitrarily small or large diameter ) that cross no loop in the loop - soup . let us now consider a domain @xmath104 as before and choose a given point @xmath38 on its boundary . the fact that there almost surely exist such small loops @xmath178 around @xmath38 that cross no loop of the full - plane loop - soup implies immediately that the carpet in @xmath104 is not empty ( because the intersection of this small loop with @xmath104 is in the carpet ) . hence , subcriticality implies that the carpets are almost surely non - empty . note also that subcriticality clearly implies that the measure @xmath17 is thin . it could be interesting to study the converse i.e. whether non - triviality of the carpet and thinness imply our subcriticality assumption . this is for instance not difficult if one assumes invariance of @xmath17 under certain rotations . in that case , subcriticality can be proven using fkg - type arguments , but this is not the purpose of the present paper . in section [ sec_intensity ] we review a coupling argument that shows that most soups of interest are subcritical as long as their intensity parameter lies below a certain value . furthermore , note that the arguments presented in @xcite in order to prove that in the `` subcritical phase , '' outer boundaries of clusters of brownian loops are indeed loops ( this is essentially only based on an fkg argument ) can be easily generalized to the present setting . this implies that the subcriticality assumption 1 implies the subcriticality assumption 2 in most cases of interest ( because the outer boundary @xmath178 of a loop - soup cluster that surrounds the origin is one closed loop that satisfies the conditions of the second assumption ) . it also indicates that in most cases , the carpet is indeed the set of points that are connected to @xmath168 by a continuous path in @xmath0 . finally , let us mention that our definition of the carpet can in fact be easily adapted to the case of the stick soups . one just needs to define the set @xmath10 as the set of points @xmath38 such that there exists a continuous path from any neighborhood of @xmath38 to @xmath168 that does not `` cross '' any stick ( in an appropriate sense ) and to modify the definition of loop - clusters similarly . the results of the present paper would still apply . other possible variants could include the possibilities that @xmath164 s are discontinuous etc . in this section we will consider a thin translation- and scale- invariant random soup on a bounded simply connected domain @xmath104 and we will suppose furthermore that the subcriticality assumption holds . we will show that the hausdorff dimension @xmath179 of the carpet is deterministic , and that it is described by `` first moment estimates '' . we will use the standard second - moment method to evaluate the dimension of a random fractal set c that is closely related to the carpet @xmath10 . the idea is to define a sequence of sets @xmath180 that converge to @xmath181 , to obtain probability estimates for these sets and to show that these yield the dimension of @xmath181 . we will use the following standard fact about first and second moments ( see @xcite for this precise statement , or @xcite for almost equivalent ones ) : [ bef ] let @xmath104 be a bounded domain and @xmath182 a family of random borel subsets of @xmath104 , so that @xmath183 if @xmath184 . define suppose that @xmath186 and define the following three conditions : 1 . there exist positive constants @xmath187 and @xmath188 such that for any small @xmath189 and @xmath190 , @xmath191 . 2 . there exists a positive constant @xmath192 such that for any @xmath189 and @xmath193 , @xmath194 . 3 . there exists a positive constant @xmath195 such that for all @xmath189 and @xmath190 , the expected area of @xmath196 , conditional on the event that @xmath197 , is at least @xmath198 . then : * if @xmath199 and both 1 . and 3 . hold , then @xmath200 almost surely . * if @xmath199 and both 1 . and 2 . hold , then @xmath201 with positive probability . * if @xmath202 and both 1 . and 3 . hold , then @xmath181 is almost surely empty . in other words , if we have first and second moments estimates for the area of @xmath180 with the correct asymptotics , then we can control the dimension of @xmath181 . we will obtain such estimates for a random set @xmath181 related to the carpet , and then use a 0 - 1 law to argue that the dimension is deterministic , and therefore is almost surely equal to @xmath203 . one natural choice for @xmath180 would be the set of all points within @xmath204 of the carpet ; however , a slightly different definition is better suited in our setting : let @xmath180 be the set of points @xmath190 with the property that there exists a path connecting @xmath34 to @xmath168 that does not cross any interior of a curve of the soup @xmath205 . in other words , we take all curves in the soup , we ignore the ones that get within distance @xmath204 of @xmath34 , and we look if there exists a path connecting @xmath34 to the boundary that does not hit the interior of any of the remaining curves . we define the * approximate carpet * @xmath206 . let @xmath207 be the union of all curves in @xmath107 . then we have @xmath208 clearly , if a point @xmath38 is in the carpet @xmath10 , then it is in @xmath181 . conversely , if a point is in @xmath181 and is not on any curve @xmath164 , then it necessarily is in the carpet . indeed , take @xmath38 in @xmath104 but not on any curve , and let @xmath189 . because the soup is thin , only finitely many curves @xmath209 in the soup intersect both circles of radii @xmath210 and @xmath204 centered at @xmath38 ( corollary [ thin_large_r ] ) . hence for @xmath211 no curve in the soup that intersects the circle @xmath212 comes @xmath213-close to @xmath38 . if @xmath38 is in @xmath181 , then by looking at @xmath214 , we conclude there exists a path joining @xmath212 to @xmath168 that crosses no curve in the soup . since this is valid for all @xmath204 , we conclude that @xmath38 is in the carpet . we first restrict ourselves to the case where @xmath104 is the unit disk @xmath215 . let @xmath98 be the random soup in the unit disk . for any @xmath216 , let @xmath217 be the set of curves in @xmath98 that are contained inside the ring @xmath218 ( including curves that wind around the inner circle without touching it ) . let @xmath219 be the event that @xmath220 i.e. there exists a path that connects the origin to the unit circle and does not cross any curve in @xmath217 . we first show that @xmath221 behaves asymptotically like some power of @xmath204 . [ pa_upper ] there exist @xmath222 and @xmath223 such that for any @xmath224 , @xmath225 the upper bound is trivial : let @xmath226 be the set of curves in @xmath98 that are contained inside the ring @xmath227 . let @xmath228 be the event that there is a path that connects the circles @xmath229 and @xmath230 and does not cross any curve in @xmath226 . by scale invariance of the soup , @xmath231 . clearly @xmath232 . since @xmath226 and @xmath217 are disjoint , @xmath228 and @xmath219 are independent . hence @xmath233 . note that either @xmath234 for all @xmath235 , or @xmath236 for all @xmath204 smaller than some @xmath237 . it is easy to verify that in this second case , the carpet is almost surely empty . we can therefore assume we are in the first case . note also that if almost all loops for @xmath17 have nonempty interiors , then with positive probability , one loop of the loop - soup will surround the origin , so @xmath238 for @xmath204 small enough . let us define the event @xmath239 that in the ring @xmath240 , there exists a closed loop that surrounds the origin and that crosses no curve in the soup ( we consider here the entire soup in the plane ) . note that because of scale - invariance , the probability @xmath241 does not depend on @xmath39 . furthermore , because soup clusters are bounded , the subcriticality assumption implies that for sufficiently small @xmath39 and large @xmath116 , @xmath242 ( consider , for example , the boundary of the cluster containing the origin ) . hence , @xmath243 is positive for large enough @xmath116 . let us now fix @xmath244 such that @xmath245 , and choose @xmath204 and @xmath246 in @xmath247 . consider the following four events : * @xmath248 is the event that for the soup in the ring @xmath249 , there exists a path joining the inner boundary to the outer boundary of the ring , that does not cross any loop of this soup . the probability of this event is @xmath250 . * @xmath251 is the event that for the soup in the ring @xmath252 , there exists a path joining the inner and outer boundary of the ring , that does not cross any loop of this soup . by scale - invariance , the probability of this event is @xmath253 . * @xmath254 is the event that in the ring @xmath255 , there exists a closed loop surrounding the origin that does not cross any curve in the entire soup @xmath256 . because of scale - invariance , the probability of this event does not depend on @xmath204 and is equal to @xmath257 . note that because @xmath258 this closed loop must intersect the paths described in the definitions of @xmath248 and @xmath251 . * @xmath259 is the event that no curve in the soup @xmath256 that intersects the ring @xmath260 has diameter greater than @xmath261 . because of scale - invariance , its probability @xmath262 does not depend on @xmath204 . it is positive because the soup is thin . + recall that an event @xmath263 depending on the realization of a soup is said to be decreasing if @xmath264 and @xmath265 implies @xmath266 . it is standard that decreasing events are positively correlated ( this is the fkg - harris inequality , see e.g. @xcite ) . here , the events @xmath248 , @xmath251 , @xmath254 and @xmath259 are all decreasing therefore , @xmath267 on the other hand , @xmath268 . indeed , if all events occur , then we can concatenate a part @xmath269 of the crossing defined by @xmath251 to a part @xmath270 of the loop defined by @xmath254 to a part @xmath271 of the crossing defined by @xmath248 to construct a crossing @xmath272 of the ring @xmath273 . then @xmath259 implies that @xmath272 does not cross any curve in @xmath274 . the lower bound follows . such a lemma implies classically up - to - constants estimates : [ pa_alpha_zero ] for some positive constants @xmath275 and @xmath276 , we have @xmath277 let @xmath278 . since @xmath279 , it follows readily from standard subadditivity that @xmath280 on the other hand , if we define @xmath281 , we get that @xmath282 so that @xmath283 but @xmath284 if we define this limit to be @xmath275 , it follows that @xmath285 and that for all @xmath204 , @xmath286 we still assume that @xmath104 is the unit disk @xmath215 . to avoid boundary effects , it is natural to define , for any @xmath287 , the set @xmath288 of points in @xmath104 that are at distance greater than @xmath213 from @xmath168 i.e. @xmath289 . [ pa_alpha ] for all small @xmath213 : * there exist positive @xmath290 and @xmath291 such that for all @xmath292 and for all @xmath293 , @xmath294 * there exists @xmath295 such that for any two points @xmath296 , and all @xmath293 , @xmath297 * there exists @xmath298 such that for all @xmath292 and @xmath299 , the expected area of @xmath300 , given that @xmath197 , is at least @xmath198 . we first estimate @xmath301 . assume that @xmath302 . for any @xmath39 define @xmath303 to be the set of all curves in @xmath98 contained in the ring @xmath304 . we have @xmath305 hence if @xmath197 , then there exists a path from @xmath306 to @xmath307 that does not cross any curve in @xmath308 . conversely , if there exists a path from @xmath306 to @xmath309 that does not cross any curve in @xmath310 , then @xmath197 . hence both the upper and lower bounds follow from corollary [ pa_alpha_zero ] and scale invariance . we now estimate the second moment . we can assume without loss of generality that @xmath311 ( otherwise the result follows from the first moment estimate ) . consider first the case when @xmath312 . define the events * @xmath248 : there is a path from @xmath34 to @xmath313 that does not cross any curve in @xmath98 contained in @xmath314 * @xmath251 : there is a path from @xmath44 to @xmath315 that does not cross any curve in @xmath98 contained in @xmath316 the balls @xmath317 and @xmath318 do not intersect , so @xmath248 and @xmath251 are independent . clearly @xmath319 , so by scale invariance @xmath320 and since @xmath321 is bounded below , ( [ pa_alpha_2 ] ) follows . when @xmath322 , let @xmath323 and consider the events * @xmath248 : there is a path @xmath324 from @xmath34 to @xmath325 that does not cross any curve in @xmath98 contained in @xmath326 * @xmath251 : there is a path @xmath327 from @xmath44 to @xmath328 that does not cross any curve in @xmath98 contained in @xmath329 * @xmath254 : there is a path @xmath330 from @xmath331 to @xmath332 that does not cross any curve in @xmath98 contained in @xmath333 the three events involve curves contained in disjoint sets , so they are independent . clearly @xmath334 , and by corollary [ pa_alpha_zero ] , @xmath335 and ( [ pa_alpha_2 ] ) follows . it now remains to check the final statement . we show first that if @xmath336 , then @xmath337 . indeed , if @xmath338 , then the soup obtained by removing all curves that are at distance less than @xmath204 from @xmath44 is contained in the soup obtained by removing all curves that are at distance less than @xmath210 of @xmath34 . hence @xmath339 . note also that since @xmath104 and @xmath288 are disks , we have that for any @xmath292 and @xmath299 , the area of the intersection @xmath340 is at least @xmath341 . hence , @xmath342 this is precisely the last statement of the lemma . we can now apply lemma [ bef ] : it shows that for any small @xmath213 , the dimension of @xmath343 is almost surely not larger than @xmath344 , and is equal to @xmath345 with nonzero probability . we conclude that almost surely , @xmath346 and that @xmath347 . we now assume that @xmath104 is a bounded non - empty open domain . the proof of the upper bound readily follows for the previous case . indeed , if @xmath38 belongs to the carpet defined by the soup in @xmath104 , and @xmath151 is any small disk with @xmath348 , then @xmath38 also belongs to the approximate carpet defined by the soup in @xmath151 . hence if we write @xmath349 as a countable union of open disks , the approximate carpet @xmath181 defined by the soup in @xmath104 will be a subset of the union @xmath350 of the approximate carpets defined inside each ball . for the lower bound on the dimension , consider a sequence @xmath351 of points in @xmath104 such that * the sequence @xmath352 converges to @xmath33 as @xmath353 . * the disks @xmath354 are disjoint . it is easy to find such a sequence : consider a for instance a sequence of points @xmath355 on @xmath168 and a positive sequence ( for instance @xmath356 ) in such a way that all the balls @xmath357 , @xmath358 etc . are disjoint . then , one just has to choose for each @xmath114 , a point @xmath351 in @xmath359 . fix any @xmath133 . let @xmath360 be the event that there exist curves in @xmath256 that intersect both @xmath361 and @xmath362 . from corollary [ thin_large_r ] and scale invariance , @xmath363 as @xmath364 , so by passing to a subsequence we can assume ( using a standard borel - cantelli argument ) that almost surely , no @xmath360 occurs for large enough @xmath114 . now for each @xmath114 , define the soups @xmath365 and @xmath366 , and consider the following events : * @xmath367 is the event that for the approximate carpet @xmath368 defined by the soup @xmath369 , the intersection @xmath370 has dimension at least @xmath344 . we have showed in the previous section that it has probability bounded below by some positive constant . * @xmath371 is the event that there exists a closed loop @xmath269 in the ring @xmath372 that does not cross any curve in @xmath369 . * @xmath373 is the event that there exists no curve in @xmath374 that intersects both @xmath375 and @xmath376 . since the soup is thin and subcritical , @xmath371 and @xmath373 also have probabilities bounded from below . the three events are decreasing and therefore positively correlated . hence , the probability of their intersection is bounded below independently from @xmath114 . since the disks @xmath377 are disjoint , it follows that the corresponding soups are independent . by borel - cantelli , there almost surely exist an infinite set of @xmath114 s such @xmath378 occurs . hence there exists @xmath114 such that all @xmath379 occur and @xmath360 does not . let @xmath380 , so there is a path @xmath271 from @xmath38 to @xmath381 that does not cross @xmath369 . the loop @xmath269 can not lie entirely inside @xmath104 , so we can concatenate parts of @xmath271 and @xmath269 to construct a path @xmath272 that connects @xmath38 to @xmath168 , lies inside @xmath382 , and crosses no curve in @xmath369 . then @xmath383 guarantees @xmath272 crosses no curve in @xmath374 , and finally , since @xmath360 does not occur , @xmath272 crosses no curve in @xmath98 . hence @xmath384 , so the approximate carpet has almost surely dimension at least @xmath344 . we have therefore completed the proof of the following fact : under our subcriticality assumption , the dimension of the approximate carpet @xmath181 defined by @xmath107 is almost surely equal to @xmath203 . finally , we consider the dimension of the carpet @xmath10 itself : under our subcriticality assumption , the dimension of the carpet @xmath10 defined by @xmath107 is almost surely equal to @xmath203 . recall that @xmath385 , where @xmath42 is the union of all the loops in the loop - soup . recall also that subcriticality implies that the soup is thin , which ensures that for any given point @xmath38 , @xmath386 . for each @xmath387 , we can define the set @xmath388 . note that @xmath389 . furthermore , for any given @xmath38 , @xmath390 . it follows easily that all first and second moment estimates that we derived for @xmath180 also hold for @xmath391 , so that @xmath392 with positive probability . the proof of the 0 - 1 law also holds essentially unchanged and we conclude that the hausdorff dimension of @xmath10 is almost surely equal to @xmath203 . consider a subcritical thin loop - soup in a bounded domain @xmath104 as before with intensity @xmath17 . recall @xmath165 are the interiors of the loops in the soup , and define @xmath393 . @xmath0 is what we informally call `` the remaining set '' . we emphasize that _ a priori _ ( and in reality ) its dimension should be larger than that of the approximate carpet . indeed , a typical point in @xmath0 will be surrounded by infinitely many chains of loops , and therefore not in the approximate carpet . this result was proved in john thacker s ph.d . thesis @xcite in the context of the brownian loop - soup . since the general proof is essentially identical , we only give an outline . the proof is a direct application of the second moment method , in the same spirit as before . things are in fact simpler here , since the loops do not interact , i.e. a point @xmath34 is in the remaining set if and only no loop in the poisson point process belongs to the set of loops @xmath396 that do not surround @xmath34 . we define @xmath397 to be the set obtained by removing from @xmath104 only the interior of the set of loops of diameter greater than @xmath204 in the soup , and we apply the second moment method ( note that @xmath398 ) . for a given @xmath38 at positive distance from the boundary of @xmath104 , the probability that @xmath399 is equal to the probability that no loop with diameter greater than @xmath204 in the loop - soup has @xmath38 in its interior i.e. to @xmath400 using lemma [ l1 ] , it follows easily that , up to multiplicative constants that depend on @xmath401 and on the distance between @xmath38 and @xmath168 , the probability that @xmath399 is comparable to @xmath402 . this is the first - moment estimate . to bound the second moment , we take two points @xmath34 and @xmath44 , we define @xmath403 and decompose the loop - soup into three pieces : those loops that remain at distance less than @xmath39 of @xmath34 , those loops that remain at distance less than @xmath39 of @xmath44 , and those loops that never come closer to @xmath404 of the midpoint between @xmath34 and @xmath44 . we then use the previous argument to deduce that the probability that both @xmath34 and @xmath44 are in @xmath397 is no larger than a constant times @xmath405 . suppose that @xmath17 is fixed measure as before , defined from a finite measure @xmath45 as in lemma [ l1 ] . we are now going to introduce a positive real parameter @xmath8 , and consider for each value of @xmath8 a soup with intensity @xmath406 ( and its approximate carpet if it is a subcritical soup ) . we can couple realizations of the soups for all @xmath8 in an increasing manner , with richer soups for larger @xmath8 . to see this , one can for instance first define a poisson point process @xmath407 with intensity @xmath408 on @xmath409 $ ] , and then for each @xmath8 , define @xmath410 and note that @xmath411 is a poisson point process with intensity @xmath412 . for each value of @xmath8 such that the approximate carpet @xmath413 is non - empty , its fractal dimension @xmath18 is almost surely constant ( the function @xmath414 of course depends on the actual choice of @xmath17 ) . it is easy to see ( this follows for instance from our estimates on the remaining set ) that when @xmath8 is large , then the approximate carpet is almost surely empty . on the other hand , comparing the soup with a deterministic well - studied fractal percolation model ( see e.g. @xcite ) , one can show that when @xmath8 is very small , the approximate carpet is almost surely not empty . hence , there exists a finite positive critical value @xmath415 that separates these two regimes ( this is the origin of the `` subcritical '' terminology ) . it is in fact not difficult ( at least when the measure @xmath45 is invariant under some rotations ) to adapt the arguments developed for fractal percolation to prove that the approximate carpet is not empty when @xmath416 @xcite . the goal of this section is to derive first - order estimates for the exponent @xmath18 when @xmath19 . intuitively , in this limit a loop in the soup will typically not intersect any other loop of comparable size . therefore , the holes in the approximate carpet will look like the interiors of the loops themselves , and this will lead to an approximation of its dimension in terms of the mean area of the interior of loops under the measure @xmath45 . the approximate carpet will be rather close to the remaining set . we will prove : consider the soup in some bounded domain @xmath104 . it contains countably many curves @xmath420 , which can be ordered in decreasing order of their diameter , so @xmath421 ( it is trivial to check that no two loops can have exactly the same diameter ) . this induces an ordering of the clusters of curves in the soup . we start with @xmath422 and consider its cluster @xmath423 , as defined in section [ carpet_cluster ] . then we take the smallest @xmath424 such that @xmath420 is not contained in @xmath425 and consider its cluster @xmath426 , and so on . this yields an ordering of the loop - clusters . again , it is possible to discover @xmath427 progressively using the properties of poisson point processes . in particular , the conditional law of @xmath428 given @xmath429 is simply that of a poisson point process of loops ( ordered according to their diameter ) with intensity @xmath430 . in fact , we are now going to make a variation of this exploration procedure , where @xmath425 , @xmath431 etc are discovered one by one . conditionally on @xmath422 and @xmath425 , the law of all other curves in the soup ( not contained in @xmath425 ) is the same as the law of a standard soup with components required ( i ) to have size smaller than @xmath422 and ( ii ) not to cross @xmath425 . hence the soup clusters admit the following equivalent description ( all the soups involved have intensity shape measure @xmath432 ) : * generate the soup @xmath226 , let @xmath433 be the largest diameter curve in @xmath226 . * generate the soup @xmath434 ( independent of @xmath226 ) , let @xmath435 be the subset of curves in @xmath434 that have diameter smaller than @xmath433 , and let @xmath436 be the cluster of @xmath433 inside @xmath437 . * let @xmath438 be the largest diameter curve in @xmath226 that does not meet @xmath425 * generate the soup @xmath439 ( independent of @xmath226 , @xmath434 ) , let @xmath440 be the subset of curves in @xmath439 that have diameter smaller than @xmath438 and do not intersect @xmath425 , and let @xmath441 be the cluster of @xmath438 inside @xmath442 * and continue inductively . [ soup_via_many_soups ] let @xmath443 be independent loop soups with intensity @xmath444 on a bounded domain @xmath104 . for each @xmath445 , we define recursively a curve @xmath446 , a set of curves @xmath374 , and a cluster @xmath447 as follows : * @xmath446 is the largest diameter curve in @xmath226 that does not intersect @xmath448 * @xmath374 is the set of curves in @xmath369 that have diameter smaller than @xmath446 and do not intersect @xmath448 * @xmath449 is the cluster of @xmath446 inside @xmath450 hence we can generate soup clusters of a soup by starting with a soup @xmath98 , selecting a subset of its curves , and `` attaching '' to each certain subsets of independent soups @xmath453 . if we attach instead a larger subset or even the whole soup @xmath453 , then this can only decrease the size of the corresponding carpet . this is what we will do in the next subsection . to formalize this idea , let us first consider the product measure @xmath454 , on pairs @xmath455 of one loop ( `` sampled '' from @xmath45 ) and one loop - soup with intensity @xmath412 * in the entire plane*. we have seen that we can choose the finite measure @xmath45 in such a way that it is supported on the set of loops of diameter @xmath13 contained in the square @xmath53 ^ 2 $ ] . let @xmath226 be the set of all curves in @xmath98 that have diameter smaller than @xmath13 and let @xmath456 be the `` filling '' of the cluster of @xmath457 that contains @xmath22 ( i.e. the closure of the complement of the unbounded connected component of the complement of the cluster ) . we denote by @xmath458 the measure under which @xmath456 is defined . * when @xmath8 is small enough , then @xmath459 . we denote this quantity @xmath460 . * when @xmath461 , @xmath462 converges to @xmath463 . * there exist @xmath464 and @xmath465 such that for all @xmath466 and all @xmath467 , the probability that there exists a cluster in @xmath226 that crosses the ring @xmath468 is bounded by @xmath469 . consider some @xmath470 so that the loop - soup @xmath471 with intensity @xmath472 is subcritical . we know there exists some @xmath473 such that the probability @xmath2 that no cluster in @xmath471 traverses the ring @xmath474 is strictly positive . for any @xmath465 , @xmath471 can also be constructed as the union of @xmath465 independent soups of intensity @xmath475 . hence we can find a sufficiently large @xmath465 so that the probability that the loop - soup @xmath476 with intensity @xmath477 contains a cluster that crosses the ring is smaller than @xmath478 . using scale - invariance , it follows that for all @xmath479 , the probability that @xmath480 contains a cluster that crosses the ring @xmath481 is smaller than @xmath482 . clearly , the same is true if one looks only at the loop - soup @xmath226 consisting only of the loops in @xmath98 of diameter smaller than @xmath13 . but the events that @xmath226 contains a cluster that crosses the ring @xmath483 for @xmath484 are in fact independent : they depend only on those loops that intersect each of the rings , and these sets of loops are disjoint since all loops have diameter smaller than @xmath13 . we conclude that the probability that a cluster of @xmath226 crosses the ring @xmath485 is bounded by @xmath486 . this immediately implies the last item of the lemma . finally , observe that if we couple the realizations of @xmath490 for all @xmath8 , then almost surely @xmath491 converges to the filling of the initial loop @xmath22 when @xmath19 , as all other loops disappear . this follows , for instance , from the fact ( proved using the same argument as above ) that for any given ring @xmath492 , the probability that a loop - soup cluster of @xmath493 crosses the ring goes to zero as @xmath8 goes to @xmath33 . we conclude , using monotone convergence , that @xmath494 we now construct the measure @xmath495 from @xmath17 similarly to the way @xmath458 was constructed from @xmath45 : consider the product measure @xmath496 , on pairs @xmath455 of one loop ( `` sampled '' from @xmath406 ) and one loop - soup with intensity @xmath412 in the entire plane . let @xmath497 be the set of all curves in @xmath98 that have diameter smaller than @xmath22 and let @xmath456 be the `` filling '' of the cluster of @xmath498 that contains @xmath22 . we denote by @xmath495 the measure under which @xmath456 is defined . clearly , scale - invariance and translation - invariance of @xmath17 ( and of the loop - soup ) imply that @xmath495 is also scale - invariant , translation - invariant , and that it can be constructed ( as in section 2 ) from @xmath458 ( note that the previous lemma ensures that @xmath499 ) . in fact , this definition makes it possible to define a poisson point process of pairs @xmath500 i.e. to couple a loop - soup with intensity @xmath406 with a loop - soup with intensity @xmath495 in such a way that for each @xmath501 , @xmath502 ( basically , each loop @xmath164 is extended by an independent soup of loops of smaller diameter in the whole plane ) . if we keep those @xmath164 that are in a domain @xmath104 , we get a loop - soup @xmath107 with intensity @xmath452 . but mind that the corresponding @xmath503 s do not necessarily stay in @xmath104 . however , they are not likely to be very large , as we now show . to keep things simple , we assume for the rest of this section that @xmath104 is the unit disk ; we can do this without loss of generality , since the dimension of the carpet does not depend on the domain . let @xmath415 and @xmath116 be defined as in the proof of lemma [ 15 ] , and consider the coupling @xmath500 defined above . if @xmath8 is small enough , then the probability that for all @xmath501 such that @xmath504 , the diameter of @xmath503 does not exceed @xmath505 is strictly positive . this is in fact a simple consequence of the last statement of lemma [ 15 ] . it suffices to show that the set @xmath506 has finite mass . note that because of scale - invariance , the @xmath17-mass of the set of loops in @xmath104 with diameter between @xmath507 and @xmath508 grows slower than @xmath509 as @xmath510 . but as @xmath511 converges , the lemma follows readily . let @xmath107 be a subcritical soup with intensity @xmath412 in the unit disk , and let @xmath180 defined as before . for @xmath8 small enough , there exists a constant @xmath465 such that for all small @xmath204 , @xmath512 , where @xmath513 . let us consider the construction of the loop soup clusters @xmath514 in @xmath104 described in proposition [ soup_via_many_soups ] . clearly , if we compare it with the coupling that we have just described , we can choose our coupling @xmath500 in such a way that each @xmath447 is in fact a subset of one of the @xmath503 s with @xmath504 . one therefore has @xmath515 where @xmath516 denotes the event that none of the sets @xmath503 with @xmath504 surround the disc of radius @xmath204 ( this is because each cluster in the initial loop - soup is contained in some @xmath503 ) . this last event depends on the poisson point process of pairs @xmath500 only . let @xmath517 denote the event that that for all @xmath501 such that @xmath504 , one has @xmath518 . recall from the previous lemma that this event has a positive probability ( provided @xmath8 is small enough ) . let @xmath519 denote the event that none of the sets @xmath503 that are subsets of @xmath520 and of diameter greater than @xmath204 surround the origin . this event is independent of @xmath517 , as they occur for disjoint set of pairs @xmath521 . note that @xmath522 . hence , @xmath523 it therefore remains to estimate @xmath524 . this is equivalent to an estimate concerning the remaining set of the poisson point process of sets @xmath525 in @xmath526 , and is obtained just as for the first - moment estimate in lemma [ coeff_c ] . we can now conclude the proof of proposition [ eq ] . recall that @xmath527 because the remaining set contains the carpet . on the other hand , the previous lemma and the definition of @xmath528 show that the @xmath529 corresponding to the loop - soup with intensity @xmath432 is not larger than @xmath530 . as we have already mentioned , in the case of the brownian loop - soup , the connection between the boundaries of clusters derived in @xcite ( see also @xcite ) and the schramm - loewner evolutions make it possible to describe the carpet via branching sle - type paths @xcite . schramm , sheffield and wilson @xcite computed in fact the `` expected dimension '' of the carpet , and our paper shows that this `` expected dimension '' ( the quantity that governs the first moment ) is equal to the almost sure hausdorff dimension @xmath18 . combining all these results , the expression derived in @xcite implies that for @xmath531 , @xmath532 where @xmath8 and @xmath14 are related by the `` usual '' relation @xmath533 . it follows that @xmath534 note that @xmath535 . as explained in @xcite ( in the context of the remaining set ) , the coefficient of the first order term @xmath536 is closely related to the mean area of the brownian loop computed in @xcite . recall @xcite that the hausdorff dimension of the sle@xmath537 curve is almost surely equal to @xmath538 . hence , in the brownian loop - soup case , we also know the dimension of the boundary of the carpet .

we present a new measurement of the-spectroscopic factor ( ) and the asymptotic normalization coefficient ( anc ) for the 6.356 mev 1/2 subthreshold state ofo through thec(b,li)o transfer reaction and we determine the-width of this state . this is believed to have a strong effect on the rate of thec(,)o reaction , the main neutron source for _ slow _ neutron captures ( the-process ) in asymptotic giant branch ( agb ) stars . based on the new width we derive the astrophysical s - factor and the stellar rate of thec(,)o reaction . at a temperature of 100 mk our rate is roughly two times larger than that by and two times smaller than that recommended by the nacre compilation . we use the new rate and different rates available in the literature as input in simulations of agb stars to study their influence on the abundances of selected-process elements and isotopic ratios . there are no changes in the final results using the different rates for thec(,)o reaction when thec burns completely in radiative conditions . when thec burns in convective conditions , as in stars of initial mass lower than and in post - agb stars , some changes are to be expected , e.g. , of up to 25% for pb in our models . these variations will have to be carefully analyzed when more accurate stellar mixing models and more precise observational constraints are available .

1208.0714

approximately half of the elements heavier than iron in the universe are produced via a series of @xmath12 neutron capture reactions and competing @xmath13-decays ( the @xmath9-process ) . during the @xmath9-process , the neutron number density is relatively low , of the order of 10@xmath14 n @xmath15 . when the flux reaches an unstable nucleus , it typically decays rather than capture another neutron and the @xmath9-process proceeds via the isotopes around the valley of @xmath13-stability ( e.g. , * ? ? ? * ) . the astrophysical sites of the @xmath9-process are core he and shell c burning in massive stars for the elements lighter than sr @xcite , and the `` he intershell '' of asymptotic giant branch ( agb ) stars for the elements between sr and bi @xcite . stars with initial masses lower than roughly 9 @xmath11 reach the agb phase in the final phases of their evolution , when both h and he have been exhausted in the core leaving c and o in electron degenerate conditions . production of nuclear energy occurs in the h and he shells , which are located between the core and the extended convective envelope and are separated by the thin he intershell layer . agb stars experience thermal pulses ( tps ) when the usually dormant he burning shell is suddenly activated . a large amount of energy is released , which drives convection in the he intershell . during tps the star expands and cools and the h burning shell is inactive . while he burning turns from the convective to the radiative regime , and eventually switches off , the convective envelope can penetrate the underlying he intershell and carry to the surface the products of he burning , in particular carbon and the elements heavier than iron made by the @xmath9-process . this mixing process is known as the `` third dredge - up '' ( tdu ) . after the tdu is ended , the star contracts and heats up again and h burning resumes until another tp occurs and the cycle is repeated . this sequence of events can occur from a few times to hundreds of times , depending on the stellar mass and the mass - loss rate . agb stars suffer from very strong stellar winds , which erode the envelope roughly within a million years and shed the newly synthesized material mixed to the surface by the tdu into the interstellar medium . eventually , the c - o degenerate core is left as a cooling white dwarf ( see * ? ? ? * for a review on agb stars ) . according to the current standard model @xcite , some protons must diffuse from the convective envelope into the he intershell at the end of each tdu in order to produce enough @xmath4c to account for the observed abundances of the @xmath9-process elements at the surface of agb stars ( see also * ? ? ? * ) . a thin layer is then produced , known as the @xmath4c `` pocket '' , which is rich in @xmath4c made via @xmath16c(@xmath17,@xmath18)@xmath4n(@xmath19)@xmath4c . when in this region the temperature reaches about @xmath20 k , the @xmath4c(@xmath0,@xmath7)@xmath8o reaction is activated and generates neutrons that trigger the @xmath9-process @xcite . considerable effort has been devoted to the direct measurement of the @xmath4c(@xmath0,@xmath7)@xmath8o cross section @xcite . these measurements have been performed at energies down to 270 kev , whereas the gamow window is at 190 @xmath21 40 kev , corresponding to a temperature of 100 mk . since this energy is far below the coulomb barrier , the reaction cross section is extremely small and direct measurement is sensitively limited by background signals and very difficult to perform in laboratories on the earth s surface . while a measurement has been proposed at the underground laboratory of luna @xcite , at present , the experimental cross sections have to be extrapolated below 270 kev . a microscopic cluster model analysis of the @xmath4c(@xmath0,@xmath7)@xmath8o and @xmath4c(@xmath0,@xmath0 ) reactions by @xcite suggested that this extrapolation is critically affected by the 1/2@xmath2 subthreshold resonance in @xmath3o ( @xmath22 = 6.356 mev , just 3 kev below the @xmath0-decay threshold ) . the contribution from this resonance depends strongly on the @xmath0-width of the 1/2@xmath2 state in @xmath3o , which can be derived from the spectroscopic factor ( @xmath1 ) or the asymptotic normalization coefficient ( anc ) of @xmath0-cluster in this state . the @xmath1 and the anc can be determined from the angular distribution of the direct @xmath0-transfer reaction using distorted wave born approximation ( dwba ) or coupled reaction channels ( crc ) analysis . although three indirect measurements via the ( @xmath23li,@xmath24 ) or the ( @xmath6li,@xmath25 ) system have been performed by @xcite , @xcite , and @xcite to study the @xmath1 or the anc of the 1/2@xmath2 state , there still exists a significant discrepancy of up to a factor of @xmath1030 in the derived @xmath1 and anc . therefore , it is interesting to perform a new measurement of the @xmath1 and the anc via an independent transfer reaction . in addition , it is necessary to understand the impact of the different resulting @xmath4c(@xmath0,@xmath7)@xmath8o rates on the @xmath9-process nucleosynthesis in agb stars . in this paper we determine a new stellar rate of the @xmath4c(@xmath0,@xmath7)@xmath8o reaction and incorporate it in calculations of the @xmath9-process nucleosynthesis in agb stars . first , we measure the angular distribution of the @xmath4c(@xmath5b,@xmath6li)@xmath3o reaction to determine the @xmath1 and the anc for the 1/2@xmath2 state in @xmath3o . using this experimental anc we derive the @xmath0-width for the 1/2@xmath2 subthreshold resonance , which is currently the most uncertain parameter for determining the @xmath4c(@xmath0,@xmath7)@xmath8o rate . finally , we use the new rate and different rates available in the literature as input for simulations of agb stars to study their influence on the abundance of some selected @xmath9-process elements and isotopic ratios . the measurement of the angular distribution for the @xmath4c(@xmath5b,@xmath6li)@xmath3o reaction was performed at the hi-13 tandem accelerator of china institute of atomic energy ( ciae ) in beijing . we used a @xmath5b beam with an energy of 50 mev leading to the production of the excited states in @xmath3o at @xmath22 = 3.055 mev , 3.843 mev , 4.554 mev and 6.356 mev . the angular distribution of the elastic scattering for the entrance channel ( @xmath5b+@xmath4c ) was also measured . we used a self - supporting @xmath4c target with a thickness of 75@xmath216 @xmath26g/@xmath27 and an initial purity of 88% . a 26 mev @xmath14li beam was also delivered for measurement of the exit channel ( @xmath14li+@xmath3o ) elastic scattering . natural silicon monoxide of 86@xmath217 @xmath26g/@xmath27 was prepared onto a 40@xmath213 @xmath26g/@xmath27 carbon foil , serving as the oxygen target . in addition , a self - supporting @xmath16c target of 66@xmath215 @xmath26g/@xmath27 was used for calibration of the focal plane and background subtraction during the whole experiment . to monitor the possible buildup of @xmath16c , the @xmath5b elastic scattering on the @xmath4c target was measured at the start and at the end of the measurement for each angle . this showed that the @xmath16c buildup was negligible compared to the initial impurity in the @xmath4c target , possibly due to the rather low gas pressure ( @xmath2810@xmath29 mb ) in the reaction chamber . to determine the absolute amount of @xmath16c in the @xmath4c target we measured the angular distributions of the @xmath5b+@xmath16c elastic scattering with both the natural @xmath16c target and the enriched @xmath4c target . the absolute amounts of @xmath16c and @xmath4c in the @xmath4c target were found to be [email protected] and 66@xmath215 @xmath26g/@xmath27 , respectively . the beam current was measured by a faraday cup covering an angular range of @xmath306@xmath31 in a laboratory frame and used for the absolute normalization of the cross sections at @xmath32 . the faraday cup was removed when measuring the cross sections at @xmath33 . a si @xmath34 telescope located at @xmath35 was employed for the relative normalization of the cross sections at @xmath33 by measuring the elastic scattering of the incident ions on the targets . in addition , the ratio of current integration in the faraday cup to the elastic scattering events was measured at the start and the end of the measurement for each angle @xmath33 by restoring the faraday cup . the ratios changed by less than 2% , which thus led to a reliable normalization of the cross sections at @xmath33 . the reaction products were focused and separated by a q3d magnetic spectrograph and recorded by a two - dimensional position - sensitive silicon detector ( pssd , 50@xmath3650 mm ) fixed at the focal plane of the spectrograph . the two - dimensional position information from the pssd enabled the products emitted into the acceptable solid angle to be recorded completely . the energy information from the pssd was used to remove the impurities with the same magnetic rigidity . due to the presence of @xmath16c in the @xmath4c target , the @xmath14li events from the @xmath4c(@xmath5b,@xmath6li)@xmath3o@xmath37 ( 6.356 mev ) reaction were mixed with those from the @xmath16c(@xmath5b,@xmath6li)@xmath8o@xmath37 ( 6.917 mev ) reaction . to evaluate this background the ( @xmath5b,@xmath6li ) reactions were measured for both the @xmath4c and @xmath16c targets at each angle with the same experimental setup . as an example , in fig . [ fig1 ] we display the focal - plane position spectra of @xmath14li at @xmath38=10@xmath31 from the ( @xmath5b,@xmath6li ) transfer reactions . the background from @xmath16c accounts for approximately half of total events from the @xmath4c target . after background subtraction and beam normalization , the angular distributions of the elastic scattering and the @xmath4c(@xmath5b,@xmath6li)@xmath3o@xmath37 ( 6.356 mev ) reaction were obtained , as presented in figs . [ fig2 ] and [ fig3 ] . in fig . [ fig3 ] we also display the angular distributions of the @xmath4c(@xmath5b,@xmath6li)@xmath3o reaction leading to the other three states ( @xmath22 = 3.055 mev , 3.843 mev , and 4.554 mev ) . note that the measurements for these three states were not affected by the @xmath16c background . the finite - range distorted wave born approximation ( dwba ) method with the fresco code @xcite was used to analyze the experimental angular distributions of the transfer reaction . the optical model potential ( omp ) parameters for the entrance and exit channels were obtained by fitting the experimental angular distributions of the @xmath5b+@xmath4c and @xmath6li+@xmath8o elastic scattering , respectively ( fig . [ fig2 ] ) . full complex remnant term interactions were included in the transfer reaction calculations . the parameters of the core - core ( @xmath6li+@xmath4c ) potential of @xcite gave reasonable account of the elastic scattering of @xmath14li from @xmath4c at 34 mev . the parameters used in the dwba calculations are listed in table [ tab1 ] . to obtain the spectroscopic factor ( @xmath1 ) and the anc of @xmath0-cluster in @xmath3o the spectroscopic amplitudes of @xmath0-cluster in the ground state of @xmath5b need to be fixed . the single - particle wave function describing the relative motion between @xmath0-cluster and @xmath14li core in the @xmath5b ground state can have two components denoted by quantum numbers @xmath39 = 3@xmath40 and 2@xmath41 , respectively , where @xmath42 is the number of radial nodes that include the origin but not the infinity and @xmath43 and @xmath44 are the orbital and total angular momenta , respectively . the spectroscopic amplitudes of these two components are @xmath450.509 and 0.629 , respectively , from a shell model calculation @xcite , and @xmath450.638 and @xmath450.422 , respectively , from the translationally invariant shell model @xcite . in the present analysis , both sets of spectroscopic amplitudes were used and the resulting difference was incorporated in the total uncertainty of our result . the @xmath0-cluster single - particle wave functions were calculated using conventional woods - saxon potentials whose depths were adjusted to reproduce the binding energies of @xmath0-cluster in the ground state of @xmath5b and in the four states of @xmath3o . the quantum numbers @xmath39 for @xmath0-cluster in the 6.356 mev 1/2@xmath2 state of @xmath3o were fixed to be @xmath46 by the oscillatory energy conservation relation @xmath47 , where @xmath48 are the corresponding single - nucleon shell quantum numbers . in addition , the quantum numbers for @xmath0-cluster in the other three states , 3.055 mev 1/2@xmath49 , 3.843 mev 5/2@xmath49 , and 4.554 mev 3/2@xmath49 , were determined to be @xmath50 , @xmath51 , and @xmath51 , respectively . the geometry parameters , radius @xmath52 and diffuseness @xmath53 , of the woods - saxon potential for @xmath0-cluster in @xmath5b were adjusted to give the root - mean - square ( rms ) radius ( @xmath54 fm ) of the @xmath0-cluster wave function . this was calculated using the following relation between cluster sizes and their mean distances within a nucleus , @xmath55 where the rms radii of @xmath56he , @xmath14li , and @xmath5b were taken to be 1.47 fm , 2.384 fm , and 2.605 fm , respectively @xcite . the resulting parameters are @xmath57 fm and @xmath58 fm . we investigated the dependence of the calculated @xmath1 and the anc on the geometry parameters for @xmath5b . with a diffuseness between [email protected] fm , the radius was adjusted to reproduce the rms radius of 3.204 fm . the impact of this change on the @xmath1 was found to be less than 2% . we also investigated the dependence of the @xmath1 and the anc on the geometry parameters for @xmath3o using the same range of @xmath52 ( [email protected] fm ) and @xmath53 ( [email protected] fm ) that was selected by maximum likelihood function set at 3@xmath59 level by @xcite . using steps of 0.02 fm for both @xmath52 and @xmath53 , 81 sets of geometry parameters were obtained and used to calculate 81 values of the @xmath1 and the anc . their standard deviations were taken as the uncertainty deriving from the geometry parameters for @xmath3o . the parameters @xmath52=1.00 fm and @xmath53=0.76 fm provide the best description for the angular distributions of the four states ( see also fig . [ fig3 ] ) . in fig . [ fig3 ] we display the calculated angular distributions normalized to the experimental data for the @xmath4c(@xmath5b,@xmath6li)@xmath3o reaction populating the 3.055 mev , 3.843 mev , 4.554 mev , and 6.356 mev states in @xmath3o . for the 3.055 mev state the @xmath1 factor was found to be [email protected] , compatible with the values obtained from the ( @xmath23li,@xmath24 ) reaction ( @[email protected]@xmath450.3 ) by @xcite and from the ( @xmath6li,@xmath25 ) reaction ( @[email protected] ) by @xcite . for the 3.843 mev and 4.554 mev states the @xmath1 factors were found to be [email protected] and [email protected] , respectively . these disagree with the values of [email protected] and [email protected] , respectively for the two states , given by @xcite , who adopted geometry parameters @xmath52=1.25 fm and @xmath53=0.65 fm for @xmath3o and used a coupled - channel calculation to fit the data . when using these geometry parameters we were not able to reproduce our measured @xmath4c(@xmath5b,@xmath6li)@xmath3o angular distributions of the four states and obtained an extremely small value of maximum likelihood function . for the 4.554 mev state our result is consistent with that of @xcite , who found an @xmath1 factor of [email protected] via the ( @xmath6li,@xmath25 ) reaction . the @xmath1 for the 6.356 mev 1/2@xmath2 state of @xmath3o was derived to be [email protected] . the error results from the statistics ( 23% ) , the target thickness ( 8% ) , the uncertainties from the spectroscopic amplitudes ( 3% ) for @xmath5b , the geometry parameters ( 2% ) for @xmath5b , and the geometry parameters ( 20% ) for the 1/2@xmath2 state of @xmath3o . the square of the coulomb modified anc ( @xmath61 ) was then extracted to be 4.0 @xmath21 1.1 @xmath62 using the relation , @xmath63 , where @xmath64 is the radial single - particle wave function for @xmath0-cluster in the 1/2@xmath2 state of @xmath3o , and @xmath65 is the coulomb modified whittaker function , with @xmath66 being gamma function and @xmath67 the coulomb parameter . the uncertainty in the anc of 27.5% is smaller than that in the @xmath1 of 32.4% . this is because the variation of the geometry parameters yields a change in @xmath64 that is opposite in sign to the change in @xmath1 so the uncertainty from the @xmath3o geometry parameters in @xmath61 of 12% is smaller than that in @xmath1 of 20% . three independent measurements in addition to the present work have been performed to date to study the anc or the @xmath1 of the 1/2@xmath2 state in @xmath3o . a very small spectroscopic factor ( @[email protected] ) was found via measurement of the @xmath4c(@xmath23li,@xmath24)@xmath3o angular distribution with an incident energy of 60 mev @xcite . this indicated that the contribution of the 1/2@xmath2 subthreshold resonance is negligible . however , a reanalysis of the same experimental data showed that the dwba analysis of @xcite could be flawed @xcite . these authors derived larger @xmath1 factors of 0.36 and 0.40 via dwba and coupled reaction channels ( crc ) calculations , respectively . @xcite measured the @xmath23li(@xmath4c,@xmath24)@xmath3o angular distribution at sub - coulomb energies ( 8 and 8.5 mev ) and derived @xmath61 = [email protected] @xmath62 . @xcite measured the @xmath4c(@xmath6li,@xmath25)@xmath3o angular distribution with two incident energies ( 28 and 34 mev ) . the @xmath1 and @xmath61 were found to be [email protected] and [email protected] @xmath62 , respectively . in table [ tab2 ] we list the @xmath1 and @xmath61 values from the present study and from the literature sources mentioned above . good agreement for the @xmath1 and @xmath61 is found between @xcite , @xcite , and the present work , which used three different transfer systems and covered an energy range of 4@xmath4510 mev / u . this indicates that the @xmath1 and anc for the 1/2@xmath2 state of @xmath3o should not significantly depend on the selection of transfer systems and incident energies . on the other hand , the @xmath61 obtained from the @xmath23li(@xmath4c,@xmath24 ) reaction at sub - coulomb energies by @xcite is about five times smaller than that obtained in the present work and by @xcite . the sub - coulomb ( @xmath23li,@xmath24 ) @xmath0-transfer cross section was recently remeasured to understand the source of this discrepancy and the data analysis is in progress ( g. v. rogachev 2012 , private communication ) . it remains to be seen if this analysis will result in a revised value of @xmath61 . the astrophysical s - factor of the @xmath4c(@xmath0,@xmath7)@xmath8o reaction via the resonances can be calculated with the breit - wigner formula , @xmath68 where @xmath26 is the reduced mass of the @xmath0+@xmath4c system ; @xmath69 represents the resonance energy ; @xmath70 , @xmath71 , and @xmath72 are the spins of the excited states in @xmath3o , @xmath0 , and @xmath4c , respectively ; @xmath73 , @xmath74 , and @xmath75 denote the @xmath0- , neutron- , and total widths , respectively ; and @xmath76 is the reaction @xmath76-value of @xmath4c(@xmath0,@xmath7)@xmath8o . for the resonances above the @xmath0-threshold , the energy dependence of these partial widths is given by @xmath77 and @xmath78 where @xmath79 and @xmath80 denote the experimental @xmath0- and neutron - widths , respectively ; @xmath81 is the @xmath0-penetrability ; and @xmath82 and @xmath83 are the orbital angular momenta for @xmath0 and neutron in the excited states of @xmath3o , respectively ( see , e.g. * ? ? ? * ) . for the subthreshold resonance , the energy dependence of the neutron - width can be also obtained by eq . [ eq4 ] , while the dependence of the @xmath0-width is expressed as @xmath84 here the reduced @xmath0-width @xmath85 can be given by @xmath86 and was extracted to be [email protected] kev at the channel radius @xmath87=7.5 fm . this large radius was chosen to reach the coulomb asymptotic behavior of @xmath64 , as suggested by @xcite and @xcite . in table [ tab3 ] we list the resonant parameters for the 1/2@xmath2 subthreshold state employed to obtain the astrophysical s - factor of the 1/2@xmath2 subthreshold resonance in @xmath4c(@xmath0,@xmath7)@xmath8o . the uncertainties in the s - factor were investigated by varying @xmath85 , @xmath69 , and @xmath80 . we also investigated the dependence of the s - factor on the channel radius by changing @xmath87 from 6 fm to 7.5 fm . the s - factor at the gamow peak of 190 kev , @xmath88 , was derived to be ( [email protected])@xmath3610@xmath89 mev b. the error is dominated by the uncertainty in the resonant parameters for the 1/2@xmath2 subthreshold state . the uncertainty from the channel radius is 4% in @xmath88 , varying slightly with the energy . we calculated the total s - factor of @xmath4c(@xmath0,@xmath7)@xmath8o by including the properties of the @xmath3o states up to 8.342 mev from the compilation of @xcite , and considering their interferences . however , the resulting total s - factor did not agree with the data from the direct measurements of @xcite , who measured the s - factor at the lowest energies to date , and of @xcite , who performed a systematical experimental verification of neutron efficiency over a range of well - defined energies . moreover , the resonant structure near the 3/2@xmath2 resonance at @xmath90=0.842 mev was not well reproduced using the parameters from @xcite . hence , we adjusted the @xmath0- and neutron - widths for this state to provide the best fitting of the experimental data . the fitting resulted in broader partial widths [ @xmath79=0.08 kev , @xmath80=340 kev ] than the recommended values of @xcite [ @xmath79=0.07 kev , @xmath80=280 kev ] , in agreement with @xcite and @xcite . all resonant parameters used in the present calculations are listed in table [ tab3 ] . in fig . [ fig4 ] we display the resulting astrophysical @xmath4c(@xmath0,@xmath7)@xmath8o s - factor as a function of the energy in the center of mass frame ( @xmath91 ) . the uncertainty mainly results from the resonant parameters of the 1/2@xmath2 subthreshold state since the resonant parameters of the states above the @xmath0-emission threshold are relatively well constrained by fitting the experimental data . the contribution of the 1/2@xmath2 resonance results in a clear increase of the s - factor at lower energies . at the gamow peak of 190 kev the 1/2@xmath2 subthreshold resonance dominates the @xmath4c(@xmath0,@xmath7)@xmath8o reaction . at this energy the contribution of this resonance accounts for 61% of total s - factor . the astrophysical @xmath4c(@xmath0,@xmath7)@xmath8o reaction rate was calculated with the present s - factor using : @xmath92}de,\end{aligned}\ ] ] where @xmath93 is avogadro s number . in table [ tab4 ] we list the present adopted rate as function of the temperature together with its upper and lower limits . in fig . [ fig5 ] we compare the present rate with the previous compilations by cf88 @xcite and nacre @xcite and with other rates available in the literature @xcite at the temperature of 100 mk . our rate agrees with that of @xcite and it is about two times smaller than the nacre recommended rate , but within its lower limit . both the rates by nacre and @xcite were derived from an extrapolation of the lowest energy experimental data available to date of @xcite . there is a large discrepancy of up to a factor of 2 between the present rate and the rates by cf88 , @xcite , and @xcite due to the fact that cf88 did not take into account the contribution of the 1/2@xmath2 subthreshold resonance and @xcite and @xcite found a significantly smaller contribution than ours of this resonance to the rate . our rate is in good agreement with that of @xcite , who found the contribution of the subthreshold resonance similar to ours , and is consistent with that of @xcite , who used an extensive multichannel @xmath94-matrix analysis to constrain the possible contribution from subthreshold resonances by taking into account all open reaction channels for the @xmath4c+@xmath0 and @xmath8o+@xmath7 systems . for conveniency , we fitted our rate with the expression used in the astrophysical reaction rate library reaclib @xcite , @xmath95\nonumber\\ + \exp[a_8+a_9t_{9}^{-1}+a_{10}t_{9}^{-1/3}+a_{11}t_{9}^{1/3 } + a_{12}t_{9}+a_{13}t_{9}^{5/3}+a_{14}\ln(t_{9})].\end{aligned}\ ] ] here @xmath96 is the temperature in units of 1 gk . the coefficients @xmath97 for our adopted value and lower and upper limits of the @xmath4c(@xmath0,@xmath7)@xmath8o reaction rate are listed in table [ tab4 ] . the overall fitting errors are less than 7% at temperatures from 0.04 to 10 gk . to assess and understand the impact of the new rates of the @xmath4c(@xmath0,@xmath7)@xmath8o reaction we used a post - processing code where a nuclear network of 320 species ( from h to bi ) and 2,336 reactions is solved simultaneously with mixing within the star when convective regions are present . we used stellar structure inputs , such as temperature , density , and convective velocity , calculated previously by the stromlo stellar structure code @xcite including mass - loss on the agb phase with the prescription of @xcite . we included the formation of the @xmath4c pocket by artificially allowing an exponentially decreasing proton profile to form just below the base of the convective envelope at the end of each tdu episode over a mass of 0.002 @xmath11 ( roughly 1/10@xmath98 of the mass of the he intershell , in the low - mass agb models considered here ) . the details of this procedure and the codes used to compute the models have been described previously ( e.g. , * ? ? ? * ; * ? ? ? we considered four models , which are summarized in table [ tab5 ] : a @xmath993 @xmath11 model of metallicity @xmath1000.02 similar to that discussed by @xcite , a @xmath991.8 @xmath11 model of @xmath1000.01 from @xcite , and @xmath991 @xmath11 and @xmath991.5 @xmath11 models of @xmath1000.0001 from @xcite . we selected these models for three reasons . ( 1 ) models of higher masses probably do not experience the @xmath4c neutron source as protons burn while being ingested in the he intershell , preventing the formation of the @xmath4c pocket @xcite . it is still not clear what the effect of this burning is on the whole stellar structure and the @xmath9-process , but it seems reasonable to us for the time being to not include a @xmath4c pocket for masses larger than @xmath104 @xmath11 . in these massive agb stars , the temperature in the tps exceeds 300 mk so that the @xmath101ne(@xmath0,@xmath7)@xmath102 mg reaction is activated and likely plays the role of the main neutron source @xcite . ( 2 ) the @xmath4c in the pocket normally burns before the onset of the following tp @xcite , however , in some cases the temperature may not be high enough for this to happen and some @xmath4c could be left over to burn in the following tp ( see also * ? ? ? * ) . in our lowest - mass models the first few @xmath4c pockets are ingested in the following tp while they still contain a relatively large amount of @xmath4c . these models are thus qualitatively different from the @xmath993 @xmath11 model where @xmath4c always burns before the onset of the next tp . in details , we have 6 , 15 , and 2 tdu episodes in the @xmath991.8 @xmath11 model of @xmath1000.01 , and the @xmath991.5 @xmath11 and @xmath991 @xmath11 models of @xmath1000.0001 , respectively . these numbers represent also the number of @xmath4c pockets introduced in each models , since a @xmath4c pocket is introduced after each tdu episode . of these @xmath4c pockets , the first 4 and 3 for the @xmath991.8 @xmath11 and @xmath991.5 @xmath11 models , respectively , are ingested in the following tp while the abundance of @xmath4c by number is still higher than 10@xmath103 . the same occurs for both the @xmath4c pockets introduced in the @xmath991 @xmath11 model . as the agb star evolves the temperature in the he intershell increases so that this effect disappears for later @xmath4c pockets . the exact number of ingested @xmath4c pockets relative to their total number increases with decreasing the stellar mass and increasing the metallicity . this is because the temperature decreases for lower masses and higher metallicities and thus the chance of some unburnt @xmath4c surviving to be ingested in the next tp is higher . ( 3 ) finally the low - mass and low - metallicity models ( 1 @xmath11 and 1.5 @xmath11 , @xmath104=0.0001 ) present a further regime for the activation of the @xmath4c neutron source . proton ingestion episodes occur in the first few tps during which a small amount of protons are ingested directly inside the tps and thus some extra @xmath4c is produced and burnt inside these convective regions @xcite . in fig . [ fig6 ] we schematically illustrate the location in time and space where the different regimes operate within the thermally - pulsating structure of an agb star and indicate the models where the regimes occur . in fig . [ fig7 ] we display the variations of the abundance of some selected @xmath9-process elements and isotopic ratios at stellar surface at the end of the evolution obtained from our models using the present @xmath4c(@xmath0,@xmath7)@xmath8o rate with respect to the two compilations , cf88 and nacre , respectively ( see fig . [ fig5 ] for a comparison of reaction rates ) . the results obtained using the rates from @xcite , @xcite , and @xcite agree with the present rate within 10% for all the models . the rate from @xcite produces larger variations , but always within those reported for the cf88 and nacre rates . using the rate from @xcite on the other hand produces variations outside of the ranges in fig . [ fig7 ] , e.g. , pb is up to 40% lower than the value obtained using the present rate . we chose to plot the elements representing the first @xmath9-process peak ( sr ) , the second @xmath9-process peak ( ba ) , and the third @xmath9-process peak ( pb ) . these can be observed in stars and provide a description of the overall @xmath9-process distribution , which is a function of the total time - integrated neutron flux . we further plot two isotopic ratios : @xmath105zr/@xmath106zr and @xmath107kr/@xmath108kr . these are sensitive to the activation of the branching points at @xmath109zr and @xmath110kr , respectively , and thus to the local value of the neutron density . these ratios can be measured in meteoritic stardust silicon carbide ( sic ) grains that originated in c - rich agb stars . we also report on the production of fluorine , however , we find that its production is not significantly sensitive ( by @xmath11110% at most ) to the choice of the @xmath4c(@xmath0,@xmath7)@xmath8o reaction in all four models ( see also * ? ? ? * ) . in the 1.8 @xmath11 , @xmath1000.01 model we find that the overall @xmath9-process distribution changes when different @xmath4c(@xmath0,@xmath7)@xmath8o rates are used . the present rate produces 16% more ba and 25% more pb than the slower cf88 rate . on the other hand , the present rate results in a 14% lower abundance of pb than the faster nacre rate . these differences are a signature of a different efficiency of the @xmath9-process . a lower efficiency produces less ba and pb than sr because for the neutron capture flux to reach ba and pb higher number of neutrons per fe seed are required . slower rate results in more @xmath4c left over in the pocket to be engulfed in the following tp and the @xmath9-process is less efficient when @xmath4c is engulfed in the tp than when @xmath4c burns radiatively for two reasons . ( i ) in the radiative @xmath4c pocket there is no mixing among the different layers of the pocket and thus @xmath112n nuclei produced in the region of the pocket where the initial number of proton is higher than 0.01 are not mixed to the @xmath4c - rich layer . this produces the highest @xmath9-process efficiency because @xmath112n is a neutron poison via the @xmath112n(@xmath113)@xmath112c reaction , which removes neutrons from being captured by fe seeds and their progeny . in the convective tp instead @xmath4c nuclei are mixed with the @xmath112n present in the pocket as well as in the h burning ashes . ( ii ) the neutrons in the radiative @xmath4c pocket are released over a very small mass ( 0.002 @xmath11 in our models ) , there are more neutrons present locally per initial fe seed and the flux of neutron captures can reach to the heaviest elements up to pb . in the convective tp instead the neutrons are released over a larger mass ( @[email protected] @xmath11 in our models ) , there are less neutrons per fe seed resulting in the production of the lighter @xmath9-process elements like sr . the isotopic ratios plotted in fig . [ fig7 ] behave in the opposite way of pb because the local neutron density is higher , and thus branching points are more activated , when @xmath4c is ingested in the tps due to the higher temperature resulting in a shorter burning timescale . interestingly , the low - metallicity @xmath114 models present the opposite result to the @xmath115 model . e.g. , in the 1 @xmath11 model the present rate produces 25% less pb than the slower cf88 rate . these stellar models have lower masses and a much larger fraction of @xmath4c burns while ingested in the tp , together with @xmath112n , due to both incomplete burning of the @xmath4c pocket before the next tp and the proton ingestion episodes . as mentioned above , @xmath112n acts a neutron poison via the ( @xmath113 ) channel , but at the same time it is destroyed by @xmath0-captures in the tp . hence , the timescale of @xmath4c ingestion and burning with respect to that of @xmath112n has an impact on the final neutron flux . this could explain why slower @xmath4c(@xmath0,@xmath7)@xmath8o rates produce more free neutrons since if neutrons are released at later times there is less @xmath112n to capture them . in these conditions , one should also consider the effect of recycling of the protons made by the @xmath112n(@xmath113)@xmath112c reaction . these protons can be captured by the abundant @xmath16c to make more @xmath4c , but also by @xmath4c itself . it is difficult to evaluate analytically the final outcome of all these combined effects . overall , our numerical models , which solve simultaneously mixing and burning in the tps , indicate that the present rate results in a lower overall neutron flux than the cf88 rate . this is accompanied by a higher or a lower neutron density , depending on the stellar mass and the exact ratio of @xmath4c burning convectively to @xmath4c burning radiatively . 17% lower and 22% higher @xmath107kr/@xmath108kr ratios , which are extremely sensitive to even small changes in the neutron density , are obtained for the 1 @xmath11 and 1.5 @xmath11 models , respectively . in the 3 @xmath11 , @xmath1000.02 model we do not find any variations in the resulting abundances within 5% . this is because in this model @xmath4c always completely burns before the onset of the next tp . the total integrated neutron flux is thus determined only by the initial amount of @xmath4c in the pocket and not by how fast it burns . we expect the same behavior for more massive agb model if a @xmath4c pocket was to be considered in these cases . the isotopic ratios in this model are also not sensitive to the @xmath4c(@xmath0,@xmath7)@xmath8o rate as they depend mainly on the neutrons released in the tps when the @xmath101ne neutron source is marginally activated . it is interesting to compare the predicted abundances of the @xmath9-process elements and isotopic ratios with the observations . we find that some differences up to 25% are predicted by the @xmath9-process models , particularly for pb , when varying the @xmath4c burning rates . pb can be measured in carbon - enhanced metal - poor ( cemp ) stars in the milky way halo , which are believed to carry the chemical signature of mass transfer from a more massive companion during its agb phase . however , the spectroscopic abundances in @xmath9-process - enhanced stars are currently determined with typical uncertainties around 80% ( see table 2 of * ? ? ? * ) , larger than the differences found here . in addition , the formation of the @xmath4c pocket itself is very uncertain and there is no agreement on exactly which mechanism drives it , and on the impact of processes such as rotation and magnetic fields . also the details of the proton ingestion episodes found in our low - metallicity models depend on the treatment of convective borders in stars , one of the largest uncertainties in stellar modeling . future observations with high resolution and high signal to noise ratio spectra of low - metallicity stars , and further understanding of the formation of the @xmath4c pocket and the proton ingestion episodes will be all important for better constraining the @xmath9-process model predictions . isotopic ratios affected by branching points are indicative of the @xmath4c burning rate , even though also in this case , other uncertainties including the neutron - capture cross section of the unstable nucleus at the branching point may have a more important effect . the @xmath105zr/@xmath106zr ratio has been measured in s - type agb stars and in meteoritic stardust sic grains that originated from agb stars of roughly solar metallicity . the data of @xmath105zr/@xmath106zr from meteoritic grains are more precise ( with uncertainties as low as 5%@xmath4510% , * ? ? ? * ) than the data from s - type stars , where often it was only possible to derive upper limits @xcite . the differences of 22% found here when changing the @xmath4c(@xmath0,@xmath7)@xmath8o rate do not drastically change the interpretation of the data , which mostly indicate strong deficits in @xmath105zr , with respect to solar , and are overall matched by agb models @xcite . in the case of @xmath107kr/@xmath108kr , stardust sic data range from @xmath100.5 to 2 of the solar ratio of 1.52 @xcite . our agb models of metallicity around solar ( @xmath104=0.01,0.02 ) , which are believed to well represent the parent stars of this stardust , only produce @xmath107kr/@xmath108kr ratios lower than solar . however , @xcite showed that the @xmath107kr atoms are probably implanted in sic by the high - velocity winds experienced during the evolution that follows the agb , the post - agb and the planetary nebula nucleus phases . the effect of the @xmath4c(@xmath0,@xmath7)@xmath8o reaction rate needs to be evaluated with regards to the composition of kr inside the he intershell , which is exposed to the surface during these final phases of the evolution instead of the surface abundances as done in the present work . in this context the possibility of a proton ingestion in the very late tp sometimes occurring during the post - agb evolution also has to be taken into account . first simulations of this proton ingestion event by @xcite have shown that the @xmath4c(@xmath0,@xmath7)@xmath8o reaction rate plays a main role in determining the final abundances ( see their fig . these models need to be further investigated in relation to the present @xmath4c(@xmath0,@xmath7)@xmath8o reaction rate and the @xmath107kr/@xmath108kr ratios in sic grains . we determined the stellar rate of the @xmath4c(@xmath0,@xmath7)@xmath8o reaction and incorporated the new reaction rate in calculations of the @xmath9-process nucleosynthesis in agb stars . the @xmath1 and anc for the 6.356 mev 1/2@xmath2 subthreshold state in @xmath3o were obtained from the measurement of the @xmath4c(@xmath5b,@xmath6li)@xmath3o angular distribution . this provided an independent examination to shed some light on the existing discrepancies in the @xmath1 and anc values derived from different authors . based on the measured anc , we extracted the @xmath0-width of the 1/2@xmath2 state in @xmath3o , which is currently the most uncertain parameter for determining the @xmath4c(@xmath0,@xmath7)@xmath8o reaction rate . by using the present @xmath0-width and considering the properties of @xmath3o states up to 8.342 mev as well as their interferences , we derived the astrophysical s - factor and the stellar rate of the @xmath4c(@xmath0,@xmath7)@xmath8o reaction . at a temperature of 100 mk the new rate is roughly two times larger than the cf88 value and two times smaller than that recommended by nacre ( see fig . [ fig5 ] ) . verification of the present result using other independent techniques is desirable , e.g. , the trojan horse approach @xcite , and the isospin symmetry approach based on a measurement of the 1/2@xmath2 6.560 mev state in @xmath3f @xcite . in addition , an extension of the experimental data of @xmath4c(@xmath0,@xmath7)@xmath8o toward lower energies is highly desirable , which can probably only be performed in an underground laboratory , e.g. , luna ( see * ? ? ? * ) . we incorporated different @xmath4c(@xmath0,@xmath7)@xmath8o reaction rates in calculations of the @xmath9-process nucleosynthesis in agb stars and found that : ( 1 ) if @xmath4c burns completely in radiative conditions during the interpulse phase ( as for stars of initial mass greater than @xmath102 @xmath11 ) , there is no change in the final results . ( 2 ) if some @xmath4c burns instead inside the convective tps ( for stars of initial mass lower than @xmath102 @xmath11 ) , we find changes of up to 25% in the @xmath9-process results , particularly for pb . there are model uncertainties related to the result of point ( 2 ) : a ) when @xmath4c burning in the tps is due to incomplete burning of the @xmath4c during the interpulse period , the exact stellar mass and metallicity range where incomplete burning of the @xmath4c during the interpulse period occurs , as well as tp numbers , and the exact amount of @xmath4c ingested , all depend sensitively on the temperature and density in the @xmath4c pocket , on the interpulse period , and on the details of the inclusion of the @xmath4c pocket . b ) when @xmath4c burning in the tps is due to ingestion of protons directly inside the tp ( as in low - mass and low - metallicity stars , as well as in post - agb stars experiencing a late tp , see * ? ? ? * ) , the amount of @xmath4c and neutrons produced strongly depends on the physical and numerical treatment of the mixing scheme adopted , which is at present uncertain . due to these model uncertainties together with the fact that the stellar observations have relatively large error bars , it is presently not possible to conclude if the new rate of the @xmath4c(@xmath0,@xmath7)@xmath8o reaction provides the best match to the available observational constraints . this may be however possible in the future , when development of recent 3d hydrodynamical models of the proton ingestion episodes @xcite will allow a better understanding of neutron production and the @xmath9-process inside tps to be compared to the composition of stellar observations and stardust grains . thanks natasha timofeyuk for helpful discussions on dwba calculation and isospin symmetry in mirror @xmath0-decays , and thanks grigory rogachev for useful discussions on the source of the discrepancy between their results , and thanks grigory rogachev and farouz hammache for providing him with their reaction rates in tabular form . m.l . acknowledges the support of the arc via a future fellowship and of monash university via a monash fellowship . we thank the anonymous referee for many helpful comments and suggestions . we acknowledge the staff of tandem accelerator for the smooth operation of the machine . this work was supported by the national natural science foundation of china under grant nos . 11021504 , 10735100 and 11035001 , the national basic research program of china ( new physics and technology at the limits of nuclear stability ) , the outstanding tutors for doctoral dissertations of s&t project in beijing under grant no . + angulo , c. , arnould , m. , rayet , m. et al . 1999 , , 656 , 3 bair , j. k. , & haas , f. x. 1973 , , 7 , 1356 burbidge , e. m. , burbidge , g. r. , fowler , w. a. , & hoyle , f. 1957 , rev . phys . , 29 , 547 busso , m. , gallino , r. , & wasserburg g. j. 1999 , , 37 , 239 busso , m. , gallino , r. , lambert , d. l. , travaglio , c. , & smith , v. v. 2001 , , 557 , 802 brune , c. r. , licot , i. , & kavanagh , r. w. 1993 , , 48 , 3119 caughlan , g. , & fowler , w. 1988 , at . data nucl . data tables , 40 , 283 cook , j. , stephens , m. n. , & kemper , k. w. 1987 , , 466 , 168 costantini , h. , formicola , a. , imbriani , g. , junker , m. , rolfs , c. , & strieder , f. 2009 , rep . 72 , 086301 cristallo , s. , straniero , o. , gallino , r. , piersanti , l. , domnguez , i. , & lederer , m. t. 2009a , , 696 , 797 cristallo , s. , piersanti , l. , straniero , o. , gallino , r. , domnguez , i. , & kppeler , f. 2009b , , 26 , 139 davids , c. n. 1968 , , 110 , 619 denker , a. , drotleff , h. w. , groe , m. , knee , h. , kunz , r. , mayer , a. , seidel , r. , soin , m. , whr , a. , wolf , g. , & hammer , j. w. 1995 , in proc . astrophys . , nuclei in the cosmos iii , 327 , 255 descouvemont , p. 1987 , , 36 , 2206 drotleff , h. w. , denker , a. , knee , h. , soin , m. , wolf , g. , hammer , j. w. , greife , u. , rolfs , c. , & trautvetter , h. p. 1993 , , 414 , 735 gallino , r. , busso , m. , picchio , g. , raiteri , c. m. , & renzini , a. 1988 , , 334 , 45 gallino , r. , arlandini , c. , busso , m. , lugaro , m. , travaglio , c. , straniero , o. , chieffi , a. , & limongi , m. 1998 , , 497 , 388 goriely , s. , & mowlavi , n. 2000 , , 362 , 599 goriely , s. , & siess , l. 2004 , , 421 , l25 harissopulos , s. , becker , h. w. , hammer , j. w. , lagoyannis , a. , rolfs , c. , & strieder , f. 2005 , , 72 , 062801 heil , m. , detwiler , r. , azuma , r. e. , couture , a. , daly , j. , grres , j. , kppeler , f. , reifarth , r. , & tischhauser , p. 2008 , , 78 , 025803 herwig , f. 2004 , , 605 , 425 herwig , f. 2005 , , 43 , 435 herwig , f. 2011 , , 727 , 89 hollowell , d. , & iben , i. , jr . 1988 , , 333 , 25 iliadis , c. 2007 , nuclear physics of stars ( weinheim : wiley - vch verlag gmbh & co. kgaa ) johnson , e. d. , rogachev , g. v. , mukhamedzhanov , a. m. , baby , l. t. , brown , s. , cluff , w. t. , crisp , a. m. , diffenderfer , e. , goldberg , v. z. , green , b. w. , hinners , t. , hoffman , c. r. , kemper , k. w. , momotyuk , o. , peplowski , p. , pipidis , a. , reynolds , r. , & roeder , b. t. 2006 , , 97 , 192701 kppeler , f. , gallino , r. , busso , m. , picchio , g. , & raiteri , c. m. 1990 , , 354 , 630 kppeler , f. , gallino , r. , bisterzo , s. , & aoki , w. 2011 , rev . , 83 , 157 karakas , a. i. , van raai , m. a. , lugaro , m. , sterling , n. c. , & dinerstein , h. l. 2009 , , 690 , 1130 karakas , a. i. , campbell , s. w. , & stancliffe , r. j. 2010 , , 713 , 374 keeley , n. , kemper , k. w. , & khoa , dao t. 2003 , , 726 , 159 kellogg , s. , vogelaar , r. , & kavanagh , r. 1989 , bull . am . soc . , 34 , 1192 kubono , s. , abe , k. , kato , s. , teranishi , t. , kurokawa , m. , liu , x. , imai , n. , kumagai , k. , strasser , p. , tanaka , m. h. , fuchi , y. , lee , c. s. , kwon , y. k. , lee , l. , ha , j. h. , & kim , y. k. 2003 , , 90 , 062501 kurath , d. 1973 , , 7 , 1390 lambert , d. l. 1995 , , 450 , 302 lattanzio , j. c. 1986 , , 311 , 708 lewis , r. , amari , s. , & anders , e. 1994 , geochim . acta , 58 , 471 liatard , e. , bruandet , j. f. , glasser , f. , kox , s. , chan , t. u. , costa , g. j. , heitz , c. , masri , y. , hanappe , f. , bimbot , r. , guillemaud - mueller , d. , & mueller , a.c . 1990 , europhys . lett . , 13 , 401 lugaro , m. , herwig , f. , lattanzio , j. c. , gallino , r. , & straniero , o. 2003a , , 586 , 1305 lugaro , m. , davis , a. m. , gallino , r. , pellin , m. j. , straniero , o. , & kppeler , f. 2003b , , 593 , 486 lugaro , m. , ugalde , c. , karakas , a. i. , grres , j. , wiescher , m. , lattanzio , j. c. , & cannon , r. c. 2004 , , 615 , 934 lugaro , m. , karakas , a. i. , stancliffe , r. j. , & rijs , c. 2012 , , 747 , 2 masseron , t. , johnson , j. a. , plez , b. , van eck , s. , primas , f. , goriely , s. , & jorissen , a. 2010 , , 509 , 93 mukhamedzhanov , a. m. , blokhintsev , l. d. , irgaziev , b. f. , kadyrov , a. s. , cognata , m. , spitaleri , c. , & tribble , r. e. 2008 , j. phys . g. , 35 , 014016 pellegriti , m. g. , hammache , f. , roussel , p. , audouin , l. , beaumel , d. , descouvemont , p. , fortier , s. , gaudefroy , l. , kiener , j. , lefebvre - schuhl , a. , stanoiu , m. , tatischeff , v. , & vilmay , m. 2008 , , 77 , 042801 pignatari , m. , gallino , r. , heil , m. , wiescher , m. , kppeler , f. , herwig , f. , & bisterzo , s. 2010 , , 710 , 1557 rauscher , t. , & thielemann , f .- k . 2001 , at . data nucl . data tables , 79 , 47 rogachev , g. v. 2012 , private communication rudchik , a. a. , rudchik , a. t. , kozeratska , g. m. , ponkratenko , o. a. , koshchy , e. i. , budzanowski , a. , czech , b. , kliczewski , s. , siudak , r. , skwirczyska , i. , szczurek , a. , mezhevych , s. yu . , kemper , k. w. , choiski , j. , czosnyka , t. , & gowacka , l. 2005 , , 72 , 034608 sekharan , k. k. , divatia , a. s. , metha , m. k. , kerekatte , s. s. , & nambiar , k. b. 1967 , phys . , 156 , 1187 spitaleri , c. , aliotta , m. , cherubini , s. , lattuada , m. , miljani , dj . , romano , s. , soic , n. , zadro , m. , & zappal , r. a. 1999 , , 60 , 055802 stancliffe , r. j. , dearborn , d. s. p. , lattanzio , j. c. , heap , s. a. , & campbell , s. w. 2011 , , 742 , 121 straniero , o. , gallino , r. , busso , m. , chiefei , a. , raiteri , c. m. , limongi , m. , & salaris , m. 1995 , , 440 , 85 thielemann , f .- k . , arnould , m. , & truran , j. 1987 , in advances in nuclear astrophysics , ed . a. vangioni - flam ( gif - sur - yvette : editions frontire ) , 525 thompson , i. j. 1988 , comput . , 7 , 167 tilley , d. r. , weller , h. r. , & cheves , c. m. 1993 , , 565 , 1 timofeyuk , n. k. , descouvemont , p. , & johnson , r. c. 2007 , , 75 , 034302 van raai , m. , lugaro , m. , karakas , a. i. , garcia - hernandez , d. a. , yong , d. 2012 , , accepted vassiliadis , e. , & wood , p. r. 1993 , , 413 , 641 verchovsky , a. b. 2004 , , 607 , 611 cccccccc channel & @xmath116 & @xmath117 & @xmath118 & @xmath119 & @xmath120 & @xmath121 & @xmath122 + @xmath5b+@xmath4c & 50.0 & 182.64 & 0.788 & 0.740 & 8.193 & 1.250 & 0.740 + @xmath6li+@xmath3o & 26.0 & 114.20 & 0.737 & 0.719 & 34.602 & 0.997 & 0.764 + @xmath6li+@xmath4c & 31.8 & 159.00 & 0.630 & 0.810 & 8.160 & 1.330 & 0.780 + ccccc reference & transfer system & incident energy&@xmath1 & @xmath61 + & & ( mev ) & & ( @xmath62 ) + @xcite & @xmath4c(@xmath23li,@xmath24)&60&0.011 & + @xcite & @xmath4c(@xmath23li,@xmath24)&60&[email protected] & + @xcite & @xmath23li(@xmath4c,@xmath24)&8.0 and 8.5&&0.89 @xmath21 0.23 + @xcite & @xmath4c(@xmath6li,@xmath25)&28 and 34&0.29 @xmath21 0.11&4.5 @xmath21 2.2 + present work & @xmath4c(@xmath5b,@xmath6li)&50&0.37 @xmath21 0.12&4.0 @xmath21 1.1 + ccccccc level&@xmath22 ( kev ) & @xmath90 ( kev ) & @xmath123 & @xmath124 ( kev ) & @xmath125 ( kev)&@xmath85 ( kev ) + 1&6356 & @xmath453 & 1/2@xmath2 & 124 & & 12.7 + 2&7165&806&5/2@xmath49&1.38&0.0033 & + 3&7201&842&3/2@xmath2&340&0.08 & + 4&7378&1019&5/2@xmath2&0.64&0.01 & + 5&7381&1022&5/2@xmath49&0.96&0.003 & + 6&7558&1199&3/2@xmath49&500&0.08 & + 7&7688&1329&7/2@xmath49&13&0.01 & + 8&7956&1597&1/2@xmath2&84&6.7 & + 9&7991&1632&1/2@xmath49&250&16 & + 10&8058&1699&3/2@xmath2&71&15 & + 11&8200&1841&3/2@xmath49&48&4.0 & + 12&8342&1983&1/2@xmath2&8.1&2.2 & + .[tab4 ] numerical values of the present @xmath4c(@xmath0,@xmath7)@xmath8o rates ( @xmath126/mol / s ) with the adopted value , upper and lower limits for the [email protected] gk temperature range and the coefficients @xmath97 in eq . [ eq8 ] for these three rates . the overall fitting errors are all less than 7% at temperatures from 0.04 to 10.0 gk . [ cols="^,^,^,^,^,^,^,^ " , ]

when two types of side chains ( a , b ) are densely grafted to a ( stiff ) backbone and the resulting bottle - brush polymer is in a solution under poor solvent conditions , an incompatibility between a and b leads to microphase separation in the resulting cylindrical brush . the possible types of ordering are reminiscent of the ordering of block copolymers in cylindrical confinement . starting from this analogy , leibler s theory of microphase separation in block copolymer melts is generalized to derive a description of the system in the weak segregation limit . also molecular dynamics simulation results of a corresponding coarse - grained bead - spring model are presented . using side chain lengths up to n = 50 effective monomers , the ratio of the lennard - jones energy parameter between unlike monomers and monomers of the same kind is varied . various correlation functions are analyzed to study the conditions when ( local ) janus cylinder - type ordering and when ( local ) microphase separation in the direction along the cylinder axis occurs . both the analytical theory and the simulations give evidence for short range order due to a tendency towards microphase separation in the axial direction , with a wavelength proportional to the side chain gyration radius , irrespective of temperature and grafting density , for a wide range of these parameters .

1105.1949

enabled by progress in chemical synthesis ( see @xcite for reviews ) , macromolecules with `` bottle - brush '' architecture where flexible side chains are densely grafted to a backbone , have found much recent interest ( e.g. @xcite ) . such molecules may be useful for various applications ( such as sensors , actuators , building blocks in supramolecular structures @xcite ) , since these systems are stimuli - responsive polymers , exhibiting large conformational changes when external conditions vary . apart from these synthetic bottle - brush polymers , also biopolymers with a related architecture are abundant in nature , e.g. proteoglycans @xcite . these brush - like polymers contain a protein backbone with carbohydrate side chains , and are held responsible for a large variety of biological functions ( cell signaling , cell surface protection , joint lubrication , etc . @xcite ) . at the same time , the interplay between entropic effects and various enthalpic forces in these `` soft '' objects makes the understanding of structure - property relationship of such bottle - brush polymers a challenging problem of statistical thermodynamics @xcite . in the present paper , we focus on the interplay between chain conformations and local order in binary a , b bottle - brush polymers , from the point of view of both theoretical arguments and molecular dynamics simulations . previous work @xcite has focused on the possibility of microphase separation in the form of `` janus cylinders '' @xcite , i.e. , the cylinder splits in two halves ( in the most symmetric case , where the lengths @xmath2 of both types of side chains and their grafting densities @xmath3 are equal ) , such that the a - b - interface contains the cylinder axis ( taken to be along the @xmath4-axis henceforth ) . then it was pointed out @xcite that due to the quasi - one - dimensional character of this ordering , for finite cylinder radius r , no true long - range order of janus - cylinder - type should be expected : rather there exists a finite correlation length over which the orientation of the interface plane decorrelates . in addition , it was speculated @xcite that also the ratio @xmath5 between the interaction strength @xmath6 of unlike monomers and monomers of the same kind @xmath7 matters : if the formation of a - b interfaces is energetically much more unfavourable rather than the formation of polymer - solvent interfaces , a phase separation along the axial ( z ) direction of the cylinder could occur also in the form of a double cylinder ( the a - rich polymers form a separate cylinder from the b - rich ones , both cylinders touch each other along the z - axis ) @xcite . also intermediate cases ( `` janus dumbbell''-like cross sections of the bottle - brush ) were suggested @xcite and some indication of such structures were observed @xcite . however , again only finite correlation lengths of such orderings along the z - axis can be expected . in addition , one needs to consider that for not very large grafting densities the bottle - brush polymer under poor solvent conditions does not form a structure that is homogeneous along the z - axis ( the direction of the backbone ; note that we restrict attention to rigid backbones only , for simplicity ) , but rather inhomogeneities in the form of a `` pearl - necklace''-structure form @xcite . the problem which we investigate in the following sections is whether it can be favorable to form a microphase - separated structure that is inhomogeneous along the z - axis . following the information from experiment that one side - chain can be grafted per backbone monomer @xcite , we assume an alternating grafting of the two types ( a and b ) of side chains @xcite . of course , this situation is essentially equivalent to the case where symmetrical block copolymers @xcite would be anchored with their a - b junction to the backbone . it is then also interesting to ask how such a situation compares to the case where such block copolymers fill a cylinder ( at the same density ) without constraint on the locations of the a - b junctions . of course , the latter problem has already been studied in other contexts @xcite . here we generalize the treatment of ref . @xcite to take this constraint on the location of the a - b junctions into account to formulate a weak segregation theory @xcite of ordering in binary bottle - brushes ( sec . 2 ) . then we present data from molecular dynamics simulations , extending the study of ref . @xcite by considering now the variation of properties with @xmath5 ( sec . we summarize the main findings of our work in the concluding section ( sec . 4 ) . the general idea of our theoretical approach is as follows . we consider the case when incompatibility of both sorts of blocks with the solvent is high enough to form a condensed state ( globule ) , which in the case of a rigid backbone would acquire the form of a cylinder uniformly filled by the monomers of the side chains . the height @xmath8 of the cylinder would be just the length of the rigid backbone and the radius @xmath9 would be related to the equilibrium monomer density @xmath10 inside the cylinder . such a condensed state could be considered as an equivalent of a bottlebrush compressed up to the monomer density @xmath10 via squeezing it into a tube of the radius @xmath9 . to consider the composition fluctuations in such a condensed ( squeezed ) bottlebrush we employ some new ideas , which were put forward in refs @xcite to generalize the seminal leibler theory @xcite ( see also refs @xcite ) to confined polymer systems . to begin with , we find , based on the theory of polymer globules by i. lifshitz et al . @xcite , the value of @xmath10 , which is to provide the minimum of the two - component bottlebrush free energy given the values of the chi - parameters @xmath11 ( a similar approach has been employed by borisov et al . @xcite for one - component bottlebrushes in a poor solvent ) . to determine the equilibrium value of a two - component bottlebrush forming a condensed state due to its overall incompatibility with the solvent we notice that , according to i. lifshitz et al @xcite , the free energy of any flexible polymer system is ( in a zeroth approximation ) just the sum of a structural ( entropic ) contribution , which allows for the effect of connectivity ( `` linear memory '' in lifshitz terms ) and an energetic contribution , which allows for the van - der - waals attraction and the excluded volume repulsion effects : @xmath12 the precise form of the structural free energy @xmath13 as a functional of the spatially non - unform distributions @xmath14 of the local number densities of the repeat units of the @xmath15-th sort ( @xmath16 ) is determined by the actual microscopic structure of the bottlebrush . we analyze it somewhat later when studying the bottlebrush stability with respect to its longitudinal , angular or helical ordering . but now , when we focus on the basically uniform cylindrical globule , the structural term could be simply disregarded as compared to the energetic contribution to the free energy : @xmath17 here @xmath18 is the volume of the cylindrical globule ( where @xmath19 and @xmath20 are the total number of both @xmath21 and @xmath22 side chains and the distance between the neighboring @xmath21 and @xmath22 chains ) , @xmath23 is the value of the local number density of the repeat units of the @xmath15-th sort , averaged over the whole volume @xmath24 of the bottlebrush , and the specific ( per unit volume ) free energy @xmath25 depends on the type of the interactions between the solvent molecules and repeat units ( monomers ) forming the bottlebrush . for simplicity , we assume in this paper that these interactions correspond to the conventional incompressible flory - huggins lattice model , in which case @xmath26 where @xmath27 is the excluded volume assumed , for simplicity , to be the same for both sorts of the repeat units and solvent molecules , @xmath28 is the local volume fraction of the particles of the @xmath15-th sort within the globular bottlebrush , @xmath29 is the total polymer volume fraction ( volume fraction of all monomers ) , @xmath30 is the volume fraction of the solvent molecules and @xmath11 are the conventional flory - huggins energetic parameters . for the homogeneous bottle brush we are considering here , @xmath31 , where @xmath32 is the fraction of the monomers of the @xmath15-th sort , and , thus , the function @xmath33 can be rewritten in the form @xmath34 where @xmath35 is the effective @xmath36-parameter describing the overall bottlebrush - solvent incompatibility and the parameter @xmath37 does not affect the equilibrium value of @xmath38 . indeed , taking into account that the polymer volume fraction within the uniform bottlebrush is @xmath39 where @xmath40 is the total average number of all monomers @xmath21 and @xmath22 per the distance @xmath41 of the backbone , one can reduce the total free energy ( [ freev ] ) to the form @xmath42 . \label{free0}%\ ] ] minimization of the free energy ( [ free0 ] ) with respect to @xmath38 results , finally , in the desired equation ( it corresponds to the so - called volume approximation of the polymer globule theory @xcite ) for the equilibrium polymer volume fraction of the two - component bottlebrush : @xmath43 obviously , the finite equilibrium polymer volume fraction of the two - component bottlebrush @xmath44 , which is the solution of eq ( [ phi_bar ] ) , depends on the value of @xmath45 only and has a physically meaningful positive value only for @xmath46 the corresponding asymptotics read @xmath47{c}% \phi=3\tau-(27/4)\tau^{2}+ ... ,\quad\;\tau\ll1\\ \phi=1-\exp\left ( 1+\widetilde{\chi}\right ) , \quad\qquad\chi\gg1 \end{array } \right . \label{asym1}%\ ] ] it is worth to notice that in so - called hildebrand approximation @xcite , in which the value of the @xmath36-parameter for the @xmath15-th and @xmath48-th species is proportional to the squared difference of their solubility parameters : @xmath49 the effective @xmath36-parameter is always positive:@xmath50 where @xmath51 is the selectivity parameter @xcite . if the @xmath52 and @xmath22 monomers are compatible ( @xmath53 ) then @xmath54 . to conclude this subsection we discuss the conditions of validity of our consideration . first , it is worth to notice that the bottlebrushes under consideration are quasi-1d systems . therefore , their equilibrium volume fraction stays finite even in the @xmath55-solvent , where @xmath56 ( here @xmath57 is the statistical segment supposed , for simplicity , to be the same for both sorts of the side chains ) . indeed , it follows from eq ( [ phi ] ) that in this case @xmath58 where we introduced the reduced distance @xmath59 between the neighboring side chains and the lifshitz number @xmath60 , which characterizes the chains flexibility . comparing eq ( [ theta ] ) and the first of eqs ( [ asym1 ] ) and taking into account that the condensed bottlebrush should be much denser than that in @xmath55-solvent we conclude that the first of the validity conditions reads @xmath61 on the other hand , the volume approximation ( [ phi_bar ] ) is known @xcite to be valid when @xmath62 where @xmath63 is the correlation radius within the globule . for dense globules ( @xmath64 or @xmath65 ) the condition holds always but for semidilute globule ( @xmath66 ) the correlation radius is known @xcite to increase with decrease of @xmath67 : @xmath68 it follows from eqs ( [ phi ] ) , ( [ asym1 ] ) and ( [ rc ] ) that close to the @xmath55-temperature the condition ( [ val1 ] ) is satisfied only when the side chains are long enough : @xmath69 condition ( [ val2 ] ) is rather severe since both quantities @xmath70 and @xmath67 are small . thus , the fact that the monomer density profile usually observed in computer simulation of bottlebrushes in poor solvents close to the @xmath55-temperature is not step - like , which is expected for a condensed ( globular ) state , but rather smooth @xcite is explained by insufficiently high degrees of polymerization of the side chains , which do not satisfy condition ( [ val2 ] ) . to consider the fluctuations within the two - component condensed bottlebrush we are to expand the total free energy ( [ freel ] ) in powers of the density fluctuations counted from the uniform globular state ( @xmath71 = @xmath72 ) : @xmath73 where @xmath74 is defined by eqs ( [ freev ] ) and ( [ free0 ] ) , the quadratic term of the expansion reads @xmath75 ( in eq ( [ f2a ] ) and thereafter we employ the rule of summation over the repeating indices ) and the next terms of the expansion ( [ f2 ] ) are defined similarly @xcite . our purpose in this subsection is to find the explicit form of the kernel @xmath76 and conditions ensuring that the quadratic form @xmath77 is positive definite , which implies that the fluctuations are small . it follows from eqs ( [ freel ] ) - ( [ f_sp ] ) and ( [ f2a ] ) that @xmath78 here we introduced new designations @xmath79@xmath80 where the number matrix @xmath81 reads @xmath82{cc}% 2\chi_{as}-\phi_{s}^{-1 } & k-\phi_{s}^{-1}\\ k-\phi_{s}^{-1 } & 2\chi_{bs}-\phi_{s}^{-1}% \end{array } \right ) \label{c}%\ ] ] and @xmath83 . thus , our problem is reduced to calculating the second derivative ( [ gam ] ) of the structural free energy functional . for this purpose we consider an auxiliary thermodynamic potential @xmath84 which has meaning of the free energy of an ideal polymer system with a specified architecture ( in our case it is the two - component bottlebrush ) affected by a set @xmath85 of external fields applied to the particles of the @xmath15-th sort . the thermodynamic potential @xmath86 could be readily calculated ( see below ) . on the other hand , it is directly related to the desired thermodynamic potential @xmath87 . indeed , the partition function @xmath88 appearing in eq ( [ phi ] ) is the integral over all possible non - uniform density distributions @xmath89 of the particles of the @xmath15-th sort in the volume of the system : @xmath90 with@xmath91 definition and calculation of the density integral @xmath88 is , generally , rather cumbersome but it becomes trivial in the saddle - point approximation:@xmath92 it follows from ( [ field_pot ] ) that if the thermodynamic potential @xmath93 is known then the thermodynamic potential @xmath94 is parametrically defined as follows : @xmath95 @xmath96 eqs ( [ l1 ] ) and ( [ l2 ] ) imply that the thermodynamic potential @xmath97 , which is a functional of all external fields @xmath98 applied to the particles of the system , is the legendre transform of the thermodynamic potential @xmath99 , which is a functional of all number densities @xmath89 of these particles . therefore , the reciprocal relationships hold : @xmath100 @xmath101 it follows from eqs ( [ f2a ] ) and ( [ l1])-([rl2 ] ) that @xmath102furthermore , if we define the structural matrix @xmath103 we obtain @xmath104 the derivation of the explicit expression for the structural matrix @xmath105 of copolymer systems under confinement is a straightforward extension of that in bulk . so , we remind the reader how the structural matrix is derived in bulk . instead of calculating the density integral ( [ z ] ) we consider an equivalent expression for the partition function @xmath106 of an ideal polydisperse @xmath107-component system of linear macromolecules affected by the external fields @xmath108 applied to the monomers of the @xmath15-th sort . the partition function @xmath106 can be written directly in terms of the corresponding discrete microscopic model : @xmath109 here @xmath110 is the macromolecular structure , which for linear chains is just a sequence of integers @xmath111 equal to the number @xmath112 of the sort of the @xmath15-th monomer , @xmath113 is the number of chains with structure @xmath110 in the whole volume @xmath24 of the system and @xmath114 is the partition function of the chain @xmath110 : @xmath115 where @xmath116 is a point of the configuration space of the chain @xmath110 and @xmath117 is the distribution function of the ideal chain @xmath110 , the effect of connectivity being taken into account by the bond functions @xmath118 , which describe the correlations between the neighboring monomers of the chains . for simplicity , we assume that the bond function does not depend on the sorts of the neighboring monomers and has the conventional normalized form @xmath119 where @xmath57 is the statistical segment length . thus , in the thermodynamic limit @xmath120 the thermodynamic potential @xmath121 reads@xmath122 where @xmath123 additivity of @xmath124 with respect to the terms characterizing macromolecules with different chemical structure implies that the matrix ( [ gam_1 ] ) is a properly weighted sum of molecular response functions:@xmath125 here @xmath126 is the average number density of the macromolecules @xmath110 and @xmath127 where we employed the definitions ( [ z1 ] ) and ( [ def1])@xmath128 summing the second field derivatives of @xmath129 and collecting similar terms one gets@xmath130 where @xmath131 @xmath132 is the number of the monomers of the @xmath15-th sort , which belong to the macromolecules @xmath110 , and @xmath133 are the numbers of pairs of monomers of the @xmath112-th and @xmath134-th sorts , which are separated by @xmath107 bonds . the structural matrix , which was first introduced in the form ( [ gab ] ) by one of us @xcite ( see also @xcite and references therein ) , is a discrete analog of the expressions for the correlators obtained in the continuous limit ( like the debye function for homopolymer and the leibler expressions @xcite for diblock copolymers ) . to calculate the structural matrix explicitly we rewrite it as the inverse fourier transform:@xmath135 where the direct fourier transform @xmath136 is an algebraic function of the fourier transform of the bond function:@xmath137 the explicit finite expressions for homo- and some block copolymers can be found in refs @xcite and @xcite . now , we make use of the fact that @xmath138 and @xmath139 are the eigen values and eigen functions of the laplace operator @xmath140 acting in the infinite space . the natural idea @xcite to generalize expression ( [ rb ] ) for the structural matrix to confined copolymers is to rewrite it in the form of a somewhat more general series@xmath141 where now @xmath142 and @xmath143 are the full sets of the eigen values and the orthonormalized eigen functions of the laplace operator @xmath140 acting in a finite volume under specified boundary conditions and @xmath144 is a matrix - function of the parameter @xmath145 . then the kernels @xmath146 and @xmath147 read@xmath148 which results in the following final expression for the desired kernel @xmath149 : @xmath150 with @xmath151 which is a generalization of the well known rpa equation for polymers in bulk : @xmath152 we conclude that the condition ensuring that the quadratic form @xmath77 is positive definite is that ensuring that the matrix @xmath153 has positive eigen values for any allowable value of @xmath145 . in particular , for confinement in a cylindrical capillary of the radius @xmath9 the eigen values and the eigen functions of the laplace operator read : @xmath154 where @xmath155 is the radius - vector in cylindrical coordinates , @xmath156 is a normalization factor and @xmath157 is location of the @xmath158-th extremum of the bessel function @xmath159 of the order @xmath107 accordingly , the quadratic form ( [ f2a ] ) takes the form @xmath160 where the coefficient @xmath161 reads @xmath162 for future comparison with ordering in two - component bottlebrushes under investigation we present here the explicit form of the structural matrix for diblock copolymers @xmath163:@xmath164 here @xmath165 , @xmath166 , @xmath167 , @xmath168 , @xmath169is location of the @xmath158-th extremum of the bessel function @xmath159 of the order @xmath107 , parameter @xmath170 is defined by eq ( [ val1 ] ) and we introduced functions @xmath171 the derivation presented above for the confined block copolymers can be now extended to the tethered copolymers . for this purpose it is sufficient to replace the expression ( [ zm ] ) for the partition function of the corresponding confined system by that of the tethered system under consideration . in this paper we consider for definiteness the two - component bottlebrush consisting of @xmath19 pairs of @xmath21 and @xmath22 side chains regularly distributed along the backbone . the partition function of such an ideal bottlebrush affected by the external fields @xmath172 reads : @xmath173 where @xmath174 is the number of side chains of each sort , @xmath175 is the partition function of an @xmath176-mer , which is attached by one of its ends to the rigid backbone aligned along the @xmath4-axes at the point @xmath177 and affected by an external field @xmath178 : @xmath179 here @xmath180 and @xmath181 is the bond function defined above by eq ( [ bond ] ) , the statistical segment length @xmath57 being the same for @xmath21 and @xmath22 chains . then the thermodynamic potential @xmath94 reads@xmath182 where @xmath183 and @xmath184 additivity of @xmath124 with respect to the terms depending on different external fields implies that the matrix ( [ gam_1 ] ) of the molecular response functions contains diagonal elements only:@xmath185 on the other hand , @xmath186@xmath187 where we employed the definitions ( [ z_s ] ) and ( [ def_s ] ) . summing the second field derivatives of @xmath188 and collecting the similar terms one gets@xmath189 where@xmath190@xmath191@xmath192 and the kernel @xmath193is defined by eq ( [ def_n ] ) . based on the previous consideration , we should use for @xmath194 the representation @xmath195 with the eigen values @xmath145 and eigenfunctions @xmath196 defined by eq ( [ tube ] ) . however , unlike the situation with free ( untethered ) copolymers under confinement , the correlators ( [ gii ] ) for bottlebrushes are , in general , not reducible to the form ( [ series ] ) . accordingly , the quadratic form ( [ f2a ] ) does not take the desired diagonal form ( [ f2 t ] ) . fortunately , it could be shown that for the condensed state the correlators can be expanded in powers of the small parameter @xmath170 defined by eq ( [ val1 ] ) . therewith , keeping only the dominant ( zeroth approximation in @xmath170 ) term of this expansion one arrives precisely at the diagonal form ( [ f2 t ] ) : @xmath197 with@xmath198 where the matrix @xmath199 is defined in eq ( [ c ] ) and @xmath200@xmath201 with all parameters and functions defined after eqs ( [ g_dib ] ) . the general condition of the multi - component block copolymer disordered state stability with respect to microphase separation is well known @xcite . for block copolymers confined in a cylindrical tube and the condensed two - component bottlebrushes it should be reformulated as follows : @xmath202 where @xmath203 are the eigen values of the correlation matrix @xmath204 defined by eq ( [ g2bb ] ) given the values of the `` quantum numbers '' @xmath107 and @xmath158 characterizing the types of the angular and radial concentration modulation and wave number @xmath205 characterizing the wave length of the longitudinal concentration oscillation @xcite , and the minimum is to be sought on the set of all possible values of @xmath107 and @xmath158 and on the half - axes @xmath206 . accordingly , the condition , which determines the surface ( in the space of the structure and energetic parameters of the system under study ) where the uniform state of the latter loses its stability ( spinodal ) , reads@xmath207 where we label with asterisks those values of @xmath208 that actually satisfy eq ( [ spin ] ) . based on the wst experience @xcite one can arrive readily at the following conclusions . _ i _ ) the value of @xmath209 defines the type of the angular modulation of the morphology arising after the uniform state looses its stability . in particular , if @xmath210 then a radially symmetric morphology arises , if @xmath211 then the janus - cylinder - type morphology would occur , if @xmath212 then the cross - like @xmath213 morphology would arise etc . _ ii _ ) if @xmath214 then the arising morphology is expected to be homogeneous along the backbone direction , otherwise a 1d morphology periodic along this direction would occur with the period @xmath215 . 3 ) the value of @xmath158 ( more precisely , @xmath157 ) defines the type of the radial modulation of the arising morphology . ( it is worth to notice that such a straightforward extension of the results of the bulk stability analysis to the quasi - one - dimensional ordering in tubes and bottlebrushes is only approximate @xcite as mentioned above . we address this issue in more detail elsewhere . ) to demonstrate the general results let us compare ordering in bottlebrushes and free diblock copolymers confined in a cylindrical tube . it follows from eqs ( [ c ] ) and ( [ g2bb ] ) - ( [ g_bbab ] ) that for bottlebrushes @xmath216{cc}% \left ( g_{aa}^{\text{bb}}\right ) ^{-1}-2\widetilde{\chi}_{as } & \phi_{s}^{-1}-k\\ \phi_{s}^{-1}-k & \left ( g_{bb}^{\text{bb}}\right ) ^{-1}-2\widetilde{\chi}_{bs}% \end{array } \right ) \label{ggt}%\ ] ] whereas for free diblocks in a tube @xmath216{cc}% \left ( \mathbf{g}^{\text{free}}\right ) _ { aa}^{-1}-2\widetilde{\chi}_{as } & \left ( \mathbf{g}^{\text{free}}\right ) _ { ab}^{-1}+\phi_{s}^{-1}-k\\ & \\ \left ( \mathbf{g}^{\text{free}}\right ) _ { ba}^{-1}+\phi_{s}^{-1}-k & \left ( \mathbf{g}^{\text{free}}\right ) _ { bb}^{-1}-2\widetilde{\chi}_{bs}% \end{array } \right ) \label{ggb}%\ ] ] with the components of the matrix @xmath217 defined in eq ( [ g_dib ] ) , @xmath83 and @xmath218 . for symmetric blocks the eigen vectors of the matrices ( [ ggt ] ) and ( [ ggb ] ) are @xmath219 and @xmath220 that correspond to the order parameters @xmath221 . the corresponding eigen values are @xmath222 for bottlebrushes ( with @xmath223 ) and @xmath224 for free diblocks in a tube . substituting eqs ( [ g_dib ] ) and ( [ g_bb0 ] ) - ( [ g_bbab ] ) into expressions ( [ l- ] ) and ( [ l ] ) we see that the functions @xmath225 are the same for free and tethered diblocks ( bottlebrushes ) in the case of a purely longitudinal ordering ( @xmath226 ) ( see the well known @xcite bold solid line _ 1 _ in fig . 1 with a minimum at @xmath227 ) . it is worth to emphasize that the location of the minimum depends only on the length @xmath176 of the side block ( which scales the dimensionless quantity @xmath228 ) and does depend neither on the grafting density nor on the value of the @xmath36-parameter ( temperature ) . this surprising similarity is due to the very nature of our approximation , which is valid only for strongly compressed bottlebrushes ( @xmath229 ) . indeed , in this case the entropic loss due to fixing one of the chain ends on the backbone is negligible as compared to that due to chain squeezing into a narrow tube , which is the same for free and tethered chains . for other concentration modes having no angular but some radial modulation ( @xmath230 ) one has @xmath231 ( see definitions after eq ( [ g_dib ] ) . in other words , for these modes the instability curve @xmath232 looks like that for the lamellar - like mode ( @xmath233 ) but shifted by @xmath234 ( thin solid lines _ 1 _ and _ 2 _ in fig . 1 ) . therefore , for the radially symmetric modes with @xmath235 the function @xmath236 reaches its minimum for @xmath237 . for @xmath238 , which is surely true within our approximation valid for @xmath229 , the lamellar - like mode is the only one for which the function @xmath239 reaches its minimum at a finite value of @xmath240 . moreover , in this case the lamellar mode is dominant among those with no angular modulation . indeed , as shown in fig . 2 , it is only the square averaged amplitude of this mode , which strongly increases when @xmath241 the situation is rather different for the modes with angular modulation ( @xmath242 ) . the relationship ( [ lam- ] ) holds for these modes of free diblock copolymers in a tube also but for bottlebrush angular concentration modes one has @xmath243 in other words , the instability condition @xmath244 looks like that for the mixture of free blocks a and b but shifted by @xmath245 ( see the dashed lines in fig . 1 . ) . as described above , passing the instability threshold ( [ spin ] ) results in formation of a lamellar - like or janus - cylinders morphology depending on which function , ( [ l- ] ) or ( [ l+ ] ) , has a deeper minimum . as shown in fig . 3 , if the shift @xmath245 is small then the instability with respect to forming of janus - cylinders appears first ( with increase of @xmath246 ) . with decrease of @xmath170 the shift increases and we arrive at the value of @xmath247 where both the lamellar and first angular modulated mode loose their stability simultaneously at @xmath248 . with further decrease of @xmath170 the lamellar phase becomes more thermodynamically stable . it follows from eq ( [ val1 ] ) that @xmath170 decreases when the distance @xmath20 between the neighboring side chains of the bottlebrush increases ( and , thus , grafting density decreases ) . thus , the janus - cylinders could be formed , in agreement with predictions of refs @xcite in bottlebrushes with high grafting density . in bottlebrushes with lower grafting density we predict , contrary to refs @xcite , longitudinal lamellar - like ( pearl - necklace ) ordering rather than forming janus - cylinder . it is worth to emphasize that our theory predicts a deep similarity between the polymer - polymer and polymer - solvent ordering in condensed bottlebrushes . indeed , comparison of eqs ( [ l- ] ) and ( [ l+ ] ) leads to conclusion that in symmetric bottlebrush the only difference in formation of the polymer concentration modulation described by the order parameter @xmath249 and solvent concentration modulation described by the order parameter @xmath250 is energetic . more precisely , depending on whether the limit ( [ ab ] ) is reachable earlier or later than @xmath251 the polymer - polymer or polymer - solvent inhomogeneities will occur but the type of morphology arising in both cases is expected to be the same . the strictly rigid backbone of the simulated bottle - brush polymer is simply taken as an immobile straight line in @xmath4-direction , where we also apply periodic boundary conditions , thus disregarding any end - effects . using units of length such that the length parameter @xmath252 of the lennard - jones potential ( see below ) is taken to be unity , @xmath253 , we graft side chains ( alternatingly of type a and type b ) at regular positions @xmath254 , with @xmath255 and a number 2 m = 50 of grafted chains . the side chain length was chosen as n = 35 . for comparison , also some results were obtained for a higher grafting density @xmath256 and two lower ones @xmath257 and @xmath258 ) ; in the latter case , three chain lengths were studied @xmath259 , @xmath260 , and @xmath261 , respectively ) . we are fully aware of the fact that our side chain lengths would not be long enough to study the asymptotic scaling behavior under good solvent conditions @xcite , of course ; however , for poor solvent conditions ( as studied here ) very large relaxation times would prevent us from reaching thermal equilibrium for longer side chains . moreover , the range of n that we explore nicely corresponds to the range that can be probed in experiments @xcite . the side chains are modeled by the standard bead - spring model @xcite already used in our previous work @xcite . all beads interact with a truncated and shifted lennard - jones potential@xmath262{r@{\ ; , \;}l}% 4\epsilon_{lj}[(\sigma_{lj}/r)^{12}-(\sigma_{lj}/r)^{6}]+c & r\leq r_{c}\\ 0 r_{c}% \end{array } \right.\ ] ] where @xmath263 , and the constant c is defined such that @xmath264 is continuous at the cutoff . units are chosen such that @xmath253 irrespective of whether we deal with aa , bb or ab pairs , and also boltzmann s constant @xmath265 , as well as the masses of the beads @xmath266 . for the energy parameters , however , we choose @xmath267 respectively . remember that for @xmath268 the problem reduces to a single - component brush . as usual @xcite the connectivity of the beads along a side chain is maintained by the `` finitely extensible nonlinear elastic '' ( fene ) potential , @xmath269 , \quad0<r \leq r_{0}\;,\ ] ] with the standard choice of parameters , @xmath270 , and @xmath271 , respectively . since @xmath272 the bond lengths are constrained . simulations were carried out controlling temperature by the langevin thermostat @xcite . the equation of motion of a bead then reads @xmath273 where @xmath274 denotes the time , @xmath275 is the total potential acting on the i - th bead ( which is at the position @xmath276 ) , @xmath277 is the friction coefficient , and @xmath278 the random force . as is well known , @xmath277 and @xmath278 are related by the fluctuation - dissipation relation @xmath279 eqs . [ eq4],[eq5 ] are numerically integrated using the gromacs package , applying the leap frog algorithm @xcite . following previous work @xcite , @xmath280 was chosen , and a time step @xmath281 where the molecular dynamics(md ) time unit was chosen as @xmath282 . as was briefly mentioned above , equilibration of our model system is rather difficult . typically we started by equilibrating the system first at a temperature @xmath283 , which is known @xcite to correspond roughly to the theta temperature of homopolymers in dilute solution . 500 different configurations at @xmath283 were equilibrated using runs extending over a time range of @xmath284 . all these configurations are then cooled down from @xmath283 to @xmath285 with temperature steps @xmath286 . at each temperature @xmath287 steps were used for further equilibration . as an example , figs . 4 and 5 show typical snapshot pictures of the simulated bottle brushes . at @xmath283 the individual side chains take irregular conformations , stretching out in the directions perpendicular to the backbone . obviously , these conformations are rather disordered , the contacts between monomers of different chains are too scarce to induce significant microphase separation , and even on local scales the composition of the bottle brush is rather randomly mixed . at @xmath288 the situation is different , however : for @xmath289 one can clearly see the presence of a pearl - necklace type structure , as far as the total density is concerned ; typically these pearls sometimes are almost exclusively formed from one type of chain only , implying a periodic concentration variation along the @xmath4-axis . for @xmath258 , however , the interpretation of many snapshot pictures rather suggest that `` janus pearls '' ( containing an a - rich and a b - rich part , separated by an interface ) have formed . for @xmath255 , already rather elongated a - rich and b - rich parts ( reminiscent of `` janus cylinders '' ) are visible , but there still occur strong undulations in the local monomer density in @xmath4-direction . these undulations for @xmath290 clearly are somewhat weaker , although the density variation still is far from a uniformly filled cylinder . while some snapshots suggest that a `` janus cylinder '' type ordering persists over the full lengths of the studied backbone length , we emphasize the caveat that in figs . 4 , 5 one can not really distinguish whether a local cross section perpendicular to the cylinder axis resembles more a sphere ( as appropriate for a cylinder ) or a dumbbell , for instance . in order to provide a quantitative analysis of the various types of short range order we define correlation functions adapted to the cylindrical symmetry of our problem , by suitable generalization of the two - point correlation function in fluid systems , @xmath291 in eq . [ eq51 ] the first sum runs over all @xmath292 particles of type @xmath112 , which have the cylindrical coordinates @xmath293 , while the second some runs over all @xmath294 particles of type @xmath134 , which have the cylindrical coordinates @xmath295 , @xmath296 . 6 shows this correlation function @xmath297 for a typical case : as expected , in the temperature region close to the @xmath298-temperature this correlation decays to its asymptotic value for large @xmath299 , @xmath300 if @xmath301 , while a nontrivial angular dependence occurs for smaller @xmath299 . basically , this structure is due to intra - chain - correlations . the correlation function @xmath302 in the region @xmath303 is predominantly smaller than unity , the chains avoid unfavorable contacts already at this rather high temperature . however , the angular dependence occurring at @xmath304 in @xmath305 for @xmath306 clearly is due to a lack of statistical accuracy , as a comparison of data for individual runs shows . similarly , the weaker deviations of @xmath307 from unity should be disregarded for the same reason either . while at low temperatures ( @xmath285 , see fig . 6 clearly a more systematic correlation over a larger range of @xmath299 develops , the systematic finite size effects due to the periodic boundary condition in @xmath4-direction become a concern , since @xmath308 where the length @xmath309 of the backbone is @xmath310 ( in the examples shown in fig . 6 we have @xmath311 , i.e. , @xmath312 is shown for @xmath313 ) . the conclusion of the discussion of fig . 6 is that a substantially larger statistical effort would be needed to sample @xmath314 with meaningful accuracy . therefore it was decided to focus on a more meaningful average information , defined as @xmath315 the choice @xmath316 simply means that the angular difference @xmath317 between the coordinates @xmath318 is not at all taken into consideration . the choice @xmath319 is used in the expectation that this correlation then will yield some information whether or not janus - cylinder type ordering occurs , while the choice @xmath320 could be useful , if `` janus - dumbbell '' and double - cylinder - like structures occur ( see also the discussion in sec . we have also used these correlations to transform to correlation functions that reflect fluctuations of total number density @xmath321 and relative concentration @xmath322 . following bhatia and thornton @xcite we obtain@xmath323 and @xmath324 where in our case @xmath325 and @xmath326 . of course , @xmath327 can be computed directly , considering a correlation @xmath328 where the sums over i and j ( eq . [ eq51 ] ) both run over all particles , irrespective of whether they are of type a or b. since it is not clear whether measurements will become possible that will yield information directly on @xmath329 , we also consider the structure factor @xmath330 of the bottle brushes ; as is well - known , for one - component bottle - brushes ample information on the structure factor @xmath331 is available from scattering experiments @xcite , and for two - component bottle - brushes relevant information could be obtained from scattering experiments as well if one species ( a or b ) is a deuterated polymer . assuming that the wave vector of the scattering @xmath332 is also oriented along the @xmath4-axis of the bottle - brush polymer , the partial structure factor @xmath330 becomes @xmath333\rangle\;,\ ] ] where @xmath334 . again it is useful to transform to structure factors that relate to total density fluctuations @xmath335 and concentration fluctuations @xmath336 , respectively , @xmath337 @xmath338 in an actual experiment , of course , it would be more natural to consider the situation when the orientation of the scattering vector @xmath332 is fixed by the experimental setup , while the orientations of the rigid backbones of the bottlebrush polymers in solutions are randomly oriented . then one would need to consider a structure factor @xmath339\rangle\;,\ ] ] where it is understood that the average @xmath340 includes an average over the orientation of @xmath205 . this structure factor @xmath341 can be interpreted as the fourier transform of a correlation function @xmath342 , where @xmath343 is the absolute value of the distance between two sites of monomers @xmath318 in the bottle brush , @xmath344 . this correlation @xmath345 was presented and discussed already in our preliminary communication @xcite , and hence will not be discussed further here . we only note that for large values of @xmath343 we essentially must have @xmath346 and then @xmath347 , since the differences @xmath348 must be relatively much smaller ( of the order of the cross - sectional radius @xmath9 of the cylindrical brush ) . however , for small @xmath343 the behavior of @xmath342 is needed to estimate the number of contacts @xmath349 which is defined as @xmath350 eq . [ eq92 ] means that a pair of particles @xmath351 is defined to have a pairwise contact if their distance is less than @xmath352 . we have followed the criterion due to stillinger @xcite to take @xmath353 . we first focus on the dependence of the number of contacts @xmath354 and @xmath355 on temperature @xmath356 and interaction energy @xmath6 ( remember @xmath357 ) , fig . 7 , to clarify what is the range of interest of those parameters , where the microphase separation occurs . we recognize that for @xmath358 all contact numbers start to become independent of @xmath356 when @xmath356 is increased . this behavior is easily interpreted by the fact that for @xmath359 the repulsive interactions among the beads dominate , the side chains essentially take on self - avoiding walk - like configurations . of course , each interior monomer of a side chain has at least two monomers of the same type ( a or b ) which are nearest neighbors along the same side chains , so we expect that @xmath360 can not be less than two . 7 shows that the actual numbers @xmath360 at temperatures slightly above @xmath361 are about twice as large , @xmath362 . in contrast , the numbers of heterocontacts between chains of different type are at least an order of magnitude smaller , from @xmath363 to @xmath364 . it is interesting to note that the number @xmath360 increases strongly with decreasing temperature for @xmath365 , which we identify as the regime of temperatures where for the chosen ( short ! ) chain length of the side chains a rather dense and approximately cylindrical structure of the bottle - brush can be identified . note that @xmath360 at these low temperatures decreases systematically with increasing @xmath6 , while @xmath355 increases with @xmath6 . however , the dependence of the total number of contacts @xmath366 ) on @xmath6 is much weaker . on the other hand , the temperature dependence of @xmath355 depends very strongly on @xmath6 : while @xmath355 for @xmath367 stays almost independent of @xmath356 , even for low @xmath356 where the brush is in a rather dense state , for @xmath368 and @xmath369 we find a strong increase of @xmath355 with decreasing @xmath356 . tentatively , this increase can be attributed to the formation of ab - interfaces in the collapsed parts of the bottle - brush . we now turn to the behavior of the correlation functions @xmath327 and @xmath370 and focus on the grafting density @xmath255 and compare two choices of @xmath371 and @xmath372 ( figs . we see that for @xmath373 ( fig . 8) in the average density fluctuation at high @xmath356 there is a rapid decay of @xmath374 to unity ( deviations from unity of order 0.01 are just statistical noise ) , while at lower temperature a clear minimum near @xmath375 followed by a maximum near @xmath376 develops , indicating rather long wavelength fluctuations in the thickness of the collapsed cylinder . this long - wavelength structure is not present in @xmath377 , fig . 8b , while @xmath378 , fig . 8c , indicates that a uniform correlation gradually develops , which does not occur for @xmath369 , however ( fig . while in the average concentration correlation @xmath379 one sees a long wavelength periodicity for @xmath373 ( fig . 9a ) , as well as for @xmath369 , ( fig . 11a ) , this periodicity is basically independent of temperature and its amplitude is very small ( of order 0.005 ) . thus , this variation clearly must be discarded as being due to somewhat insufficient preparation of the initial states . in contrast , fig . 9b shows that @xmath380 develops much stronger order of uniform sign , indicating that this system shows a tendency to develop janus - cylinder type short range order . while at @xmath368 ( not shown ) this tendency is also present , though less developed , for @xmath369 ( fig . 11b ) the ordering tendency is much weaker : this indicates that the effective @xmath381-parameter for phase separation between a and b at @xmath369 and @xmath382 may be too weak to cause micro - phase separation . the correlation function @xmath383 , figs . 9c and 11c show a rapid decay to zero in both cases , independent of temperature , indicative of the fact that this correlation measures only some intra - chain correlation effects in these cases , and no collective effects are detected by @xmath384 at all . when we study still larger grafting densities , such as @xmath385 ( figs . 12 , 13 ) , these conclusions are confirmed : there is no significant long wave length correlation in either uniform density or concentration , but a pronounced tendency to form janus - cylinder like short - range order develops , together with a long range correlation in the density , seen in @xmath378 , which is due to a cross - section of the cylindrical brush which is locally non - spherical . the obvious interpretation of fig . 8c ) and 12c ) is that the local janus - type order is not strictly of janus cylinder type but rather of janus dumbbell type , for @xmath373 , as it was suggested already in our preliminary communication @xcite where other quantities for the case @xmath386 were studied . this conclusion is corroborated by the concentration correlations ( fig . 13 ) , which reveal neither a significant long range correlation in @xmath379 nor in @xmath387 while @xmath388 exhibits the development of very pronounced janus - type order . for considerably lower grafting density , such as @xmath289 ( not shown ) or @xmath258 ( figs . 14,15 ) the behavior is very different , however : both @xmath374 and @xmath389 do not exhibit much structure at high temperature , but develop an oscillation ( minimum near @xmath375 , maximum near @xmath390 ) when @xmath356 is lowered towards @xmath285 . for longer chain lengths ( n = 50 , not shown ) the behavior is qualitatively similar , but the minimum now occurs near @xmath391 and the maximum near @xmath392 , and for n = 20 ( not shown ) @xmath393 has its minimum at @xmath394 , the maximum at @xmath391 . thus , there is a clear dependence on this weak correlation seen in figs . 14a , 15a on the length of the side - chains . in @xmath395 there occurs only a fast decay to zero for n=20 ( not shown ) , while for n=35 a minimum develops for @xmath375 ( fig.14b ) and for n=50 this minimum occurs for @xmath391 , i.e. , these structures are strongly correlated with the structure that develops for @xmath374 . taken together with the observation ( fig . 5b ) that the chains form a pearl - necklace structure , we conclude that typically the `` mass '' in the pearls is not distributed spherically symmetric around the backbone ( the @xmath4-axis ) . this feature is also suggested by visual inspection of the snapshots ( fig . 5a , 5b ) , but fig . 14 implies that it is not an accidental observation of some particular snapshots , but a statistically significant physical effect . turning to the concentration correlation @xmath388 we notice that for high temperatures a rapid decay to zero occurs but this decay becomes significantly slower as @xmath356 is lowered ( fig . again the comparison with the corresponding data for @xmath396 ( not shown ) and n=50 ( not shown ) reveals a qualitatively similar picture ( for n = 50 the decay is so slow , however , that our choice of 2 m = 50 would not be large enough to avoid artefacts due to the periodic boundary conditions ) . turning finally attention to @xmath378 and @xmath384 , figs . 14c and 15c , we notice a monotonous decay to zero , which becomes somewhat slower when @xmath356 is decreased . for n = 20 ( not shown ) these correlations are almost independent of @xmath356 , decay to zero for @xmath397 , while for n = 50 ( not shown ) @xmath378 develops a slow decay ( which is absent in @xmath398 , however ) . clearly these data are delicate to interpret , since upon lowering the temperature for this larger chain length one presumably crosses the boundary from an essentially uniform cylindrical structure ( as function of @xmath4 ) at higher temperatures to rather elongated clusters @xcite . if one compares the behavior to the case @xmath289 ( here some snapshot pictures , fig . 5a , suggest a clear alternation of a - rich and b - rich regions along the z - axis ) , one does not find any significant structure of longer range in either @xmath399 or @xmath400 [ in order to save space , the corresponding data are not shown ] , and only both @xmath374 and @xmath379 exhibit the development of periodic structures , similar to what is seen in figs . 14a , 15a . finally fig . 16 shows a few typical results of structure factors @xmath401 . note that due to the periodic boundary conditions @xmath205 is `` quantized '' in multiples of @xmath402 , with @xmath403 in fig . 16 ( a , b ) while @xmath404 in fig . 16 ( c , d ) . we find that the position of the main peak occurs at @xmath405 , ( which corresponds to @xmath406 , where @xmath407 would be the gyration radius of a diblock macromolecule @xmath408 with @xmath409 , as used here , at the theta point , @xmath410 ) in all cases , it is only the peak intensity that strongly grows with increasing @xmath411 or changing the temperature . such a finding is quite nontrivial , and without guidance by the theoretical treatment of the previous section , would have been somewhat unexpected . moreover , it is worth to notice that the theoretical and simulation values for @xmath412 ( @xmath413 and @xmath414 , respectively ) are in remarkable agreement . since the theory is done for temperatures slightly below the theta temperature but very long side chain lengths , while the simulation is done for temperatures down to one half the theta temperature but rather short side chains , the agreement is perhaps better than one would have expected . in previous work on cylindrical bottle - brushes with a single type of side chain under poor solvent conditions @xcite it was found that for @xmath415 clusters elongated in @xmath4-direction occur , which have a gyration radius of order @xmath416 . these elongated clusters are also visible in the snapshot pictures ( figs . 5a , b ) . for larger @xmath411 the range over which the concentration distribution is uniform ( as measured by @xmath417 see fig . 9b ) involves a similar length scale . however , the lack of a more pronounced temperature dependence of @xmath401 is disturbing , and lacks an explanation . we conclude that in any case it seems difficult to clarify the nature of short range order in binary bottle - brush polymers from an analysis of @xmath401 . in order to get a more precise picture of the mesophase ordering in reciprocal space from the simulations , which could be compared to the corresponding results of the analytical calculations of sec . 2 ( e.g. , figs . 2 , 3 ) , it is useful to carry out numerical fourier transforms of the data on the correlations in real space , presented in figs . for example , fig . 17 shows the fourier transforms @xmath418 and @xmath419 which are the fourier transforms of @xmath420 and @xmath421 , for a representative selection of our data , while fig . 18 shows the fourier transform of the corresponding concentration fluctuation correlations \{@xmath422 is the fourier transform of @xmath423 , while @xmath424 is the fourier transform of @xmath425}. one can see that for @xmath426 the density fluctuation structure factors @xmath418 and @xmath419 show a very pronounced increase at small @xmath205 as the temperature is lowered , while for @xmath427 this increase is much weaker , and for @xmath418 it is evident that a small peak grows at nonzero @xmath205 only @xmath428 , compatible with the @xmath429 extracted from @xmath430 as shown in fig . we conclude from these data that density variations play a central role for mesophase formation for @xmath426 , but less so for @xmath427 . the concentration fluctuation structure factor @xmath431 exhibits for @xmath432 a strongly growing peak near @xmath433 ( not shown since the data are qualitatively similar to fig . 16b ) , while @xmath431 for @xmath427 exhibits a very weak temperature dependence only ( not shown ) , as it must be , considering that @xmath434 in this case is basically independent of temperature ( fig . 9 ) . however , the trend seen for @xmath422 , which exhibits a growth for small @xmath205 with decreasing temperature for all the cases included in fig . 18 , indicates a tendency for local janus cylinder type order in these cases , while long range order clearly is not established . we have not included @xmath424 into this study , since the data for @xmath425 \{fig . 9c , 11c , 13c } indicated already hardly any interesting temperature dependence to be present . the data shown in figs . 16 - 18 indicate that in the case @xmath426 density takes the leading role in the mesophase ordering of the cylindrical brush , and the concentration ordering adjusts to it , while for @xmath427 concentration variations along the backbone drive the mesophase ordering , with a strong coupling between density variations and concentration variations still being present . in this case , a periodic modulation of density along the z - axis still grows when the temperature is lowered ( fig . 17a ) , and correspondingly also @xmath401 has a pronounced peak near @xmath435 ( fig . 16 c , d ) . on the other hand , for @xmath436 a rather different behavior is found ( fig . 19 ) : while for @xmath418 only a weak peak at about @xmath437 grows when @xmath356 is lowered ( fig . 19a ) , the growth of both @xmath438 and @xmath422 for @xmath439 is very strong . we interpret these observations as a tendency towards the formation of janus dumbbell - type structures . in this paper the microphase separation in binary ( a , b ) bottlebrush polymers with rigid backbones driven by decrease of temperature ( i.e. , variation of the solvent quality from theta solvents to poor solvents ) was studied , considering also the variation of grafting density along the backbone of these cylindrical brushes . two complementary theoretical methods were used , namely , the first approach has the merit that arbitrarily long chain lengths of the side chains can be considered , as well as infinitely long backbones ( so that end effects or effects of boundary conditions along the backbone do not matter ) . besides , the approach results in some explicit expressions for the correlation functions under study . however , the disadvantage of the approach is that it is basically a linear stability analysis around the homogeneous state , strongly nonlinear effects are out of consideration . the md approach , on the other hand , in principle can take both nontrivial correlations and nonlinear effects into account , but in practice ( to avoid excessive requests for computer time resources ) is limited to rather short length of the side chains , and also the length of the ( rigid ) backbone is finite ( end effects then are avoided by a periodic boundary conditions , but the price to be paid is that reciprocal space in the z - direction along the backbone is discretized ) . also , non - negligible statistical errors ( and systematic errors due to insufficient length of the md trajectories , which affect the establishment of full thermal equilibrium ) restrict the accuracy of the results that can be gotten . whereas the rpa treatment assumes that the monomer density in the radial direction perpendicular to the cylinder axis is uniform and constant ( up to the cylinder radius , while it is zero outside of the cylinder surface ) , in the model studied by md this assumption clearly does not apply , as has been shown in our previous work @xcite . rather , it was found that there occur density oscillations very close to the cylinder axis ( reminiscent of the layering of fluid particles adjacent to the hard wall ) , and for the chain lengths one can study only for a rather small regime of radial distances the density profile then is approximately constant . one approaches rather fast the regime where the density continuously drops to zero , similar to the density profiles across interfaces in phase - separated polymer solvent systems , where the density decreases from its value in the melt towards zero over a distance of several monomer diameters . this low temperature behavior in the simulation comes closest to the situation assumed in the rpa calculations . at higher temperatures , close to the @xmath55 temperature , the interfacial profile is so broadened that a flat region where the density has reached the melt density no longer can be identified . given the different temperature and chain length regimes for which theory and simulation access the microphase separation conditions , it is very gratifying that there is not only a good qualitative agreement between both , but furthermore the wavelength of instability towards the microphase separation agrees quantitatively between both . qualitatively , the analytical theory describes similar trends as seen in the simulation : there is a competition between microphase separation developing periodic order in axial direction , i.e. , the system tends to develop long range order with a particular periodicity ( the wavelength of this periodic modulation being controlled by the gyration radius of the side chains , while the strength of the modulation depends on temperature and grafting density , as the analytical theory compellingly shows , and as is confirmed by our simulations ) and microphase separation of the janus - type , characterized by @xmath440 ( the cross section of the dumbbell , depending on the ratio @xmath5 , in the strongly segregated regime ) . the theory shows that the instability leading to the latter type of order prevails if the grafting density is sufficiently high ( and/or the side chain length sufficiently long ) . both the snapshot pictures of the simulated bottle - brushes ( figs . 4,5 ) and the quantitative analysis of the simulations in terms of the correlation functions of density fluctuations @xmath327 [ with @xmath441 and concentration fluctuations ( or their fourier transforms ) are compatible with such an interpretation . we emphasize that both the simulation and the analytical theory have considered a particularly symmetric case , where both types of side chains have equal length ( @xmath442 ) and also their number was taken exactly the same ( and moreover the same solvent quality was chosen for both types of chains , @xmath443 ) . clearly , when one would like to discuss particular experimental systems , these rather special conditions need to be relaxed . also , we have restricted attention to the case where the side chains are perfectly flexible while the backbone was assumed to be rigid . for many real bottle - brush systems it is probably more realistic to assume backbones that are semi - flexible rather than completely rigid , and also the local intrinsic stiffness of the side chains may play a role . thus , the treatment presented in our paper can be taken as a first step towards a more complete description of less complex but interesting systems only . * acknowledgments*. i.e. thanks the alexander von humboldt foundation and the ministry of education and science of the russian federation for support of this work . p.e.t . thanks the max planck institute for polymer research for supporting him with a max planck fellowship . ( color online ) snapshot picture of bottle - brush polymers at @xmath444 for @xmath283 ( upper part ) and @xmath285 ( lower part ) . a and b monomers are distinguished by different color ( or light grey vs. dark grey , respectively),title="fig:",height=234 ] ( color online ) snapshot picture of bottle - brush polymers at @xmath444 for @xmath283 ( upper part ) and @xmath285 ( lower part ) . a and b monomers are distinguished by different color ( or light grey vs. dark grey , respectively),title="fig:",height=226 ] i.m . lifsitz , a.yu . grosberg and a.r . khokhlov , rev . phys . * 50 * , 683 ( 1978 ) ; a. yu . grosberg and a. r. khokhlov , _ statistical physics of macromolecules _ , american institute of physics , new york , 1994 .

we develop a variational approximation to the entanglement entropy for scalar theory in 1 + 1 , 2 + 1 , and 3 + 1 dimensions , and then examine the entanglement entropy as a function of the coupling . we find that in 1 + 1 and 2 + 1 dimensions , the entanglement entropy of theory as a function of coupling is monotonically decreasing and convex . while theory with positive bare coupling in 3 + 1 dimensions is thought to lead to a trivial free theory , we analyze a version of with infinitesimal negative bare coupling , an asymptotically free theory known as _ precarious _ theory , and explore the monotonicity and convexity of its entanglement entropy as a function of coupling . within the variational approximation , the stability of precarious theory is related to the sign of the first and second derivatives of the entanglement entropy with respect to the coupling . 5cm(11cm,-12.2 cm )

1509.05685

entanglement is a pervasive phenomenon in quantum physics , and has been studied extensively in finite - dimensional systems using the tools of quantum information theory . more elusive is the entanglement structure of infinite - dimensional systems , namely quantum field theories . the last twenty years has seen the development of a large literature on the entanglement entropy of qft s and has shed light on the entanglement between spatial degrees of freedom @xcite . currently , there are only a few classes of field theories for which entanglement entropy can be computed exactly . these include certain exactly solvable qft s , free field theories , and conformal field theories . cft s with holographic duals are particularly interesting because their entanglement entropies can be computed using the ryu - takayanagi formula @xcite . despite this vast body of work , the entanglement entropy of qft s remains rather mysterious . entanglement entropies for qft s typically contain a combination of divergent and finite terms which depend on the parameters of the qft . often it can be difficult to gain insight into the entanglement structure of a qft by analyzing the form of its entanglement entropy . more insightful is how the entanglement entropy _ changes _ as a qft undergoes dynamics such as a quench , or as parameters of the qft are tuned . conspicuously missing from the list of qft s with exactly computable entanglement entropies are those interacting field theories which comprise the standard model . even conventional qft s such as @xmath0 theory and yukawa theory have not had much presence in the entanglement entropy literature . while it may not be possible to study the exact entanglement entropy of these theories , it is possible to study perturbative or variational approximations . in this paper we embark on a program to analyze the entanglement entropies of more conventional interacting qft s , and begin with interacting scalar field theories and @xmath0 theory in particular . our plan of attack is to use a variational principle to determine a non - perturbative approximation to the ground state of @xmath0 theory for arbitrary coupling , and then compute the entanglement entropy of the approximation . our variational ansatz will be a gaussian wave functional , for which we can compute the entanglement entropy exactly . since the ground state of a massive free scalar field theory is exactly a gaussian wave functional , our approximation will be accurate in a neighborhood of zero coupling . we will also analyze the variational approximation at larger values of the coupling , for which the variational method is well - defined . the validity and accuracy of the gaussian variational approximation compares favorably to one - loop computations and large n approximations for theories where such comparisons make sense . our consideration of the approximation at larger values of the coupling will augment our understanding of the small coupling case , and provide evidence for features of entanglement entropy that may hold for all values of the coupling . once we have computed the variational approximation to the entanglement entropy in @xmath0 theory , we will analyze how it changes as we tune the coupling . we will find that the derivatives of entanglement entropy with respect to coupling are often finite and independent of regularization , and represent meaningful physical quantities . by studying the dependence of entanglement entropy on the coupling , we find several surprising features which suggest new insights into the nature of entanglement in qft s . the structure of the paper is as follows : first we discuss the variational methods used to approximate the ground state and other more general states of @xmath0 theory . then we explain how to compute the entanglement entropy of the variational ansatz using the replica trick . finally , we analyze the dependence of the entanglement entropy on the coupling in various dimensions for @xmath0 theory within the variational approximation . in this section we introduce and review the essential aspects of the variational methods we will use for the approximate calculations of entanglement entropy in interacting quantum field theories . the framework of variational methods that employ a gaussian trial wave functional in the computation of the effective action and effective potential ( in particular the 2pi effective action ; see cornwall , jackiw , and tomboulis @xcite ) is referred to as the _ gaussian effective potential _ , or gep . the gep for @xmath0 theory , as well as for other quantum field theories , has been studied extensively by stevenson and his collaborators @xcite beginning in the 1980s . that work built on earlier work of barnes and ghandour @xcite and others in the 1970s , which in turn extended and clarified the pioneering work of schiff @xcite in the early 1960s . we will trace the development of the gep and the closely related gap equation along a slightly different route , using a special case of a time - dependent variational principle based on work of kerman and koonin @xcite , and applied to quantum field theory by jackiw , kerman , and others @xcite . in considering the special case of a static uniform system , our results reproduce those of stevenson et al . for scalar @xmath0 theories . we will use the schrdinger picture description of quantum field theory ( see jackiw @xcite for an excellent review and additional references ) , where quantum field operators @xmath1 and their canonical momenta @xmath2 are time - independent , and the wave functional @xmath3 = \langle \phi(\mathbf{x } ) | \psi(t ) \rangle$ ] is in general time - dependent , although we will not need to consider the full time - dependence in this paper . our goal is to approximately describe the quantum field theory using variational methods . we define the variational hamiltonian functional @xmath4 by @xmath5 where @xmath6 is the hamiltonian operator for the quantum field theory . the variational method consists of minimizing this variational hamiltonian functional with respect to a set of variables which parameterize the wave functional @xmath7 . we make the ansatz that the variational wave functional is a gaussian , parameterized by a real - valued function @xmath8 and a real - valued symmetric kernel @xmath9 @xmath10 = n \exp \big [ - \frac{1}{4 } \int d^d\mathbf{x } \int d^d\mathbf{x ' } \left ( \phi(\mathbf{x } ) - \bar{\phi}(\mathbf{x } ) \right ) g^{-1}(\mathbf{x},\mathbf{x ' } ) \left ( \phi(\mathbf{x ' } ) - \bar{\phi}(\mathbf{x ' } ) \right ) \big]\ ] ] where @xmath11 is the spatial dimension . the kernel @xmath9 must be positive - definite in order for the variational wave function to be normalizable , and choosing @xmath12^{-1/4}$ ] assures that @xmath7 has unit norm . we will frequently describe the kernel @xmath9 as matrix elements of an operator @xmath13 acting on the space of single - particle wave functions . for example , in position space , @xmath14 . it is important to distinguish the state space of the quantum field theory where the states @xmath15 are defined and upon which the quantum field operators @xmath1 and @xmath2 act , and the space of single - particle wave functions where the functions @xmath16 and @xmath17 are defined and upon which operators such as @xmath13 and @xmath18 act . using the more compact operator notation , we may write the variational wave functional as @xmath10 = [ \det(2\pi g)]^{-1/4 } \exp \big [ \ ! - \frac{1}{4 } \ , ( \phi - \bar{\phi } ) \ , g^{-1 } ( \phi - \bar{\phi } ) \big]\ ] ] as it stands , this variational wave functional is a formal expression that will be given more concrete meaning through the regularization and renormalization procedures discussed in subsequent sections . although the vacuum state of the theory corresponds to @xmath19 , other values of the scalar field lead to a variety of physically meaningful and interesting results . considering states other than the vacuum is necessary to compute ( the variational approximation to ) the effective action for the quantum field theory . similar wave functionals are considered in @xcite , where those authors were motivated by studying entanglement entropy in the context of free field theory in 1 + 1 dimensions . the schrdinger picture field operators @xmath1 and their canonically conjugate momentum operators @xmath2 act on the wave functional @xmath20 $ ] in the canonical way @xmath21 & = \phi(\mathbf{x ) } \psi[\phi ] \\ \hat{\pi}(\mathbf{x } ) \psi[\phi ] & = - i \frac{\delta}{\delta \phi(\mathbf{x } ) } \psi[\phi]\end{aligned}\ ] ] it follows immediately that the variational parameters @xmath8 are the expectation values of the schrdinger picture operator in the gaussian variational state @xmath15 @xmath22 and we can also immediately see that @xmath9 is the two - point function or green function @xmath23 \ , [ \hat{\phi}(\mathbf{x ' } ) - \bar{\phi}(\mathbf{x ' } ) ] \ , | \psi \rangle\ ] ] we now focus on a specific interacting scalar quantum field theory using the variational method just discussed . the hamiltonian operator for this quantum field theory is given by @xmath24\ ] ] where @xmath25 is the scalar field potential given in terms of the bare parameters of the theory we will be concerned with a quartic potential of the form @xmath26 where @xmath27 is a bare mass parameter and @xmath28 is a bare coupling parameter . the coupling @xmath28 has mass dimension 2 , 1 , and 0 in 1 + 1 , 2 + 1 , and 3 + 1 dimensions , respectively . using the gaussian variational wave functional @xmath7 and the hamiltonian operator @xmath6 , a straightforward calculation gives @xmath29 for the gradient - squared term in the hamiltonian , we find @xmath30 where @xmath31 is the momentum operator on the space of single - particle wavefunctions , given by @xmath32 in the position basis . finally , the expectation value of the potential operator @xmath33 is defined by @xmath34 where , for the quartic potential , a short computation ( equivalent to applying wick s theorem ) shows that @xmath35 where we have defined @xmath36 . only this position - space - diagonal part of the green function appears in @xmath37 , because in a local quantum field theory all of the schrdinger picture field operators @xmath1 appearing in the hamiltonian operator @xmath6 are evaluated at the same point in space . as in any continuum quantum field theory we expect @xmath38 to be divergent , which we will address in subsequent discussions of renormalization . it is worth emphasizing that the last term in the above expression is the square of a matrix element @xmath39 , _ not _ the matrix element of the square of the operator @xmath13 , which we would write as @xmath40 . combining the results above , the full variational hamiltonian is @xmath41 + \int d^d\mathbf{x } \ , \left [ \frac{1}{2 } ( \nabla \bar{\phi})^2 + \overline{v}(\bar{\phi},\overline{g } ) \ , \right]\ ] ] the equations for the variational parameters , for an arbitrary potential function @xmath25 , follow directly from minimizing this functional with respect to @xmath16 and @xmath13 @xmath42 \frac{\delta \bar{h}}{\delta g } & = 0 \quad\quad \longrightarrow \quad\quad\quad \frac{1}{4 } \ , g^{-2 } = \omega^2\end{aligned}\ ] ] where the last equation is written in the condensed operator form . the operator @xmath43 is defined by its kernel in position space as @xmath44 where the quantity @xmath45 is defined , for an arbitrary potential function @xmath25 , by @xmath46 and for the quartic potential in particular , it is @xmath47 the quantity @xmath48 will play a significant role in subsequent developments . in addition to depending on the bare mass @xmath27 and bare coupling @xmath28 , @xmath48 is a function of @xmath8 and @xmath38 , so it may depend on position in general . in the case of a uniform system , @xmath48 will be a constant parameter . we now turn our attention to such static uniform systems . if we now restrict ourselves to spatially uniform systems , which are translation invariant , then @xmath16 does not depend on position and @xmath49 . additionally , the green function @xmath9 will only depend on the difference @xmath50 . because of translation invariance , the analysis simplifies considerably in momentum space . defining the fourier transforms @xmath51 \omega(\mathbf{x},\mathbf{x ' } ) & = \int \frac{d^d\mathbf{p}}{(2\pi)^d } \ , \tilde{\omega}(\mathbf{p } ) \ , e^{i \mathbf{p } \cdot ( \mathbf{x } - \mathbf{x'})}\end{aligned}\ ] ] and using the definition of of @xmath43 , its fourier transform takes the simple form @xmath52 where we now indicate explicitly the dependence @xmath48 . the second of the two non - trivial hamilton equations , for @xmath53 in terms of @xmath54 , is now @xmath55 this important equation , together with @xmath56\ ] ] is called the _ gap equation_. the expression for @xmath57 , now given in terms of @xmath48 , is @xmath58 the gap equation gives the functional form of the green function @xmath13 in the static uniform case _ and _ an equation connecting the mean field @xmath16 to the mass parameter @xmath48 . we will frequently refer to just this relationship between @xmath16 and @xmath48 in ( [ gapequation0 ] ) as the gap equation , although the full meaning is as just discussed . the green function describes a particle in the interacting quantum field theory propagating with a mass @xmath48 . we have arrived at this interpretation of the green function @xmath13 and the mass @xmath48 via the variational principle and the gaussian form of the variational wave functional ; it is not a separate assumption . the variational hamiltonian simplifies in the uniform case to @xmath59 + \int d^d\mathbf{x } \ , \overline{v}(\bar{\phi},\overline{g})\ ] ] computing the trace in the momentum space representation , the variational hamiltonian density @xmath60 , defined by @xmath61 , is given by @xmath62 + \overline{v}(\bar{\phi},\overline{g})\ ] ] if we look back through the derivation of this expression , we see that the first term , with @xmath18 , comes from the expectation value of the @xmath63 term of the hamiltonian operator ; the second term , with @xmath64 , comes from the expectation value of the @xmath65 term of the hamiltonian operator ; and the third term @xmath37 comes from the expectation value of the potential @xmath33 term of the hamiltonian operator . the sum of the first two terms is less divergent than either one separately , which is a consequence of the lorentz invariance of the theory . we now use the relationship of the momentum space green function @xmath66 to @xmath67 via the gap equation , so the momentum space integrals may be written as @xmath68 = \int \frac{d^d\mathbf{p}}{(2\pi)^d } \left [ \ , \frac{1}{4 } \ , \tilde{g}(\mathbf{p},m)^{-1 } - \frac{m^2 } { 2 } \ , \tilde{g}(\mathbf{p},m ) \ , \right ] \\[6pt ] = & \int \frac{d^d\mathbf{p}}{(2\pi)^d \ , 2 \tilde{\omega}(\mathbf{p},m ) } \left [ \ , \tilde{\omega}(\mathbf{p},m)^2 - \frac{m^2}{2 } \right ] = i_1(m ) - \frac{m^2}{2 } \ , i_0(m)\end{aligned}\ ] ] where , following @xcite , we have defined the integrals @xmath69 we note in particular that @xmath70 is the integral appearing in the gap equation and the @xmath37 part of the variational hamiltonian . most of these integrals will be divergent in the cases we will be studying , and the renormalization prescriptions we are about to discuss will address this in detail . the variational hamiltonian density @xmath60 , with @xmath48 given as a function of the mean field @xmath16 via the gap equation , is known as the _ gaussian effective potential _ ( gep ) , and is given by @xmath71 and @xmath37 for the quartic potential was given earlier by @xmath72 where we have used @xmath70 . we now turn to the renormalization of the theory and a physical interpretation of the formal expressions for the gap equation and the gep . the vacuum state , with @xmath19 , corresponds to a minimum of @xmath73 @xmath74 following stevenson @xcite as well as barnes and ghandour @xcite , we define the renormalized mass @xmath75 @xmath76 and the renormalized coupling @xmath77 @xmath78 computing these derivatives is routine , and the results are quite simple when the derivatives are evaluated in the vacuum state @xmath19 . for the renormalized mass @xmath75 we have @xmath79 the second equality above follows because @xmath80 satisfies the gap equation @xmath81 with @xmath19 . so , the renormalized mass @xmath75 is identified with the vacuum solution of the gap equation : @xmath82 . the renormalized mass sets the scale for all dimensionful quantities in the renormalized theory , and it will often be convenient to work in units such that @xmath83 . for the renormalized coupling @xmath77 , we have @xmath84 with these renormalization prescriptions , we can now derive the renormalized forms of the gap equation and the gep in various dimensions . the detailed discussion of the 1 + 1 and 2 + 1 dimensional cases are quite similar . in these dimensions there are ultraviolet divergences that require mass renormalization , but coupling renormalization is merely a finite coupling reparameterization , and in these lower dimensions we will express results in terms of the original finite bare coupling @xmath28 . we refer the reader to the papers of stevenson and collaborators for more detail , but we will review and summarize the most important results here . in 1 + 1 and 2 + 1 dimensions , we will make use of the following identities for the divergent integrals @xmath85 and @xmath86 : @xmath87 i_0(m ) - i_0(m_r ) & = f_0(m)\end{aligned}\ ] ] where the finite quantities @xmath88 and @xmath89 are @xmath90 f_1(m ) & = -\frac{1}{24\pi } \ , ( m - m_r)^2 ( 2 m + m_r ) \quad\quad\quad \text{2 + 1 dimensions}\end{aligned}\ ] ] and @xmath91 f_0(m ) & = - \frac{1}{4\pi } ( m - m_r ) \quad\quad\quad\quad \text{2 + 1 dimensions}\end{aligned}\ ] ] the renormalization of the gap equation proceeds in essentially the same way in 1 + 1 and 2 + 1 dimensions . starting with the bare gap equation and using the definition of the renormalized mass , we have @xmath92 & = m_b^2 + \frac{\lambda_b}{2 } i_0(m_r ) + \frac{\lambda_b}{2 } ( i_0(m ) - i_0(m_r ) ) + \frac{\lambda_b}{2 } \bar{\phi}^2 \\[4pt ] & = m_r^2 + \frac{\lambda_b}{2 } f_0(m ) + \frac{\lambda_b}{2 } \bar{\phi}^2\end{aligned}\ ] ] as a check , note that when @xmath16 takes its vacuum value @xmath19 , the mass parameter @xmath48 also takes its vacuum value @xmath93 , since @xmath94 . so , the renormalized gap equations for each dimension are @xmath95 \bar{\phi}^2 & = \frac{1}{4\pi } ( m - m_r ) + \frac{2}{\lambda_b } ( m^2 - m_r^2 ) \quad\quad\quad\quad\quad \text{2 + 1 dimensions}\end{aligned}\ ] ] in either case , the renormalized gap equation is of the form @xmath96 we see the similarity of the form of the gap equations across different dimensions . only the first term has a different form in different dimensions and has no dependence on the form of the potential in the original hamiltonian . the last term contains the coupling @xmath28 and depends on the detailed form of the potential . as we will see , a very similar result holds in 3 + 1 dimensions , where the gap equation will be expressed in terms of the renormalized coupling instead of the bare coupling . the renormalization of the gaussian effective potential @xmath97 also proceeds in essentially the same way in 1 + 1 and 2 + 1 dimensions . consider the first two terms of @xmath97 : @xmath98 & = i_1(m_r ) + \frac{1}{2 } ( m^2 - m_r^2 ) i_0(m_r ) + f_1(m ) - \frac{m^2}{2 } i_0(m ) \\[6pt ] & = i_1(m_r ) - \frac{m_r^2}{2 } \ , i_0(m_r ) + f_1(m ) - \frac{m^2}{2 } ( i_0(m ) - i_0(m_r ) ) \\[6pt ] & = i_1(m_r ) - \frac{m_r^2}{2 } \ , i_0(m_r ) + f_1(m ) - \frac{m^2}{2 } f_0(m)\end{aligned}\ ] ] the first two terms contribute an additive divergent constant to @xmath97 , but we will require the energy density to be zero in the vacuum state , using the freedom to add a constant to the original energy density to make this so . using the above expressions for @xmath99 and @xmath100 , the finite parts are quite simple : @xmath101 f_1(m ) - \frac{m^2}{2 } f_0(m ) & = \frac{1}{24\pi } ( m^3 - m_r^3 ) \quad\quad\quad \ \ \text{2 + 1 dimensions}\end{aligned}\ ] ] turning to the third and final term of @xmath97 , namely @xmath37 , a useful form is given by ( recall that @xmath70 ) @xmath102 & = \frac{\lambda_b}{4 ! } \ , \bar{\phi}^4 + \frac{1}{2 } ( m_b^2 + \frac{\lambda_b}{2 } \ , \overline{g}(m ) ) \bar{\phi}^2 + \frac{1}{2 \lambda_b } ( m_b^2 + \frac{\lambda_b}{2 } \ , \overline{g}(m ) ) ^2 - \frac{m_b^4}{2 \lambda_b } \\[4pt ] & = \frac{\lambda_b}{4 ! } \ , \bar{\phi}^4 + \frac{1}{2 } ( m^2 - \frac{\lambda_b}{2 } \bar{\phi}^2 ) \ , \bar{\phi}^2 + \frac{1}{2 \lambda_b } ( m^2 - \frac{\lambda_b}{2 } \bar{\phi}^2 ) ^2 - \frac{m_b^4}{2 \lambda_b } \\[4pt ] \label{vbar1 } & = -\frac{\lambda_b}{12 } \ , \bar{\phi}^4 + \frac{m^4}{2 \lambda_b } - \frac{m_b^4}{2 \lambda_b}\end{aligned}\ ] ] where we have once again used the bare gap equation . this expression ( [ vbar1 ] ) holds in any dimension . note that the last term above is also an additive divergent constant . combining the results above for the two parts of @xmath97 , and choosing the overall additive constant such that @xmath103 in the vacuum state characterized by @xmath19 and @xmath93 , we have the final result for the renormalized gep in each dimension @xmath104 v_{ge}(\bar{\phi } ) & = \frac{1}{24\pi}(m^3 - m_r^3 ) + \frac{1}{2 \lambda_b } ( m^4 - m_r^4 ) - \frac{\lambda_b}{12 } \bar{\phi}^4 \quad\quad \text{2 + 1 dimensions}\end{aligned}\ ] ] as already noted for the gap equations , this way of writing things clearly shows the similarity of the gep across different dimensions . only the first term has a different form in different dimensions ; it arises purely from expectation value of the operators @xmath63 and @xmath65 terms in the hamiltonian operator @xmath6 , and has no dependence on the form of the potential operator @xmath33 . the last two terms , which contain the coupling @xmath28 , depend only on the choice of potential . of course , the dependence of @xmath48 on @xmath16 is given by the appropriate gap equation in each dimension , and these have similar but not identical forms . in 3 + 1 dimensions , we will restrict our attention to a version of @xmath0 theory which is non - trivial in 3 + 1 dimensions meaning it is physically well - defined and is not a free field theory . this is the theory which stevenson @xcite has given the name _ precarious _ @xmath0 theory . in this theory , the bare coupling is taken to be _ negative _ and arbitrarily close to zero in a limit in which the ultraviolet cutoff goes to infinity . this theory , while unstable , is in a very specific sense arbitrarily metastable as the ultraviolet cutoff goes to infinity , and constitutes a physically sensible effective field theory . we do not attempt to review precarious @xmath0 theory and refer the reader to the very clear discussion in stevenson @xcite . as noted there , and observed much earlier by symanzik @xcite and parisi @xcite , this version of @xmath0 theory has the intriguing property of being asymptotically free . we do not consider the `` standard '' version of @xmath0 theory with positive bare coupling because it is trivial using the variational methods of the gep meaning it is a free field theory with the renormalized coupling driven to zero from above as the ultraviolet cutoff goes to infinity . this is in harmony with other triviality " results for standard @xmath0 theory in 3 + 1 ( and higher ) spacetime dimensions . according to stevenson and tarrach @xcite , in addition to the precarious version there is one other physically sensible version of @xmath0 theory in 3 + 1 dimensions which has a positive bare coupling but differs from the standard theory . they call this theory _ autonomous _ @xmath0 theory . we will not consider this version in this paper . in 3 + 1 dimensions the ultraviolet divergences require coupling renormalization as well as mass renormalization . the prescription for the renormalized coupling in any dimension , as formulated earlier , is @xmath105 in the precarious @xmath0 theory , since the bare coupling @xmath28 is negative and driven to zero as the cutoff goes to infinity , the coupling renormalization is given by @xmath106 where we have dropped terms of order @xmath107 or smaller . the following identities for the integrals @xmath108 in 3 + 1 dimensions are useful for the renormalization procedure : @xmath109 & \quad i_{-1}(m ) - i_{-1}(m_r ) = f_{-1}(m)\end{aligned}\ ] ] where the finite parts are @xmath110 f_0(m ) & = - \frac{1}{16\pi^2 } ( m^2 - m_r^2 - m^2 \log \frac{m^2}{m_r^2 } ) \\[4pt ] f_{-1}(m ) & = - \frac{1}{8\pi^2 } \log \frac{m^2}{m_r^2}\end{aligned}\ ] ] using these identities and the coupling renormalization , it is now straightforward to derive the renormalized gap equation , which is @xmath111 following similar derivations already discussed for lower dimensions , a straightforward calculation shows that the renormalized gep is @xmath112 it is worth noting that there is a non - trivial cancellation of logarithmic divergences between the kinetic and potential pieces of the hamiltonian , associated with the coupling renormalization . this is the key to the renormalizability of the theory in 3 + 1 dimensions . it does not occur and is not required in lower dimensions . we now express all dimensionful quantities in units of the renormalized mass @xmath75 , so that @xmath83 . it is convenient to define the normalized coupling @xmath113 and the negative normalized inverse coupling @xmath114 , and a re - scaled version of the scalar field by @xmath115 so that @xmath116 . then the renormalized gap equation takes the form @xmath117\ ] ] and the renormalized gep is @xmath118 following @xcite , for each value of the coupling we define the _ critical value _ @xmath119 as the largest value of the scalar field where the gap equation has a solution . these are given by @xmath120 the _ break value _ @xmath121 is defined as the largest value of @xmath16 where the solution of the gap equation gives the global minimum of the gep . when @xmath122 , the gep is a constant @xmath123 , corresponding to the minimizing value of @xmath124 . @xmath125 it is always the case that @xmath126 and @xmath127 . the region between these values corresponds to a metastable region of the gep , where the solution to the gap equation corresponds to a local but not global minimum . as discussed in detail in @xcite , when the coupling is strong enough , specifically when @xmath128 , the break value @xmath121 goes to zero and the gep becomes flat for all values of @xmath129 , corresponding to an interacting theory with a massless particle . keeping track of the critical values and break values will be important in the discussion of our results for the entanglement entropy in 3 + 1 dimensions . we need to make some final observations before we move on to entanglement entropy , concerning the sign of the coupling in each dimension . once again , stevenson @xcite provides a clear discussion of these issues , and we merely summarize his findings . in 1 + 1 and 2 + 1 dimensions , the concern is over infrared divergences . in these cases it is essential that the bare coupling be non - negative : @xmath130 . otherwise , the gaussian effective potential is either completely ill - defined ( in 1 + 1 ) or is unbounded below ( in 2 + 1 ) . in 3 + 1 dimensions for precarious @xmath0 theory the concern is not infrared divergences and a gep unbounded below , but ultraviolet divergences that require coupling renormalization . however , something even more interesting takes place : there is a duality relation that connects positive renormalized couplings to negative renormalized couplings , along with a rescaling of the mass parameter @xmath48 and the scalar field @xmath16 , and with an additive constant shift of the gep . the explicit formulas defining the duality transformation are @xmath131 where the asterisk denotes the dual values , and the scale factor @xmath132 is the solution of @xmath133 the entire range of positive couplings @xmath134 ( or equivalently negative inverse couplings @xmath135 ) gets mapped to a range of negative couplings @xmath136 ( or equivalently a range of positive inverse couplings @xmath137 ) under duality , so the duality mapping is one - to - one but not onto . note the somewhat counter - intuitive results that @xmath138 is the self - dual point , indicating that the two formulations with infinitely negative coupling and infinitely positive coupling are dual to each other , and that arbitrarily weak positive coupling is _ not _ dual to arbitrarily weak negative coupling . this duality relation means we need only consider negative values of the renormalized coupling @xmath139 with no loss of generality . in later sections on entanglement entropy we will express our results in terms of the negative renormalized couplings , although we will briefly discuss how some of these results look for positive couplings which are dual to the negative couplings for purposes of comparison with results in lower dimensions . on the following pages we plot the solutions of the renormalized gap equation and the renormalized gaussian effective potential in 1 + 1 , 2 + 1 , and 3 + 1 dimensions , for different values of the coupling . we also plot the duality relations for coupling and scale . we emphasize that all dimensionful quantities in these plots are measured in units of the renormalized mass scale @xmath75 , which sets the scale for the entire theory . in these units , @xmath83 . since our variational ansatz for @xmath0 theory takes the same form as the ground state of a massive free scalar field , we would like to know how to compute the entanglement entropy of a massive free scalar field . we will use the replica trick , which provides an efficient way of computing the von neumann entropy of a density matrix . let @xmath140 denote an un - normalized density matrix , and let @xmath141 denote the corresponding normalized density matrix @xmath142 . then the replica trick @xcite gives us @xmath143 equation ( [ replica1 ] ) is particularly useful since it normalizes our density matrix of interest , @xmath140 . this allows us not to worry about the normalization of our density matrix prior to applying the replica trick . with the help of the replica trick , computing the entanglement entropy is conceptually straightforward but technically difficult . given a wave functional @xmath144 $ ] of our quantum field theory , we start by tracing out the degrees of freedom in a spacetime region @xmath145 from the density matrix @xmath144 \psi^*[\phi(\textbf{x})]$ ] to obtain the reduced density matrix @xmath140 , which need not be normalized . then we compute @xmath146 , and use the replica trick to obtain the desired entanglement entropy . in our case , we consider the gaussian wave functional discussed in detail earlier @xmath10 = [ \det(2\pi g)]^{-1/4 } \exp \big [ \ ! - \frac{1}{4 } \ , \phi \ , g^{-1 } \ , \phi \big]\ ] ] although we are actually interested in wave functionals which take the form @xmath147 = \psi[\phi(\textbf{x } ) + \bar{\phi}]$ ] , letting @xmath148 be the reduced density matrix of @xmath147 \widetilde{\psi}^*[\phi(\textbf{x } ) ] = \psi[\phi(\textbf{x } ) - \bar{\phi } ] \psi^*[\phi(\textbf{x } ) - \bar{\phi}]$ ] , we see that @xmath149 by a change of variables . therefore , it suffices to consider the entanglement entropy corresponding to @xmath150 $ ] . in the context of free field theories , the entanglement entropy for such gaussian states has been considered by a number of authors , originating in the work of sorkin and collaborators @xcite . for this paper , we will restrict ourselves to the case in which @xmath145 is the half - space . when @xmath145 is the half - space , it can be shown that @xmath146 is proportional to the partition function @xmath151 of a massive free scalar field living on an @xmath152-sheeted covering of the euclidean plane , which is a cone of deficit angle @xmath153 . this deficit angle is non - positive because each power of the density matrix adds another sheet which contributes @xmath154 radians of angle surplus . " since @xmath155 , when @xmath145 is the half - space equation ( [ replica1 ] ) can be written as @xcite @xmath156 using heat kernel methods , it is possible to compute @xmath157 , which will give us the leading area law contributions to the entanglement entropy @xcite . using equation ( [ deficit1 ] ) we find that @xmath158\ ] ] where @xmath159 is the length scale which serves as our uv cutoff , @xmath160 is the boundary of @xmath145 , and @xmath161 is the area of @xmath160 . the expansion of equation ( [ ee ] ) contains divergent terms due to the uv cutoff . in this paper , we will be interested in the cases @xmath162 . for @xmath163 , we have @xmath164 & = - \frac{a}{12 } \left(\log(m^2 \epsilon^2 ) + \gamma_e \right ) + \mathcal{o}(m^2 \epsilon^2 ) \nonumber \\[4pt ] & = - \frac{a}{12 } \left ( \log \varepsilon^2 + \log m^2 \right ) + \mathcal{o}(m^2 \epsilon^2 ) \quad\quad\quad\quad \ \ \text{1 + 1 dimensions}\end{aligned}\ ] ] where @xmath165 is the exponential integral and @xmath166 is the euler - mascheroni constant , and without loss of generality we have rescaled @xmath159 to @xmath167 . in @xmath163 dimensions only , we have that @xmath168 . for @xmath169 , @xmath170 where @xmath171 is the error function . finally , for @xmath172 , @xmath173 where , as in 1 + 1 dimensions , we have again rescaled @xmath159 to @xmath167 . we note that there are other sub - leading area terms which can be obtained using more sophisticated methods @xcite , but they will not affect our analysis . in particular , when we consider derivatives of the entanglement entropy with respect to the coupling , these other sub - leading area terms will either vanish in the @xmath174 limit , or if they diverge will be sub - leading to other divergences . one may notice that the divergences in @xmath175 in @xmath11 spatial dimensions take the same form as the divergences of the effective potential for the massive free scalar field in @xmath11 spatial dimensions . indeed , this is no coincidence . noting that @xmath176 is the effective action for a free massive scalar field with deficit angle @xmath177 , we can rewrite equation ( [ deficit1 ] ) as @xmath178 since the effective action contains the effective potential and thus the divergences of the effective potential , it is unsurprising that the entanglement entropy also contains divergences of the same form . in fact , the entanglement entropy takes the following simple unifying form in all dimensions : @xmath179 where we recall that @xmath180 is the divergent diagonal piece of the green function in position space . let us define @xmath181 by the finite part of the entanglement entropy @xmath175 plus an additive constant which makes @xmath181 equal to zero in the free field vacuum . then in all dimensions we have @xmath182 for our variational approximation , we will simply replace @xmath48 in the above equations with the variational mass . in this section , we will compute the entanglement entropy of @xmath0 theory within the variational approximation , and examine how the entanglement entropy depends on the coupling . in a preceding section on the variational method and gaussian effective potential , we found that the renormalized gap equations in spatial dimensions @xmath183 take very similar forms , which we repeat here : @xmath184 \bar{\phi}^2 & = \frac{1}{4\pi } ( m - m_r ) + \frac{2}{\lambda_b } ( m^2 - m_r^2 ) \quad\quad\quad\quad\quad\quad\quad\quad\quad\quad \text{2 + 1 dimensions } \\[4pt ] \bar{\phi}^2 & = - \frac{1}{16\pi^2 } \ , ( m_r^2 - m^2 + m^2 \log \frac{m^2}{m_r^2 } ) - \frac{6}{\lambda_r } ( m_r^2 - m^2 ) \quad\quad \text{3 + 1 dimensions}\end{aligned}\ ] ] also , in the previous section on the formalism of entanglement entropy , we found that expressions for @xmath181 in spatial dimensions @xmath183 also take very similar forms , expressed in all cases by @xmath185 : @xmath186 \delta s / a & = -\frac{1}{12 } \ , ( m - m_r ) \quad\quad\quad\quad\quad\quad\quad\quad\quad\quad \text{2 + 1 dimensions } \\[4pt ] \delta s / a & = -\frac{1}{48\pi } \ , ( m^2 - m_r^2 - m^2 \log \frac{m^2}{m_r^2 } \ , ) \quad\quad\quad \text{3 + 1 dimensions}\end{aligned}\ ] ] to put all of these equations in the exactly the same form is now just a question of normalization , so we define the normalized dimensionless coupling @xmath139 and the normalized dimensionless scalar field @xmath129 for each dimension as @xmath187 \alpha & = \frac{\lambda_b}{8\pi m_r } \quad\quad\quad\quad\quad \varphi^2 = 4 \pi \bar{\phi}^2 \quad\quad\quad\quad\quad \ \text{2 + 1 dimensions } \\[4pt ] \alpha & = \frac{\lambda_r}{96\pi^2 } \quad\quad\quad\quad\quad \ \ , \varphi^2 = 16 \pi^2 \bar{\phi}^2 \quad\quad\quad\quad \ \ , \text{3 + 1 dimensions}\end{aligned}\ ] ] and the normalized dimensionless quantities appearing in the gap equation @xmath188 f & = m - 1 \quad\quad\quad\quad\quad\quad\quad\quad\quad\quad \ \text{2 + 1 dimensions } \\[4pt ] f & = - ( 1 - m^2 + m^2 \log m^2 ) \quad\quad\quad \text{3 + 1 dimensions}\end{aligned}\ ] ] where @xmath48 and other dimensionful parameters are now expressed in units of the renormalized mass scale @xmath75 , so in effect we have chosen units such that @xmath83 . finally , define the normalized entanglement entropy per unit area @xmath189 by @xmath190 \sigma = 12 \ , \delta s / a \quad\quad\quad\quad\quad\quad \text{2 + 1 dimensions } \\[4pt ] \sigma = 48\pi \ , \delta s / a \quad\quad\quad\quad\quad\quad \text{3 + 1 dimensions}\end{aligned}\ ] ] the term `` area '' is used to refer to the measure of the surface of spatial co - dimension 1 in all cases , and as with other dimensionful quantities it is measured in units of the appropriate inverse power of the renormalized mass scale @xmath75 . in terms of these quantities , the dimensionless gap equation and the normalized entanglement entropy per unit area take the same simple form in any dimension : @xmath191 we note the ranges of the relevant quantities in various dimensions are @xmath192 \alpha \geq 0 \quad\quad\quad m^2 \geq 1 \quad\quad\quad f(m^2 ) \geq 0 \quad\quad\quad\quad \text{2 + 1 dimensions } \\[4pt ] \alpha \leq 0 \quad \ \ 0 \leq m^2 \leq 1 \quad \ -1 \leq f(m^2 ) \leq 0 \quad\quad \text{3 + 1 dimensions}\end{aligned}\ ] ] the system of equations given by and allows us to determine the variational approximation to the entanglement entropy for fixed values of the coupling and vacuum expectation value @xmath129 . in particular , for a fixed value of the coupling and a fixed value of @xmath129 , we solve the gap equation in eqn . for the variational mass @xmath48 , and then plug @xmath48 into equation to obtain the normalized entanglement entropy per unit area @xmath189 . before delving into a full discussion of the dependence of entanglement entropy on the coupling , we briefly point out a connection between our variational calculation of entanglement entropy in @xmath0 theory and a perturbative calculation . for @xmath172 , @xmath0 theory is widely believed to be a sick theory for finite positive coupling in a non - perturbative sense . however , one can make sense of the theory for infinitesimal positive coupling , and treat the theory perturbatively . to compute the entanglement entropy in the vacuum of such a theory , we consider equation for @xmath172 and @xmath193 in the @xmath194 limit , and find that @xmath195 . temporarily restoring the renormalized mass scale @xmath75 , we find that the solution to the gap equation in the @xmath194 limit is @xmath196 . then the approximate normalized entanglement entropy per unit area is @xmath197 , which agrees exactly with a perturbative calculation of hertzberg to first order in perturbation theory @xcite . this result demonstrates that our variational framework can reproduce perturbative results in the small coupling limit . however , we note that in the case of @xmath0 theory with infinitesimal positive coupling , we learn nothing about the explicit dependence of the entanglement entropy on the coupling since @xmath197 does not depend on the normalized coupling @xmath139 . now we will analyze the dependence of the normalized entanglement entropy per unit area @xmath189 on the coupling . the results presented in this section demonstrate the utility of the variational techniques , and explore the relationship between entanglement and coupling . in order to determine @xmath189 as a function of coupling , there are two approaches . the first approach is to solve the gap equation at fixed coupling and fixed @xmath129 for the variational mass @xmath48 , and then plug @xmath48 into equation to obtain @xmath189 . in fact , the gap equation can be solved exactly in terms of lambert @xmath198 functions for @xmath163 and @xmath172 , and by the quadratic formula for @xmath169 . however , the form of the corresponding normalized entanglement entropy per unit area @xmath189 is not particularly enlightening since it is difficult to parse , although we can graph it numerically and examine the plot . the normalized entanglement entropy per unit area @xmath189 is plotted in each dimension on the following pages . while we can guess some trends from the plots , it is best to be more quantitative . a second approach to characterizing the dependence of @xmath189 on the coupling is to compute @xmath199 and @xmath200 implicitly using the gap equation , and then examine the behavior of these derivatives . following this approach , let us first examine @xmath199 . first we note that @xmath201 while we can easily compute @xmath202 using the explicit form of @xmath203 in each dimension , we need to use the gap equation to compute @xmath204 . considering @xmath48 as a function of the scalar field and coupling given by the gap equation , and differentiating the dimensionless gap equation with @xmath129 held fixed leads to the general result @xmath205 specifically , in each dimension , @xmath206 \frac{\partial m^2}{\partial \alpha } & = \frac{m^2 - 1}{\alpha \ , ( 1 + \alpha / 2 m ) } \quad\quad\quad\quad\quad\quad\quad\quad \text{2 + 1 dimensions } \\[4pt ] \label{m^2derivative3 + 1 } \frac{\partial m^2}{\partial \alpha } & = \frac{m^2 - 1}{\alpha \ , ( 1 + \alpha \log m^2 ) } \quad\quad\quad\quad\quad\quad\quad \text{3 + 1 dimensions}\end{aligned}\ ] ] calculation of the first derivative of the normalized entanglement entropy per unit area @xmath189 is now simple : @xmath207 using the ranges of the relevant quantities noted above , we find @xmath208 \label{monotonicity2 + 1 } \left ( \frac{\partial \sigma}{\partial \alpha } \right)_{\varphi } & \leq 0 \quad\quad\quad\quad\quad \text{2 + 1 dimensions } \\[4pt ] \label{monotonicity3 + 1,1 } \left ( \frac{\partial \sigma}{\partial \alpha } \right)_{\varphi } & \leq 0 \quad\quad\quad \text{3 + 1 dimensions , $ m^2 \geq e^{1/\alpha}$ for $ \alpha \leq 0 $ } \\[4pt ] \label{monotonicity3 + 1,2 } \left ( \frac{\partial \sigma}{\partial \alpha } \right)_{\varphi } & \geq 0 \quad\quad\quad \text{3 + 1 dimensions , $ m^2 \leq e^{1/\alpha}$ for $ \alpha \leq 0 $ } \ ] ] we also note the results for _ positive _ coupling @xmath139 in 3 + 1 dimensions , which are related to the negative coupling results by the duality relation we discussed previously @xmath209 \label{monotonicity3 + 1,2.5 } \left ( \frac{\partial \sigma}{\partial \alpha } \right)_{\varphi } & \leq 0 \quad\quad\quad \text{3 + 1 dimensions , $ m^2 \leq e^{1/\alpha}$ for $ \alpha \geq 0$}\end{aligned}\ ] ] we will now examine the results in equations ( [ monotonicity1 + 1 ] ) through ( [ monotonicity3 + 1,2 ] ) . first , for equations ( [ monotonicity1 + 1 ] ) and ( [ monotonicity2 + 1 ] ) , we note that @xmath210 in the above equation , we can replace @xmath181 with @xmath175 since the divergent pieces of @xmath175 have no dependence on the coupling for @xmath163 and @xmath169 . thus , while @xmath175 is divergent and regularization dependent , @xmath211 is finite and independent of regularization , and so represents a meaningful physical quantity . on the other hand , for precarious @xmath0 theory where @xmath172 , @xmath211 is not finite and so is not proportional to @xmath199 . however , the divergent term in @xmath211 is proportional to @xmath212 and so in light of equation ( [ m^2derivative3 + 1 ] ) , for small @xmath159 we find that @xmath211 has the same sign as @xmath199 . so to analyze the monotonicity of @xmath175 for precarious @xmath0 theory , it suffices to analyze the monotonicity of @xmath189 . in equations ( [ monotonicity1 + 1 ] ) and ( [ monotonicity2 + 1 ] ) , we see that for @xmath163 and @xmath169 the variational approximation to the entanglement entropy is monotonically _ decreasing _ with respect to coupling for fixed @xmath129 . this result is surprising , since one might have expected the entanglement entropy to increase as the coupling is increased . we can trust our result in a neighborhood of zero coupling for which the variational approximation is accurate . thus , a conservative interpretation of equations ( [ monotonicity1 + 1 ] ) and ( [ monotonicity2 + 1 ] ) is that for @xmath163 and @xmath169 the entanglement entropy is monotonically decreasing in a neighborhood of zero coupling . more ambitiously , we might take the monotonicity of the variational approximations of the entanglement entropies as an indication that the _ true _ entanglement entropies for @xmath163 and @xmath169 are monotonically decreasing for all values of the coupling . this amounts to little more than a conjecture at the moment , but is nonetheless suggestive . now we turn to the precarious @xmath0 theory for @xmath172 , and consider equations ( [ monotonicity3 + 1,1 ] ) , ( [ monotonicity3 + 1,1.5 ] ) , ( [ monotonicity3 + 1,2 ] ) and ( [ monotonicity3 + 1,2.5 ] ) . for @xmath213 , the theory is stable and asymptotically free and equation ( [ monotonicity3 + 1,1 ] ) indicates that for @xmath214 , @xmath189 is monotonically _ increasing _ for fixed @xmath129 as we increase the magnitude of the negative coupling ( i.e. , make the coupling more negative ) . similarly for @xmath213 , equation ( [ monotonicity3 + 1,1.5 ] ) indicates that for @xmath215 , @xmath189 is monotonically _ increasing _ for fixed @xmath129 as we increase the magnitude of the positive coupling . for @xmath216 the theory is unstable , and interestingly equation ( [ monotonicity3 + 1,2 ] ) indicates that for such an unstable theory when @xmath214 , @xmath189 is monotonically _ decreasing _ for fixed @xmath129 as we increase the magnitude of the negative coupling . likewise for the unstable theory , equation ( [ monotonicity3 + 1,2.5 ] ) tell us that for @xmath215 we have that @xmath189 is monotonically _ increasing _ for fixed @xmath129 as we increase the magnitude of the positive coupling . thus , it appears that the direction of the monotonicity of @xmath189 is tied to the stability of the theory . while we understand that the variational approximation is accurate for small values of @xmath217 , we might trust in the variational approximation and conjecture that the sign of the derivative of an _ exact _ entanglement entropy with respect to coupling depends on the stability of the qft in question . having examined the first derivative @xmath199 to analyze the monotonicity of the entanglement entropy with respect to coupling , we now turn to @xmath200 to analyze the convexity of the entanglement entropy with respect to coupling for a fixed value of the expectation value of the scalar field . using equations ( [ m^2_first_derivative ] ) and ( [ sigma_first_derivative ] ) , we find that the second derivative of the normalized entanglement entropy per unit area @xmath189 is @xmath218 again using the ranges of the relevant quantities noted above , we have @xmath219 \label{convexity2 + 1 } \left ( \frac{\partial^2 \sigma}{\partial \alpha^2 } \right)_{\varphi } & \geq 0 \quad\quad\quad\quad\quad \text{2 + 1 dimensions } \\[4pt ] \label{convexity3 + 1,1 } \left ( \frac{\partial^2 \sigma}{\partial \alpha^2 } \right)_{\varphi } & \geq 0 \quad\quad\quad \text{3 + 1 dimensions , $ m^2 \geq e^{1/\alpha}$ for all $ \alpha$ } \\[4pt ] \label{convexity3 + 1,2 } \left ( \frac{\partial^2 \sigma}{\partial \alpha^2 } \right)_{\varphi } & \leq 0 \quad\quad\quad \text{3 + 1 dimensions , $ m^2 \leq e^{1/\alpha}$ for all $ \alpha$}\end{aligned}\ ] ] similar to equation ( [ deltas_to_s_1 ] ) , we note that @xmath220 and so @xmath221 is a finite and regularization independent quantity for @xmath163 and @xmath169 . by equations ( [ convexity1 + 1 ] ) and ( [ convexity2 + 1 ] ) , for @xmath163 and @xmath169 we have that @xmath189 is convex in the coupling for fixed @xmath129 . since for @xmath163 and @xmath169 , @xmath189 is monotonically decreasing , as the coupling increases the entanglement decreases less and less rapidly . now let us examine precarious @xmath0 theory where @xmath172 . for this theory , @xmath221 is not finite and is not proportional to @xmath200 . in fact , the divergent term in @xmath221 is proportional to @xmath222 . since for @xmath172 we have @xmath223 for small @xmath159 we see that @xmath221 has the same sign as @xmath200 . therefore , to analyze the convexity of @xmath175 for precarious @xmath0 theory it is sufficient to analyze the convexity of @xmath189 . by equation ( [ convexity3 + 1,1 ] ) , in precarious @xmath0 theory for the stable region @xmath213 we have that @xmath189 is convex in the coupling for fixed @xmath129 . however , by equation ( [ convexity3 + 1,2 ] ) , in the unstable region where @xmath216 we have that @xmath189 is concave in the coupling for fixed @xmath129 . we conclude that in the variational approximation , the convexity of the entanglement entropy is linked with the stability of the theory . one might conjecture that this connection between convexity and stability holds for exact calculations of the entanglement entropy . in this section we compute the renormalization group equation for the normalized entanglement entropy per unit area @xmath224 , and show that it is simply related to the usual beta function for the renormalized coupling . for 1 + 1 and 2 + 1 dimensional @xmath0 theory , as discussed earlier , there is merely a finite reparameterization of the coupling , with no coupling renormalization depending on a renormalization mass scale @xmath225 , so the beta function vanishes . however , for precarious @xmath0 theory in 3 + 1 dimensions , there is a non - vanishing beta function . if we differentiate the coupling renormalization equation with respect to the renormalization scale @xmath225 for fixed bare coupling and fixed cutoff , then in terms of the normalized coupling @xmath226 the beta function is given by @xmath227 since for precarious @xmath0 theory we have @xmath228 , the beta function in equation ( [ betafunction1 ] ) tells us that the theory is asymptotically free . in particular , @xmath229 . using the simplified forms of the gap equation and entanglement entropy formulas in equations and , we obtain the renormalization group equation for the normalized entanglement entropy per unit area @xmath230 } \ , \frac{\mu}{\alpha^2 } \frac{d\alpha}{d\mu } = \frac{\left(1 - \frac{m^2}{\mu^2 } \right ) f'(m^2/\mu^2)}{\left[\frac{1}{\alpha } + f'(m^2/\mu^2 ) \right ] } \ , \frac{1}{\alpha^2 } \ , \beta(\alpha)\ ] ] we will refer to this quantity as the _ entanglement entropy beta function _ , although any such quantity obviously depends on the geometry of the region that defines the entanglement entropy as well as the quantum field theory itself . in 1 + 1 and 2 + 1 dimensional @xmath0 theory , since the beta function vanishes , the renormalization group equation for the normalized entanglement entropy per unit area is trivial . simplifying equation ( [ eerg ] ) in the case of precarious @xmath0 theory , we obtain @xmath231 since @xmath232 , this entanglement entropy beta function for precarious @xmath0 theory is `` asymptotically free '' , meaning that @xmath233 , but only in the stable region of precarious @xmath0 where @xmath234 . equality @xmath235 holds only in the vacuum where @xmath236 . the sign is reversed for the unstable region , with singular behavior when @xmath237 where precarious @xmath0 theory goes from being stable to unstable . it will be interesting to see which properties of the entanglement entropy beta function for precarious @xmath0 are shared by other theories with asymptotic freedom , such as the gross - neveu model and non - abelian gauge theories . we have applied variational methods to approximate entanglement entropies for scalar @xmath0 theories . within our approximation scheme we have found that in @xmath163 and @xmath169 dimensions , the entanglement entropy is monotonically decreasing and convex with respect to the coupling . this result is unexpected since the authors anticipated that larger coupling would lead to more entanglement . our results are accurate for small coupling , although the approximation scheme is well - defined for all values of the coupling . inspired by the variational results , we conjecture that the _ exact _ entanglement entropy in @xmath163 and @xmath169 is monotonically decreasing and convex with respect the coupling for the entire range of positive couplings . indeed , it would be strange if the derivative of the entanglement entropy switched signs for some particular value of the coupling . if this were the case , then this value of the coupling would be of immense physical interest since it would correspond to a local minimum of entanglement entropy . for precarious @xmath0 theory in the stable range , we have also demonstrated that the variational approximation to the entanglement entropy is monotonic in @xmath139 for @xmath214 and for the dual values @xmath215 , and is convex for all @xmath139 . this theory is of particular interest since it is asymptotically free . as we tune the parameters of the theory so that it becomes unstable , the derivative of the variational approximation to the entanglement entropy changes sign exactly at the transition between quasi - stability and instability . likewise , the second derivative changes sign . we have also shown the the entanglement entropy beta function " for precarious @xmath0 theory in 3 + 1 dimensions is asymptotically free . it would be interesting to see whether or not monotonicity and convexity with respect to coupling are generic features of entanglement entropy for qft s , and to explore why certain theories appear to have entanglement entropies which are monotonically _ decreasing _ with respect to coupling . it would also be interesting to examine the entanglement entropy beta functions of asymptotically free theories , and see if the behavior of precarious @xmath0 theory is generic . in future work , we plan to apply our techniques to quantum field theories with fermions as well as bosons . in addition , we hope to explore the entanglement structure of gauge theories with variational techniques . this would be a valuable step towards analyzing entanglement entropies of quantum field theories appearing in the standard model . the authors would like to thank arthur kerman , mark hertzberg , frank wilczek , hong liu , julian sonner , roman jackiw , jeffrey goldstone , brian swingle , john mcgreevy , eric tonni , and paul stevenson for valuable discussions and feedback . jordan cotler is supported by the fannie and john hertz foundation and the stanford graduate fellowship program . mark mueller is supported by the u.s . department of energy under grant contract number de - sc00012567 . jackiw , r. , analysis on infinite - dimensional manifolds - schrdinger representation for quantized fields " , _ brazil summer school , 78 - 143 ( 1989 ) ; reprinted in `` diverse topics in theoretical and mathematical physics '' , world scientific ( 1995 ) . _ sorkin , r. on the entropy of the vacuum outside a horizon " , originally published in _ b. bertotti , f. de felice and a. pascolini ( eds . ) , tenth international conference on general relativity and gravitation ( held padova , 4 - 9 july , 1983 ) , contributed papers , vol . 734 - 736 ( roma , consiglio nazionale delle ricerche , 1983 ) . _

since the discovery of the first double neutron star ( dns ) system in 1975 by hulse and taylor , there are currently 8 confirmed dns in our galaxy . for every system , the masses of both neutron stars , the orbital semi - major axis and eccentricity are measured , and proper motion is known for half of the systems . using the orbital parameters and kinematic information , if available , as constraints for _ all system _ _ , we investigate the immediate progenitor mass of the second - born neutron star and the magnitude of the supernova kick it received at birth , with the primary goal to understand the core collapse mechanism leading to neutron star formation . compared to earlier studies , we use a novel method to address the uncertainty related to the unknown radial velocity of the observed systems . for psr b1534 + 12 and psr b1913 + 16 , the kick magnitudes are 150 - 270 km / s and 190 - 450 km / s ( with 95% confidence ) respectively , and the progenitor masses of the 2nd born neutron stars are 1.3 - 3.4 m and 1.4 - 5.0 m ( 95% ) , respectively . these suggest that the 2nd born neutron star was formed by an iron core collapse supernova in both systems . for psr j0737 - 3039 , on the other hand , the kick magnitude is only 5 - 120 km / s ( 95% ) , and the progenitor mass of the 2nd born neutron star is 1.3 - 1.9 m ( 95% ) . because of the relatively low progenitor mass and kick magnitude , the formation of the 2nd born neutron star in psr j0737 - 3039 is potentially connected to an electron capture supernova of a massive white dwarf . for the remaining 5 galactic dns , the kick magnitude ranges from several tens to several hundreds of km / s , and the progenitor mass of the 2nd formed neutron star can be as low as.5 m , or as high as m . therefore in these systems , it is not clear which type of supernova is more likely to form the 2nd neutron star . _ _ _

1008.2397

over the years , computation modeling and hydrodynamic calculations of supernova explosions due to core - collapse of massive stars have improved our theoretical understanding of how compact objects form . at present , two types of supernovae are actively discussed in the literature : ( i ) after iron is synthesized within a massive star , the core completes its final nuclear burning state . it then becomes degenerate and keeps growing in mass due to silicon shell burning outside the core . eventually , it crosses the chandrasekhar mass limit and ensues collapse . when the core reaches nuclear density , it bounces back and explodes the star as a supernova . ( bethe 1990 ; mezzacappa 2005 ; woosley & bloom 2006 ; kotake et al 2006 ; janka et al . ( ii ) on the other hand , if the core is not massive enough to ignite neon , nuclear burning within the core stops . the core becomes degenerate due to cooling by neutrino emission . because of carbon shell burning outside the core , the mass of the core approaches the chandrasekhar limit . when the core density and temperature rise above a certain threshold , electron capture by nuclei ( such as @xmath3 , @xmath4 , @xmath5 , and @xmath6 ) takes place , decreasing degenerate electron pressure support . as a result , the core contracts rapidly and eventually bounces back , driving a supernova explosion . this scenario is referred to as electron capture supernova ( ecs ) ( miyaji et al . 1980 ; nomoto 1987 ) . although the general idea of ecs has been established for several decades , the mechanism of how electron capture occurs in these relatively less massive cores is not yet settled . regardless of the supernova mechanism , it is expected that some asymmetries develop during the collapse , imparting a recoil kick to the nascent neutron star ( see harrison et al . 1993 ; frail et al . 1994 ; lyne & loimer 1994 ) . the observed proper motions of 233 galactic pulsars show that there is a class of neutron stars receiving high recoil velocities at birth ( hobbs et al . 2005 ) . on the other hand , pfahl et al . ( 2002 ) studied the orbital parameters of low eccentric be / x - ray binaries , and pointed out that these neutron stars could only have received low kick velocities ( < 50 km / s ) at the time of formation . there is thus evidence for the existence of two different neutron star formation mechanisms ( van den heuvel 2004 , 2007 ) , potentially distinguishable by the magnitude of the natal kick of the neutron star . currently , there are 8 dns observed in our galaxy . their orbital parameters are summarized in table 1 . the magnitude of the natal kick and the progenitor mass of the second - born neutron star ( ns2 ) in these systems , except psr j1906 + 0746 , have been studied previously by willems et . al . ( 2004 , 2006 ) , thorsett et . al . ( 2005 ) , stairs et al . ( 2006 ) , and wang et al . ( 2006 ) . in their study , wang et al . ( 2006 ) studied the formation of _ all _ dns except psr j1906 + 0746 , and derived constraints on the natal kick magnitude and progenitor mass of ns2 by studying the orbital dynamics of asymmetric supernova explosions , but not their kinematic history . the other authors focused on psr b1534 + 12 , psr b1913 + 16 , and psr j0737 - 3039 and used both orbital dynamics of asymmetric supernova explosions and the measured proper motion of these systems to constrain the natal kick magnitude and progenitor mass of ns2 . the authors traced the motion of the systems in the galaxy backwards in time and assumed each crossing of the binary through the galactic disk within the age of the dns to be a possible birth site of the dns . however , the backward calculation of the motion in the galaxy requires the knowledge of the radial velocity which can not be measured . willems et al . ( 2004 , 2006 ) , thorsett et al ( 2005 ) and stairs et al . ( 2006 ) therefore considered possible radial velocities drawn from either a uniform or gaussian radial velocity distribution . _ _ in this paper , we avoid the assumption of a present - day radial velocity distribution by carrying out monte carlo simulations of the motion of the observed dns in the galaxy that is forward in time instead of backward . for this purpose , we distribute populations of dns progenitors in the galaxy according to a double exponential distribution function . each system in the populations is assigned a velocity equal to the vector sum of the local galactic rotational velocity and an isotropic kick velocity with a magnitude generated from the kick velocity distribution function , derived from the supernova orbital dynamics constraints for each of the observed systems . compared to willems et al.(2004 ) , thorsett et al . ( 2005 ) , and wang et al . ( 2006 ) , we also relax the constraint that ns2 s progenitor needs to be more massive than 2.1 m@xmath7 , which is a convectional limit for a helium star that explodes in a core collapse supernova . instead , we restrict the progenitor to be more massive than ns2 only . furthermore we study _ all _ eight known dns in our galaxy , using the up - to - date observational parameters listed in table 1 . _ _ the methodology of the adopted analysis is outlined in more detail in 2 , while results for the individual systems and comparison with earlier studies are discussed in 3 . in 4 , we summarize our results and discuss their implications for the supernova forming the second - born neutron star in the observed dns . @l@ccccc@ccc@c@c@c@ psr b1534 + 12 & 15 37 09.96 & 11 55 55.55 & 1.02 & 1.34(1 ) & -25.05(2 ) & 250 & 1.3332(10 ) & 1.3452(10 ) & 3.28 & 0.274 & 25(155 ) @xmath8 3.8 + psr b1913 + 16 & 19 15 28.00 & 16 06 27.40 & 8.3(1.4 ) & -3.27(35 ) & -1.04(42 ) & 110 & 1.4408(3 ) & 1.3873(3 ) & 2.80 & 0.617 & 18(162 ) @xmath8 6 + psr j0737 - 3039 & 07 37 51.25 & -30 39 40.71 & 1.15 & -3.82(62 ) & 2.13(23 ) & 210 & 1.337(5 ) & 1.250(5 ) & 1.26 & 0.0878 & < 15 + psr j1518 + 4904 & 15 18 16.80 & 49 04 34.25 & 0.625 & -0.67(4 ) & -8.53(4 ) & 20000 & @xmath9 & @xmath10 & 24.7 & 0.249 & + psr j1756 - 2251 & 17 56 46.63 & -22 51 59.40 & 2.5 & -0.7(2 ) & & 443 & 1.312(17 ) & @xmath11 & 2.70 & 0.181 & + psr j1811 - 1736 & 18 11 55.03 & -17 36 37.70 & 6.0 & & & 1830 & @xmath12 & @xmath13 & 40.7 & 0.828 & + psr j1829 + 2456 & 18 29 34.60 & 24 56 19.00 & 1.2 & & & 12400 & @xmath14 & @xmath15 & 6.36 & 0.139 & + psr j1906 + 0746 & 19 06 48.67 & 07 46 28.60 & 5.4 & & & 0.112 & 1.365(18 ) & 1.248(18 ) & 1.75 & 0.0853 & + tauris @xmath16 van den heuvel ( 2004 ) reviewed the general scenario for forming dns . it starts with a binary consisting of two massive zams stars ( @xmath17 ) . the primary star , which is initially the more massive one , will leave the main sequence first and explode as the first supernova , in which it forms the first - born neutron star ( ns1 ) in this binary . there may be mass transfer from the primary to the secondary star before that supernova takes place . as the secondary star leaves the main sequence , wind mass transfer from the secondary to ns1 can occur , leading to the formation of a x - ray binary . while ns1 is accreting mass from its companion , it is spun up or `` recycled '' , and its spin axis is expected to align with the orbital angular momentum . later on , the x - ray binary may end up in a common envelope phase , during which ns1 is engulfed by the extended envelope of its companion . the orbit shrinks and circularizes rapidly due to the frictional forces acting on ns1 as it moves through the envelope of the companion . the lost orbital energy is deposited into the gas envelope , causing it to become gravitationally unbound and get ejected , leaving behind the companion s bare helium core . the helium star eventually explodes in a supernova and becomes ns2 , orbiting the recycled , milli - second pulsar ns1 in a compact binary orbit . mass transfer from the helium star to ns1 can also occur prior to the second supernova explosion . here , our goal is to use the current dns binary properties listed in table 1 to constrain as best possible the magnitude of the natal kick and the mass of the progenitor of ns2 in the 8 known dns in our galaxy at the time of the second supernova . we also predict the unknown radial velocity of each system , as well as the total transverse velocity and ns1 spin tilt angle for those systems where these properties have not been measured yet . we achieve these goals by means of monte carlo simulations , incorporating orbital dynamics during supernova explosions and kinematic of binaries in the galactic potential , as detailed in what follows . we start our analysis by deriving a probability distribution function ( pdf ) for the natal kick of ns2 , which depends on the semi - major axis and orbital eccentricity immediately after the second supernova ( @xmath18 and @xmath19 respectively ) . as the tidal interaction between two neutron stars in dns is extremely weak ( bildsten & cutler 1992 ) , the major mechanism that changes the orbital semi - major axis and eccentricity of dns is the emission of gravitational waves . to account for gravitational radiation driven orbital evolution , we use the current observational parameters as initial conditions and integrate equations ( 35 ) and ( 36 ) in junker & schfer ( 1992 ) backwards in time . because of the uncertainty in the true age of the observed dns ( see kiziltan et al . 2009 for details ) , we randomly draw the integration time @xmath20 from a uniform distribution of ages between 0 and @xmath21 , where @xmath21 is the characteristic age of the dns . if @xmath21 is greater than the age of our galaxy , which is about 10 gyr , we use 10 gyr as the upper limit for @xmath20 instead . for psr b1534 + 12 , psr b1913 + 16 , and psr j0737 - 3039 , different upper limits are adopted for @xmath20 , as discussed in more detail in the next section . once @xmath18 and @xmath22 are known from the gravitational radiation integration , we randomly draw the natal kick velocity ( @xmath23 ) from a uniform distribution of values between 0 and 2500 km / s ( i.e. , we assume no prior kick distribution derived from pulsar samples analysis ) , and the kick direction @xmath24 from an isotropic distribution . here @xmath25 is the polar angle between the natal kick velocity and the instantaneous orbital velocity of ns2 s progenitor at the time of the second supernova explosion , and @xmath26 is the corresponding azimuthal angle ( see figure 1 in kalogera 2000 for a graphical representation ) . using the conservation laws of orbital energy and angular momentum expressed by equations ( 19 ) and ( 20 ) in willems et al . ( 2005 ) , we obtain the orbital semi - major axis ( @xmath27 ) immediately before the second supernova and the mass of ns2 s helium star ( @xmath28 ) . the pre - supernova orbit is assumed to be tidally circularized at the time of the second supernova explosion . on the other hand , the spin tilt angle of ns1 , which is the angle between ns1 axis and the post supernova orbital angular momentum axis , can be calculated by following equation ( 3 ) in kalogera ( 2000 ) . also , using equations ( 12 ) in the same paper , we find the polar angle @xmath29 between the natal kick and the pre - supernova orbital angular momental axis . immediately before the second supernova explosion , there may be mass transfer from ns2 s progenitor to ns1 . if the radius @xmath30 of ns2 s progenitor is greater than its roche lobe radius @xmath31 , which is approximated by equation ( 2 ) in eggleton ( 1983 ) , roche lobe overflow ( rlo ) takes place . since ns2 s progenitor is a bare helium star before the second supernova , we use equation ( a10 ) in kalogera & webbink ( 1998 ) to calculate @xmath30 . note that the left hand side of that equation should be @xmath32 instead of @xmath33 . in order to have dynamically stable rlo , so that ns2 s progenitor does not coalesce with ns1 to form an isolated black hole , the mass ratio @xmath28 to @xmath34 needs to be less than 3.5 ( ivanova et al . even if ns2 s progenitor does not overflow its roche lobe , there may be mass transfer by stellar wind . in addition , @xmath28 is restricted to be less than 8 m@xmath7 , which is a conventional limit that a helium star will become a neutron star instead of a black hole ( see , e.g. , figure 1 in belczynski et al . 2002 ; table 16.4 in tauris & van den heuvel 2004 ) . the conservation laws of orbital energy and angular momentum allow for real solutions only for @xmath28 , @xmath23 , @xmath25 , @xmath26 , and @xmath27 values satisfying the orbital dynamics constraints given by equation ( 21)(27 ) in willems et al . ( 2005 ) . for psr b1534 + 12 , psr b1913 + 16 , and psr j0737 - 3039 , these constraints are supplemented with the measured ns1 spin tilt angle constraints . applying these constraints together with the constraints imposing stability of any mass transfer taking place right before the second supernova explosion yields a pdf of the natal kick velocity of ns2 . a corresponding pdf for the kick velocity @xmath35 imparted to the binary s center of mass is obtained using equations ( 28)(32 ) in willems et al . in all past studies , kinematic analyses have either been skipped ( wang et al . 2006 ) or simplified to one dimension ( piran & shaviv 2004 ) or have suffered by ad hoc assumptions about the un - measurable radial velocities . willems et al . ( 2006 ) used general dns population synthesis models to guide this assumption . in this sutdy we employ a new method that still uses forward in time kinematic simulations , but they are tied to each observed system instead of being general . we perform a monte carlo simulation of the motion of dns in the galactic potential . for this purpose , we adopt a reference frame with origin at the galactic center , with @xmath36-axis pointing to the northern galactic pole , and with the @xmath37-axis pointing in the direction from the sun to the galactic center . with respect to this reference frame , the sun is located at @xmath38 kpc ( ghez et al . 2008 , gillessen et al . 2009 , reid et al . 2009 , joshi 2007 ) and has a peculiar velocity of @xmath39km / s ( bienaym 1999 ) . with the given @xmath40 and peculiar velocity , the sun is moving at a velocity @xmath41 km / s . to simulate the motion of dns in the galaxy , we randomly distribute a population of newly formed dns in the galactic disk according to a double exponential distribution function @xmath42 here r and z are the cylindrical galactic coordinates , @xmath43 and @xmath44 are the galactic scale length and height respectively . the distribution is normalized to unity by setting @xmath45 . since the progenitors of dns are distributed in the same way as the massive stellar binaries in our galaxy , we choose @xmath43 = 2.8 kpc and @xmath44 = 0.07 kpc ( see joshi 2007 ) . for all of the observed dns except psr b1534 + 12 , we use a population of @xmath210@xmath46 simulated binaries . we use a population of @xmath210@xmath47 in the analysis of psr b1534 + 12 , due to the limiting chance of kicking it to the currently high galactic altitude and simultaneously satisfying the constraints derived from the orbital dynamics of supernova explosions . if the scale height @xmath44 were larger than 70 pc , it would reduce the difficulty in satisfying the galactic dynamics constraints of psr b1534 + 12 . a smaller population of simulated binaries would then be needed for the analysis of this system . the initial center - of - mass velocity of the dns is obtained by summing up the kick velocity @xmath48 imparted to the binary center of mass during the second supernova explosion and the local galactic rotational velocity @xmath49 . given the unknown orientation of the binary orbit in the galactic reference frame , the systemic kick velocity @xmath48 is assumed to be distributed isotropically in space . the magnitude of @xmath48 is drawn from the velocity distribution derived in the previous step specific to each of the observed systems . starting from the randomly generated initial position and velocity , the motion of the dns is calculated by numerically integrating the equations of motion from the dns birth time to the current epoch . the equations of motion are derived from the galactic potential of carlberg & innanen ( 1987 ) with updated model parameters of kuijken & gilmore ( 1989 ) . the dns birth time is chosen according to the dns age @xmath20 generated in the first step of the analysis . lastly , we use the measured current position of the observed dns to further constraint the ranges of possible @xmath23 and @xmath28 . we set the tolerance of this constraint to be within 100 pc from the observed position . we test the dependence of our results on this tolerance level by repeating the analysis with a 10 pc , or 50 pc tolerance , but our results do not show any significant changes . for psr j1518 + 4904 and psr j1756 - 2251 , the position constraint is supplemented with the measured current proper motion constraint . for psr b1534 + 12 , psr b1913 + 16 , and psr j0737 - 3039 , the position constraint is supplemented with both the measured proper motion and ns1 spin tilt angle constraints . the analysis outlined in the previous section allows us to derive pdfs for the kick velocity @xmath23 , ns2 s immediate pre - supernova progenitor mass @xmath28 , the spin tilt @xmath50 of ns1 , the systemic radial velocity @xmath51 , and the current transverse velocity @xmath52 . our primary interest is in @xmath23 and @xmath28 , because they can shed light on the type of supernova that formed ns2 , which will be discussed in more detail in the next section . the remaining quantities give us some predictions on currently unavailable parameters of the 8 known dns in our galaxy . instead of focusing on the most likely values as studies have done in the past , we derive confidence levels for the unknown parameters and they are summarized in table 2 . since the @xmath23 and @xmath28 constraints are correlated , we present our results in the form of confidence level plots of a 2 dimensional joint probability distribution of @xmath23 - @xmath28 , for each dns binary , in addition to the corresponding 1 dimensional pdf s of @xmath23 and @xmath53 . as pointed out by willems et al ( 2006 ) , physical parameter constraints are affected by the dimensionality of the pdf used , because of inherent correlations between parameters and projection effects . this variation is stronger when one solely examines the most likely values ( i.e. , values at the peak of the pdf ) and weakens as broader confidence levels are considered . in what follows , the confidence levels of @xmath23 and @xmath28 are derived from the 2d joint pdf . to calculate the 2d confidence levels , we first bin the data into a 2d grid and normalize the total probability within the 2d grid to unity . then , we add the probability of each bin from the highest to the lowest , until the sum best matches the desired confidence level . on the other hand , the ranges of @xmath51 , @xmath52 , and @xmath50 are derived from the corresponding 1d pdf . in this case , the confidence level is found by having the shortest range of bins that has a total probability that best matches the desired confidence level . lccccc psr b1534 + 12 & 170 - 260 ( 150 - 270 ) & 2.00 - 2.90 ( 1.34 - 3.40 ) & @xmath54 - 50 ( @xmath55 - 145 ) & 121 @xmath8 6 & 25.0 @xmath8 3.8 + psr b1913 + 16 & 200 - 410 ( 190 - 450 ) & 1.40 - 3.30 ( 1.39 - 5.00 ) & @xmath56 - 85 ( @xmath57 - 315 ) & 135 @xmath8 25 & 18 or 162 @xmath8 6 + psr j0737 - 3039 & 5 - 50 ( 5 - 120 ) & 1.25 - 1.55 ( 1.25 - 1.90 ) & 8 - 44 ( @xmath58 - 76 ) & @xmath59 & 0 - 2.0 ( 0 - 7.8 ) + psr j1518 + 4904 & 20 - 80 ( 5 - 110 ) & 1.80 - 3.30 ( 1.49 - 4.70 ) & @xmath60 - 8 ( @xmath61 - 48 ) & 25.4 @xmath8 3.6 & 2 - 18 ( 0 - 32 ) + psr j1756 - 2251 & 5 - 80 ( 5 - 185 ) & 1.25 - 1.90 ( 1.25 - 2.65 ) & @xmath62 - 32 ( @xmath55 - 98 ) & 6.5 - 28.5 ( 6.0 - 64.5 ) & 0 - 3.4 ( 0 - 16.4 ) + psr j1811 - 1736 & 0 - 170 ( 0 - 310 ) & 1.11 - 4.00 ( 1.11 - 8.00 ) & 0 - 112 ( @xmath63 - 210 ) & 30 - 114 ( 4 - 202 ) & 0 - 10.2 ( 0 - 57.6 ) + psr j1829 + 2456 & 5 - 85 ( 5 - 225 ) & 1.40 - 2.70 ( 1.36 - 6.10 ) & @xmath60 - 26 ( @xmath64 - 66 ) & 12 - 54 ( 4 - 102 ) & 0 - 8.5 ( 0 - 30.0 ) + psr j1906 + 0746 & 5 - 170 ( 5 - 510 ) & 1.25 - 2.90 ( 1.25 - 4.80 ) & 12 - 103 ( @xmath65 - 297 ) & 89 - 174 ( 21 - 376 ) & 0 - 6.7 ( 0 - 51.5 ) + in this binary , the masses of ns1 and ns2 are 1.33 @xmath66 and 1.35 @xmath66 respectively . the orbital period is @xmath67 days , with an eccentricity @xmath68 . the angle between ns1 s spin axis and the orbital angular momentum axis is @xmath69 or @xmath70 . the binary is currently located out of the galactic plane , at @xmath71 and @xmath72 , and 1.02 kpc away from us . the observed proper motion @xmath73 mas yr@xmath74 and @xmath75 mas yr@xmath74 . at the measured distance this implies @xmath76 km s@xmath74 and @xmath77 km s@xmath74 , in the reference frame of the sun . also , we use the spin down age , which is 210 myr given by arzoumanian et al . 1999 , as the upper limit of psr b1534 + 12 s age , instead of the characteristic age . furthermore , the current orientation of the orbital angular momentum axis is also constrained for this sytem . the orientation axis can be described by the inclination angle @xmath78 and the angle of the orbital ascending node on the plane of the sky ( @xmath79 ) . the sine of the inclination angle is @xmath80 = 0.975 ( stairs et al . the angle @xmath79 is @xmath81 or @xmath82 , reckoned north through east . the two solutions correspond to @xmath83 and @xmath84 respectively ( bogdanov et al . 2002 ) . since tidal effects between the two neutron stars are negligible , the orbital angular momentum axis keeps a fixed orientation in space . from the kinematic analysis of the galactic motion , the constraints on the present - day orbital inclination and proper motion can therefore be used to determine the systemic kick component @xmath85 parallel to the post - supernova orbital angular momentum axis . following kalogera ( 1996 ) and wex et al . ( 2000 ) , @xmath86 can also be expressed analytically as @xmath87 where @xmath88 @xmath89 @xmath90 here , @xmath91 ( @xmath92 ) is the @xmath37 , @xmath93 , or @xmath36 component of the natal kick velocity in the frame centered on ns2 s progenitor ( see figure 1 in kalogera for a graphical representation ) . note that the sign of @xmath86 needs to be consistent with the sign of @xmath94 . hence , through equation ( 2 ) , the @xmath86 derived from observations gives us a constraint on the y and z components of the natal kick , as well as @xmath50 . the results are shown in figures 1 and 2 . right before the second supernova , there must have been rlo from the progenitor of ns2 to ns1 , as noted previously by thorsett et al . as seen in the 2d @xmath23-@xmath28 joint probability distribution in figure 1 , the kick imparted to ns2 at birth is between 150 and 270 km / s , while the progenitor mass @xmath28 is between 1.34 and 3.40 m@xmath0 , both at 95% confidence . as seen in figure 2 , polar kicks ( i.e. , @xmath95 or @xmath96 ) are unlikely , and the possible kick directions are asymmetric about the pre - supernova orbital plane ( i.e. , @xmath29 = 90@xmath97 ) . this asymmetry comes from the constraint on the orientation of the post - supernova orbital angular momentum axis on the sky . moreover , we arrive at the same conclusion as thorsett et al . ( 2005 ) that @xmath98 is very unlikely . the current radial velocity of this binary is between -110 and 145 km / s , at 95% confidence , and it is likely to be moving away from us . within the observationally constrained range , the @xmath52 pdf is roughly flat . when comparing with recent analyses by willems et al . ( 2004 ) , thorsett et al . ( 2005 ) , and wang et . 2006 ) , our limits on @xmath23 and @xmath28 are consistent wtih thorsett et al . ( 2005 ) , and are more constrained than willlems et al . ( 2004 ) and wang et al . ( 2006 ) , because willems et al . ( 2004 ) did not have the spin tilt @xmath50 measurement and wang et al . ( 2006 ) did not have the proper motion and position constraints in their analysis . our limits on the angle @xmath29 , which is the polar angle between the natal kick direction and the pre - supernova orbital angular momentum , are not as constrained as the limits shown in thorsett et al . ( 2005 ) . through recent further investigation , it has been found that this difference is due to a code typo related to the galactic latitude of the source , which fortunately affected significantly only the @xmath29 results ( dewey & stairs 2010 , private communication ) . the masses of ns1 and ns2 in this binary are 1.44 @xmath66 and 1.38 @xmath66 respectively . the orbital period is @xmath99 days , with an eccentricity @xmath100 . the angle between the ns1 spin axis and the post supernova orbital angular momentum axis is @xmath101 or @xmath102 . at the current time , this binary is about 300 pc away the galactic plane , at @xmath103 and @xmath104 , and 8.3 kpc away from us . the observed proper motion in right ascension and declination is @xmath105 mas yr@xmath74 and @xmath106 mas yr@xmath74 . at the measured distance , this implies @xmath107 km s@xmath74 and @xmath108 km s@xmath74 , in the reference frame of the sun . in addition , instead of the characteristic age , we use the spin down age , which is 80 myr ( arzoumanian et al . 1999 ) , as the upper limit on psr b1913 + 16 s age . the results are displayed in figures 3 and 4 . as shown in the 2d @xmath23-@xmath28 joint probability distribution in figure 3 , the kick imparted to ns2 at birth is between 190 and 450 km s@xmath74 , while the progenitor mass @xmath28 is between 1.4 and 5.0 m@xmath0 , both at 95% confidence . as seen from figure 4 , both polar and planar kick directions are allowed . the current radial velocity is between @xmath57 and 315 km s@xmath74 at 95% confidence . within the observationally constrained range , the @xmath52 pdf is roughly flat , with a slight preference for transverse velocities of @xmath2120 km / s . furthermore , although observations show a degeneracy in @xmath50 at @xmath109 and @xmath110 , our analysis shows that @xmath50 is much more likely to be @xmath109 . in addition , although our confidence intervals of @xmath23 and @xmath28 are consistent with willems et al . ( 2004 ) and wang et . al . ( 2006 ) , our reported intervals are tighter , because willems et al . ( 2004 ) and wang et al . ( 2006 ) reported the allowed ranges of @xmath23 and @xmath28 instead of confidence intervals . the masses of the two neutron stars in this binary are 1.34 and 1.25 m@xmath0 , where the more massive one is ns1 . they are orbiting each other with a period of 0.102 days , and an eccentricity of 0.0878 . at the current time , the binary is close to the galactic plane , at @xmath111 and @xmath112 . through direct measurement of geometric parallax , deller et al . ( 2009 ) measured the distance to psr j0737 - 3039 to be 1.15 kpc , and the proper motion in r.a . and declination to be @xmath1133.82 and 2.13 mas yr@xmath74 respectively . at the measured distance , this implies @xmath114 km s@xmath74 and @xmath115 km s@xmath74 , in the reference frame of the sun . in addition , lorimer et al . ( 2007 ) constrained the possible age range to be either @xmath116 or @xmath117 myr . we adopt these age ranges in our analysis of psr j0737 - 3039 . as the radio pulse profile of psr j0737 - 3039a ( i.e. , ns1 ) has not changed significantly through six years of observation , the spin tilt angle of j0737 - 3039a with respect to the post supernova orbital angular momentum axis is believed to be small . however , there is no well constrained value for this angle . ferdman et al . ( 2008 ) and breton ( 2008 ) argue that there are two possible models to explain the observed pulse profile : the single - cone and two - cone emission model . in this context , the number of emission cones means the number of emitting cones that intersect with our line of sight . each model gives a different spin tilt angle estimate . ferdman et al . ( 2008 ) estimate this angle to be < 15@xmath97 for the single - cone model and < 6.1@xmath97 for the two - cone model , both at 68.3 % confidence . due to this uncertainty , we take the spin tilt angle @xmath50to be < 15@xmath97 in our work , which corresponds to 68.3 % confidence value in the single - cone model and 95.4 % confidence value in the two - cone model in ferdman et al ( 2008 ) . the results are displayed in figures 5 - 7 . as shown in the 2d @xmath23-@xmath28 joint probability distribution , the kick recieved by ns2 ( i.e. , psr j0737 - 3039b ) at birth is between 5 and 120 km s@xmath74 , while the progenitor mass @xmath28 is between 1.25 and 1.90 m@xmath0 , both at 95% confidence . the probability distribution of the kick direction is symmetric about the pre - supernova orbital plane ( i.e. , @xmath118 = 90@xmath97 ) , and planar kicks are more favorable than polar kicks . the pdf of the semi - major axis immediately before the second supernova ( @xmath27 ) peaks at two narrow ranges : @xmath119 and @xmath120 @xmath121 . the first peak corresponds to an age of @xmath116 myr , and the second peak to an age of @xmath117 myr . at the current time , psr j0737 - 3039 is more likely to be moving away from us , with a radial velocity between @xmath58 and 76 km / s with 95% confidence . within the observationally constrained range , the @xmath52 pdf peaks at @xmath220 km / s . furthermore , rlo from ns2 s progenitor to ns1 before the second supernova explosion can not be avoided . to test the dependence of the presented results on the adopted upper limit for @xmath50 , we repeated the analysis for the upper limits of @xmath122 and @xmath123 . the results do not show any significant differences compared to those obtained with the @xmath124 upper limit . this is not surprising , because the @xmath50 pdf as shown in figure 8 , which is obtained from the same constraints as before except raising the upper limit of @xmath50 to 60@xmath97 , peaks strongly at 0@xmath97 and decays exponentially with increasing values of @xmath50 . comparatively , our confidence intervals on @xmath23 and @xmath28 are consistent with stairs et al . ( 2006 ) and willems et al . ( 2006 ) , but are much more constrained than wang et al . ( 2006 ) , mainly because they did not have the proper motion constraint in their analyses . the masses of the two neutron stars in this binary are 0.72 and 2.00 m@xmath0 , where the less massive one is ns1 . however , each one has an uncertainty of @xmath20.5 m@xmath0 . in our analysis , we choose @xmath34 to be 1.23 and @xmath125 to be 1.49 , so that @xmath34 is greater than the smallest measured neutron star mass , which is about 1.2 @xmath126 . the characteristic age is 20 gyr , which is larger than a hubble time . thus , we set the upper limit on psr j1518 s age to be 10 gyr instead . the two neutron stars are orbiting each other with a period of 8.63 days , and an eccentricity of 0.249 . at the current time , this binary is out of the galatic plane , at @xmath127 and @xmath128 , and at a distance of 0.625 kpc from the sun . the proper motion in r.a . and dec . is @xmath129 and @xmath130 mas yr@xmath74 . at the measured distance , this implies @xmath131 and @xmath132 km s@xmath74 , in the reference frame of the sun . there is no spin tilt measurement available for this system . the results are shown in figures 9 and 10 . as seen in the 2d @xmath23-@xmath28 joint probability distribution , the kick imparted to ns2 at birth is between 5 and 110 km / s , while the progenitor mass @xmath28 is between 1.5 and 4.7 m@xmath0 , both at 95% confidence . both polar and planar kick directions are possible , but the planar kicks have a higher probability . the spin tilt angle @xmath50 is smaller than 32@xmath97 at 95% confidence . the radial velocity is found to be between -62 and 48 km / s at 95% confidence , and the system is currently likely moving towards us . within the observationally constrained ranges , the @xmath52 pdf is roughly flat . our limits of @xmath23 and @xmath28 are much more constrained than those of wang et al . ( 2006 ) , primarily because they did not have the proper motion and galactic position constraints included in their analysis . the masses of the two neutron stars in this binary are 1.31 and 1.26 m@xmath0 , where the more massive one is ns1 . the characteristic age is 443 myr . the two neutron stars are in an orbit with a period of 0.320 days , and an eccentricity of 0.181 . at the current time , this binary is close to the galactic plane , at @xmath133 and @xmath134 , and at a distance of 2.5 kpc away from us . only proper motion @xmath135 is measured , which is -0.7 mas / yr . at the measured distance , this implies @xmath136 km / s , in the reference frame of the sun . there is currently no spin tilt measurement available for this system . the results are shown in figures 11 and 12 . from the 2d @xmath23-@xmath28 joint probability distribution , the kick received by ns2 at birth is between 5 and 185 km s@xmath74 , while the progenitor mass @xmath28 is between 1.25 and 2.65 m@xmath0 , both with 95% confidence . both polar and planar kicks are allowed , but planar kicks are more favorable than polar ones . the spin tilt angle of ns1 is smaller than 16.4@xmath97 with 95% confidence . at the current time and at the measured distance , psr j1756 - 2251 has a total transverse velocity between 6.0 and 64.5 km s@xmath74 , and a radial velocity between @xmath55 and 98 km s@xmath74 , both with 95% confidence . furthermore , this binary is equally likely to be moving towards us as away from us . when comparing with wang et al . ( 2006 ) , we have tighter constrained limits on @xmath23 and @xmath28 , due to the fact that we have the additional constraint on the proper motion @xmath135 . the two neutron stars have masses of 1.62 and 1.11 m@xmath0 , where the more massive one is ns1 . however , each mass measurement has an uncertainty of @xmath20.5 m@xmath0 . psr j1811 has a characteristic age of 1830 myr . the neutron stars are in an orbit with a period of 18.8 days , and an eccentricity of 0.828 . at the current time , this binary is close to the galactic plane , at @xmath137 and @xmath138 , and at a distance of 6.0 kpc away from us . like psr j1756 - 2251 , there are currently no spin tilt nor proper motion measurements available . the results are displayed in figures 13 and 14 . the 2d @xmath23-@xmath28 joint probability distribution shows that the kick given to ns2 at birth is less than 310 km s@xmath74 , while the progenitor mass @xmath28 is between 1.11 and 8.0 m@xmath0 , both with 95% confidence . both polar and planar kicks are allowed , but the planar kicks are more favorable than the polar ones . the spin tilt angle @xmath50 is smaller than 57.6@xmath97 with 95% confidence . at the measured distance , the total transverse velocity of psr j1811 - 1736 at the current epoch is between 4 and 202 km s@xmath74 , and the radial velocity between -90 and 210 km s@xmath74 with 95% confidence , which means this binary is likely moving away from us . although our upper boundary of the 95% confidence intervals of @xmath28 is the same as the upper limit found by wang et al . ( 2006 ) , the value of our lower boundary is less than their lower limit . this is because wang et al . ( 2006 ) put a conservative lower limit of 2.1 m@xmath0 on the progenitor mass of a neutron star , which we did not . on the other hand , our limits on @xmath23 are more constrained than those of wang et al . ( 2006 ) , as we compute confidence levels instead of allowed ranges of solution as wang et al . ( 2006 ) . the two neutron stars of this binary have masses of 1.14 and 1.36 m@xmath0 , where the less massive one is ns1 . however , there is an uncertainty of @xmath20.5 m@xmath0 in the mass measurements . the characteristic age is 12.4 gyr . for the same reason as psr j1518 + 4904 , we set an the upper limit of 10 gyr on the age of this binary in our analysis . the neutron stars are orbiting each other with a period of 1.18 days , and an orbital eccentricity of 0.139 . at current time , this binary is out of the galactic plane , at @xmath139 and @xmath140 , and it is 1.2 kpc away from us . although there is no spin tilt nor proper motion measurements available at current time , lorimer et al . ( 2005 ) derived that the transverse velocity is smaller than 118 km s@xmath74 . the results are shown in figure 15 . the @xmath23-@xmath28 joint probability distribution shows that the kick given to ns2 at birth is between 5 and 225 km s@xmath74 , while the progenitor mass @xmath28 is between 1.4 and 6.1 m@xmath0 with 95% confidence . the pdfs of @xmath50 and @xmath29 are similar to those of psr j1756 - 2251 . both polar and planar kicks are allowed , but planar kicks are more probable . the spin tilt angle @xmath50 is less than 30.0@xmath97 with 95% confidence . at the known distance , the total transverse velocity of psr j1829 + 2456 at the current time is between 4 and 102 km s@xmath74 , and the radial velocity between @xmath64 and 66 km s@xmath74 with 95% confidence . in addition , the probability that this system is moving towards the sun is @xmath220% higher than that it is moving away from the sun . comparing our @xmath23 and @xmath28 results with wang et al . ( 2006 ) , we have more constrained limits , as they do not include any kinematic constraints . the masses of the neutron stars are 1.37 and 1.25 m@xmath0 , where the more massive one is ns1 . unlike the other 7 binaries , the observed pulsar is ns2 . this implies the binary is relatively young , with a charateristic age of 0.112 myr only . the neutron stars are in an orbit with a period of 0.166 days , and an orbital eccentrirctiy of 0.0853 . at current time , this binary is close to the galactic plane , at @xmath141 and @xmath142 , and it is 5.4 kpc away from us . again , there are no spin tilt nor proper motion measurements available yet . the results are displayed on figure 16 . the @xmath23-@xmath28 joint probability distribution shows that the kick given to ns2 at birth is between 5 and 510 km s@xmath74 , while the progenitor mass @xmath28 is between 1.25 and 4.80 @xmath126 with 95% confidence . the @xmath29 and @xmath50 pdf s are similar to those of psr j1756 - 2251 . both polar and planar kicks are possible , with the planar kicks having a higher probability . the polar angle @xmath50 is less than 51.5@xmath97 with 95% confidence . at the measured distance , the total transverse velocity of psr j1906 + 0746 at the current time is between 21 and 376 km s@xmath74 , and the radial velocity is between @xmath65 and 297 km s@xmath74 with 95% confidence . it is most likely that this binary is moving away from us . furthermore , right before its supernova , the progenitor of ns2 must have been overflowing its roche lobe , transferring mass to ns1 . our limits on @xmath23 and @xmath28 are tighter constrained than wang et al . ( 2006 ) , because they did not have a detailed analysis on this system . instead , they calculated the allowed @xmath23 with an assumption that @xmath28 was between 2.1 and 8.0 m@xmath0 . the anisotropic mass and neutrino ejection in a supernova explosion gives a recoil natal kick to the newly formed neutron star . iron core collapse supernova and ecs produce natal kicks of different strength because of the different physical conditions of the nascent neutron star immediately before the supernovae . there are several reviews ( bethe 1990 , mezzacappa 2005 , woosley & bloom 2006 , kotake et al 2006 , and janka et al . 2007 ) that summarize our current understanding of the iron core collapse supernova . after iron is synthesized in the core , the massive star reaches the final stage of hydrostatic nuclear burning , because the synthesis of any heavier elements requires energy instead of releasing energy . when the iron core grows by silicon shell burning around the core to a mass above the chandrasekhar mass limit , electron degeneracy pressure can not support the core any more and it starts to collapse . the iron core continues to collapse until it reaches nuclear density ( i.e. , @xmath143 g @xmath144 ) . since nuclear matter has a much lower compressibility , the collapse decelerates and the core bounces back because of the increased nuclear matter pressure . this drives a shock wave and eventually explodes the outer layers of the massive star away in a supernova explosion . on the other hand , the ecs scenario is described in the reviews of miyaji et al . ( 1980 ) and nomoto ( 1987 ) . after carbon burning within the core of a post main sequence massive star , a @xmath145 core is formed . neon can not be ignited if the mass of the @xmath145 core is less than the critical mass of 1.37 m@xmath0 for neon ignition . as the @xmath145 core is cooled by neutrino emissions , it becomes strongly degenerate . when the mass of the core grows by carbon shell burning and approaches the chandrasekhar mass limit , electron captures onto @xmath6 and @xmath5 take place . this leads to a decrease in the electron degeneracy pressure and the chandrasekhar mass limit , which induces a rapid contraction of the core . the rapid core contraction ignites the oxygen deflagration , which incinerates materials into nuclear statistical equilibrium ( nse ) . due to the rapid electron capture onto nse elements , the collapse of the core accelerates . similar to the iron core collapse picture , when the core reaches the nuclear matter density , it bounces back and drives a shock wave that eventually explodes the star in a supernova . gutirrez et al . ( 2005 ) show that the abundance of @xmath6 in @xmath145 core needs to be greater than 15% , in order to have nse developed through electron captures . however , in simulations of evolving massive agb stars by siess ( 2007 ) , the abundance of @xmath6 after carbon burning is smaller by a factor of @xmath210 than what is required to drive an explosion by electron captures . therefore , the mechanism of developing an ecs is not a settled issue yet . in addition , podsiadlowski et al . ( 2004 ) studied the binary evolution and the dynamics of core collapse , and suggested that ecs could only occur in interacting binaries , but not in single stars . in recent simulations of supernova explosions from the collapse of @xmath145 cores by kitaura et al . ( 2006 ) , they found these supernovae are powered by neutrino heating and neutrino - driven wind of the nascent neutron star . scheck et al . ( 2006 ) showed that the shock wave in these supernovae can propagate outwards on a relatively short timescale after the rebound of the core , which means the non - radial hydrodynamics instabilities do not have time to merge and grow to global asymmetry before the anisotropic pattern freezes out in the accelerating outward motion of the shock wave . as a result , _ _ the natal kick of an ecs due to anisotropic mass ejection is expected to be fairly small(see scheck et al . furthermore , as the @xmath145 core that eventually explodes in an ecs is not massive enough to ignite neon , _ _ the mass of the neutron star progenitor should be less than the progenitors that explode in an iron core collapse supernova . _ _ _ _ the present analysis is the first to include _ all _ currently known dns systems and account of all orbital and kinematic constraints . we also employ a novel method for dealing with the uncertainty due to the un - measurable radial velocities and we focus on the derived pdfs and associated confidence levels , instead of just the most likely values , which can be misleading at times . our results are summarized in table 2 ; the derived constraints are consistent with earlier studies ( when available ) but typically limit parameters to narrower ranges , and in this sense they represent the best available constraints on the formation of neutron stars in dns systems . _ _ in the context of our current understanding of massive star core collapse , we can use our results to draw a number of conclusions : + ( 1 ) psr j0737 - 3039 has a @xmath23 upper limit of 120 km / s and a @xmath28 upper limit of 1.9 @xmath126 at 95% confidence . therefore , the formation of ns2 ( i.e. , pulsar b ) likely occurred through an ecs event . this is consistent with the speculation of podsiadlowski et al . ( 2005 ) based on the space velocity and orbital eccentricity of psr j0737 - 3039 . for psr b1534 + 12 and psr b1913 + 16 , @xmath23 are 150 - 270 km / s and 190 - 450 km / s ( 95% ) respectively , which means ns2 in both system must have received a significant recoil natal kick at birth . psr b1534 + 12 has a @xmath28 upper limit of 3.4 @xmath126 ( 95% ) , while psr b1913 + 16 has a @xmath28 upper limit of 5.0 @xmath28 ( 95% ) . because of the relatively high @xmath23 ranges and @xmath28 upper limits , ns2 in both systems are probably formed through an iron core collapse supernova event . + ( 2 ) psr j1518 + 4904 has a @xmath146 upper limit of 110 km / s and a @xmath28 upper limit of 4.7 @xmath126 at 95% confidence . even though ns2 likely received a low recoil velocity at birth , we can not firmly conclude which type of supernova occurred during the formation of ns2 , due to the relatively high @xmath28 upper limit . + ( 3 ) psr j1756 - 2251 has a @xmath23 upper limit of 80 km / s and a @xmath28 upper limit of 1.90@xmath126 at 60% confidence , hence the formation of ns2 might possibly relate to an ecs event . however , the @xmath23 and @xmath28 upper limits are 185 km / s and 2.65 @xmath126 respectively at 95% confidence , which means the likelihood of ns2 formed through an ecs event is lower than that of psr j0737 - 3039 . + ( 4 ) psr j1811 - 1736 , psr j1829 + 2456 , and psr j1906 + 0746 have @xmath23 upper limits of several hundreds of km / s , and @xmath28 upper limits ranging from 4.80 to 8.0 @xmath126 ( i.e. , the conventional limit on neutron star progenitor mass ) at 95% confidence . since both low and high @xmath23 and @xmath28 are possible , we can not conclude which type of supernova event is more favorable for the formation history of ns2 in these systems . + ( 5 ) for all of the known dns except psr b1534 + 12 , all available constraints are consistent with imparting a polar or planar kick to ns2 at birth . for psr b1534 + 12 , the kick direction is constrained to be polar . + ( 6 ) for psr b1534 + 12 , psr j0737 - 3039 and psr j1906 + 0746 , the pre - supernova orbit is so tight that they can not avoid a rlo from ns2 progenitor to ns1 just before the second supernova explosion . + ( 7 ) furthermore , despite the low orbital eccentricity in psr j1829 + 2456 and psr j1906 + 746 , ns2 could have a high progenitor mass and have received a high recoil natal kick at birth , as shown in the 2d joint pdf of these systems ( see figures 15 , and 16 ) . + ( 8) we also tested the dependence of our psr j0737 - 3039 results on the current upper limit of the spin - orbital misalignment angle @xmath50 of ns1 ( i.e. , pulsar a ) . as shown in figure 8 , the @xmath50 pdf peaks strongly at @xmath147 , so choosing an upper limit of 15@xmath97 or 60@xmath97 does not affect our results noticably . arzoumanian , z. , cordes , j. m. , & wasserman , i. 1999 , , 520 , 696 belczynski , k. , kalogera , v. , bulik , t. 2002 , , 572 , 407 bethe , h. a. 1990 , rev . phys . , 62 , 801 bienaym , o. 1999 , , 341 , 86 bildsten , l. , cutler , c. 1992 , , 400 , 175 bogdanov , s. , pruszyska , m. , lewandowski , w. , & wolszczan , a. 2002 , , 581 , 495 breton , r. p. 2009 thesis at mcgill university burgay , m. , damico , n. , possenti , a. , manchester , r. n. , lyne , a. g. , et al . 2003 , , 426 , 531 champion , d. j. , lorimer , d. r. , mclaughlin , m.a . , cordes , j. m. , arzoumanian , z. , et al . 2004 , , 350 , l61 champion , d. j. , lorimer , d. r. , mclaughlin , m. a. , xilourix , k. m. , arzoumanian , z. , et al . 2005 , , 363 , 929 carlberg , r. g. , & innanen , k. a. 1987 , , 94 , 666 corongiu , a. , kramer , m. , stappers , b. w. , lyne , a. g. , jessner , a. , et al . 2007 , , 462 , 703 damour , t. , & taylor , j. h. 1991 , , 366 , 501 deller , a. t. , bailes , m. , & tingay , s. j. 2009 , science , 323 , 1327 eggleton , p. p. 1983 , , 268 , 368 faulkner , a. j. , kramer , m. , lyne , a. g. , manchester , r. n. , mclaughlin , m. a. , et al . 2005 , , 618 , l119 ferdman , r.d . 2008 , phd . thesis at the university of british columbia ferdman , r. d. , stairs , i. h. , kramer , m. , manchester , r. n. , lyne , a. g. , et al . in 40 years of pulsars : millisecond pulsar , magnetars and more . aip conference proceedings , 2008 , pp . 474 - 478 frail , d. a. , goss , w. m. , & whiteoak , j. b. z. 1994 , , 437 , 781 ghez , a. m. , salim , s. , weinberg , n. n. , lu , j. r. , do , t. , et al . , 2008 , , 689 , 1044 gillessen , s. , eisenhauer , f. , trippe , s. , alexander , t. , genzel , r. , et al . 2009 , , 692 , 1075 gutirrez , j. , canal , r. , garca - berro , e. 2005 , , 435 , 231 harrison , p. a. , lyne , a. g. , & anderson , b. 1993 , , 261 , 113 hobbs , g. , lorimer , d. r. , lyne , a. g. , kramer , m. 2005 , , 360 , 974 hulse , r. a. , & taylor , j. h. 1975 , , 195 , l51 ivanova , n. , belczynski , k. , kalogera , v. , rasio , f. a. , taam , r. e. 2003 , , 592 , 475 janka , h .- th . , langanke , k. , marek , a. , martnez - pindeo , g. , & mller , b. 2007 , , 442 , 38 janssen , g. h. , stappers , b. w. , kramer , m. , nice , d. j. , jessner , a. , et al . 2008 , , 490 , 753 joshi , y. c. 2007 , , 378 , 768 junker , w. , & schfer , g. 1992 , , 254 , 146 kalogera , v. 1996 , , 471 , 352 kalogera , v. 2000 , , 541 , 319 kalogera , v. , & webbink , r. f. 1998 , , 493 , 351 kasian , l. in 40 years of pulsars : millisecond pulsars , magnetars and more . aip conference proceedings , 2008 , pp . 485 - 487 kitaura , f. s. , jankka , h .- th . , & hillebrandt , w. 2006 , , 450 , 345 kiziltan b. , & thorsett s. e. 2009 , arxiv0909.1562 konacki , m. , wolszczan , a. , & stairs , i. h. 2003 , , 589 , 495 kotake , k. , sato , k. , & takahashi , k. 2006 , rep . , 69 , 971 kuijken , k. , & gilmore , g. 1989 , , 239 , 571 lorimer , d. r. , stairs , i. h. , freire , p. c. , cordes , j. m. , camilo , f. , et al . 2006 , , 640 , 428 lorimer , d. r. , freire , p. c. c. , stairs , i. h. , kramer , m. , mclaughlin , m. a. , et al . 2007 , , 379 , 1217 lyne , a. g. , & loimer , d. r. 1994 , , 369 , 127 lyne , a. g. , camilo , f. , manchester , r. n. , bell , j. f. , kaspi , v. m. , et al . 2000 , , 312 , 698 lyne , a. g. , burgay , m. , kramer , m. , possenti , a. , manchester , r. n. , et al . 2004 , science , 303 , 1153 mezzacappa , a. 2005 , annu . nucl . part . , 55 , 467 miyaji , s. , nomoto , k. , yokoi , k. , & sugimoto , d. 1980 , , 32 , 303 nomoto , k. 1987 , , 332 , 206 pfahl , e. , rappaport , s. , podsiadlowski , p. , spruit , h. 2002 , , 574 , 364 podsiadlowski , ph . , langer , n. , poelarends , a. j. t. , rappaport , s. , heger , a. , et al . 2004 , , 612 , 1044 podsiadlowski , ph . , dewi , j. d. m. , lesaffre , p. , miller , j. c. , newton , w. g. , et al . 2005 , , 361 , 1243 piran , t. , shaviv , n. j. 2004 , , 94 , 051102 reid , m. j. , menten , k. m. , zheng , x. w. , brunthaler , a. , moscadelli , l. , et al . 2009 , , 700 , 137 scheck , l. , plewa t. , janka , h .- th . , kifonidis , k. , & mller , e. 2004 , , 92 , 011103 scheck , l. , kifonidis , k. , janka , h .- th . , & mller , e. 2006 , , 457 , 963 siess , l. 2007 , , 476 , 893 stairs , i. h. 2004 , science , 304 , 547 stairs , i. h. in 40 years of pulsars : millisecond pulsars , magnetars and more . aip conference proceedings , 2008 , pp . 424 - 432 stairs , i. h. , thorsett , s. e. , & arzoumanian , z. 2004 , , 93 , 141101 stairs , i. h. , thorsett , s. e. , dewey , r. j. , kramer , m. , & mcphee , c. a. 2006 , , 373 , l50 stairs , i. h. , thorsett , s. e. , taylor , j. h. , & wolszczan , a. 2002 , , 581 , 501 tauris , t. m. , & van den heuvel e. p. j. 2004 , in compact stellar x - ray sources , ed . w. h. g. lewin & m. van der klis ( cambridge : cambridge univ . press ) , arxiv : astro - ph/0303456 taylor , j. h. , fowler , l. a. , & mcculloch , p. m. 1979 , , 277 , 437 taylor , j. h. , hulse , r. a. , fowler , l. a. , gullahorn , g. e. , & rankin , j. m. 1976 , , 206 , l53 taylor , j. h. , & weisberg , j. m. 1982 , , 253 , 908 taylor , j. h. , & weisberg , j. m. 1989 , , 345 , 434 thorsett , s. e. , & chakrabarty , d. 1999 , , 512 , 288 thorsett , s. e. , dewey , r. j. , & stairs , i. h. 2005 , , 619 , 1036 van den heuvel , e. p. j. , in the proceedings of the 5th integral workshop on the integral universe , 2004 , p. 185 van den heuvel , e. p. j. , in the multicolored landscape of compact objects and their explosive origins . aip conference proceedings , 2007 , pp . 589 - 606 wang , c. , lai , d. , & han , j. l. 2006 , , 639 , 1007 wex , n. , kalogera , v. , & kramer , m. 2000 , , 528 , 401 willems , b. , kalogera , v. , & henninger , m. 2004 , , 616 , 414 willems , b. , henninger , m. , levin , t. , ivanova , n. , kalogera , v. , et al . 2005 , , 625 , 324 willems , b. , kaplan , j. , fragos , t. , kalogera , v. , & belczynski , k. 2006 , , 74 , 043003 wolszczan , a. 1991 , , 350 , 688 woosley , s. e. , & bloom , j. s. 2006 , , 44 , 507

extensive muon spin relaxation (sr ) measurements have been performed to determine the magnetic field penetration depth in high cuprate superconductors with simple hole doping , zn - doping , overdoping , and formation of static sdw nano islands . system dependence of ( superconducting carrier density / effective mass ) reveals universal correlations between and in all these cases with / without perturbation . evidence for spontaneous and microscopic phase separation into normal and superconducting regions was obtained in the cases with strong perturbation , i.e. , zn - doping ( swiss cheese model ) , overdoping , and coexisting magnetic and superconducting states ( sdw nano islands ) . the length scale of this heterogeneity is shown to be comparable to the in - plane coherence length . we discuss implication of these results on condensation mechanisms of htsc systems , resorting to an analogy with purehe andhe/he mixture films on regular and porous media , reminding essential features of bose - einstein , bcs and kosterlitz - thouless condensation / transition in 3-d and 2-d systems , and comparing models of be - bcs crossover and phase fluctuations . combining thesr results on and the pseudo - gap behavior , we propose a new phase diagram for htsc , characterized by : ( 1 ) the line that represents pair formation ; ( 2 ) disappearance of this line above the critical hole concentration ; ( 3 ) in the underdoped region between and , there exists another line which corresponds to the onset of dynamic superconductivity with superconducting phase fluctuations ; and ( 4 ) the overdoped region being phase separated between hole - poor superfluid and hole - rich normal fermion metal regions . finally , we elucidate anomalous reduction of superfluid spectral weight in the crossover from superconducting to metallic ground states found not only in overdoped htsc cuprates but also in pressurized organic bedt and ac fulleride superconductors . * superfluid density of high- cuprate systems : implication on condensation mechanisms , heterogeneity and phase diagram *

cond-mat0212643

muon spin relaxation ( @xmath0sr ) technique [ 1 - 3 ] has made significant contributions to studies of high-@xmath2 superconductors ( htsc ) . measurements in transverse - fields ( tf-@xmath0sr ) provide a reliable way to determine the magnetic field penetration depth @xmath1 of type - ii superconductors and to infer details of the flux vortex lattice , while measurements in zero - filed ( zf-@xmath0sr ) allow studies of static magnetism , leading to determination of the phase diagrams , local spin structures , and volume fraction of static magnetism . in this article , we provide a perspective view on development of these measurements in htsc systems and their implications on condensation mechanisms , heterogeneity and phase diagrams , focusing on the behavior of @xmath3 ( superconducting carrier density / effective mass ) . since the parameter @xmath3 in htsc systems plays a role similar to the superfluid density in superfluid he systems , we refer to it as `` superfluid density '' in the title and the text of this paper . in the following section , section ii , we start with the results of the penetration depth measurements of hole - doped cuprate systems , which exhibit universal correlations between @xmath2 and @xmath3 . we discuss these correlations in terms of energy scales of superconducting carriers . then , we look into the cases involving heterogeneity , i.e. , ( cu , zn ) substitution and overdoping in section iii , and systems with coexisting superconductivity and static magnetism , with the formation of magnetic nano - islands , in section iv . in all these cases , the superfluid density has a trade - off with the volume of non - superconducting regions created by perturbation and/or phase separation . we shall see that the universal correlations between @xmath2 and the superfluid density @xmath3 are followed not only by the simple hole doped cuprates but also by these htsc systems with microscopic phase separation . thin films of superfluid @xmath4he and @xmath4he/@xmath5he mixtures represent another set of systems where the superfluid transition temperature @xmath2 is strongly correlated with the superfluid density . in section v , we compare the results of htsc systems with superfluid he films in normal and porous media . based on the correlations between @xmath2 and @xmath3 and the pseudogap phenomena in the underdoped region , some models / conjectures have been presented in terms of the condensation mechanisms of htsc systems . to elucidate these models , in section vi , we will first consider differences and similarities among bose - einstein ( be ) condensation , bcs condensation and kosterlitz - thouless ( kt ) transition . then , we will compare models of be - bcs crossover and phase fluctuations , taking into account relevance to the kt transition and distinction between pair - formation and dynamic superconductivity in the pseudo - gap regime . we will also introduce a phase diagram involving phase separation in the overdoped region in section vii , followed by a summary in section viii . in type - ii superconductors , the external magnetic field , between @xmath11 and @xmath12 , forms a lattice of flux vortices , resulting in inhomogeneous internal field distributions . the width of this distribution is proportional to the muon spin relaxation rate @xmath13 in tf-@xmath0sr , which is related to the penetration depth @xmath1 and @xmath3 as @xmath14 \times [ 1/(1+\xi/{\it l\/})]\eqno{(1)},\ ] ] as given by the london equation . in superconducting systems in the clean limit , where the coherence length @xmath15 is much shorter than the mean free path _ l _ , the relaxation rate @xmath13 is proportional to @xmath3 . the inhomogeneity of the internal field is due to partial screening of the external field by the supercurrent . so , it is quite natural that the field width and @xmath13 are proportional to the `` supercurrent density '' @xmath3 , in the same sense as the conductivity of normal metals is proportional to @xmath16 , where @xmath17 denotes the normal state carrier density . generally , tf-@xmath0sr studies in type - ii superconductors provide information on : ( a ) the temperature dependence of @xmath1 ; ( b ) absolute values of @xmath1 at @xmath18 ; ( c ) the coherence length @xmath15 via analyses of local field distribution ; and ( d ) vortex lattice properties . among these , use of single crystal specimens is essential in ( a ) , ( c ) and ( d ) . for example , early @xmath0sr results of @xmath1 on ceramic samples of htsc systems mostly exhibited behavior consistent with isotropic energy gap , and it is only after studies of high - quality single crystal specimens that the d - wave nature was confirmed in temperature dependence of @xmath1 . for details of ( a ) , ( c ) and ( d ) , readers are referred to refs . [ 3 - 4 ] . in contrast , the results on ceramic specimens have been very useful in comparing absolute values of @xmath3 in different systems . ceramic specimens provide advantages over single crystal specimens in homogeneity of doped hole concentrations and in being less affected by crossover from the 3-d to 2-d vortex lattice . in this article , we shall focus on the system and doping dependence of the superfluid density , i.e. , ( b ) . in htsc systems with a large anisotropy , the relaxation rate @xmath13 measured in ceramic specimens is determined predominantly by the supercurrent flowing in the conducting ab - planes [ 5 ] . thus , the results of @xmath13 from ceramic specimens should be regarded as reflecting the in - plane penetration depth @xmath19 and the in - plane effective mass @xmath20 . figure 1 shows a plot of @xmath21 versus @xmath2 of various different htsc systems [ 6 - 11 ] . in this figure , results from simple hole - doped htsc systems [ 6,7 ] are shown with open symbols . with increasing hole concentration , @xmath2 increases linearly with @xmath3 in the underdoped region , and then shows a saturation . the slope of this linear relationship in the underdoped region is common to various different series of htsc systems . these results have also been confirmed in several other @xmath0sr measurements [ 12 - 16 ] . this observation alone already implies that @xmath3 can be an important determining factor for @xmath2 in htsc systems . we also find in fig . 1 that the ratios of @xmath2 versus @xmath3 for some other superconductors [ 17,18 ] are not quite far from the value for htsc systems . if an independent estimate for the effective mass @xmath22 is available , the @xmath0sr measurements give the superconducting carrier density @xmath23 . then , one can calculate the number of carries existing in a region of coherence length squared on the conducting plane . for typical bcs superconductors , such as sn or pb , one finds more than 10,000 pairs overlapping with one another in the @xmath24 area . for superfluid he , which is the well - known case for bose - einstein condensation , we find that each boson exist without overlapping with each other : i.e. , one pair per @xmath24 . the systems following the linear relationship in fig . 1 have 3 to 6 pairs overlapping within the @xmath24 area , as illustrated in fig . 2 . this feature encourages us to consider htsc and some novel superconductors in a crossover from be to bcs condensation . by knowing the average distance @xmath25 between the conducting cuo@xmath26 planes , one can obtain @xmath27 where @xmath28 denotes the 2-dimensional carrier density . we remind here that the fermi energy @xmath29 of 2-dimensional electron gas is proportional to @xmath30 . thus , we can consider @xmath27 as a parameter representing kinetic energy for translational motion of superconducting carries . for 3-d systems without phase separation , in which @xmath31 , one can combine the muon relaxation rate @xmath32 with the sommerfeld constant @xmath33 , or pauli susceptibility @xmath34 , to deduce @xmath35 . the `` effective '' fermi energy @xmath29 obtained in this way does not necessarily correspond to the real fermi energy in band structure calculations . since @xmath36 corresponds to the drude spectral weight in optical conductivity which condenses into a delta function at @xmath37 below @xmath2 , @xmath29 might also be called as a drude energy scale . figure 3 shows a plot of @xmath2 versus @xmath38 thus obtained from the results of @xmath3 [ 7 ] . for a@xmath9c@xmath10 where a reliable value of _ l _ is not available , we used the clean - limit value of @xmath3 without corrections regarding @xmath15/_l _ [ 17,18 ] . we find that htsc , organic bedt [ 19 ] , and some other systems [ 17 - 20 ] have very high and nearly equal ratios of @xmath39 . the @xmath40 line in this figure shows the bose - einstein condensation temperature for a non - interacting 3-d bose gas having the boson density @xmath41 and mass @xmath42 . compared to @xmath40 , the actual transition temperatures @xmath2 of htsc systems are reduced by a factor of 4 to 5 . this reduction is natural in view of overlapping pairs which would reduce @xmath2 , and in view of 2-dimensional character of htsc systems . however the parallel behavior of @xmath40 and the observed results of @xmath2 suggests that the origin of the linear relationship between @xmath2 and @xmath3 could be deeply related to be condensation . figure 3 serves as an empirical way to classify various superconductors in a crossover from be to bcs condensation . when a very small amount of zn is substituted for cu on the cuo@xmath26 planes of htsc systems , @xmath2 is reduced . figure 4(a ) shows the reduction of @xmath21 as a function of zn concentration @xmath43 substituting in - plane cu . the superfluid density @xmath3 decreases with increasing @xmath43 . to account for this reduction , we proposed `` swiss cheese model '' [ 9 ] where each zn is assumed to destroy superconductivity in the surrounding region with the area @xmath44 , as illustrated in fig . the solid lines in fig . 4(a ) , which represent predictions of this model , are obtained just from @xmath43 and @xmath45 deduced from the upper critical field @xmath12 , _ without any fitting _ to the data . sufficiently good agreement between these lines and the observed results indicate that carriers within the @xmath44 area around each zn no longer contribute to the superfluid density @xmath3 . subsequently , this picture was directly confirmed by pan et al . [ 21 ] in scanning tunneling microscopy studies , where the local density of states around zn showed features characteristic to normal regions , as shown in fig . these @xmath0sr and stm results are consistent with the non - zero @xmath46 term of the specific heat , which increases with increasing zn concentration [ 22 ] . in the plot of @xmath13 versus @xmath2 in fig . 1 , the results of zn - doped systems ( closed triangle and star symbols ) follow the trajectory of simple hole doped htsc systems ( open symbols ) . tl@xmath26ba@xmath26cuo@xmath47 ( tl2201 ) systems have been extensively studied by various experimental methods as a prototype of overdoped htsc systems . tl2201 compounds have very small residual resistivity , which assures that the system lies well within the clean limit . to our surprise , with increasing overdoping from the nearly optimal @xmath2 = 85 k sample , the relaxation rate @xmath21 in tf-@xmath0sr decreased [ 8,15 ] as shown by the closed circle symbols in fig . this behavior is also shown in a plot of @xmath21 versus doping factor @xmath48 in fig . in view of no anomaly in @xmath22 , the different behaviors of the normal state carriers @xmath17 and the superfluid density @xmath3 suggest that carriers are spontaneously separated into those which are involved in superconductivity and those which remain unpaired fermions , as illustrated in fig . 5(a ) . by defining the `` gapped '' and `` ungapped '' responses in the specific heat measurements , as illustrated in fig . 5(a ) , we plotted the @xmath48 dependence of the gapped response obtained from the data of loram _ et al . _ [ 22 ] in tl2201 in fig . the good agreement between the @xmath0sr superfluid density @xmath21 ( closed circles ) and the `` gapped '' response in specific heat ( open symbols ) provides a support to our view [ 8,23 ] with spontaneous phase separation between superconducting and normal regions in overdoped htsc systems . the volume fraction of the superconducting region can also be estimated from the `` specific heat jump '' @xmath49 . in bcs superconductors @xmath50 , where @xmath51 denotes the normal - state specific heat , and @xmath52 stands for the sommerfeld constant derived in the normal state . then , in the plot of @xmath53 versus @xmath54 as in the inset of fig . 5(a ) , one would expect that the specific heat jump @xmath55 should be independent of @xmath2 , for systems having a common @xmath52 value . in loram s specific heat results of tl2201 [ 22 ] , @xmath52 does not depend on doping , while the jump @xmath55 decreases with increasing doping . this observation further supports decreasing superfluid volume fraction with increasing doping ( decreasing @xmath2 ) in tl2201 . residual normal response in the overdoped cuprate was also found in measurements of optical conductivity [ 24 ] . furthermore , we calculated an expected doping dependence of @xmath3 based on a simple model assuming phase separation [ 23 ] ( as described in section vii ) , and obtained a good agreement with the observed results in ( y , ca)ba@xmath26cu@xmath9o@xmath56 [ 25 ] , as shown in fig . so far , there is no direct observation reported regarding the size of these phase - separated regions in overdoped htsc . magnetic order of the parent compound la@xmath26cuo@xmath57 of the 214 cuprates was first confirmed by zf-@xmath0sr measurements [ 26 ] . figure 6(a ) shows the time spectra obtained in the zf measurements of antiferromagnetic la@xmath26cuo@xmath57 ( af - lco ) which has the n/eel temperature @xmath58 k. as in most other magnetic systems , we find that muon spin precession sets in below @xmath59 , with the precession amplitude independent of temperature , while the frequency increasing with decreasing temperature as the sub - lattice magnetization builds up . in the la214 systems with the hole concentration near 1/8 per cu , incommensurate static spin correlations have been found by neutron scattering [ 27 ] . time spectrum of zf-@xmath0sr in la@xmath60ba@xmath61cuo@xmath57 ( lbco:0.125 ) [ 28 ] is shown in fig . in this case , we find a bessel function line shape characteristic of zf-@xmath0sr in incommensurate magnetic systems [ 29,30 ] , such as the one observed in ( tmtsf)@xmath26pf@xmath62 [ 29 ] . la@xmath26cuo@xmath63 ( lco:4.11 ) is a system with oxygen intercalated in a stage-4 structure . this system is superconducting with @xmath2 = 42 k , which is the highest among the la214 family , while also exhibiting static incommensurate magnetism below @xmath59 = 42 k [ 31 ] this incommensurate modulation has a very long correlation length ( @xmath64 ) , as determined from a very sharp satellite magnetic bragg peak in neutron scattering results . zf-@xmath0sr spectra of this system [ 10,32 ] , shown in fig . 6(c ) , has a bessel - function line shape , as expected for an incommensurate spin structure . the amplitude of this precession , however , increases gradually below @xmath59 with decreasing temperature , while the frequency is almost independent of temperature below @xmath59 . furthermore , the amplitude of precessing signal at @xmath65 is less than half of that in lbco:0.125 ( fig . 6(b ) ) , which indicates that the static magnetism exists in less than a half of the total volume . in fig . 7(a ) and ( b ) , we show temperature dependences of the volume fraction @xmath66 of muons in the region with static magnetic freezing , derived from the precession amplitude , and the frequency @xmath67 observed in lco:4.11 , ( la@xmath68sr@xmath69)cuo@xmath57 ( lsco:0.12 ) , lbco:0.125 and ( la@xmath70nd@xmath71sr@xmath61)cuo@xmath57 ( lnsco:0.125 ) [ 33 ] . in lco:4.11 and lsco:0.12 , where superconductivity coexists with static magnetism , the static magnetism is confined to a partial volume fraction . as shown in fig . 7(c ) , the temperature dependence of the neutron bragg intensity in lco:4.11 is consistent with the behavior of @xmath72 . unlike usual magnetic systems , however , the @xmath54 dependence is mostly due to the change of the volume fraction @xmath66 . the internal field at the muon site in htsc systems is due to dipolar field from neighbouring static cu spins . depending on the range of this field , the volume fraction of muons @xmath66 subject to static field is somewhat larger than the volume fraction @xmath73 of frozen cu spins . figure 8(a ) shows our simulation results for the relationship between @xmath73 and @xmath66 for cases with incommensurate static magnetism in island regions having a radius @xmath74 [ 10 ] . here we see that about 70 - 80% population of static cu spins is enough to create observable static fields at the sites of all the muons . thus , our previous result [ 33 ] on la@xmath75nd@xmath71sr@xmath76cuo@xmath57 ( lnsco:0.15 ) , which is a superconductor with @xmath77 with @xmath78% is still compatible with a picture that superconductivity and static magnetism occur in mutually exclusive regions of cuo@xmath26 plane . although the relationship between @xmath66 and @xmath73 is nearly independent of the radius @xmath74 of static sdw islands , the damping rate @xmath79 of the bessel oscillation depends on @xmath74 , as shown in fig . 8(b ) . by converting observed @xmath66 into @xmath73 using fig . 8(a ) , and then plotting observed @xmath79 in fig . 8(b ) , we find that our zf-@xmath0sr results in lco:4.11 are consistent with the island size @xmath80 15 - 30 [10 ] . note that this length scale is comparable to @xmath45 . this model of `` static sdw nano islands '' can be reconciled with the long range spin correlations found by neutron bragg peaks , if we consider percolation of these islands via a help of interplaner correlations , as proposed in [ 10 ] . to study the relationship between the magnetic volume fraction @xmath73 and the superfluid density @xmath3 , we performed @xmath0sr measurements in ( la@xmath81eu@xmath56sr@xmath76)cuo@xmath57 ( lesco ) [ 11,34 ] where the concentration of doped hole carriers is fixed to 0.15 per cu , while the magnetic volume changes with increasing eu concentration @xmath82 . the volume fraction with static magnetism was determined by the zf-@xmath0sr , while @xmath3 was measured in tf-@xmath0sr . the results shown in fig . 9 clearly demonstrate a trade - off between @xmath73 and the superfluid density @xmath3 . this implies that superconductivity does not exist in the volume which has static magnetism . in the plot of @xmath21 versus @xmath2 in fig . 1 , the results of these systems with static stripe magnetism , shown by the striped square symbols , follow the trajectory of simple hole - doped and zn - doped systems . this indicates that @xmath3 is again a determining factor of @xmath2 in htsc systems where superconductivity and magnetism coexist . we also note that the results in overdoped tl2201 can be viewed as following a monotonic relationship between @xmath2 and the superfluid density , with the slope roughly comparable to cuprates with / without other types of perturbations . superfluid he films represent another set of systems where @xmath2 is strongly correlated with the 2-dimensional superfluid density @xmath28 at @xmath65 . using the published results , we made fig . 10 [ 35 ] which shows correlations between @xmath2 and @xmath27 for @xmath4he films on mylar substrate ( open circle symbol ) [ 36 ] , on vycor glass which represents a porous media ( star symbol ) [ 37,38 ] , as well as for @xmath4he/@xmath5he mixtures adsorbed on fine alumina powders ( closed diamond symbol ) [ 39 ] . the results on mylar films show linear relationship , consistent with the behavior expected for the kt transition [ 40 ] ( see next section for details ) as indicated by the solid line . with the lowest level of perturbation , this case corresponds to simple hole - doped htsc systems in comparison between the cuprates and he films . the results on vycor glass is analogous to the zn - doped cuprates : in both cases some normal regions are formed as a `` healing region / layer '' , while @xmath2 is still strongly correlated with the superfluid density despite non - trivial geometry of the superfluid . mixture of bosonic @xmath4he and fermionic @xmath5he liquids exhibits a phase diagram shown in the inset of fig . 10 . with increasing @xmath5he fraction , @xmath2 decreases , being roughly proportional to the volume fraction of @xmath4he . in bulk geometry , the mixture undergoes macroscopic phase separation into @xmath4he - rich superfluid and @xmath5he - rich normal fluid , the heavier superfluid existing underneath the lighter normal fluid in a container . this phase separation can be confined into a microscopic length scale by adsorbing the mixture onto porous media or fine powders [ 39,41 ] , where superfluidity remains up to a large @xmath5he fraction . the results for the @xmath4he/@xmath5he mixture in fig . 10 exhibit a behavior very similar to that of overdoped tl2201 in the cuprates in fig . 1 . in both the he mixture and the overdoped htsc , inclusion of too many fermions ( doped holes and @xmath5he ) results in microscopic phase separation and reduction of both @xmath2 and the superfluid density . thus , we find a remarkable similarities [ 35 ] in the plot of @xmath2 versus superfluid density in cuprates ( fig . 1 ) and he films ( fig . 10 ) for the cases with / without perturbation . figure 1 shows correlations between @xmath2 and 3-dimensional ( 3-d ) superfluid density while fig . 10 shows @xmath2 versus 2-d superfluid density . by knowing the average interlayer distance @xmath25 between the cuo@xmath26 planes , we can generate a plot of @xmath2 versus @xmath27 for the cuprate systems . the resulting plot , fig . 11 , clearly indicates that in htsc systems , @xmath2 is higher for shorter @xmath25 for a given 2-d superfluid density @xmath27 [ 42 ] . this observation is consistent with the variation of @xmath2 reported for multilayer films with alternating layers of superconducting yba@xmath26cu@xmath9o@xmath83 ( ybco ) and insulating prba@xmath26cu@xmath9o@xmath83 ( pbco ) [ 43 ] , as shown in the inset of fig . these results indicate that the interlayer coupling is essential for obtaining higher @xmath2 in htsc systems . the observed variation of @xmath2 can not be explained by the simplest version of kt transition where @xmath2 should be determined solely by @xmath27 . correlations between @xmath2 and the superfluid density shown in fig . 1 and fig . 11 help consideration of condensation mechanisms in htsc systems . in be condensation [ 44 ] , the bosons are pre - formed at a very high `` paring '' temperature @xmath84 . at a given temperature @xmath85 each boson has kinetic energy of @xmath86 , which defines the thermal wave length @xmath87 representing the spread of the wave function of this boson due to the uncertainty principle . with decreasing @xmath54 , @xmath87 increases . when @xmath87 becomes comparable to the inter - particle distance @xmath88 , the wave functions of neighbouring bosons start to overlap , as illustrated in fig . thanks to the tendency of bosons to fall into the same state by building up phase coherence of wave functions , bose condensation occurs at @xmath54 where @xmath89 . from this , we have the be condensation temperature @xmath90 , where @xmath91 and @xmath92 denote the density and mass of the boson . in this way , the superfluid density and @xmath2 is directly related in be condensation . in bcs condensation [ 45 ] , the energy scale of attractive interaction determines the energy gap and @xmath2 . the effective fermi energy @xmath93 is much larger than @xmath94 in bcs condensation , i.e. , there is a sufficient density of fermions above @xmath2 . however , the system stays in the normal state until the temperature is reduced to the pairing energy scale @xmath84 , where bosons are formed . once pairs are formed , their boson density is high enough at @xmath2 , and thus the condensation occurs immediately at @xmath95 . in bcs condensation , @xmath2 depends on carrier density through the density of states at the fermi level which determines the strength of electron - phonon interaction . however , this dependence is rather indirect . suppose we have a bcs superconductor with the carrier density @xmath23 . if the debye frequency is doubled , @xmath2 and the energy gap @xmath96 would be doubled . however the carrier density stays unchanged , as can be found from the illustration of fig . this example shows that the superfluid density is not a direct determining factor for @xmath2 in bcs condensation . kosterlitz - thouless transition [ 40 ] is a phenomenon in 2-dimensional superfluids / superconductors . the elementary excitation of superfluid he film is the formation of vortex anti - vortex pairs . the energy required for this is governed by the phase stiffness , which is proportional to the superfluid density . at the transition temperature @xmath97 , the thermal energy becomes sufficient to form un - bound vortices , which results in dissipation and destruction of superfluidity / phase coherence . thermodynamic arguments lead to the universal relationship between the 2-d superfluid density at @xmath98 and the value of @xmath2 as @xmath99 , which should be independent of material . this relationship was confirmed as the jump of the superfluid density at the superfluid transition of he films [ 46 ] . figure 13(a ) illustrates evolution of various energy scales in superfluid he film and a thin film of bcs superconductor . the be condensation temperature of he in the 3-d limit corresponds to the lambda temperature @xmath100 k , which is close to @xmath101 , except for some reduction due to departure of real system from an ideal non - interacting bose - gas limit . with decreasing film thickness @xmath102 , @xmath2 starts to change when the thickness becomes comparable to a few times the coherence length @xmath103 . for @xmath104 , @xmath105 . in superfluid he film on mylar , the superfluid density at @xmath65 is very close to that at just below @xmath2 , as shown in fig . thus the kosterlitz - thouless relation holds for @xmath28 at @xmath65 , as can be seen in fig . 10 . between @xmath2 and @xmath100 k , the order parameter exhibits phase fluctuations , i.e. , superfluidity is dynamic . the pair formation energy scale is much higher than @xmath2 in he for any thickness value @xmath102 . in a thin film of a bcs superconductor , @xmath2 should show a similar reduction from the value @xmath106 in the bulk 3-d system , when the film thickness becomes comparable to or smaller than a few times @xmath15 in the clean limit ( and _ l _ in the dirty limit ) , as shown in fig . the superfluid density @xmath3 at @xmath65 would follow a thickness dependence similar to that of @xmath2 . note that @xmath3 is very large , as it is related to @xmath107 in the 3-d limit . there should be difference between @xmath16 and @xmath3 in the region @xmath108 ( or _ l _ ) , due to difficulty in forming pairs in a restricted geometry . on the other hand , the general argument of kt relationship should still hold . this is possible only if @xmath27 at @xmath109 for @xmath108 is much smaller than @xmath27 at @xmath65 , as illustrated in fig . therefore , the `` jump '' of the superfluid density at @xmath97 is practically invisible in a bcs thin film . for @xmath104 , the region @xmath110 is characterized by phase fluctuations . however , the pair formation occurs at @xmath111 for any thickness . consequently , bosons exist only below @xmath106 , as illustrated in fig . 13(b ) . in 1989 [ 6 ] and 1991 [ 7 ] , we suggested the relevance of be condensation to the universal relationship between @xmath2 and @xmath3 in htsc systems shown in fig . 1 [ 6 ] . by combining this phenomenon with the pseudogap behavior then observed by nmr and conductivity , we proposed a picture of be to bcs crossover in 1993 - 94 [ 47,48 ] , as illustrated in fig . 14(a ) [ 42,47,48 ] . general concept of the be - bcs crossover has been considered earlier by several scientists [ 49 ] , including randeria and co - workers [ 50 ] who adopted this concept to the interpretation of the susceptibility / nmr results for the pseudo gap . within our knowledge , however , our proposal was the first to combine the results of superfluid density with the pseudogap behavior in the underdoped cuprates . as shown in fig . 14(a ) , we regarded the pseudogap temperature @xmath6 as the pair formation temperature @xmath84 . the reduction of the c - axis dc conductivity below @xmath6 can be interpreted as resulting from reduced interplaner tunneling probability for paired bosonic carriers having 2e charges , while reduction of magnetic susceptibility can be attributed to the formation of a spin singlet bosonic ( pairing ) state in this picture . in 1995 , emery and kivelson ( ek ) [ 51 ] proposed a model based on phase fluctuations , as illustrated in fig . the be - bcs model and the phase fluctuation model share many features in common . however , there are several important differences . based on the linear relationship between @xmath2 and the superfluid density at @xmath65 , ek presented arguments basically parallel to that for the kt transition in 2-d systems . they pointed out that @xmath2 of the underdoped cuprates is determined by the energy scale for the phase fluctuations . in their picture , the entire region of pseudogap below @xmath6 is characterized by phase fluctuations . they argued that the 2-dimensional aspect is essential for obtaining high @xmath2 . in order to calculate @xmath112 which denotes the energy scale for phase fluctuations to destroy superconductivity , ek multiplied @xmath113 to the interplaner distance @xmath25 for htsc and some other 2-d systems . this is equivalent of obtaining @xmath114 in 2-d . for 3-d systems , ek multiplied @xmath3 and the coherence length @xmath15 , which leads to an energy scale much higher than @xmath115 . this energy scale is unrealistically high , and irrelevant to condensation arguments . if one substitutes the interparticle distance @xmath116 instead of @xmath15 , we recover @xmath115 for 3-d systems . then , table 1 in ref [ 51 ] by ek becomes essentially equivalent to fig . 3 of our 1991 paper [ 7 ] , shown as a part of fig . 3 of this article , where we plotted @xmath2 versus @xmath115 . we consider that the distinction between 2-d and 3-d systems by ek is an artefact resulting from the overestimate of @xmath112 in 3-d systems . in general , the 2-dimensional aspect does not help increasing @xmath2 of any superconductor , as can be found in fig . 13 . figure 15 shows the current estimates of @xmath6 from various methods as plotted versus hole concentration . if the entire region below @xmath6 is supposed to have superconducting phase fluctuations , as conjectured by ek , the situation is rather similar to a thin - film bcs superconductor shown in fig . in contrast , if @xmath6 is solely representing the pair formation , as proposed in our be - bcs crossover picture , and if there is a 2-dimensional aspect remaining in htsc systems , there should be two different energy scales above @xmath2 , as we discussed in ref [ 35 ] . they are the temperature @xmath8 at which the phase coherence of bosons completely disappear and dynamic superconductivity vanishes , in addition to the temperature @xmath84 at which pairs are dissolved into fermions . this situation is similar to the case of he films shown in fig . one of important differences between be - bcs and ek conjectures lies in this point . since high-@xmath2 cuprates have a highly 2-dimensional electronic structure , several experiments have detected the `` dynamic superconductivity '' existing above @xmath2 : ( 1 ) in ybco - pbco films with a thick pbco layers separating a single layer ybco , @xmath2 is reduced to 15 - 20 k. between this temperature and the @xmath117 k of the bulk ybco , one observes reduction of the normal state conductivity , which follows the predictions of the kosterlitz - thouless theory [ 52 ] . ( 2 ) corson _ et al . _ [ 53 ] measured dynamic superfluid response in underdoped bi2212 , and found that the response depends on the measuring frequency @xmath118 above a certain `` branch - off '' temperature @xmath119 , as shown in fig . furthermore , the superfluid density @xmath3 at @xmath120 agrees well with the universal value @xmath121 expected for the kt transition . the critical temperature @xmath2 , at which @xmath122 , increases with increasing @xmath118 , suggesting that this phenomenon corresponds to the `` dynamic superfluidity '' expected above @xmath123 and below @xmath8 . it should be noted , however , that @xmath8 for the measuring frequency of 600 ghz is limited to @xmath124 k. these results indicate that certain cuprate systems with high anisotropy , such as underdoped bi2212 and single - layer ybco film , exhibit dynamic superconductivity as expected from the kt theory . however , these results do not necessarily provide explanation to the origin of the pseudo gap at @xmath125 , since @xmath8 observed in bi2212 is significantly lower than the pseudo - gap temperature @xmath6 . namely , the phase fluctuations alone can not explain the entire pseudo - gap phenomena . dependence to an external magnetic field should be quite different if these two distinct energy scales above @xmath2 correspond respectively to pairing and dynamic superconductivity . the former would not depend much on external fields , while the latter should be very sensitive to the field . indeed , the pseudo gap below @xmath6 was found to be insensitive to the applied field in tunneling measurements [ 54 ] . recently , lavrov _ et al . _ [ 55 ] found that c - axis conductivity shows negative magneto - resistance upon cooling at a temperature well below @xmath6 but above @xmath2 . this negative magneto - resistance may be due to the onset of dynamic superconductivity below @xmath8 . the results of the nernst effect [ 56 ] , which appear in the underdoped region of la214 systems above @xmath2 but well below @xmath6 can also be interpreted as possible evidence for dynamic superconductivity below @xmath8 . other phenomena potentially related to this energy scale include : ( a ) superfluid response of the arpes coherence peak [ 57 ] ; and ( b ) 41 mev neutron resonance mode [ 58 ] . based on these considerations , we propose a new phase diagram , with two distinct lines of @xmath6 and @xmath8 in the underdoped region as shown in fig . note that the pair formation at @xmath6 is necessary for superconductivity in cuprates , but is not enough to support any phase coherence . it is only when the thermal energy scale becomes less than the energy scale representing the number density of bosons that the phase coherence can set in . the onset temperature @xmath8 of dynamic superfluidity represents this energy scale . in highly 2-dimensional cases , one should further cool down below @xmath8 before achieving long - range phase coherence at @xmath2 . in bcs superconductors in the clean limit , all the carriers in the fermi sphere contribute to superfluid . there should be no normal carriers remaining at @xmath18 . therefore , neither the be - bcs nor the phase fluctuation pictures can explain the reduction of @xmath23 with increasing hole doping observed in several overdoped cuprate systems [ 8,15,25 ] described in section iii - b . tallon , loram and co - workers [ 59 ] have noticed that the @xmath6 line may be heading towards @xmath126 at a `` critical hole concentration '' around @xmath127 . @xmath0sr studies of panagopoulos _ et al . _ [ 60 ] show that static magnetism , either spontaneously existing or induced by zn doping , disappears at @xmath128 . generally , increasing hole doping would tend to destroy magnetic interactions in the cuprates . let us make the following three assumptions : ( a ) the paring in htsc systems is due to a magnetic interaction ; ( b ) the @xmath6 line represents the pair formation , and ( c ) the @xmath6 line disappears at @xmath129 . then , no genuine superconducting pairing exists in the overdoped region at @xmath128 . however , if the energy loss for charge disproportionation is overcome by the gain of pairing and condensation energies , the system can spontaneously phase separate into a `` hole - poor '' superfluid with the local hole concentration @xmath130 and a `` hole - rich '' normal fermion region with @xmath131 . if this phase separation remains microscopic with the length scale comparable to @xmath45 , we can expect superconductivity in the overdoped region , similarly to the case of @xmath4he/@xmath5he mixture . the energy loss of charge disproportionation can be estimated following coulomb blockade type calculation and/or formation of nano - islands serving as capacitor allays . our crude estimate [ 23 ] shows that this energy becomes comparable to paring / condensation energy gain in the cuprates . based on these considerations , we proposed a phase diagram shown in fig . 17 , with phase separation in the overdoped region . as shown in fig . 5(b ) , the doping evolution of the superfluid density @xmath23 , calculated using this model [ 23 ] , shows a good qualitative agreement with the observed results . we also note that this model can reproduce a very sharp temperature dependence of @xmath3 and @xmath12 at @xmath65 found in the heavily overdoped region of tl2201 [ 8,15,61 ] . tallon and loram [ 59 ] presented a view that the existence of superconductivity in the overdoped cuprates at @xmath128 , where the @xmath6 line does not exist , implies that the magnetic interaction below the @xmath6 line is not required for superconductivity but rather is a competing factor which weakens superconductivity . the present model with phase separation provides an alternative view where the pseudogap phenomena , as a necessary factor ( pair formation ) for superconductivity , can be compatible with superconductivity in the overdoped cuprates . in htsc cuprates , as well as in 2-d organic ( bedt - ttf)@xmath26-x and 3-d a@xmath9c@xmath10 superconductors , superconductivity appears in the vicinity of the insulator to metal transition . superconductivity is taken over by a presumably fermi - liquid metallic state by increasing carrier doping in cuprates , or by application of chemical and/or external pressure in bedt and a@xmath9c@xmath10 . in all these cases , we expect that the normal state spectral density ( or drude weight ) @xmath16 to increase in the process towards a simple fermi liquid . we showed that the superconducting spectral weight _ decreases _ in this process for the case of cuprates . we performed magnetization measurements of the in - plane penetration depth in ( bedt - ttf)@xmath26cu(ncs)@xmath26 under applied pressure [ 62 ] , and found that @xmath3 decreases with increasing pressure as shown in fig . 18(a ) , contrary to the behavior of @xmath16 found in quantum oscillation measurements [ 63 ] . this system is known to be well in the clean limit . these results indicate that not only in cuprates but also in bedt , the crossover from superconducting to metallic ground state is associated with anomalous reduction of the superfluid spectral weight . similar behavior is also seen in a@xmath9c@xmath10 systems . as shown in fig . 1 and fig . 18(b ) , the muon relaxation rate @xmath132 measured in a@xmath9c@xmath10 decreases with decreasing lattice constant [ 17,18 ] , in the approach towards a simple metallic state . in view of rather large residual resistivity in the normal state , the results for these fullerides might be subject to correction related to the mean free path . however , the un - corrected raw data indicate that @xmath36 again shows anomalous reduction when the system approaches presumably a simple metallic ground state . these three systems exhibit similar phase diagrams : superconductivity appears in the evolution from magnetic and insulating ground state to presumably simple fermi liquid state ; in bedt systems , a pseudo - gap like behavior near the magnetic phase was found in susceptibility . these features suggest a possibility that the anomalous results in the overdoped cuprates may be a generic behavior shared by a wider range of superconductors based on correlated electron systems . we note that all these systems follow the universal linear relationship in the plot of @xmath2 versus @xmath3 , with approximately the same slope as shown in fig . 1 and in ref . in this paper , we showed that @xmath2 in htsc cuprate systems exhibits universal correlations with the superfluid spectral weight @xmath3 , for the cases of simple hole doping , as well as for more complicated cases with zn - doping , overdoping and static sdw nano island formation , where the system undergoes spontaneous phase separation between superconducting and normal regions with the length scale comparable to the in - plane coherence length . robustness of this relation for the case with / without perturbation is analogous to the case of superfluid @xmath4he and @xmath4he/@xmath5he mixture films in non - porous and porous media . in all these cases , the superfluid density is the determining factor for @xmath2 . this is the basic feature of be condensation . by slightly revising the be - bcs crossover picture , we have proposed a new phase diagram for the cuprates which has ( a ) two separate lines of @xmath6 and @xmath8 above @xmath2 in the pseudogap state in the underdoped region , ( b ) disappearance of the @xmath6 line at the critical hole concentration @xmath133 , and ( c ) phase separation in the overdoped region . in this model , the @xmath6 line represents pair formation , whereas the @xmath8 line corresponds to the onset of dynamic superconductivity . low dimensionality prevents formation of long - range phase coherence between @xmath8 and @xmath2 . we also noted anomalous reduction of the superfluid spectral weight @xmath3 in the overdoped cuprates as well as in the 2-d organic bedt and 3-d fulleride superconductors , when the system approach simple metallic ground state . the author would line to acknowledge collaboration with g.m . luke , k.m . kojima , s. uchida , r.j . birgeneau , k. yamada and discussions with o. tchernyshyov and m. randeria . this work is supported by a grant from us nsf dmr-01 - 02752 and che-01 - 17752 . for general aspects and historical development of @xmath0sr , see proceedings of nine international conferences on muon spin rotation / relaxation/ resonance in hyperfine interact . * 6 * ( 1979 ) ; * 8 * ( 1981 ) ; * 17 * -*19 * ( 1984 ) ; * 31 * ( 1986 ) ; * 63 * -*65 * ( 1990 ) ; * 85 * -*87 * ( 1994 ) ; * 104 * -bf 106 ( 1997 ) ; physica * b 289 * -*290 * ( 2000 ) ; and physica * b * ( proceedings for @xmath0sr 2002 , williamsburg , virginia , june , 2002 ) , in press . for a recent reviews of @xmath0sr studies in topical subjects , see _ muon science : muons in physics , chemistry and materials _ , proceedings of the fifty first scottish universities summer school in physics , st . andrews , august , 1988 , ed . lee , s.h . kilcoyne , and r. cywinski , inst . of physics publishing , bristol , 1999 . uemura , g.m . luke , b.j . sternlieb , j.h . brewer , j.f . carolan , w.n . hardy , r. kadono , j.r . kempton , r.f . kiefl , s.r . kreitzman , p. mulhern , t.m . riseman , d.ll . williams , b.x . yang , s. uchida , h. takagi , j. gopalakrishnan , a.w . sleight , m.a . subramanian , c.l . chien , m.z . cieplak , gang xiao , v.y . lee , b.w . statt , c.e . stronach , w.j . kossler , and x.h . yu , _ universal correlations between @xmath2 and @xmath3 ( carrier density over effective mass ) in high-@xmath2 cuprate superconductors _ , phys . lett . * 62 * , ( 1989 ) 2317 . uemura , l.p . luke , b.j . sternlieb , w.d . brewer , t.m . riseman , c.l . seaman , m.b . maple , m. ishikawa , d.g . hinks , j.d . jorgensen , g. saito , and h. yamochi , _ basic similarities among cuprate , bithmuthate , organic , chevrel - phase and heavy - fermion superconductors shown by penetration - depth measurements _ , phys . lett . * 66 * , ( 1991 ) 2665 . uemura , a. keren , l.p . luke , w.d . wu , y. kubo , t. manako , y. shimakawa , m. subramanian , j.l . cobb , and j.t . markert , _ magnetic field penetration depth in tl@xmath26ba@xmath26cuo@xmath47 in the overdoped regime _ , nature ( london ) * 364 * ( 1993 ) 605 . b. nachumi , a. keren , k. kojima , m. larkin , g.m . luke , j. merrin , o. tchernyshov , y.j . uemura , n. ichikawa , m. goto , and s. uchida , _ muon spin relaxation studies of zn - substitution effects in high-@xmath2 cuprate superconductors _ , phys . * 77 * ( 1996 ) 5421 . savici , y. fudamoto , i.m . gat , t. ito , m.i . larkin , y.j . uemura , g.m . luke , k.m . kojima , y.s . lee , m.a . kastner , r.j . birgeneau , k. yamada , _ muon spin relaxation studies of incommensurate magnetism and superconductivity in stage-4 la@xmath26cuo@xmath63 and la@xmath68sr@xmath69cuo@xmath57 _ , phys . rev . * b66 * ( 2002 ) 014524 . kojima , s. uchida , y. fudamoto , i.m . gat , m.i . larkin , y.j . uemura , g.m . luke , _ superfluid density and volume fraction of static magnetism in stripe - stabilized la@xmath81cu@xmath56sr@xmath76cuo@xmath57 _ , proceedings of @xmath0sr2002 , physica * b * , in press . b. pmpin , h. keller , w. kndig , i.m . savi , j.w . schneider , h. simmer , p. zimmermann , e. kaldis , s. rusiecki , c. rossel , _ @xmath0sr in oxygen deficient yba@xmath26cu@xmath9o@xmath134 ( 6.5 @xmath135 7.0 ) _ , hyperfine interact . * 63 * ( 1990 ) 25 . seaman , j.j . neumeier , m.b . maple , l.p . luke , b.j . sternlieb , y.j . uemura , j.h . brewer , r. kadono , r.f . kiefl , s.r . kreitzman , t.m . riseman , _ magnetic penetration depth of y@xmath136pr@xmath134ba@xmath26cu@xmath9o@xmath137 measured by muon - spin relaxation _ , phys . rev . * b42 * ( 1990 ) 6801 . tallon , c. bernhard , u. binninger , a. hofer , g.v.m . williams , e.j . ansaldo , j.i . budnick , ch . niedermayer , _ in - plane anisotropy of the penetration depth due to superconductivity on the cu - o chains in yba@xmath26cu@xmath9o@xmath138 , y@xmath26ba@xmath57cu@xmath83o@xmath139 , and yba@xmath26cu@xmath57o@xmath140 _ phys . * 74 * ( 1995 ) 1008 . niedermayer , c. bernhard , u. binninger , h. glckler , j.l . tallon , e.j . ansaldo , and j.i . budnick , _ muon spin rotation study of the correlation between @xmath2 and @xmath3 in overdoped tl@xmath26ba@xmath26cuo@xmath47 _ , phys . rev . lett . * 71 * ( 1993 ) 1764 . a. keren , a. kanigel , j.s . lord , a. amato , _ superconductivity and magnetism in ( ca@xmath134la@xmath136)(ba@xmath141la@xmath142)cu@xmath9o@xmath56 : a @xmath0sr investigation _ , in this proceedings , solid state comm . uemura , a. keren , g.m . luke , l.p . sternlieb , w.d . wu , j.h . brewer , r.l . whetten , s.m . huang , sophia lin , r.b . kaner , f. diederich , s. donnovan , g grner , k. holczer , _ magnetic field penetration depth in k@xmath9c@xmath10 measured by muon spin relaxation _ , nature * 352 * ( 1991 ) 605 . uemura , a. keren , l.p . le , g.m . luke , w.d . tsai , k. tanigaki , k. holczer , s. donovan , r.l . whetten , _ system dependence of the magnetic - field penetration depth in c@xmath10 superconductors _ , physica * c235 * -*240 * ( 1994 ) 2501 . luke , b.j . sternlieb , w.d . wu , y. j. uemura , j.h . brewer , t.m . riseman , c. e. stronach , g. saito , h. yamochi , h.h . wang , a.m. kini , k.d . carlson , j.m . williams _ muon - spin - relaxation measurements of magnetic penetration depth in organic superconductors ( bedt - ttf)@xmath26-x : x = cu(ncs)@xmath26 and cu[n(cn)@xmath26]br _ phys . * 68 * , ( 1992 ) 1923 . uemura , y. fudamoto , i.m . gat , m.i . larkin , g.m . luke , j. merrin , k.m . kojima , k. itoh , s. yamanaka , r.h . heffner , d.e . maclaughlin , _ @xmath0sr studies of intercalated hfncl superconductors _ , physica * b289 * ( 2000 ) 389 pan , e.w . hudson , k.m . lang , h. eisaki , s. uchida , j.c . davis , _ imaging the effects of individual zinc impurity atoms on superconductivity in bi@xmath26sr@xmath26cacu@xmath26o@xmath144 _ , nature ( london ) * 403 * ( 2000 ) 746 . j. schtzmann , s. tajima , s. miyamoto , s. tanaka , _ c - axis optical response of fully oxygenated yba@xmath26cu@xmath9o@xmath138 : observation of dirty - limit - like superconductivity and residual unpaired carriers _ , phys . * 73 * ( 1994 ) 174 . c. bernhard , j.l . tallon , th . blasius , a. golnik , ch . niedermayer , _ anomalous peak in the superconducting condensate density of cuprate high-@xmath2 superconductors at a unique doping state _ , phys . * 86 * ( 2001 ) 1614 . y.j uemura , w.j . kossler , x.h . kempton , h.e . schone , d. opie , c.e . stronach , d.c . johnston , m.s . alvarez , d.p . goshorn , _ antiferromagnetism of la@xmath26cuo@xmath145 studied by muon spin rotation _ rev . lett . * 59 * ( 1987 ) 1045 . luke , l.p . sternlieb , w.d . uemura , j.h . brewer , t.m . riseman , s. ishibashi , s. uchida , _ static magnetic order in la@xmath60ba@xmath61cuo@xmath57 _ , physica * c185 - 189 * ( 1991 ) 1175 . le , a. keren , g.m . luke , b.j . sternlieb , w.d . uemura , j.h . brewer , t.m . riseman , r.v . upasani , l.y . chiang , w. kang , p.m. chaikin , t. csiba , g. grner , _ muon spin rotation / relaxation studies in ( tmtsf)@xmath26-x compounds _ , phys . rev . * b48 * ( 1993 ) 7284 . heffner , j.d . thompson , g.j . nieuwenhuys , d.e . maclaughlin , p.c . canfield , b.k . chyo , a. amato , r. feyerherm , f.n . gygax , a. schenck , _ @xmath0sr studies of borocarbides _ , hyperfine interact . * 104 * ( 1997 ) 49 . lee , r.j . birgeneau , m.a . kastner , y. endoh , s. wakimoto , k. yamada , r.w . erwin , s.h . lee , g. shirane , _ neutron - scattering study of spin - density wave order in the superconducting state of excess oxygen - doped la@xmath26cuo@xmath146 _ , phys . rev . * b60 * ( 1999 ) 3643 . savici , y. fudamoto , i.m . gat , m.i . larkin , y.j . uemura , k.m . kojima , y.s . lee , m.a . kastner , r.j . birgeneau , _ static magnetism in superconducting stage-4 la@xmath26cuo@xmath146 _ physica * b289 - 290 * ( 2000 ) 338 . b. nachumi , y. fudamoto , a. keren , k.m . kojima , m. larkin , g.m . luke , j. merrin , o. tchernyshyov , y.j . uemura , n. ichikawa , m. goto , h. takagi , s. uchida , m.k . crawford , e.m . mccarron , d.e . maclaughlin , r.h . heffner , _ muon spin relaxation study of the stripe phase order in la@xmath147nd@xmath71sr@xmath134cu@xmath26o@xmath57 and related 214 cuprates _ phys * b58 * ( 1998 ) 8760 . kojima , h. eisaki , s. uchida , y. fudamoto , i.m . gat , a. kinkhabwara , m.i . larkin , g.m . luke , y.j . uemura , _ magnetism and superconductivity of high-@xmath2 cuprates ( la , eu , sr)@xmath26cuo@xmath57 _ , physica * b289 * ( 2000 ) 373 . . fisher , j.m . triscone , o. brunner , l. antognazza , m. affronte , o. eibt , l. mi/eville , t. boichat , m.g . karkut , _ superconductivity in artificially grown copper - oxide superlattices _ , physics * b169 * ( 1991 ) 116 . uemura , _ energy scales of exotic superconductors _ , in _ polarons and bipolarons in high - t@xmath148 superconductors and related materials _ , proceedings of the international workshop on polarons and bipolarons in high - t@xmath148 superconductors and related materials , cambridge , uk , april , 1994 , ed . by e. salje , a.s alexandrov and y. liang , cambridge university press ( 1995 ) , p.p . 453 - 460 . uemura , _ energy scales of high-@xmath2 cuprates , doped fullerenes , and other exotic superconductors _ , proceedings of international symposium / workshop on high-@xmath2 superconductivity and the c@xmath10 family , may 1994 , beijing , ed . ren , gordon and breach ( new york ) , ( 1995 ) , p.p . 113 - 142 . m. randeria , _ crossover from bcs theory to bose - einstein condensation _ , in _ bose - einstein condensation _ , ed . by a. griffin , d.w . snoke , s. stringari , cambridge university press , cambridge , 1995 , p.p . 355 - 391 , and references therein . kransov , a.e . kovalev , a. yurgens , d. winkler , _ magnetic field dependence of superconducting gap and the pseudogap in bi2212 and hgbr@xmath26-bi2212 studied by intrinsic tunneling spectroscopy _ , phys . rev . lett . * 86 * ( 2001 ) 2657 . z.a . ong , y. wang , t. kakeshita , s. uchida , _ vortex - like excitations and the onset of superconducting phase fluctuation in underdoped la@xmath149sr@xmath134cuo@xmath57 _ , nature * 406 * ( 2000 ) 486 . feng , d.h . shen , c. kim , h. eisaki , a. damascelli , r. yoshizaki , j. shimoyama , k. kishio , g.d . gu , s. oh , a. andrus , j. odonnell , j.n . eckstein , z.x . shen , _ signature of superfluid density in the single - particle excitation spectrum of bi@xmath26sr@xmath26cacu@xmath26o@xmath144 _ , science * 289 * ( 2000 ) 277 . c. panagopoulos , j.l . tallon , b.d . rainford , j.r . cooper , c.a . scott , t. xiang _ low frequency spins and the ground state in high-@xmath2 cuprates _ , contribution to this proceedings , solid state commun . mackenzie , s.r . julian , g.g . lonzarich , a. carrington , s.d . hughes , r.s . liu , d.c . sinclair , _ resistive upper critical - field of tl@xmath26ba@xmath26cuo@xmath62 at low temperature and high magnetic - fields * 71 * ( 1993 ) 1238 . larkin , a. kinkhabwala , y.j . uemura , y. sushko , g. saito , _ pressure dependence of the magnetic penetration depth in @xmath151-(bedt - ttf)@xmath26cu(ncs)@xmath26 _ , phys . * b64 * ( 2001 ) 144514 . j. caulfield , w. lubczynski , f.l . pratt , j. singleton , d.y.k . ko , w. hayes , m. kurmoo , p. day , _ magnetotransport studies of the organic superconductor @xmath151- ( bedt - ttf)@xmath26cu(ncs)@xmath26 under pressure the relationship between carrier effective - mass and critical temperature _ j. phys . condens .- matter * 6 * ( 1994 ) 2911 .

we study the properties of one - dimensional hypergeometric integral solutions of the-difference ( `` quantum '' ) analogue of the knizhnik zamolodchikov bernard equations on tori . we show that they also obey a difference kzb heat equation in the modular parameter , give formulae for modular transformations , and prove a completeness result , by showing that the associated fourier transform is invertible . these results are based on transformation properties parallel to those of elliptic gamma functions . supported in part by the swiss national science foundation]supported in part by nsf grant dms-9801582 ] _department of mathematics , eth - zentrum , _ _ 8092 zrich , switzerland _ _department of mathematics , university of north carolina at chapel hill , _ _ chapel hill , nc 27599 - 3250 , usa _ october 2001

math0110081

in this paper we continue the study of the q - analogue of the knizhnik zamolodchikov bernard ( qkzb ) equations on elliptic curves and their solutions initiated in @xcite . in @xcite , hypergeometric solutions of qkzb equations were introduced . in @xcite , the monodromy of hypergeometric solutions was calculated , and a symmetry between equations and monodromies was discovered : the equations giving the monodromy are again qkzb equations but with modular parameter and step of the difference equations exchanged . in @xcite we introduced the @xmath0-analogue of the kzb heat equation , which governs the change in the modular parameter of the elliptic curve , proved that it is compatible with the other equations and explained how to recover the kzb heat equation in the semiclassical limit . in this paper we prove three results about our hypergeometric solutions in the case where the sum of highest weights is two : we show that the hypergeometric solutions also obeys the qkzb heat equation of @xcite , see theorem [ t - he1 ] . we give a formula ( theorem [ t - mod ] ) for the transformation properties of the hypergeometric solutions under the modular group @xmath4 . we prove a completeness result , corollary [ completeness ] , by showing that the associated `` fourier transform '' is invertible . then we show that these results are part of a bigger picture , in which the modular group combines with the transformation defined by the qkzb heat equation to give a set of quadratic identities for our generalized hypergeometric integrals . in fact , this picture can already be seen in a simpler situation , in which the elliptic gamma function @xcite , @xcite plays the role of the hypergeometric integral . the elliptic gamma function is a function @xmath5 of three complex variables obeying identities @xcite involving its values at points related by an action of @xmath6 . these identities mean that @xmath7 is a `` degree 1 '' generalization of a jacobi modular function , see @xcite . these identities are a scalar version of the identities obeyed by the hypergeometric solutions of the qkzb equation . their meaning is that the hypergeometric integrals define a discrete projectively flat @xmath1-connection ( i.e. , a lift of the action to the projectivization ) on a vector bundle over `` regular '' orbits of @xmath1 acting on the variety of pairs ( point in @xmath8 , plane through @xmath9 containing it ) . this is the content of theorem [ t - pf ] . the results on the elliptic gamma function are used in the proof , since the `` phase function '' which appears in the integrand of hypergeometric solutions is a ratio of gamma functions . in fact , we see `` experimentally '' that there seems to be a principle stating that to every identity obeyed by the gamma function , there corresponds an identity for hypergeometric integrals . the proofs of the identities consist in applying the gamma function identity to the phase function in the integrand , and then use a version of stokes theorem to relate the integrals . the second step is relatively simple in the case of the one - dimensional integrals to which we restrict ourselves here , but becomes exceedingly involved in the higher dimensional case . proving our identities in the higher dimensional case , i.e. , if the sum of highest weights is larger than two , remains a challenging open problem . an alternative approach to this problem is based on representation theory : in @xcite a representation - theoretic interpretation of a degenerate version of the qkzb equations was established . it was shown that traces of intertwining operators for quantum groups satisfy a version of the qkzb equations and are eigenfunctions of analogues of macdonald operators . this fact indicates that our theorem [ t - he1 ] is an elliptic analogue of the macdonald - mehta identities proved by cherednik , @xcite , @xcite , and the present work is a glimpse into an elliptic version of macdonald s theory . in fact theorem [ t - pf ] concerns the case of generic parameters , where an infinite - dimensional vector bundle is preserved by the projectively flat connection . in a next paper we will restrict our attention to special @xmath1-orbits . the projectively flat connection can be then defined on a finite - dimensional vector bundle of theta - functions which are a @xmath0-deformed version of the space of conformal blocks in conformal field theory . in this setting the analogy with macdonald s theory will appear more explicitly . another degeneration of the @xmath1 symmetry of our hypergeometric integrals are the @xmath1 symmetries of the ordinary fourier transform indicated in @xcite . we use the definitions and notations of @xcite . the elliptic @xmath10 qkzb equations are a compatible system of difference equations for a function @xmath11 of @xmath12 , @xmath13 and @xmath14 , @xmath15 , taking values in the zero weight space @xmath16 $ ] of a tensor product of @xmath17 evaluation modules . this space comes with a basis of eigenvectors of commuting operators @xmath18 , @xmath19 and depends on parameters @xmath20 . the qkzb equations have the form @xmath21 they are supplemented by the qkzb heat equation @xmath22 the step @xmath23 is a complex parameter and @xmath24 is the standard basis of @xmath25 . here @xmath26 is viewed as a function of @xmath27 and @xmath14 with values in the space @xmath28)$ ] of meromorphic @xmath16$]-valued functions of a complex variable @xmath29 . the qkzb operators @xmath30 are difference operators in @xmath29 and can be expressed in terms of a product of ( dynamical ) @xmath31-matrices . the last equation is the qkzb heat equation and involves the integral operator @xmath32 . the latter is expressed in terms of hypergeometric integrals : let @xmath33\otimes v_{{\mathbf{\lambda}}}[0]$ ] be the universal hypergeometric function as in @xcite . we may view it as a function @xmath34 taking its values in the @xmath35\otimes v_{{\mathbf{\lambda}}}[0]$]-valued functions of the two `` dynamical variables '' @xmath29 and @xmath36 . then it is a projective solution ( i.e. , a solution up to a constant factor ) of the qkzb equations in the first factor , and of the mirror qkzb equations in the second : @xmath37 the mirror qkzb operators @xmath38 are obtained from the qkzb operators by exchanging @xmath14 and @xmath23 and `` reversing the order of factors '' , see @xcite . the operators @xmath39 , @xmath40 are operators of multiplication by certain functions of the @xmath18 and the dynamical variable @xmath36 and @xmath29 , respectively : in terms of the function @xmath41 we have , for @xmath42 , @xmath43 the integral operator @xmath32 is then @xmath44 @xmath45 is the operator of multiplication by the function @xmath46 and @xmath47 is a bilinear form on @xmath16$]-valued functions , whose kernel is the shapovalov bilinear form on @xmath16 $ ] : @xmath48 if @xmath49 , the universal hypergeometric function is given in terms of jacobi s first theta function @xmath50 , and the phase function , see appendix [ app - a ] , by the following formulae : @xmath16 $ ] has a basis @xmath51 , with @xmath52 in the @xmath53th factor ( @xmath54 ) and @xmath55 . then the _ weight functions _ are @xmath56 , with @xmath57 on the other hand , @xmath58 , where @xmath59 the universal hypergeometric function @xmath60 is expressed in terms of these weight functions and the phase function @xmath61 we then have @xmath62 @xmath63 , with @xmath64 the integral is defined as the analytic continuation from the region where re@xmath65 , @xmath66 and im@xmath67 . in this region , the integration cycle is just the interval @xmath68 $ ] . after the analytic continuation , the cycle is deformed to go above the pole at @xmath69 and below the pole at @xmath70 , as in fig . [ fig1 ] . the shapovalov form is @xmath71 with @xmath72 our first result is that our hypergeometric projective solutions of the qkzb equations are also projective solutions of the quantum heat equation : [ t - he1 ] suppose that @xmath49 . let us view @xmath73 as a function @xmath34 with values in the space of @xmath35\otimes v_{{\mathbf{\lambda}}}[0]$]-valued functions of @xmath29 and @xmath36 . then @xmath60 is a projective solution of the qkzb heat equation @xmath74 where @xmath75 here @xmath32 acts on the first dynamical variable @xmath29 and @xmath76 is the operator of multiplication by the function @xmath77 of the second dynamical variable @xmath36 . it is then easy to construct true solutions to the system ( [ e - qkzb ] ) , ( [ e - heat ] ) from these projective solutions : for any @xmath36 , @xmath78 , viewed as a function of @xmath79 , obeys ( [ e - qkzb ] ) and ( [ e - heat ] ) in the first factor . in more explicit terms , we have the following statement . [ q - heat - n ] let for @xmath80 @xmath81 and let the _ fundamental hypergeometric solution _ @xmath82 be defined by @xmath83 then , for any @xmath84 , the @xmath16$]-valued function @xmath85 is a solution of the system @xmath86 where @xmath87 . _ proof : _ a consequence of ( [ eq1 ] ) , theorem [ t - he1 ] and the identity @xmath88 the exponential function on the right - hand side is independent of @xmath79 and therefore does not affect the statement of the theorem . @xmath89 * remarks . the constant @xmath90 could also be eliminated by including a factor @xmath91 in @xmath92 , but it is simpler to consider @xmath90 as part of the heat equation . 2 . in the definition of the fundamental hypergeometric solution above , we have included an additional factor , see ( [ e - d99 ] ) , with respect to the obvious choice . this leads to simpler formulae in the next section , since @xmath93 appears in the modular transformations of the qkzb operators . the coefficients of the qkzb equations are quasi - periodic functions of @xmath94 with periods @xmath95 and @xmath14 . it is therefore natural to consider the qkzb equations as equations for sections of a certain vector bundle over a cartesian power of the elliptic curves @xmath96 . since , for any @xmath97 , the elliptic curves with modulus @xmath14 and @xmath98 are isomorphic , the corresponding qkzb equations are related . for generators , the formulae relating solutions are the following . let for @xmath80 @xmath99 then we have : 1 . suppose that @xmath100 is a solution of the qkzb equations with parameters @xmath101 . then @xmath102 is a solution of the qkzb equations with parameters @xmath103 . 2 . suppose that @xmath100 is a solution of the qkzb equations with parameters @xmath104 . then @xmath105 is a solution of the qkzb equations with parameters @xmath103 . the proof of this proposition is based on the formulae for transformation properties of the qkzb equations under modular transformations , see appendix [ app - b ] . now the question about monodromy is well - posed : can one express the fundamental solution at the transformed values of the parameters in terms of the fundamental solution at the original values ? the answer is provided by the following result . 1 . @xmath106 2 . suppose that @xmath107 , @xmath108 . @xmath109 the monodromy matrix @xmath110 is @xmath111 the integration over @xmath36 is over the path @xmath112 , @xmath113 for any generic real @xmath114 . the first statement of the theorem is trivial to check . the second is the first of the two _ modular relations_. introduce functions @xmath115 : @xmath116 then the modular relations are the identities : [ t - mod ] \(i ) suppose that @xmath117 , @xmath108 . then the universal hypergeometric function @xmath60 satisfies the following relation , @xmath118 the integration over @xmath36 is over the path @xmath112 , @xmath113 for any generic real @xmath114 . \(ii ) suppose that @xmath119 , @xmath120 . then the universal hypergeometric function @xmath60 satisfies the following relation , @xmath121 @xmath122 @xmath123 the integration over @xmath36 is over the path @xmath112 , @xmath113 for any generic real @xmath114 . this theorem is proved in [ ss - modular ] and [ new - modular ] . throughout this section , we assume , for definiteness , that @xmath124 , and that @xmath125 is sufficiently small . the results hold for a more general range of parameters by analytic continuation . the qkzb heat equation is based on an integral transformation . in this section we give a space on which this integral transformation is defined and invertible . let us fix our parameters @xmath126 . then the shapovalov form has poles at the points @xmath127 where @xmath128 as well as at the translates of this points by the lattice of periods . at these points , the hypergeometric integrals obey the following `` resonance relations '' @xcite . [ p - resrel ] let @xmath129 , @xmath130 be integers and let @xmath131 . then : 1 . if @xmath132 , then @xmath133 2 . @xmath134 3 . if @xmath135 , then @xmath136 4 . @xmath137 _ proof : _ ( i ) using the functional relation @xmath138 , we see that @xmath139 and @xmath140 thus we have equality at the level of integrands . \(ii ) similarly , we have @xmath141 and using the relation @xmath142 we obtain @xmath143 the last product of ratios of theta functions may be absorbed using the functional relation ( [ eq - zwz ] ) for @xmath144 . this gives @xmath145 the result is then obtained by shifting the integration variable @xmath146 by @xmath147 and using the relation @xmath148 the remaining claims ( iii ) , ( iv ) are proved in a similar way . @xmath89 . * definition . * let @xmath149 be the space of holomorphic functions @xmath150 $ ] , @xmath151 such that 1 . if @xmath132 , then @xmath152 2 . @xmath153 let @xmath154 be the space of functions @xmath155 such that 1 . there exist constants @xmath156 ( depending on @xmath157 ) such that @xmath158 for all @xmath80 . examples of functions in @xmath154 can be constructed using the universal hypergeometric function : [ p - uine0 ] if @xmath159 then the function @xmath160 belongs to @xmath154 for all values of the remaining parameters . if @xmath161 then the function @xmath162 belongs to @xmath154 for all values of the remaining parameters . _ proof : _ prop . [ p - resrel ] implies that these functions belong to @xmath149 . the bound ( iii ) follows from lemma [ l - est ] below . @xmath89 [ l-925 ] @xmath163 if and only if @xmath164 . _ proof : _ it is clear that @xmath165 obeys ( i ) in the definition of @xmath166 . property ( ii ) follows from the identity @xmath167 for @xmath168 . @xmath89 for the construction of the integral transform , we will need the following variant of prop . [ p - uine0 ] . [ c - uine ] the function @xmath169 belongs to @xmath170 for all @xmath36 . the function @xmath171 belongs to @xmath172 for all @xmath29 . _ proof : _ an immediate consequence of prop . [ p - uine0 ] and lemma [ l-925 ] . @xmath89 we define two integral transforms @xmath173 and @xmath174 , then show that they are inverse to each other . the first is @xmath175 with @xmath176 the second is @xmath177 with @xmath178 these transformations depend on the choice of a real parameter @xmath179 or @xmath180 . we say that @xmath179 is an _ admissible shift _ for @xmath181 if it does not lie in any interval @xmath182 $ ] , @xmath183 . we say that @xmath180 is an _ admissible shift _ for @xmath184 if it is a generic real number . [ thm - fourier ] suppose that @xmath124 . if @xmath185 then the integral ( [ eq - f1 ] ) is absolutely convergent , independent of the choice of admissible shift and defines a linear map @xmath186 . if @xmath187 then the integral ( [ eq - f2 ] ) is absolutely convergent , independent of the choice of admissible shift and defines a linear map @xmath188 . moreover @xmath189 the proof of the first part of the theorem is contained in the next subsection . the proof of the inversion formula is deferred to [ ss - inv ] . this theorem implies a completeness result for the qkzb equations . for generic @xmath190 , we may consider the qkzb equations on the set @xmath191 . any solution is uniquely determined by its initial condition at @xmath192 . a class of solutions is given by taking linear combinations of components of the fundamental hypergeometric solution : let us say that a solution @xmath26 is _ of hypergeometric type _ if it is of the form : @xmath193 for some functions @xmath194 and some cycle @xmath195 . [ completeness ] any solution @xmath196 with initial condition in @xmath197 is of hypergeometric type . more precisely , let @xmath190 be generic . suppose @xmath198 and let @xmath199 . then , for all @xmath200 , @xmath201 , viewed as a function of @xmath36 , belongs to @xmath202 and the function @xmath203 , with @xmath204 is , for any generic @xmath205 , a solution of hypergeometric type of the qkzb equations with modulus @xmath14 , step @xmath23 and initial condition @xmath206 . moreover , for all @xmath207 , @xmath102 is independent of the choice of @xmath180 , and @xmath208 belongs to @xmath209 . _ proof : _ let @xmath210 denote the eigenvalues of @xmath211 : @xmath212 they grow at most exponentially as @xmath213 . the coordinates of @xmath214 are then @xmath215 by the theorem , @xmath216 . then @xmath214 clearly obeys the bound ( iii ) of the definition of @xmath202 . the resonance relations ( i),(ii ) are checked by inserting the definitions . for example , for @xmath132 we have ( using the fact that @xmath217 has integral coordinates ) @xmath218 on the other hand , the expression in curly brackets is equal to @xmath219 if @xmath220 , to @xmath221 if @xmath222 , and to @xmath95 otherwise . it then follows by theorem [ thm - fourier ] that @xmath223 is in @xmath209 and that the initial condition @xmath224 is satisfied . @xmath26 is a solution of hypergeometric type since the @xmath225-dependent part of the kernel of integration is @xmath226 which is equal to the fundamental solution @xmath92 up to @xmath225-independent factors , cf . ( [ e - d99 ] ) . @xmath89 * remarks . * 1 . the result may be expressed in the following terms : our generalized fourier transform maps the qkzb equations to the difference equations @xmath227 which are easily solved , since @xmath228 is a diagonal multiplication operator . so the solution given in the corollary is the fourier transform of the solution of this simple system of equations with initial condition given by the inverse fourier transform of the given initial condition . an initial condition at @xmath192 uniquely determines a solution only on the set @xmath229 , since solutions defined for all @xmath27 can always be changed by multiplying by @xmath23-periodic functions without changing the initial condition . however the formula in corollary [ completeness ] gives a solution for all @xmath27 . on the other hand it is only for @xmath230 that we know that the integral is independent of the choice of @xmath180 . for general @xmath27 there is no cancellation of residues at pairs of poles that would allow us to move the integration contours , and one should expect that the solution depends on @xmath180 . in fact , if we move @xmath180 just a little bit , then the integration contour crosses infinitely many poles ( for generic @xmath125 ) , so one would expect that the solution depends discontinuously on @xmath180 . it would be interesting to understand this dependence in more detail . we prove here that the integral transformations are well - defined on the considered spaces . we start by estimating the integrands . [ l - est ] @xmath231 for some @xmath156 depending on @xmath232 . for every @xmath233 there exist constants @xmath234 such that if the distance between @xmath29 and the singularities of @xmath235 is at least @xmath114 , then @xmath236 _ proof : _ the first bound is obtained by applying the estimates of lemma [ l - c1 ] to the integral ( [ eq-467 ] ) . the second follows using the lower bound in lemma [ l - c1 ] . @xmath89 in order to show that the integral is independent of the choice of admissible shift and to bound the integral transform , we will need to shift the integration contour . the next result shows that this is possible since poles occur in pairs with opposite residues . [ l - res ] 1 . suppose @xmath237 . then for all @xmath168 , @xmath238 and @xmath239 , @xmath240 2 . suppose @xmath241 . then for all @xmath168 , @xmath238 and @xmath239 , @xmath242 _ proof : _ a straightforward calculation . @xmath89 _ proof of the first part of theorem [ thm - fourier ] . _ it follows from lemma [ l - res ] ( iii ) that the integrand in ( [ eq - f1 ] ) has only simple poles at @xmath243 . if @xmath179 is an admissible shift then the integration contour stays away from the singularities and we can use the estimates of lemma [ l - est ] . the integrand is then the product of a function growing at most like @xmath244 times @xmath245 which for @xmath124 converges very rapidly to zero in the real direction . so the integrand is an @xmath246 function . moreover by lemma [ l - res ] ( iv ) the residue at @xmath247 is opposite to the residue at @xmath248 . it follows that the integration contour can me moved across each of these pairs of singularities without changing the value of the integral . this shows that the integral is independent of the choice of admissible shift . let us next show that @xmath249 defined by ( [ eq - f1 ] ) belongs to @xmath202 . properties ( i ) ( ii ) follow from corollary [ c - uine ] , and we are left with the proof of ( iii ) . for this we shift variables : let @xmath250 . then @xmath251 the integration contour is moved by this change of variable , but using the independence on the choice of @xmath179 we may choose it to lie in the interval @xmath252 $ ] say . then lemma [ l - est ] yields @xmath253 the triangle inequality @xmath254 and the gaussian integral over @xmath255 give the estimate ( iii ) in the definition of @xmath202 . the proof that ( [ eq - f2 ] ) defines a function in @xmath209 is analogous with a little difference . now the integration contour is parallel to the line @xmath256 . so if the shift @xmath180 is generic it will not meet the singular set @xmath257 of the integrand . however it will come arbitrarily close to these singularities . still , the integral is absolutely convergent , since the distance to the singular set decreases polynomially : dist@xmath258 for some @xmath259 . this implies that the divergence coming from the poles close to the integration contour is at most polynomial . this does not spoil the integrability which is due to the exponential decay of @xmath260 . here we prove theorem [ t - he1 ] . the statement of the theorem is @xmath261 we proceed to evaluate the right - hand side . the @xmath36-dependent part may be simplified by using the following identity : [ l-89 ] @xmath262 _ proof : _ consider each term in the sum on the left as a function of @xmath36 : it is periodic with period one and as @xmath36 is replaced by @xmath263 , it is multiplied by @xmath264 . the poles of the term labeled by @xmath265 are at @xmath266 and at @xmath267 modulo @xmath268 . the most general form of a meromorphic function of @xmath36 with these properties is @xmath269 the coefficients are determined by comparing the residues at the poles : @xmath270 similarly one determines @xmath271 which turns out to be equal to @xmath272 . it follows that in the sum over @xmath265 only two terms are not canceled and we obtain our claim . @xmath89 the products of ratios of theta functions in the above identity may be absorbed into the phase functions @xmath273 by means of their functional relation ( [ eq - fromega ] ) . the right - hand side of ( [ claim1 ] ) is then @xmath274dt\,ds\,d\mu . \nonumber\end{aligned}\ ] ] the next thing to notice is that if we change variables in the first of these two terms by replacing @xmath146 by @xmath275 , @xmath276 by @xmath277 and @xmath36 by @xmath278 we obtain exactly the second term up to a sign ! indeed , the shift of @xmath146 in @xmath279 produces a factor @xmath280 , the shift of @xmath276 in @xmath281 produces a factor @xmath282 and we get an additional @xmath283 from shifting the argument @xmath284 in the ratio of theta function in the square bracket . these factors are canceled by the shift of @xmath36 in the exponential function . to compute this integral we therefore have to carefully consider the deformation of integration contours involved in the change of variables . as we shall see , this deformation produces a residue at the pole @xmath285 . for these considerations we assume that im@xmath286 , @xmath54 , for some @xmath287 large compared to @xmath14 , @xmath23 , and that the points @xmath288 are on the real axis . the general case can then be obtained by analytic continuation . in this range of parameters , the integration cycles for the @xmath146 and @xmath276 integration is the interval @xmath68 $ ] . the integrand , viewed as a function of @xmath146 or @xmath276 , is then regular in the strip im@xmath289,im@xmath290 . the first term in ( [ eq-88 ] ) , however , has additional poles at @xmath291 , ( @xmath292 ) . to deal with these poles we move slightly the @xmath276-integration cycle into the upper half plane . after the change of variable , in the first term in ( [ eq-88 ] ) , @xmath293 , @xmath294 , the new variables are integrated over @xmath295 $ ] , @xmath296 $ ] for some small @xmath297 . the first term becomes then equal to the minus the second term in ( [ eq-88 ] ) after deforming the integration cycles to the original position , but during this deformation we encounter the pole at @xmath285 . therefore we obtain a residue @xmath298 using the identity ( [ eq - om1 ] ) and @xmath299 we see that this expression reduces to @xmath300 , completing the proof of theorem [ t - he1 ] . here we present the proof of theorem [ t - mod ] ( i ) . the assumption that @xmath107 implies that the integration over @xmath36 on the path @xmath301 , @xmath113 , converges absolutely : the singularities of @xmath302 are avoided for generic @xmath114 , and at infinity the factor @xmath303 converges very fast to zero . to get the identity as stated in the theorem , one uses the following simple properties of @xmath304 : @xmath305 let us proceed with the proof . we insert in the left - hand side the hypergeometric integral for @xmath306 and @xmath307 and call the integration variables @xmath146 and @xmath276 , respectively . it will be convenient to make the change of variables @xmath308 . to apply lemma [ l-89 ] we use the transformation properties of weight functions : @xmath309 a similar formula involving @xmath310 gives the transformation behavior of @xmath311 . using lemma [ l-89 ] and the functional relation ( [ eq - zwz ] ) of @xmath312 , we get : @xmath313 \,dt\,ds\,e^{-\frac{\pi i\tau}{4\eta p}\mu^2 } d\mu,\nonumber\end{aligned}\ ] ] where @xmath314 as in the proof of the heat equation , the two terms in this equation cancel after formally changing variables @xmath315 , @xmath316 , @xmath317 in the first term . however , we have to carefully see what happens to the integration cycles after this change of variables . we consider the region of parameters where @xmath66 , @xmath318 and @xmath319 . then the original integration contours may be chosen to be the interval @xmath68 $ ] . for @xmath320 , let us denote by @xmath321 the path @xmath322.\ ] ] after the change of variable @xmath308 , the @xmath276 integration contour becomes the path @xmath323 . [ residue ] suppose that @xmath324 are complex numbers in the upper half plane . let @xmath325 be a meromorphic function of two variables such that @xmath326 and such that @xmath327 is regular in a domain containing @xmath328\}.\ ] ] then @xmath329 and all integrals are absolutely convergent . _ proof : _ since @xmath23 is not real the integration contours intersect transversally , so that @xmath325 ( which has at most a simple pole on the diagonal ) is absolutely integrable . more generally , with our assumption on the regularity of @xmath330 , the integral @xmath331 is absolutely convergent for all @xmath332 $ ] . let , for @xmath297 , @xmath333 be the integration domain obtained from @xmath334 by removing the points where @xmath335 and set @xmath336 then @xmath337 . we have @xmath338 from which the claim follows immediately . in the application of stokes theorem above the other boundary components of @xmath339 do not contribute since the function is periodic . the remaining component is homotopic to a circle with negative orientation , which explains the change of sign . @xmath89 applying this lemma to our situation after shifting @xmath36 by @xmath340 in the first term yields : @xmath341 now the gaussian integral over @xmath36 along @xmath342 may be performed explicitly , and our claim follows from ( [ eq - om4 ] ) and ( [ eq - tpwf ] ) . here we present the proof of theorem [ t - mod ] ( ii ) which is parallel to the proof of theorem [ t - mod ] ( i ) . the assumption that @xmath343 implies that the integration over @xmath36 over the path @xmath301 , @xmath113 , converges absolutely : the singularities of @xmath302 are avoided for generic @xmath114 , and at infinity the factor @xmath344 converges very fast to zero . we insert in the left - hand side the hypergeometric integral for @xmath306 and @xmath307 and call the integration variables @xmath146 and @xmath276 , respectively . it will be convenient to make the change of variables @xmath345 . to apply lemma [ l-89 ] we use the transformation properties of weight functions : @xmath346 we have @xmath347 \,dt\,ds\,e^{-\frac{\pi i p}{4\eta \tau}\mu^2 } d\mu,\nonumber\end{aligned}\ ] ] where @xmath348 the two terms in the right hand side of this equation cancel after formally changing variables @xmath349 , @xmath350 , @xmath351 in the second term . however , we have to carefully see what happens to the @xmath352 integration cycle after this change of variables . we consider the region of parameters where @xmath66 , @xmath318 and @xmath353 . in this case the @xmath352 integration cycle is the product of paths @xmath354 . applying to this situation lemma [ residue ] after decreasing @xmath36 by @xmath340 we see that the left hand side of the new modular relation is equal to @xmath355 @xmath356 now the gaussian integral over @xmath36 along @xmath342 may be performed explicitly , and our claim follows from ( [ eq - om4 ] ) and ( [ eq - tpwf ] ) . here we conclude the proof of theorem [ thm - fourier ] . we have to prove the identities @xmath357 @xmath358 we give a proof of the first of these identities . the second is treated in a similar way . as in the proof of the heat equation , the @xmath36-dependent part of the integrand may be simplified by collecting the terms with the same poles , and using the functional relation of @xmath312 , with the result : @xmath359\,dt\,ds\,d\mu.\nonumber\end{aligned}\ ] ] as in the proof of the heat equation , the two terms in this equation cancel after formally changing variables @xmath315 , @xmath316 , @xmath317 in the second term . however , we have to carefully see what happens to the integration cycles after this change of variables . we consider a region of parameters where @xmath66 and @xmath360 are small compared to 1 , im@xmath361 and to the distance between the @xmath288 . we study the singularities of the integrand and the form of the integration cycles in the vicinity of @xmath288 . the poles of the integrand of @xmath362 are at @xmath363 and at @xmath364 ( @xmath365 ) . the cycle for the integration over @xmath276 goes below @xmath70 and above @xmath69 . the poles of the integrand of @xmath366 in the vicinity of @xmath288 are at @xmath367 and at @xmath368 ( @xmath365 ) . the @xmath146 integration cycle goes above @xmath70 and below @xmath69 . these integration contours are depicted in fig . [ fig2 ] . to treat the two terms on the right - hand side of ( [ eq-90 ] ) separately , we split the contour for the @xmath146 integration into two pieces , as shown in fig . then the integration cycle does not meet the additional poles at @xmath285 of both terms . let us first consider the first term in ( [ eq-90 ] ) . as a function of @xmath146 it is actually regular at @xmath69 so that the circle around this point does not contribute . ( [ fig4 ] ) the remaining integration cycle is shown , along with the position of the poles for @xmath146 and @xmath276 . after changing variables @xmath315 , @xmath369 , @xmath317 in the second term of ( [ eq-90 ] ) the integrand of the second term becomes equal to the first one , but the integration contour are shifted : the @xmath36-integration cycle can be deformed back to the original one without encountering singularities . the cycle for the @xmath146 and @xmath276 integration is depicted in fig . [ fig5 ] . we have to deform the integration cycle to the position shown in fig . [ fig4 ] . in so doing , we pick the residue at @xmath370 of the residue at @xmath371 . at this point , we may use the identity ( [ eq - om5 ] ) . the result is @xmath372 the residue may be computed at @xmath69 since its argument is @xmath14-periodic . our claim follows then from the @xmath373 _ proof : _ this formula can be deduced from the more general result in @xcite , formula ( c.4 ) . for the sake of completeness , we include the simple proof in this special case . the product of weight functions of which we compute the residue is an elliptic function of @xmath146 whose simple poles in a fundamental domain are in the set @xmath374 . if @xmath375 , then the function is regular at @xmath69 , @xmath54 , so the residues vanish . if @xmath376 , then the function is regular at @xmath70 . therefore the sum of the residues in the claim is the sum over the residues at _ all _ poles and vanishes by the residue theorem . if @xmath377 , only the residue at @xmath378 gives a non - vanishing contribution , which is easily evaluated , and gives @xmath379 . in this section we recast our results into a form which shows the analogy with the identities discovered in @xcite for the elliptic gamma function . in @xcite we showed that the elliptic gamma function is a `` degree 1 '' generalized jacobi modular function for the group @xmath380 . this amounts to the identities of theorem [ t - jm ] below . the hypergeometric integral @xmath60 obeys a non - abelian version of these identities . we also discuss the geometric interpretation of these identities : they can be interpreted as the projective flatness of a discrete connection on a vector bundle over a suitable @xmath381-space , see also @xcite . as in the case of gamma functions these properties are more transparent if we rewrite the identities , which we formulate in [ ss - ide ] in homogeneous variables @xmath382 defined by @xmath383 , @xmath384 . the @xmath381-action is then linear . the next step , as in the case of @xmath7 , is to extend the range of parameter to a @xmath381-space , as the condition of positivity of imaginary parts is not preserved by the action . this can be done in a surprisingly easy way by reflection arguments . we do it in the simplest case of one tensor factor in [ ss - newrange ] . after this is done , we rewrite the identities as the projective flatness of an @xmath1-connection in [ ss - profl ] . we first recall the properties of the elliptic gamma function that are relevant here . more details are included in appendix [ app - a ] . @xcite[t - jm ] the elliptic gamma function , defined by the formula @xmath385 for @xmath386 , obeys the following identities @xmath387 @xmath388 for some @xmath389 $ ] given in appendix [ app - a ] . in fact @xmath7 is also defined for negative imaginary parts of @xmath14 and @xmath23 by a reflection procedure , see [ a - ext ] , and the above identities continue to hold in this wider range of parameters . let @xmath390 be the space of functions @xmath155 such that there exist constants @xmath156 ( depending on @xmath157 ) such that @xmath391 for all @xmath80 . for @xmath392 we have constructed examples of functions in @xmath393 in prop . [ p - uine0 ] . let @xmath394 , @xmath395 be the operators acting on @xmath16$]-valued functions @xmath396 as @xmath397 then these operators restrict to isomorphisms @xmath398 then theorem [ thm - fourier ] about the fourier transform may be reformulated , in a slightly generalized form , as follows : [ invert ] the operator @xmath399 is an invertible linear map from @xmath400 onto @xmath401 . the properties of the universal hypergeometric function may then be expressed as relations for these operators : first of all the analogue of the identities ( [ e - gf2])([e - gf4 ] ) are the qkzb and mirror qkzb equations ( [ eq1 ] ) . the remaining identities are given by the following result . [ t - relations ] 1 . @xmath402 , if @xmath168 . the identities @xmath403 hold in @xmath404 . the identity @xmath405 holds in @xmath406 with @xmath407 3 . the identity @xmath408 holds in @xmath409 with @xmath410 and @xmath411 _ proof : _ the first statement ( i ) is trivial . the second statement ( ii ) is a reformulation of theorem [ t - he1 ] and ( iii ) is a reformulation of theorem [ t - mod ] . @xmath89 for simplicity we assume from now on that the tensor product @xmath412 of @xmath10 representations consists of one factor which is the three - dimensional representation @xmath413 . in this case the universal hypergeometric function does not depend on @xmath414 and is a scalar function , @xmath415 the integration cycle @xmath90 is as in sect . [ s - hyp ] : in this case it can be defined to be the interval @xmath68 $ ] if @xmath125 has positive imaginary part . for general @xmath125 the integral is defined by analytic continuation , with the effect of deforming the contour as in fig . the resulting function is a meromorphic function on @xmath416 , where @xmath417 denotes the open upper half - plane . this function is regular as long as the integration cycle is not pinched between poles of the integrand . as these poles are at @xmath418 , @xmath419 , @xmath60 is regular if @xmath340 does not belong to the set @xmath420 . the following is an interesting alternative formula for the function @xmath60 , @xmath421 } e^{-\frac{\pi i}{2\eta}(\lambda+2\eta)(\mu+2\eta)}\\ & & \times\ , \sum_{j , k=0}^\infty e^{-2\pi i(j\lambda+k\mu)+(2j+1)(2k+1)\eta } \theta_0(\lambda+2\eta+kp,\tau ) \theta_0(\mu+2\eta+j\tau , p ) \\ & & \times\,\prod_{l=0}^{k-1 } \frac { \theta_0(lp+4\eta,\tau ) } { \theta_0((l+1)p,\tau ) } \prod_{l=0}^{j-1 } \frac { \theta_0(l\tau+4\eta , p ) } { \theta_0((l+1)\tau , p)}\,,\end{aligned}\ ] ] where @xmath422 , see app . [ app - a ] . this formula can be proved by moving @xmath90 to infinity and picking up the residues at the poles , and is valid in a certain region of parameter space . we will not use this formula in this paper . the shapovalov form is @xmath423 consider @xmath424 and the projectivization of the dual space , @xmath425 . consider @xmath426 where @xmath427 the natural projection @xmath428 is a projective line bundle . the group @xmath429 acts on @xmath430 , @xmath431 for @xmath432 . for @xmath433 such that @xmath434 introduce the space of functions @xmath435 thus @xmath436 consists of entire holomorphic functions @xmath437 obeying the resonance relation @xmath438 , where @xmath179 is the sign of @xmath439 , and the bound @xmath440 for some constants @xmath156 . note that @xmath441 for @xmath442 , introduce an integral operator @xmath443 acting on functions in @xmath444 as @xmath445 the integration path is given as follows . suppose first that @xmath446 . then the integration is over a straight line which does not intersect the segments joining @xmath447 to @xmath448 ( @xmath168 ) and such that the integrand decays exponentially at infinity along it . such a path exists as the integrand behaves as @xmath449 at infinity , which decays for @xmath36 in a cone . the resonance relation implies that the integrand has opposite residue at the pairs @xmath450 , so that , up to sign , the integral is independent of the choice of path . for @xmath451 general , the integral is defined by analytic continuation . the operator is defined up to multiplication by @xmath452 since one needs to choose a square root and an orientation of the path . we assume that @xmath453 this ensures that @xmath443 is well - defined . indeed , the integration kernel defining @xmath454 is regular at points obeying this condition by the discussion of [ ss - onefactor ] above . moreover this condition implies that in the integral over @xmath36 defining @xmath454 , the integration contour is not pinched between poles . proposition [ invert ] says that @xmath443 is an invertible linear map @xmath455 if @xmath456 introduce operators @xmath457 , @xmath458 where @xmath459 the @xmath29-independent terms in @xmath45 and @xmath460 were added to simplify the relations ( 1)-(4 ) below , which would otherwise only hold up to factors depending on @xmath382 . [ p - alphabeta ] the operators @xmath45 and @xmath460 are isomorphisms @xmath461 @xmath462 for all @xmath463 for which they are defined . _ proof : _ the statement for @xmath45 is easily checked inserting the definitions . the fact that @xmath460 respects the resonance relation is also easily checked . we are left to prove that @xmath26 obeys the bound if and only if @xmath464 does . let us prove the only if part in the case where @xmath465 has positive imaginary part . the other cases are proved in the same way . so we assume that @xmath466 . it follows using the relation @xmath467 that @xmath468 obeys the bound @xmath469 . the claim then follows from the identity @xmath470 @xmath89 theorems [ q - heat - n ] , [ t - mod ] , [ thm - fourier ] take the form of identities : 1 . the _ q - heat equation _ , @xmath471 holds on @xmath444 , if @xmath442 . the _ first modular equation _ , @xmath472 holds on @xmath473 if @xmath474 . the _ second modular equation , @xmath475 holds on @xmath444 if @xmath476 . the _ inversion relation _ , @xmath477 holds on @xmath478 if @xmath442 . notice that each of the operators @xmath454 in these identities is defined up to multiplication by @xmath452 , so the right hand side of each of the identities is equal to the left hand side up to multiplication by @xmath452 . extend the definition of the operator @xmath443 from the domain @xmath442 to the domain @xmath479 by the formulas @xmath480 [ t - hexagons ] 1 . for any @xmath433 such that @xmath479 and @xmath481 , the operator @xmath482 defines an invertible linear map @xmath483 . 2 . moreover , the q - heat equation , first and second modular equations , the inversion equation hold for @xmath484 such that @xmath485 and @xmath481 . the six - term relations of theorem [ t - hexagons ] are the commutativity of the following diagrams . the _ @xmath0-heat hexagon _ : @xmath486 the _ first modular hexagon _ : @xmath487 the _ second modular hexagon _ : @xmath488 the proof of theorem [ t - hexagons ] is done by reduction to the case where the imaginary parts of @xmath489 and @xmath465 are positive . in fact it is straightforward to see that in all cases the identities for general imaginary parts can be rewritten using the definitions as identities at some other values of @xmath490 where the imaginary parts are positive . we give here an interpretation of our relations in terms of projectively flat discrete @xmath1-connections . we start by introducing the notion of discrete connections . let @xmath491 be a @xmath381-space with a fixed presentation by generators @xmath492 , @xmath493 and relations @xmath494 , @xmath495 . a _ discrete @xmath381-connection _ on a complex vector bundle @xmath496 , with fibers @xmath497 , @xmath498 , assigns to each generator @xmath492 a collection of linear isomorphisms @xmath499 , @xmath498 . the _ parallel translation _ along an element @xmath500 of the free group @xmath501 generated by the @xmath492 s is the collection of maps @xmath502 uniquely defined by the properties @xmath503 , @xmath504 . here @xmath505 is the canonical projection @xmath506 . the _ curvature _ of a discrete @xmath381-connection is the collection of parallel translations @xmath507 along the relations . a connection is called _ projectively flat _ if @xmath508 for all @xmath509 and @xmath382 . let @xmath510 be a vector bundle over a subset @xmath511 of a @xmath381-space @xmath491 . then a _ discrete connection defined on @xmath511 _ assigns to each generator @xmath492 a collection of linear isomorphisms @xmath499 , for all those @xmath512 such that @xmath513 . the parallel translation @xmath514 is then defined on some subset @xmath515 . a discrete connection defined on @xmath511 is projectively flat if @xmath508 for all @xmath495 and @xmath516 . in our case , @xmath517 and for @xmath491 we take certain orbits in @xmath430 . the group @xmath518 is generated by the elementary matrices @xmath519 , ( @xmath520 ) with ones in the diagonal and at @xmath521 and zeros everywhere else . the relations can be chosen @xcite to be @xmath522 an @xmath1-orbit @xmath491 in @xmath430 is called _ regular _ if for all @xmath523 1 . @xmath524 and @xmath525 and 2 . @xmath481 . let @xmath491 be a regular orbit in @xmath430 . for @xmath526 , set @xmath527 for @xmath528 , set @xmath529 [ p - iguanodon ] let @xmath530 be a regular orbit . 1 . the assignment @xmath531 with @xmath532 defines a projectively flat @xmath1-connection defined on @xmath533 . the assignment @xmath534 with @xmath535 defines a projectively flat @xmath1-connection defined on @xmath536 . 3 . for all @xmath537 and all @xmath538 such that @xmath539 , we have @xmath540 _ proof : _ it first follows from theorem [ t - hexagons ] ( i ) and prop . [ p - alphabeta ] that @xmath541 is in all cases a well - defined isomorphism from @xmath542 to @xmath543 and similarly for @xmath214 . the other claims are then simple consequences of the q - heat , modular and inversion relations ( theorem [ t - hexagons ] ( ii ) ) , and the relations @xmath544 which follow from the fact that @xmath545 is 1-periodic in @xmath14 and @xmath23 . the easiest way to do the computations is to first check ( iii ) , which can be easily deduced from our three term relations . this identity implies that if the curvature associated to a relation is scalar for one of the connections @xmath546 or @xmath214 , then it is scalar ( and equal ) also for the other connection . then the curvature can be computed using @xmath546 or @xmath214 , whichever is simpler . for example , to compute the curvature @xmath547 associated to the relation @xmath548 , it is better to use the connection @xmath214 . one gets @xmath549 where @xmath550 @xmath89 in particular , if an orbit does not contain any point with @xmath551 or @xmath552 , which is true for generic orbits , we have @xmath553 and the connections are defined everywhere . the trouble is that we do not know if the spaces @xmath554 , @xmath555 are nontrivial for these orbits . by contrast , we have infinitely many examples of linearly independent functions in @xmath556 with @xmath552 or @xmath543 with @xmath551 . indeed , these spaces are isomorphic to spaces @xmath557 with @xmath392 , which contain the functions @xmath558 with any fixed @xmath36 , see prop . [ p - uine0 ] . these functions are non - zero for generic @xmath36 since they are the meromorphic kernel of an invertible integral operator . there are infinitely many linearly independent functions among them since they behave differently under shifts by 1 : @xmath559 . in this case we may construct a connection defined everywhere by gluing the two partially defined connections : [ t - pf ] let @xmath491 be a regular orbit containing a point @xmath560 with @xmath551 . let @xmath561 , @xmath562 be the subsets as above , on which the connections @xmath546 , @xmath214 are defined . we have @xmath563 . let @xmath564 be the vector bundle on @xmath491 obtained from @xmath555 and @xmath554 by identifying the fibers over @xmath565 via @xmath566 . let a connection @xmath567 be defined as @xmath541 , if both @xmath560 and @xmath568 are in @xmath569 , and as @xmath570 if both @xmath560 and @xmath568 are in @xmath571 . then @xmath572 is a well - defined projectively flat connection on @xmath491 . _ proof : _ since any two points @xmath573 related by a generator belong both to @xmath569 or both to @xmath571 , the connection @xmath574 is defined in all cases . moreover , by prop . [ p - iguanodon ] ( iii ) , @xmath572 is well - defined on @xmath564 , at least up to sign . it is easy to check that that the curvatures along the relations involve products of isomorphisms @xmath567 mapping between fibers at points @xmath560 which are all in @xmath569 or all in @xmath561 . thus the claim that @xmath572 is a projectively flat connection follows from the projective flatness of @xmath546 and @xmath214 . @xmath89 we give here a formula for the curvature of the discrete connection @xmath546 ( and @xmath214 ) . it is convenient to do a gauge transformation @xmath575 , with @xmath576 @xmath577 @xmath578 then the curvature of the connection @xmath579 has components @xmath580 , @xmath581 associated to the relations . by definition , they are given by @xmath582 @xmath583 @xmath584 where @xmath585 and @xmath586 we have @xmath587 @xmath588 and all other @xmath580 as well as @xmath581 are equal to 1 . in @xcite , a nontrivial 2-cocycle of @xmath1 with values in @xmath589)$ ] was obtained by a similar construction involving the elliptic gamma function instead of @xmath454 . in the language of discrete connections we use here , one defines an @xmath1 connection on the trivial line bundle over a dense subset of @xmath424 by setting @xmath590 for our purpose , @xmath591 may be considered here as a complex parameter on which @xmath1 acts trivially . the curvature of this connection was computed in @xcite . it has the form @xmath592 , @xmath593 , for some cubic polynomials @xmath594 in @xmath591 with coefficients in @xmath595 . the relation to the curvature of @xmath579 is @xmath596 the curvature of a projectively flat connection defines an extension of the group and thus a characteristic class , see @xcite . conjecturally , @xmath597 defines a nontrivial class in the group cohomology @xmath598 . this means ( if the conjecture is true ) that there is no gauge transformation given by exponentials of rational functions that can make the curvature trivial . we summarize some formulae about theta functions , gamma functions and phase functions , see @xcite for more details . jacobi s first theta function is defined by the series @xmath599 it is an entire holomorphic odd function such that @xmath600 and obeys the heat equation @xmath601 its transformation properties with respect to @xmath4 are described in terms of generators by the identities : @xmath602 we choose the square root in the right half plane . let @xmath603 with @xmath604 . the function @xmath605 is a solution of the functional equation @xmath606 let @xmath607 and @xmath608 . then @xmath609 we will also need the following variant of the theta function @xmath50 : @xmath610 this function obeys @xmath611 its modular properties are @xmath612 , and if @xmath613 , @xmath614 , @xmath615 here we consider two parameters @xmath14 and @xmath616 in the upper half plane , and set @xmath608 , @xmath617 , and consider the function of @xmath607 , @xmath618 it is a solution of the functional equations @xmath619 the _ elliptic gamma function _ @xcite , @xcite is @xmath620 it obeys the identities @xmath621 and is normalized by @xmath622 . the zeros of @xmath623 are at @xmath624 , @xmath625 . they are all simple . thus @xmath7 has only simple zeros and simple poles . the zeros are at @xmath626 and the poles are at @xmath627 . here @xmath628 run over nonnegative integers and @xmath629 over all integers . in fact , @xmath5 is , up to normalization , the unique 1-periodic meromorphic solution of @xmath630 holomorphic in the upper half plane . we consider the transformation properties of the elliptic gamma function under modular transformations of @xmath616 and @xmath14 . we have the identities @xmath631 since many operations we perform do not preserve the upper half plane , it is important to extend the range of values @xmath14 and @xmath616 can take . we set @xmath632 these formulae define an extension of the functions @xmath633 , @xmath623 to meromorphic functions on @xmath634 . it is clear that the functional relations @xmath635 still hold in this larger domain . correspondingly , we extend the definition of @xmath636 and the elliptic gamma function by using the same formulae in terms of the infinite products . we obtain : @xmath637 an easy check gives the following result : all identities for @xmath7 and @xmath636 of the preceding subsections continue to hold for all @xmath638 such that @xmath639 . however , the statements about the position of zeros and poles are no longer valid . we keep the notation of the previous subsection and introduce a new variable @xmath640 , and set @xmath641 . the phase function is @xmath642 we have @xmath643 the properties of this function follow from those of the gamma function : [ p-1 ] the function @xmath644 obeys the identities @xmath645 @xmath646 @xmath647 @xmath648 @xmath649 @xmath650 @xmath651 @xmath652 @xmath653 @xmath654 here we summarize some formulae giving the transformation properties of the @xmath31-matrix . fix @xmath655 and let @xmath656 be the @xmath31-matrix of @xmath17 associated to the evaluation verma modules @xmath657 , @xmath658 . we have @xmath659 and @xmath660 here @xmath661 with @xmath662.\ ] ] these formulae can be deduced from the functional realization of representations @xcite : the @xmath31-matrix may be defined as the unique linear map such that @xmath663 the weight function @xmath664 and the mirror weight function @xmath665 take values in the tensor product @xmath666 of evaluation verma modules . they are given explicitly in terms of ratios of theta functions , see @xcite . from the above formulae , we deduce the transformation properties of the qkzb operators . one finds @xmath667 where we set @xmath668 if @xmath669 and @xmath670 if @xmath671 . in this somewhat simplified notation , the operators @xmath672 depend on @xmath29 and are to be viewed as multiplication operator by these functions of @xmath29 ; @xmath673 taken at @xmath674 really means that we conjugate @xmath673 by the operator of dilation of @xmath29 by a factor @xmath675 . [ l - c1 ] 1 . for all @xmath14 in the upper half plane , there exists a constant @xmath676 such that for all @xmath13 , @xmath677 2 . for all @xmath297 and @xmath14 in the upper half plane , there exists a constant @xmath678 such that for all @xmath29 such that @xmath679 , @xmath680 aaaa i. cherednik , _ difference macdonald - mehta conjecture , _ q - alg/9702022 , internat . notices 1997 , no . 10 , 449467 p. etingof , a. kirillov jr , _ on cherednik - macdonald - mehta identities , _ q - alg/9712051 , electron . announc . math.soc . 4 ( 1998 ) , 4347 p. etingof , a. varchenko , _ traces of intertwiners for quantum groups and difference equations , i , _ math.qa/9907181 , duke math . j. 104 ( 2000 ) , no . 3 , 391432 g. felder , v. tarasov and a. varchenko , _ solutions of the elliptic qkzb equations and bethe ansatz , i _ , 180 ( 1997 ) , 4575 g. felder , v. tarasov and a. varchenko , _ monodromy of solutions of the elliptic knizhnik - zamolodchikov - bernard difference equations , _ q - alg/9705017 , internat . j. math . 10 ( 1999 ) , no . 8 , 943975 g. felder and a. varchenko , _ the q - deformed knizhnik zamolodchikov bernard equation _ , math/9809139 , commun . math . 221 ( 2001 ) , 549571 g. felder and a. varchenko , _ the elliptic gamma function and @xmath6 _ , adv . math . 156 ( 2000 ) , 4476 g. felder and a. varchenko , _ resonance relations for solutions of the elliptic qkzb equations , fusion rules , and eigenvectors of transfer matrices of restricted interaction - round - a - face models _ , contemp . math . 1 ( 1999 ) 3 , 335403 g. felder and a. varchenko , _ special functions , conformal blocks , bethe ansatz , and @xmath1 _ , math.qa/0101136 , phil . trans . r. soc . lond . a 359 ( 2001 ) , 13651373 j. milnor , _ introduction to algebraic @xmath691-theory _ , annals of mathematics studies , no . 72 , princeton university press , princeton , n.j . ; university of tokyo press , tokyo , 1971 s. n. m. ruijsenaars , _ first order analytic difference equations and integrable quantum systems _ , j. math . 38 ( 1997 ) , 10691146 v. tarasov and a. varchenko , _ geometry of @xmath0-hypergeometric functions as a bridge between yangians and quantum affine algebras _ , invent . 128 ( 1997 ) , no . 3 , 501588

we report the analysis of 22 b - band light curves of the dwarf nova v4140 sgr obtained with soi / soar during two nights along the decline of a superoutburst in 2006 sep 12 - 24 and in quiescence over 50 days following the superoutburst . three - dimensional eclipse mapping of the outburst light curves indicates that the accretion disc is elliptical ( eccentricity ) and that superhump maximum occurs when the mass donor star is aligned with the bulge of the elliptical disc . the accretion disc is geometrically thin both in outburst and in quiescence ; it fills the primary roche lobe in outburst and shrinks to about half this size in quiescence . the stability of the eclipse shape , width and depth along quiescence and the derived disc surface brightness distribution indicate that the quiescent accretion disc is in a high - viscosity , steady - state . flickering mapping of the quiescent data reveal that the low - frequency flickering arises from an azimuthally - extended stream - disc impact region at disc rim and from the innermost disc region , whereas the high - frequency flickering originates in the accretion disc . assuming the disc - related flickering to be caused by fluctuations in the energy dissipation rate induced by magneto - hydrodynamic turbulence , we find that the quiescent disc viscosity parameter is large at all radii . the high - viscosity quiescent disc and the inferred low disc temperatures in superoutburst are inconsistent with expectations of the disc - instability model , and lead to the conclusion that the outbursts of v4140 sgr are powered by mass transfer bursts from its donor star . [ firstpage ] accretion , accretion discs binaries : eclipsing novae , cataclysmic variables stars : dwarf novae stars : individual ( v4140 sagitarii )

1609.03453

in dwarf novae , mass is transferred from a late - type star ( the secondary ) to a companion white dwarf ( wd ) via an accretion disc . they show recurrent outbursts on timescales of days - months , in which the disc brightens by factors 20 - 100 for @xmath2 days . outbursts are explained in terms of either a thermal - viscous disc - instability model ( dim , e.g. , * ? ? ? * ; * ? ? ? * ) or a mass - transfer instability model ( mtim , e.g. , * ? ? ? dim predicts matter accumulates in a low viscosity , where @xmath3 is the non - dimensional viscosity parameter , @xmath4 is the local sound speed and @xmath5 is the disc scaleheight . ] disc during quiescence ( @xmath6 ) which switches to a high - viscosity regime during outbursts , whereas in mtim the disc viscosity is always the same ( @xmath7 , from the decline timescale of outbursting dwarf novae , e.g. , * ? ? ? therefore , estimating @xmath3 of a quiescent disc is key to gauge which model is at work in a given dwarf nova . aside of the normal outbursts , the short - period dwarf novae of the suuma type show longer , slightly brighter and more regular superoutbursts , during which a hump - shaped brightness modulation ( named _ superhump _ ) with period slightly longer than the orbital period @xmath8 is seen in their light curves @xcite . the most promising explanation for superhumps is given by the tidal resonance instability model @xcite : during a superoutburst , the accretion disc expands beyond the 3:1 resonance radius , @xmath9 , and a tidal instability sets in , giving rise to an elliptical , slowly precessing disc . superhump modulation then arises from the periodic tidal interaction between the outer elliptical disc and the mass - donor star ( normal superhumps ) or by the varying release of gravitational energy at the point where the gas stream hits the precessing elliptical disc outer edge ( late superhumps ) , at the beat period between the orbital and the disc precession period , @xmath10 . eclipse mapping @xcite is a powerful tool to test for the presence of elliptical discs in outbursting dwarf novae as well as to check whether the orientation of the ellipse is according to the expectation of the tidal instability model ( e.g. , * ? ? ? * ; * ? ? ? flickering is the intrinsic brightness fluctuation of 0.01 - 1 mag on timescales of seconds to dozens of minutes seen in dwarf novae light curves . optical studies ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) suggest there may be three different sources of flickering in dwarf novae and novalike systems , the relative importance of which varies from system to system : ( i ) the stream - disc impact region ( possibly because of unsteady mass transfer or post - shock turbulence , * ? ? ? * ; * ? ? ? * ) , ( ii ) the innermost disc regions around the wd [ ] ej82,bruch92 , and ( iii ) the accretion disc itself ( probably as a consequence of magneto - hydrodynamic ( mhd ) turbulence or events of magnetic reconnection at the disc atmosphere , * ? ? ? * ; * ? ? ? * ; * ? ? ? the power density spectrum ( pds ) of the flickering is characterized by a continuum power - law , @xmath11 , with @xmath12 @xcite , which flattens below a given cut - off frequency @xmath13 . currently , the most promising explanation for the anomalously large viscosity of accretion discs is related to mhd turbulence in the differentially rotating disc gas ( driven by the magnetorotational instability , mri , see * ? ? ? * ; * ? ? ? most of the studies on this subject over the last two decades focused on confirming that mri leads to self - sustained , turbulent and efficient outward flow of angular momentum and inward flow of disc matter ( e.g. , * ? ? ? * ; * ? ? ? * and references therein ) , and on the comparison of the numerically derived values of @xmath3 with those inferred by the decline timescale of outbursting dwarf novae ( e.g. , * ? ? ? * ; * ? ? ? on the other hand , the study of @xcite focused on the time variability of the viscous energy release in a mhd turbulent disc . they found that mhd turbulence leads to large fluctuations in the energy dissipation rate per unit area at the disc surface , @xmath14 , which they suggested could be a source of flickering in mass - exchanging binaries . indeed , the pds of the fluctuations in their study resemble those of flickering sources , with a power - law dependency of similar index range and a flat slope at low - frequencies . the interpretation of the fluctuations in @xmath14 as the stochastic and statistically independent release of energy from a large number of turbulent eddies leads to a direct relation between the relative amplitude of the energy fluctuations @xmath15 and the disc viscosity parameter , @xmath16 , providing an interesting observational way to estimate the local accretion disc viscosity parameter ( see , e.g. , * ? ? ? * ) . for a thin accretion disc ( @xmath17 ) , the above relation predicts that low - viscosity discs ( @xmath18 ) should show low - amplitude disc flickering ( hardly detectable at a level @xmath19 per cent ) , whereas high - viscosity discs ( @xmath20 0.1 - 1 ) should display detectable flickering with relative amplitudes in the range 2.5 - 7.5 per cent . there is observational support for this prediction : e.g. , the accretion disc seems the dominant source of flickering in the high - viscosity discs of the novalike systems rwtri @xcite and uxuma @xcite , whereas there is no evidence of disc - related flickering in the low - viscosity accretion discs of the dwarf novae ugem @xcite and ippeg @xcite in quiescence . v4140 sgr is an 88-min period eclipsing su uma type dwarf nova @xcite showing low - amplitude ( @xmath21 mag ) , 5 - 10d long outbursts recurring every 80 - 90d and longer , brighter superoutbursts where superhumps appear in its light curve @xcite . here we report the analysis of a sample of light curves of v4140 sgr with eclipse mapping techniques to trace the evolution of the surface brightness of its accretion disc during decline from a superoutbursts , to locate the sources of flickering in the binary and to estimate the radial run of the quiescent disc viscosity parameter . section[observa ] reports the observations and data reduction procedures . data analysis and results are presented in section[analysis ] , discussed in section[discuss ] , and summarized in section[conclusions ] . time series of b - band ccd photometry of v4140 sgr were obtained along 2006 with the soar optical imager ( soi ) at the 4.1 m soar telescope , in cerro pachn , chile . the soi camera has a mini - mosaic of two back illuminated e2v @xmath22 pixels ccds covering a 5.26 arcminute square field of view . the soi detectors are mounted with their long sides parallel and spaced 102 pixels apart , resulting in a @xmath23 gap between the individual ccd images . all observations were performed in its @xmath24 pixels binning and fast - readout mode ( 6.1s total readout time ) at a spatial scale of @xmath25 . the observations are summarized in table [ tab1 ] . column 3 lists the number of points in the light curve ( @xmath26 ) , column 4 gives the exposure time in seconds ( @xmath27 ) . column 5 lists the eclipse cycle number ( e ) ; observations that do not cover the eclipse are indicated in parenthesis , and those with incomplete eclipse phase coverage are indicated by a colon after the cycle . column 7 gives an estimate of the quality of each run . the observations comprise 22 light curves obtained with the same instrument and telescope , ensuring a high degree of uniformity to the data set . ccrccccc date ( ut ) & hjd start & @xmath26 & @xmath28 & e ) . ] & phase & night & seeing + [ -0.5ex ] ( 2006 ) & ( 2.450.000 + ) & & ( s ) & ( cycle ) & range & quality & ( @xmath29 ) + sep 12 & 3991.53678 & 528 & 3.7 & 125833 & @xmath30 & a & 0.8 - 2.0 + `` & 3991.58268 & 446 & 3.7 & 125834 & @xmath30 & b + '' & 3991.64405 & 467 & 3.7 & 125835 & @xmath31 & b + sep 24 & 4003.60079 & 205 & 3.7 & ( 126029 ) & @xmath32 & a & 0.7 - 1.0 + `` & 4003.62291 & 548 & 3.7 & 126030 & @xmath30 & b + '' & 4003.68432 & 385 & 3.7 & 126031 & @xmath33 & b + sep 27 & 4006.50008 & 77 & 3.7 & ( 126076 ) & @xmath34 & c & 0.7 - 0.9 + `` & 4006.51006 & 547 & 3.7 & 126077 & @xmath30 & b + '' & 4006.57151 & 557 & 3.7 & 126078 & @xmath30 & b + `` & 4006.64293 & 548 & 3.7 & 126079 & @xmath30 & b + '' & 4006.69433 & 270 & 3.7 & 126080 : & @xmath35 & b + sep 29 & 4008.49251 & 428 & 3.7 & 126109 & @xmath36 & a & 0.7 - 0.9 + oct 01 & 4010.54703 & 95 & 5.0 & ( 126142 ) & @xmath37 & b & 0.8 - 1.1 + `` & 4010.56445 & 442 & 5.0 & 126143 & @xmath30 & b + '' & 4010.62589 & 207 & 5.0 & 126144 : & @xmath38 & b + oct 21 & 4030.48861 & 359 & 4.0 & 126467 & @xmath39 & a & 0.7 - 0.8 + nov 14 & 4054.52175 & 202 & 4.0 & ( 126858 ) & @xmath40 & a & 0.8 - 1.0 + `` & 4054.54806 & 499 & 4.0 & 126859 & @xmath41 & a + nov 15 & 4055.51412 & 131 & 4.0 & ( 126874 ) & @xmath42 & a & 0.8 - 1.1 + '' & 4055.53092 & 405 & 4.0 & 126875 & @xmath43 & a + nov 16 & 4056.51947 & 370 & 4.0 & 126891 & @xmath44 & a & 0.9 - 1.2 + nov 17 & 4057.51611 & 359 & 4.0 & 126907 & @xmath45 & b & 1.0 - 1.6 + + [ -8ex ] data reduction procedures included bias subtraction , flat - field correction , cosmic rays removal and aperture photometry extraction . time series were constructed by computing the magnitude difference between the variable and a bright reference comparison star @xmath46 s of the variable ( star c1 in the finding chart of * ? ? ? * ) with scripts based on the aperture photometry routines of the apphot / iraf package . light curves of other comparison stars in the field were also computed in order to check the quality of the night and the internal consistency and stability of the photometry over the time span of the observations . the magnitude and colors of the reference star were tied to the johnsons - cousins bvri system @xcite from observations of this star and of standard stars @xcite made on a photometric night . the reference star has a calibrated magnitude of @xmath47 mag . we used the relations of @xcite to transform its @xmath48-band magnitude to flux units and to convert the light curves of the variable star from magnitude difference to absolute flux with an estimated photometric accuracy of 5 per cent . moreover , the analysis of the light curves of field comparison stars of brightness comparable to that of the variable indicates that the internal error of the photometry is less than 2 per cent . the error in the photometry of the variable is derived from the photon count noise and is transformed to flux units using the same relation applied to the data . the individual light curves have typical signal - to - noise ratios of @xmath49 out - of - eclipse and @xmath50 - 20 at mid - eclipse . fig.[fig1 ] shows the 2006 observations of v4140 sgr . it was caught in outburst in sep 12 ( @xmath51mag brighter than in quiescence ) , while the sep 24 observations frame the decline from that outburst . the object was back to its quiescent state from sep 27 onwards . given the typical shape and length of the v4140 sgr outbursts @xcite , the sep 12 data probably correspond to the outburst plateau phase ( i.e. , past outburst maximum ) . as previously noted by @xcite , the eclipses in outburst are shallower than in quiescence , indicating that the brightness increase is mainly from the outer and partially eclipsed disc regions . the presence of superhumps in the outburst light curves ( sect.[super ] ) indicates that this was a superoutburst . the light curves were phase - folded according to the linear plus sinusoidal ephemeris of baptista et al ( 2003 ) , @xmath52 \ , d \ , . \label{efem}\end{aligned}\ ] ] where @xmath53 is the cycle number and @xmath54 gives the wd mid - eclipse times . we checked whether this ephemeris correctly centred the wd eclipse by median combining all quiescent data , computing the derivative of the combined light curve , and measuring the phases of minimum / maximum in the derivative on the assumption that they indicate the mid - ingress / egress phases of the wd eclipse . this leads to a wd eclipse width of @xmath55 cycle , in good agreement with the value derived by @xcite . we also find a wd mid - eclipse phase of @xmath56 cycle , indicating that the observed eclipses occurred 20.5s after the prediction of the ephemeris of eq.([efem ] ) . all light curves were then offset by @xmath57 cycle to make the wd eclipse centre coincident with phase zero . fig.[fig2 ] shows phase - folded light curves of v4140 sgr grouped per brightness state . the median quiescent light curve displays a steep , narrow eclipse with a long egress shoulder and a weak orbital hump with an apparent maximum at phase @xmath58 cycle , signaling a small contribution from the bright spot at disc rim to the total flux . the superoutburst light curves show strong flickering ( at @xmath59 per cent level ) and a pronounced , broad superhump centred at phase zero . in the decline light curves , flickering amplitude is reduced ( in absolute and relative terms ) and the superhump maximum is displaced to phase @xmath60 cycle . flares and flickers occurring right before ingress and just after egress of the wd suggest that a significant fraction of the flickering in superoutburst and decline arises from the inner disc regions . unfortunately , the small number of light curves covering superoutburst and decline prevents us from applying flickering mapping techniques ( see sect.[quies ] ) to derive the spatial distribution of the flickering sources at these brightness states . the total width of the eclipses decreases with decreasing brightness level suggesting that the accretion disc shrinks along the decline from superoutburst , being smallest in quiescence . furthermore , the depth of the eclipses increases with decreasing brightness signaling that the relative contribution of the outer disc regions to the total flux decreases along the decline and is also smallest in quiescence . was the 2006 sep outburst a normal one or a superoutburst ? a straightforward way to address this questions is by searching for superhumps in the outburst and decline light curves . in order to detect non - orbital periodicities , we removed the eclipses ( phase range @xmath61 to @xmath62 ) from the light curves and computed lomb - scargle periodgrams @xcite separately for the sep 12 and sep 24 data . the power spectrum of the latter data set shows a periodicity at frequency @xmath63 , corresponding to a period of @xmath64 d , where the quoted uncertainty is derived from the width at half power of the observed peak . the same periodicity is found in the sep 12 periodgram , although the corresponding power spectrum peak is less pronounced in this case . this period is consistently longer than the orbital period , and is the same in both nights within the uncertainties . moreover , the maximum of the observed modulation coincides with phase zero on sep 12 and has shifted to phase @xmath60 cycle on sep 24 ( fig.[fig2 ] ) , indicating that it is not the consequence of anisotropic emission from a bright spot at disc rim , and confirming that @xmath65 is different from the orbital period . also , the length of the 2006 sep outburst is at least 13 days ( it probably started before sep 12 ) , which is clearly longer than the typical 5 - 10d duration of the normal outbursts of this dwarf nova @xcite . these evidences give us confidence that the observed modulations in sep 12 and sep 24 light curves are indeed superhumps and , therefore , that these observations framed a superoutburst of v4140 sgr . as a consequence of the short time span of the observations ( only 2 - 3 binary orbits in each date ) , the uncertainty in @xmath65 is too large to allow a meaningful derivation of a superhump period excess from @xmath65 . we attempted to increase the accuracy of the measurement of @xmath65 by computing a power spectrum of the combined light curve of both nights . however , no periodicity other than harmonics of the orbital period appears in this case , suggesting a significant change in phase of the superhump signal along this time interval as is usual in superoutbursting dwarf novae ( e.g. , * ? ? ? we grouped the data per outburst stage to generate phase - binned , median - averaged light curves for the outburst ( sep 12 data ) , decline ( sep 24 data ) and quiescence ( the remaining data sets ) brightness states . the resulting light curves have phase resolutions of 0.003 cycle during eclipse ( phase range @xmath61 to @xmath62 cycle ) and of 0.009 cycle outside of eclipse . only orbital phases in the range @xmath66 to @xmath67 cycle are included because our eclipse mapping algorithm does not handle a possible eclipse of the secondary star by the accretion disc ( at phase @xmath68 ) . three - dimensional ( 3d ) maximum entropy eclipse mapping techniques were applied to these light curves in order to investigate the source of the superhump and to follow the evolution of the disc surface brightness distribution along the superoutburst decline . the reader is referred to @xcite for the historical presentation , @xcite for a description and performance tests of the 3d eclipse mapping algorithm , and @xcite for a more recent review on eclipse mapping techniques . our 3d eclipse map consists of a grid of @xmath69 pixels on a conical surface with side @xmath70 ( where @xmath71 is the distance from disc centre to the inner lagrangian point ) centred at the wd position and inclined at a half - opening angle @xmath72 with respect to the orbital plane , plus a circular rim of 102 pixels orthogonal to the orbital plane at a distance @xmath73 ( @xmath74 ) from the disc centre . the disc rim allows us to model out - of - eclipse modulations ( such as anisotropic emission from the bright spot at stream - disc impact ) as the fore - shortening of an azimuthally - dependent brightness distribution in the disc rim pixels @xcite . the eclipse geometry is defined by the mass ratio @xmath75 and the inclination @xmath76 , and the scale of the map is set by @xmath71 . we adopted @xmath77 , @xmath78 and @xmath79 @xcite , which correspond to a wd eclipse width of @xmath80 cycle @xcite . this combination of parameters ensures that the wd is at the centre of the map . the light curves were analyzed by the 3d eclipse mapping algorithm to solve for a map of the disc + rim surface brightness distribution . the reconstructions were performed with a polar gaussian default function @xcite with radial blur width @xmath81 and azimuthal blur width @xmath82 , and reached a final reduced chi - square @xmath83 for all light curves . the uncertainties in the eclipse maps were derived from monte carlo simulations with the light curves using the bootstrap technique @xcite , generating a set of 20 randomized eclipse maps ( see * ? ? ? * ) . these are combined to produce a map of the standard deviations with respect to the true map . a map of the statistical significance ( or the inverse of the relative error ) is obtained by dividing the true eclipse map by the map of the standard deviations @xcite . the uncertainties obtained with this procedure are used to draw the contour maps of fig.[fig4 ] , and to estimate the uncertainties in the derived disc rim intensity ( fig.[fig4 ] ) and radial brightness temperature distributions ( fig.[fig5 ] ) . an entropy landscape technique was used to derive the best - fit disc rim radius @xmath73 and half - opening angle @xmath72 for each brightness state ( see appendix ) . we obtained reconstructions for disc rims in the range @xmath84 at steps of @xmath85 , and half - opening angles in the range @xmath86 at steps of @xmath87 , and investigated the resulting space of parameters in search of the combination @xmath88 that yields the eclipse map of highest entropy . the space of parameters is well behaved with a single , well defined entropy maximum in each case . @xmath73 is best constrained in outburst , while @xmath72 is better constrained in quiescence : the entropy of the eclipse maps drops quickly for @xmath89 in outburst , and for @xmath90 in quiescence . the accretion disc fills the primary roche lobe at outburst ( @xmath91 ) . it shrinks to @xmath92 during decline and reaches @xmath93 in quiescence . in each case . ] these results are in line with the qualitative inferences drawn in sect.[lcurve ] and in agreement with the @xmath94 quiescent disc radius found by @xcite . there is marginal evidence that the disc half - opening angle is larger in outburst ( @xmath95 ) than in quiescence ( @xmath96 ) . the disc is geometrically thin in all three cases . in order to check the consistency of the inferred @xmath73 values , we applied the method described by @xcite to estimate the disc radius from the half - width of the eclipse @xmath97 . by assuming a spherical secondary star it is possible to derive the disc radius @xmath73 in units of the orbital separation @xmath98 from the analytical expression , @xmath99 where @xmath100 is the radius of a sphere containing the same volume as the roche lobe of the secondary star , given by the relation @xcite , @xmath101 values of @xmath102 are transformed to @xmath103 assuming @xmath104 @xcite . measurements of @xmath97 were obtained by fitting a spline function to the out - of - eclipse phases of each light curve , finding the phases where the fitted spline deviates from the data by 1@xmath105 of the local , out - of - eclipse median flux , and dividing the resulting phase range by two . we find @xmath97 values of @xmath106 cycle , @xmath107 cycle and @xmath108 cycle , respectively for the outburst , decline and quiescence light curves . these measurements lead to corresponding disc radii of @xmath109 , @xmath110 , and @xmath111 , in very good agreement with the values derived from the entropy landscape procedure . the measured ingress / egress eclipse phases are depicted as pairs of vertical tick marks in the left - hand panels of fig.[fig4 ] . we used the expression of @xcite to compute a circularization radius ( the smallest possible radius for an inviscid accretion disc ) of @xmath112 . thus , the measured quiescent disc radius of v4140 sgr is larger than its circularization radius at a @xmath113 confidence level . eclipse maps corresponding to the best - fit ( @xmath114 ) values are shown in fig.[fig4 ] , together with the data and model light curves . the orientation of the eclipse maps is such that the secondary star rotates counter - clockwise around the wd at the centre of the map ( in the observer s reference frame ) , while the observer rotates clockwise ( in the binary reference frame ) . at phase zero , the secondary and the observer are on the right side of the map ( with the projected shadow of the secondary star covering most of the disc ) . at phase @xmath115 cycle the observer is at the bottom ( in the binary frame ) while the secondary star is at the top of the eclipse map ( in the observer s frame ) . the inner disc regions remain roughly at the same brightness level through outburst , while significant brightness changes take place in the outer disc regions . in order to emphasize these changes and the brightness asymmetries in the outer disc regions , the inner disc regions of the eclipse maps ( @xmath116 ) were masked in the middle - right panels of fig.[fig4 ] . at outburst , the surface brightness distribution of the outer disc is skewed towards the l1 point indicating that the disc is elliptical at this stage . two dotted circles are overplotted on the eclipse map to help visualize the elliptical disc brightness distribution . the outer circle corresponds to the disc rim @xmath117 , while the inner circle has @xmath118 . from these radii we estimate a disc eccentricity @xmath119 . for our outburst dataset the phase of superhump maximum coincides with mid - eclipse , right when the bulge of the elliptical disc is aligned with the secondary star . although less conspicuous , a similar brightness asymmetry is also present in the decline eclipse map . in this case the major axis of the ellipse is oriented at an angle @xmath120 with respect to the line joining both stars , implying that the alignment of the secondary star with its bulge occurs at binary phase @xmath121 cycle . this is in good agreement with the observed phase of superhump maximum in the decline light curve , @xmath122 cycle . from this eclipse map we find @xmath123 and @xmath124 , also leading to a disc eccentricity @xmath0 . these results are in line with the tidal resonance instability model , which explains superhumps in terms of enhanced tidal dissipation ( and extra disc emission ) as the secondary star passes closer to the bulge of a slowly precessing elliptical disc @xcite . for v4140 sgr , the 3:1 resonance radius is @xmath125 . at outburst , the disc is comfortably larger than @xmath9 ( @xmath126 ) , whereas in the decline dataset the disc radius is slightly smaller than @xmath9 ( @xmath127 ) . the measured disc eccentricities are in good agreement with the @xmath128 value found in numerical simulations of the tidal resonance instability by @xcite . since the timescale of disc precession ( @xmath129 2 - 3 days , see table3.3 of * ? ? ? * ) is much longer than @xmath8 , the change in orientation of the elliptical disc along the time span of the outburst and decline datasets ( @xmath130 h ) is @xmath131 , smaller than the azimuthal smearing effect adopted for the maximum entropy reconstructions ( @xmath82 ) . therefore , there is no significant azimuthal ` blurring ' of the outburst and decline eclipse maps caused by the slow precession of the elliptical disc . the right - hand panels of fig.[fig4 ] show the brightness distributions along the disc rim . a word of caution is on demand before we attempt to interpret these distributions . one of the basic assumptions of the eclipse mapping method is that the intensities in the eclipse map are independent of binary phase , and that all changes in the light curve are caused by variable visibility and/or aspect of pixels with phase . superhumps clearly violate this assumption if they correspond to true , physical changes in intensity of accretion disc regions with binary phase . therefore , the disc rim intensity distributions in outburst and decline should not be interpreted as if there is enhanced emission from extended regions at disc rim . instead , they just signal the azimuths ( or binary phases ) where enhanced emission from the superhump light source occurs in each light curve . thus , the disc rim distributions tell us that superhump emission is produced over a large fraction of the orbital period and is centreed at phase @xmath132 in outburst and decline , respectively ( of course , this can also be inferred by direct inspection of the corresponding light curves ) . however , an additional useful information is that , aside of the elliptical disc bulge ( responsible for the disc rim emission at azimuth @xmath133 ) , there are two other superhump light sources in outburst , at azimuths @xmath134 when the secondary star is roughly orthogonal to the major axis of the elliptical disc . a similar inference can be drawn from the decline disc rim distribution , although in this case the two sources are closer in azimuth than in outburst and the disc bulge contribution seems considerably reduced or blended in azimuth with that from the two side sources . these results are reminiscent to those of @xcite . he applied a modified eclipse mapping technique to investigate the superhump light source in z cha when superhump maximum is centreed in eclipse , and found that the extra emission arises from three sources in the outer disc regions , one towards the l1 point and two sources symmetrically located in the orthogonal directions . these orientations coincide with the predicted positions where the elliptical orbits intersect @xcite . let us turn our attention to the quiescence light curve . its eclipse map is dominated by emission from an extended asymmetric source with no distinguished contribution from the wd ( i.e. , no sharp ingress / egress features in eclipse shape ) , indicating that efficient accretion through a high - viscosity disc is taking place in quiescence . the long egress shoulder in the eclipse shape translates into enhanced emission along the outer disc regions ahead of the stream - disc impact point , suggesting inefficient downstream cooling in the disc flow @xcite . on the other hand , the weak orbital hump maps into a disc rim narrow spot ( @xmath135 ) at zero azimuth . the apparent hump phase of maximum @xmath136 cycle is an illusion , as the spot starts being eclipsed together with the accretion disc soon after that phase and is behind the shadow of the secondary star when seen face on . the azimuth of maximum emission of this spot ( @xmath137 ) does not coincide with the radial direction of the stream - disc impact point ( @xmath138 ) , but with the direction of the infalling gas stream at that point . this is in contrast with results from the dwarf novae ugem @xcite , oycar @xcite and ippeg @xcite , where the maximum emission of the quiescent bright spot lies in the direction normal to a plane roughly midway between the disc and stream flows . in the dim framework , a dwarf nova accretion disc cycles between a hot , high - viscosity state where hydrogen is fully ionized ( outburst ) and a faint , low - viscosity state where hydrogen is mostly neutral ( quiescence ) @xcite . the instability that sets this limit - cycle behavior is tied to the partial ionization of hydrogen which leads to tight constraints in disc gas temperature along the outburst cycle . namely , there is a critical effective temperature @xmath139 above which the disc gas should stay while in outburst , and a critical effective temperature @xmath140 below which it should remain while in quiescence @xcite , @xmath141 @xmath142 decline from outburst is understood in terms of the inwards propagation of a cooling front from the outer disc regions , progressively transitioning the disc from its hot state back to the cool quiescent state . disc regions ahead of the cooling front should still have @xmath143 , while those behind the cooling front should show @xmath144 . this prediction can be tested with eclipse mapping . if the distance to the binary is known , the usual practice is to convert the intensities in the eclipse maps to blackbody brightness temperatures @xmath145 , which can then be compared to @xmath146 and to the @xmath147 law of opaque , steady - state disc models . a valid criticism about this procedure is that @xmath145 may not be a proper estimate of the disc effective temperature @xmath148 . as pointed out by @xcite , a relation between these two quantities is non - trivial , and can only be properly obtained by constructing self - consistent models of the vertical structure of the disc . on the other hand , observations indicate that the colors of optically thick disc regions are in between those of blackbody and main - sequence stars @xcite in which cases a broad - band , blackbody brightness temperature becomes a fair and useful approximation to @xmath148 . @xcite applied a method similar to cluster main - sequence fitting to an accretion disc color - magnitude diagram of v4140sgr to infer a distance of @xmath149pc . their eclipse maps lead to radial brightness temperatures in outburst and quiescence which are consistent at the 1@xmath105 limit in the @xmath48 , @xmath150 and @xmath151 passbands , suggesting that the accretion disc is indeed optically thick and that @xmath152 is a reasonable approximation in this case . fig.[fig5 ] shows azimuthally - averaged radial brightness temperature distributions for the outburst ( solid squares ) , decline ( crosses ) and quiescence ( open squares ) accretion disc maps for assumed distances of 170pc ( bottom ) and 260pc ( top , the upper 3@xmath105 limit on the distance estimate of * ? ? ? the intensities in the eclipse maps were scaled to the assumed distances and corrected for interstellar extinction effects before being converted to brightness temperatures . from the galactic interstellar contour maps of @xcite , we estimate a reddening of @xmath153 towards the direction of v4140sgr . . dashed lines show the 1@xmath105 limits on the average temperatures . dotted lines correspond to steady - state optically thick disc models for mass accretion rates of @xmath154 , respectively from top to bottom . dot - dashed lines show the critical effective temperatures @xmath139 ( above which the disc gas should stay while in outburst ) and @xmath140 ( below which it should remain while in quiescence ) , according to the dim ( e.g. , * ? ? ? the distributions are shown for assumed distances of 260pc ( top ) and 170pc ( bottom ) . ] the temperatures in the inner disc regions ( @xmath155 ) are consistently the same throughout outburst . the outer disc regions ( @xmath156 ) progressively cool from outburst to quiescence . at a distance of 170pc the whole outburst cycle proceeds at temperatures below @xmath139 . increasing the distance to the binary increases the intrinsic intensities and corresponding disc brightness temperatures . but even at a distance of 260pc most disc regions ahead of the inwards moving cooling wave show temperatures below @xmath139 . in order to reconcile the observations with dim , one needs to increase the distance above 300pc . however , this steepens the radial temperature distribution in quiescence , bringing it into good agreement with the @xmath157 law of high - viscosity , opaque steady - state discs in contradiction to the dim expectation that quiescent disc of dwarf novae should have low viscosities ( @xmath158 ) , long viscous timescales and , therefore , would hardly have time to reach a steady - state in between consecutive outbursts . furthermore , this also increases the disc temperatures in quiescence , bringing an increasing ( and significant ) fraction of the quiescent disc into contradiction with the dim requirement @xmath159 . in other words , the disc temperatures do not differ sufficiently in outburst and quiescence that one can have both @xmath159 everywhere in quiescence and @xmath160 over most of the disc in outburst . it seems clear that _ it is not possible to reconcile the outbursts of v4140 sgr with dim_. figure[fig6 ] shows the individual quiescent light curves of v4140 sgr superimposed in phase . along the @xmath161d time interval covered by the quiescence observations , v4140 sgr showed no systematic change or long - term trend in eclipse depth , out - of - eclipse brightness or eclipse width . therefore , the light curves of that period can be assigned to a single brightness state , and all observed variability can be attributed to flickering . the top panel depicts light curves of a comparison star of similar brightness . the scatter with respect to the mean flux is perceptibly larger in v4140sgr than in the comparison star and is caused by flickering . the scatter is larger close to orbital hump maximum ( suggesting that the bright spot contributes to the flickering ) and is smaller during eclipse ( indicating that flickering sources are occulted at these phases ) . we applied the `` single '' @xcite and `` ensemble '' @xcite techniques to the quiescent light curves of v4140sgr to derive the orbital dependency of its steady - light and of the low- and high - frequency flickering components . the reader is referred to @xcite for a detailed description and combined application of both techniques . while the ensemble technique samples flickering at all frequencies , the typical power law dependency of the flickering in cvs implies that an ensemble curve is dominated by the low - frequency flickering components . on the other hand , the high - pass filtering process of the single technique leads to curves which sample the high - frequency flickering . the combination of both methods allows one to separate the low- ( ensemble ) and high - frequency ( single ) components of the flickering sources . because all light curves are at the same brightness level , applying the ensemble technique becomes a matter of computing the mean flux at each phase bin ( the steady - light curve ) and the standard deviation with respect to the mean ( the scatter curve , @xmath162 , with contributions from the poisson noise , @xmath163 , and from the flickering , @xmath164 ) . the angular coefficient ( which measures the long - term brightness changes ) is consistently zero within the uncertainties . the ensemble flickering curve is obtained by correcting the scatter curve from the poisson noise contribution , @xmath165 the average steady - light curve was subtracted from each individual light curve to remove the dc component , and a lomb - scargle periodgram @xcite was calculated . the periodgrams of all light curves were combined to yield a mean periodgram and a standard deviation with respect to the mean . figure[fig7 ] shows the resulting average power density spectrum ( pds ) binned to a resolution of 0.02 units in log(frequency ) . limits on the average power . the best - fit power law @xmath166 is shown as a dashed line . a vertical tick marks the low - frequency cut - off of the filtering process applied to derive the single scatter curve . the keplerian frequencies corresponding to @xmath167 and to the quiescent disc radius @xmath73 are illustrated as vertical dashed ticks . ] the pds is well described by a power law @xmath166 , which becomes flat for @xmath168mhz ( @xmath169 minutes ) and disappears in the white noise for @xmath170 mhz ( @xmath171s ) . the slope of the pds distribution is consistent with the general trend found in other cvs , which can be well described by power laws with average exponent @xmath172 @xcite . we followed @xcite and subtracted the average steady - light curve from each individual light curve before applying the single filtering process in order to eliminate the steep gradients produced by the eclipse in the light curve . our single light curve includes flickering components with frequencies @xmath173mhz ( timescales shorter than 500s ) . single curves obtained with higher cutoff frequencies show the same morphology but are noisier ( because of the reduced power ) and lead to less reliable results , while for those obtained with lower cutoff frequencies the excess of filtering starts smearing out the eclipse of the flickering . the single flickering curve is obtained by correcting the corresponding scatter curve from the poisson noise contribution ( cf . eq.[eq - flick ] ) . as the disc opening angle in quiescence is negligible ( @xmath174 , sect.[discrim ] ) , the steady - light and the flickering curves were analyzed by a standard eclipse mapping algorithm @xcite , with a flat cartesian grid of @xmath69 pixels centreed on the wd and side @xmath70 in the orbital plane , to solve for a map of the disc surface brightness distribution in each case . because this version of the eclipse mapping method does not take into account out - of - eclipse brightness changes , these were removed from the light curves by fitting a spline function to the phases outside eclipse , dividing the light curve by the fitted spline , and scaling the result to the spline function value at phase zero . the adopted eclipse geometry ( @xmath175 ) and default map parameters ( @xmath176 ) are the same as in sect.[erup ] , and the reconstructions also reached a final reduced chi - square @xmath83 in all cases . monte carlo simulations with the bootstrap technique were similarly performed to obtain the uncertainties in the eclipse maps . these are used to draw the contour maps of fig.[fig8 ] , and to estimate the uncertainties in the radial distributions shown in fig.[fig9 ] . the resulting steady - light , ensemble and single curves have phase resolution of 0.003cycle and cover the phase range ( @xmath177 ) cycle . these light curves and corresponding eclipse maps are shown in fig.[fig8 ] . the steady - light curve is the quiescence light curve of sect.[erup ] after removal of the weak orbital hump . consistently , the steady - light eclipse map is very similar to the quiescence eclipse map ( fig.[fig4 ] ) , showing a broad brightness distribution of an asymmetric accretion disc which extends up to @xmath178 . this brightness distribution was already discussed in sect.[evol ] . it is worth noting that , although derived from essentially the same light curve , the steady - light map is of higher statistical significance than its cousin quiescence map . this reflects a subtle but relevant difference between the standard and the 3d eclipse mapping . standard eclipse mapping is simpler and more robust to noise in the light curve ( e.g. , flickering ) as only a relatively narrow phase range of the light curve covering the eclipse shape is used to derive the disc brightness distribution . because 3d eclipse mapping uses the whole orbital light curve to derive the disc + rim brightness distributions , it is more sensitive to noise in the light curve and any flare / flicker outside of eclipse contributes to the uncertainty of all visible pixels at that phase . nevertheless , the intrinsic lower statistical significance of the 3d eclipse mapping seems a reasonable price to pay in return for the extra information about the disc radius and half - opening angle delivered by this technique . on the other hand , the higher statistical significance with which maps are obtained with the standard eclipse mapping is key to allow statistically meaningful results in the derivation of the radial distribution of the flickering amplitude ( sect.[flick ] ) . the ensemble and single flickering curves have similar shapes , showing the deeper eclipse of a central source plus an extended egress shoulder . this eclipse shape maps into an asymmetric source along the disc rim ahead of the stream - disc impact point ( @xmath179 , stream - disc impact flickering ) and an extended central source several times larger in radius than the wd at disc centre ( boundary layer + disc - related flickering ) . the solid contour line overplotted on each eclipse map depicts the @xmath113 confidence level region as derived from the map of statistical significance in each case . the two flickering sources are statistically significant at or above the @xmath113 confidence level both in the ensemble and in the single map . the disc - related flickering component is not centred at the wd position , but is slightly skewed towards the ballistic stream trajectory . we tested whether this was an artifact caused by an error in the choice of phase zero by obtaining eclipse maps for versions of the light curves with small offsets in phase and by allowing changes in the adopted eclipse geometry within the uncertainties in the binary parameters @xmath75 and @xmath76 . these simulations indicate that it is not possible to simultaneously centre the steady - light brightness distribution and the disc flickering component , and we conclude that the slight offset of the disc flickering component is real . azimuthally - averaged radial intensity distributions for the steady - light , ensemble and single flickering maps are shown in the right - hand panels of fig.[fig8 ] . each eclipse map is separated in radial bins of width @xmath180 and the median intensity is computed for each bin . the associated uncertainty is obtained by computing the median absolute deviation with respect to the median intensity for the set of bootstrap reconstructions of the corresponding map and multiplying the result by 1.48 to find the standard deviation ( e.g. , * ? ? ? fig.[fig9]a compares the relative amplitude of the ensemble and single flickering in v4140 sgr . the radial run of the relative amplitudes are obtained by dividing the radial intensity distribution of these two flickering maps by that of the steady - light . dashed lines show the @xmath181 limits on the average flickering amplitudes . the distributions are not reliable for @xmath182 because of the reduced statistical significance of the flickering maps and the rapidly declining intensities in the steady - light map , and are not shown in fig.[fig9]a . limits on the average values . ( b ) the disc relative thickness ( @xmath183 ) derived from the @xmath184 distribution assuming @xmath185 ( mtim case ) for distances of 170pc ( black ) and 260pc ( green ) . dashed lines show the @xmath181 limits on the @xmath183 distribution derived from the uncertainties in the @xmath184 distribution . the dotted line corresponds to the @xmath183 distribution assuming @xmath186 ( dim case ) . horizontal dot - dashed lines depict the @xmath183 value derived from the measured disc half - opening angle ( sect.[discrim ] ) and its @xmath181 upper limit . ( c ) the disc viscosity parameter @xmath3 derived using the gertseema & achterberg ( 1992 ) model . the solid line corresponds to the h / r distribution for the mtim case ( @xmath185 ) ; dashed lines show the @xmath181 limits on the derived @xmath3 value . a dotted line depicts the corresponding distribution for the dim case ( @xmath186 ) . ] the difference between the ensemble and single distributions tells us where the low - frequency flickering comes from . an excess of ensemble over single flickering at @xmath187 indicate the existence of a low - frequency flickering central source at the @xmath188 confidence level . the radial extent of this source is too large to be accounted for by the radial smearing effect of the entropy on the eclipse map . ] ( see * ? ? ? * ) , indicating that the source is larger than the wd by a factor @xmath189 . this central flickering component may possibly arise from optical reprocessing of intermitent x - ray and uv irradiation by the boundary layer at the atmosphere of the innermost disc regions . additional excess of ensemble over single flickering is seen in the outer disc regions ( @xmath190 , at the @xmath113 confidence level ) , which is readily associated to the stream - disc impact region . this stream - disc flickering component is relatively stronger ( @xmath191 per cent amplitude ) than the central source flickering ( @xmath130 per cent amplitude ) . there is no low - frequency flickering produced in the intermediate disc regions ( @xmath192 ) . the high - frequency flickering is spread over the accretion disc , with a @xmath193 per cent amplitude roughly independent of disc radius . this is the disc - related flickering component . the slight increase in high - frequency flickering amplitude at the innermost and outer disc regions might be due to some high - frequency contribution from the central and stream - disc flickering components . our ensemble and single flickering maps allow us to disentangle three different sources of flickering in v4140sgr . the presence of a central flickering and of a stream - disc impact flickering sources are in good agreement with the scenario put forward by @xcite . if the central flickering component arises in the boundary layer its spectrum may be much bluer than the inner disc spectrum , and our optical flickering mapping experiment is underestimating its contribution to the total flickering . in this case , this boundary layer flickering should dominate the uv light curve of v4140sgr , with fast reduction in flickering activity as the wd goes into eclipse . our flickering mapping experiment also reveals the existence of a disc flickering component in v4140sgr , which allow us to probe the viscosity in its quiescent disc ( see sect.[discuss ] ) . this disc flickering component accounts for 2/3 of the total optical flickering in v4140sgr . the results bear some resemblance with those from the quiescent dwarf nova v2051oph , where the low - frequency flickering is associated with an overflowing gas stream while the high - frequency flickering arises from the accretion disc @xcite . the identification of a disc - related flickering component enables us to follow the steps of @xcite and to use the model of @xcite to estimate the magnitude of the viscosity parameter @xmath3 for the quiescent disc of v4140sgr . this allows us to critically test the largely different predictions of dim and mtim for a quiescent disc of a dwarf nova . as the expression relating @xmath3 and @xmath15 involves the disc relative thickness @xmath183 and because this quantity is loosely constrained by the entropy landscape analysis , we start our exercise by using the measured radial temperature distribution for an independent ( and more precise ) measurement of @xmath183 as well as to check the consistency of the thin disc result of sect.[discrim ] . in the atmosphere of an opaque , @xmath3-disc model , @xmath194 where @xmath195 is the mid - plane disc temperature , @xmath196 is the square root of the mid - plane to the effective temperature ratio and the final step is justified by the discussion in sect.[radtemp ] . an estimate of the @xmath196 value is obtained by comparing the analytical fits of the critical effective temperatures ( @xmath197 ) and of @xmath195 ( see , e.g. , eqs . a5-a8 in * ? ? ? * ) ; for the upper turning point ( @xmath198 ) , @xmath199 , while for the lower turning point ( @xmath158 ) , @xmath200 . @xcite computed detailed equilibrium vertical atmospheres of @xmath3-discs subject to mhd turbulence for a set of physical conditions of interest to dwarf nova accretion discs . their model for a high - viscosity ( @xmath201 ) , hot and opaque disc with @xmath202 ( representative of the range of disc temperatures found in v4140sgr ) consistently has @xmath203 ( or @xmath185 ) . thus , we adopt @xmath185 for the case of a high - viscosity quiescent disc ( as expected in the mtim framework ) and we take @xmath186 as corresponding to a low - viscosity quiescent disc ( as expected in the dim framework ) . we may then use the inferred @xmath184 distribution and the thin disc approximation to estimate the radial run of the relative disc thickness @xmath183 @xcite , @xmath204^{1/2 } , \label{eq - thick}\ ] ] where @xmath205 and @xmath206 are , respectively , the boltzmann and the gravitation constants , @xmath207 is the mean molecular weight , @xmath208 is the hydrogen atom mass , and we adopt @xmath209 ( adequate for a fully ionized cosmic misture of gases ) , @xmath210 @xcite . the disc relative thickness obtained with eq.([eq - thick ] ) assuming @xmath185 ( mtim case ) is plotted in fig.[fig9]b for distances of 170pc ( black ) and 260pc ( green ) . an @xmath183 distribution computed assuming @xmath186 ( dim case ) and distance of 170pc is shown as a dotted line in fig.[fig9]b . the uncertainties in the @xmath183 distribution are dominated by the uncertainty in the temperature scaling factor @xmath196 . for a given @xmath148 value , a quiescent dim disc is thinner than its mtim counterpart because its mid - plane temperature is lower and , therefore , @xmath4 and @xmath5 are also lower . the fact that the @xmath183 distributions increase with radius imply that the disc face is concave and , therefore , there is room for reprocessing of x - ray and uv radiation from the boundary layer in the disc atmosphere ( e.g. , * ? ? ? * see sect.[flick ] ) . the disc relative thickness derived from the disc half - opening angle ( @xmath211 ) is shown as horizontal dot - dashed lines for comparison . disc relative thicknesses derived independently from the disc temperatures and from the disc half - opening angle are consistent with each other at the @xmath181 level , and indicate that the quiescent disc of v4140sgr is geometrically thin . once we have the radial runs of the relative amplitude of the disc flickering and of the disc thickness , we may use the mhd turbulence model of @xcite to estimate the radial run of the viscosity parameter in the quiescent disc of v4140sgr . in their model , the number of turbulent eddies that contribute to the local fluctuation is , @xmath212 where @xmath213 is the size of the largest turbulent eddy . the local rms value of the fluctuations @xmath214 in the average energy dissipation rate per unit area @xmath215 is found to be , @xmath216 while the disc viscosity parameter can be written as , @xmath217 where @xmath218 is the local average shear stress ( maxwell + reynolds ) , @xmath219 is the local pressure , and @xmath220 is the local ( turbulent ) disc viscosity . combining eqs.([eq - ga1]-[eq - ga3 ] ) , we obtain the radial run of the disc viscosity parameter , @xmath221^{-1 } \left [ \frac{\sigma_\mathrm{d}(r)}{0.05\,\langle d(r ) \rangle } \right]^2 \ , . \label{eq - flick - ampl}\ ] ] if the disc - related flickering is caused by fluctuations of the energy dissipation rate induced by mhd turbulence in an optically thick disc , the relative amplitude of this disc flickering component yields a good estimate of the ratio @xmath15 . the resulting @xmath222 distributions are shown in fig.[fig9]c . because of the dependency @xmath223 , the influence of errors in distance and temperature on @xmath3 is negligible . errors in @xmath196 also have a relatively small effect to the uncertainty of @xmath3 in comparison with the large contribution of the errors in the flickering relative amplitude curves ( because of the dependency @xmath224 ^ 2 $ ] and the relatively lower accuracy of the latter quantity ) . large values of @xmath1 at found all disc radii , in good agreement with the mtim expectations . in order to model the long ( 80 - 90d ) time interval between outbursts of v4140sgr in the dim framework , an @xmath225 is required @xcite . this corresponds to the lower y - axis end of fig.[fig9]c . the derived @xmath3 s are systematically larger than the dim expectation by at least an order of magnitude . outside the region @xmath226 , the discrepancy is significant at the @xmath227 limit . computing @xmath222 from the @xmath183 distribution of the dim case ( @xmath186 , shown as a dotted line in fig.[fig9]c ) only increases the discrepancy , as the thinner dim disc leads to even larger @xmath3 values . if the mhd turbulence model of @xcite is correct , the above results indicate that the quiescent accretion disc of v4140sgr has high - viscosity . the v4140sgr quiescent data allows an additional test of dim , which is independent of any mhd turbulence model . one of the achilles heel of dim is the prediction ( unmatched by observations ) that the average brightness of a dwarf nova should increase between successive outbursts as matter piles up in a low - viscosity quiescent disc , steadily increasing the disc surface density , temperature and , therefore , brightness ( e.g. , * ? ? ? in an attempt to overcome the discrepancy , @xcite proposed that the unobserved increase in disc brightness along quiescence could be compensated for by the gradual cooling of a small , hot and high - viscosity inner disc region . their two - dimensional time - dependent numerical models of dwarf nova accretion discs seem to support this proposal . could the observed high - viscosity quiescent disc of v4140 sgr be the gradually cooling hot inner disc region proposed by @xcite ? in an eclipsing dwarf nova , the gradual cooling of a hot inner disc region combined with the simultaneous increase in brightness of the outer regions plus the gradual contraction of a low - viscosity , non - steady quiescent disc should make its eclipse progressively narrower ( because of disc shrinkage ) and shallower ( because of the reduced brightness contrast between inner and outer disc ) . there is no observational support for any of these effects in our v4140 sgr quiescent data . the shape , width and depth of the quiescent eclipse is consistently the same within the uncertainties along the @xmath161d time interval covered by our quiescence observations . this observational window corresponds to more than 50 per cent of the typical v4140 sgr outburst recurrence time @xcite and started right after the end of a superoutburst . the stability of our observations not only do not support the prediction of @xcite , but they also indicate that the quiescent accretion disc of v4140 sgr is in a steady - state regime . again , this is in excellent agreement with the picture of a high - viscosity quiescent accretion disc . if the quiescent accretion disc of v4140 sgr is already in the high - viscosity regime , there is no room for disc instabilities to set in and it is not possible to explain its outbursts in terms of dim . moreover , it is not possible to reproduce its short ( 5 - 10d ) outburst length in comparison with the relatively long time interval between sucessive outburst ( 80 - 90d ) in the limit - cycle of the dim framework since there is no further room to increase @xmath3 by an order of magnitude from quiescence to outburst . in combination with the observed outburst disc temperatures ( below those required by dim , @xmath228 , sect.[radtemp ] ) , these results strengthen the conclusion that the outbursts of v4140 sgr are driven by bursts of mass transfer and not by disc instabilities ( c.f . v4140 sgr is not the only dwarf nova showing observational evidence for outbursts of mass transfer , but the younger member of an increasing list including v2051 oph @xcite , ex dra and ht cas @xcite , v513 cas and iw and @xcite , and possibly ex hya @xcite . we applied 3d eclipse mapping techniques to follow the evolution of the surface brightness of the accretion disc of v4140 sgr in a superoutburst . we find that the disc is elliptical in outburst and decline , with an eccentricity @xmath0 . in both outburst stages , the disc orientation is such that superhump maximum occurs when the secondary star is aligned with the bulge of the elliptical disc . this lends observational support for the tidal resonance instability model of superhumps . the accretion disc fills the primary roche lobe at outburst ( @xmath126 ) , shrinks to @xmath229 during decline and reaches a lower value of @xmath178 in quiescence . there is marginal evidence that the disc half - opening angle is larger in outburst ( @xmath95 ) than in quiescence ( @xmath96 ) , but the disc is geometrically thin in all three cases . the superoutburst occurs at disc temperatures too low to be accounted for by the disc - instability model even at the upper @xmath230 limit on the inferred distance to the binary . the stability of the eclipse shape , width and depth along 50d in the quiescent period following the superoutburst and the derived disc surface brightness distribution ( outshining any contribution from the white dwarf ) indicate that the quiescent accretion disc of v4140 sgr is in a high - viscosity , steady - state regime . flickering mapping of the quiescent data reveal three different sources of flickering in v4140 sgr : an azimuthally - extended stream - disc impact region at disc rim ( statistically significant at the @xmath113 level ) and the innermost disc region ( statistically significant at the @xmath188 level ) , responsible for the low - frequency flickering , and an extended , high - frequency disc component responsible for 2/3 of the total optical flickering in v4140sgr . assuming that the disc - related flickering is caused by fluctuations in the energy dissipation rate induced by mhd turbulence according to the model of @xcite , we find that the quiescent disc viscosity parameter in v4140 sgr is large ( @xmath231 ) at all disc radii , in agreement with mtim predictions and in marked contrast with dim predictions . the discrepancy between the inferred @xmath3 values and dim predictions is statistically significant at the @xmath227 confidence level outside the region @xmath226 . the high - viscosity , steady - state quiescent disc of v4140 sgr and the inferred low disc temperatures in superoutburst are inconsistent with expectations of the disc - instability model , and lead to the conclusion that the outbursts of v4140 sgr are powered by the only other mechanism considered so far , namely , bursts of enhanced mass transfer rate from its donor star . the dominant source of uncertainty in the computed @xmath222 distribution are the errors in the relative amplitude of the high - frequency , disc - related flickering component . these errors can be reduced by increasing the @xmath232 of the flickering curves , i.e. , by increasing the statistics and/or the @xmath232 of the individual light curves included in the data sample . the benefits of this observational effort are double : it will allow a more meaningful and statistically robust estimate of the magnitude of the disc viscosity parameter and it will open the possibility to properly probe changes in @xmath3 with radius . we thank an anonymous referee for useful comments and suggestions which helped to improve the presentation of our results . rb acknowledges financial support from cnpq / brazil through grant 308946/2011 - 1 . aso acknowledges fapesp for financial support under grant 03/12618 - 7 . here we report simulations aimed to show the performance of our 3d eclipse mapping code in reproducing accretion disc brightness distributions with out - of - eclipse modulations . we also show how the entropy of the eclipse map can be used to constrain the disc half - opening angle @xmath72 and the disc rim radius @xmath73 . we illustrate the performance of the 3d eclipse mapping code with reconstructions from a synthetic eclipse map comprised of a steady - state ( @xmath157 ) , axi - symmetric disc brightness distribution plus a bright spot `` painted '' in the disc rim at an azimuth of @xmath233 ( azimuths are counted counter - clockwise from the line joining both stars ) . this model distribution is shown in fig.[figsim3](a ) . . the lower inset shows the intensity distribution along the disc rim . azimuths are counted from the line joining both stars and increase counter - clockwise . ( b ) a reconstruction using the correct geometry , @xmath234 . ( c ) a reconstruction with correct @xmath73 but underestimated @xmath235 . ( d ) the same as in ( c ) for an overestimated @xmath236 . ( e ) a reconstruction with correct @xmath72 but underestimated @xmath237 . ( f ) the same as in ( e ) for an overestimated @xmath238 . ] it was convolved with the eclipse geometry of v4140 sgr ( @xmath239 ) and pairs of values ( @xmath240 ) to generate synthetic light curves to which gaussian noise was added in order to simulate real data . each of these light curves were then subjected to the 3d eclipse mapping code . if an eclipse mapping reconstruction is performed with the wrong choice of @xmath72 and @xmath73 , the code will develop artifacts in the brightness distribution in order to compensate for the incorrect parameters . this is illustrated in fig.[figsim3 ] . a synthetic light curve with added gaussian noise ( @xmath49 ) was generated from the brightness distribution of fig.[figsim3](a ) and reconstructions were obtained for different combinations of @xmath72 and @xmath73 . fig.[figsim3](b ) shows a reconstruction with the correct geometry @xmath241 . the disc surface brightness distribution and the azimuth of the spot at disc rim are well recovered . the azimuthal blur of the spot at disc rim is intrinsic to the eclipse mapping method and is controlled by the choice of @xmath242 ( see sect.[erup ] ) . figs.[figsim3](c ) and ( d ) show the effects of errors in the choice of @xmath72 . underestimating ( overestimating ) the disc half - opening angle forces the code to artificially increase the brightness of the disc hemisphere farther away from ( closest to ) the l1 point to compensate for their lower effective area during eclipse . eclipse maps with these spurious , additional structures have lower entropy than the map with the correct choice of @xmath72 and @xmath73 . measures the amount of structure in an eclipse map . the more structured an eclipse map is , the lower its entropy is . the map of highest entropy @xmath243 is the one with the least amount of structures allowed by the data . ] figs.[figsim3](e ) and ( f ) show the effects of errors in the choice of @xmath73 . again , artifacts develop in the disc brightness distribution in order to compensate for the wrong radial position of the disc rim . in both cases , the additional asymmetry leads to eclipse maps of lower entropy . because of these artifacts , reconstructions obtained with a wrong choice of ( @xmath244 ) have lower entropy than that with the correct combination of parameters . therefore , it is possible to use the entropy of the eclipse map as a tool to gauge the correct choice of ( @xmath240 ) . in order to do this , we obtain reconstructions covering the space of parameters ( @xmath240 ) with a suitable sampling and we search for the combination which leads to the reconstruction of highest entropy ( @xmath245 ) . this is called ` entropy landscape ' , a technique that has been successfully used to find best - fit parameters with other image reconstruction methods ( e.g. , * ? ? ? fig.[figsim2 ] illustrates the procedure . we generated an input synthetic light curve ( @xmath49 ) from the brightness distribution of fig.[figsim3](a ) for a given combination of parameters ( @xmath246 ) and obtained reconstructions of that light curve for a range of ( @xmath244 ) values . the space of parameters is covered by sampling @xmath72 in the range @xmath86 at steps of @xmath87 , and by sampling @xmath73 in the range @xmath84 at steps of @xmath247 . the left - hand panels show the entropy of the reconstruction as a function of @xmath72 for light curves obtained with three different @xmath248 values and fixed @xmath249 , whereas the right - hand panels show the entropy of the reconstruction as a function of @xmath73 for light curves obtained with three different @xmath250 values and fixed @xmath251 . it is possible to recover the correct value of @xmath72 with an accuracy of @xmath252 , and the correct value of @xmath73 with an accuracy of @xmath253 . these results are confirmed with a series of simulations with several different brightness distributions and light curves with @xmath232 comparable to those of the v4140 sgr data .

we prove the knneth formula in floer ( co)homology for manifolds with restricted contact type boundary . we use viterbo s definition of floer homology , involving the symplectic completion by adding a positive cone over the boundary . the knneth formula implies the vanishing of floer ( co)homology for subcritical stein manifolds . other applications include the weinstein conjecture in certain product manifolds , obstructions to exact lagrangian embeddings , existence of holomorphic curves with lagrangian boundary condition , as well as symplectic capacities . department of mathematics , ethz , rmistrasse 101 , 8092 zrich ( ch ) . email : [email protected]

math0403376

the present paper is concerned with the floer homology groups @xmath0 of a compact symplectic manifold @xmath1 with contact type boundary , as well as with their cohomological dual analogues @xmath2 . the latter were defined by viterbo in @xcite and are invariants that take into account the topology of the underlying manifold _ and _ , through an algebraic limit process , all closed characteristics on @xmath3 . their definition is closely related to the symplectic homology groups of floer , hofer , cieliebak and wysocki @xcite . throughout this paper we will assume that @xmath4 is exact , and in particular @xmath5 . this last condition will be referred to as _ symplectic asphericity_. the groups @xmath0 are invariant with respect to deformations of the symplectic form @xmath4 that preserve the contact type character of the boundary and the condition @xmath6 . the groups @xmath0 actually depend only on the _ symplectic completion _ @xmath7 of @xmath8 . the manifold @xmath7 is obtained by gluing a positive cone along the boundary @xmath3 and carries a symplectic form @xmath9 which is canonically determined by @xmath4 and the conformal vector field on @xmath8 . we shall often write @xmath10 instead of @xmath0 . the grading on @xmath10 is given by minus the conley - zehnder index modulo @xmath11 , with @xmath12 the minimal chern number of @xmath8 . there exist canonical maps @xmath13 @xmath14 which shift the grading by @xmath15 . * theorem a ( knneth formula ) . * _ let @xmath16 and @xmath17 be compact symplectic manifolds with restricted contact type boundary . denote the minimal chern numbers of @xmath8 , @xmath18 and @xmath19 by @xmath20 , @xmath21 and @xmath22 respectively . _ 1 . for any ring @xmath23 of coefficients there exists a short exact sequence which splits noncanonically @xmath24 & fh_k(m \times n , \ , \omega \oplus \sigma ) \ar@{->>}[d ] \\ & { \bigoplus_{\widehat r+\widehat s = k-1 } } { \text{\rm tor}}_1^a \big ( fh_{\widehat r}(m , \ , \omega ) , \ fh_{\widehat s}(n , \ , \sigma ) \big ) } } \ ] ] + the morphism @xmath25 induces a morphism of exact sequences whose source is the knneth exact sequence of the product @xmath26 and whose target is ( [ suite exacte kunneth floer ] ) . 2 . for any field @xmath27 of coefficients there is an isomorphism @xmath28^{\qquad \sim } & fh^k(m \times n , \ , \omega \oplus \sigma ) \ , } } \ ] ] + the morphism @xmath29 induces a commutative diagram with respect to the knneth isomorphism in cohomology for @xmath30 . in the above notation we have @xmath31 , @xmath32 , @xmath33 and the @xmath34 symbol associates to an integer its class in the corresponding @xmath35 ring . the reader can consult ( * ? ? ? * vi.12.16 ) for a construction of the knneth exact sequence in singular homology . the algebraic properties of the map @xmath29 strongly influence the symplectic topology of the underlying manifold . our applications are based on the following theorem , which summarizes part of the results in @xcite . * theorem ( viterbo @xcite ) . * _ let @xmath36 be a manifold with contact type boundary such that @xmath37 . assume the map @xmath38 is not surjective . then the following hold . _ 1 . the same is true for any hypersurface of restricted contact type @xmath39 bounding a compact region ; 2 . any hypersurface of contact type @xmath39 bounding a compact region carries a closed characteristic ( weinstein conjecture ) ; 3 . there is no exact lagrangian embedding @xmath40 ( here @xmath8 is assumed to be exact by definition ) ; 4 . for any lagrangian embedding @xmath40 there is a loop on @xmath41 which is contractible in @xmath8 , has strictly positive area and whose maslov number is at most equal to @xmath42 ; 5 . for any lagrangian embedding @xmath40 and any compatible almost complex structure @xmath43 there is a nonconstant @xmath43-holomorphic curve @xmath44 ( of unknown genus ) with non - empty boundary @xmath45 . viterbo @xcite introduces the following definition , whose interest is obvious in the light of the above theorem . a symplectic manifold @xmath16 which verifies@xmath46 is said to satisfy the strong algebraic weinstein conjecture ( sawc ) if the composed morphism below is not surjective @xmath47 we shall still denote the composed morphism by @xmath29 . theorem a now implies that the property of satisfying the sawc is stable under products , with all the geometric consequences listed above . * theorem b. * _ let @xmath1 be a symplectic manifold with restricted contact type boundary satisfying the sawc . let @xmath48 be an arbitrary symplectic manifold with restricted contact type boundary . the product @xmath49 satisfies the sawc and assertions ( a ) to ( e ) in the above theorem of viterbo hold . in particular , the weinstein conjecture holds and there is no exact lagrangian embedding in @xmath50 . _ the previous result can be applied for subcritical stein manifolds of finite type . these are complex manifolds @xmath51 which admit proper and bounded from below plurisubharmonic morse functions with only a finite number of critical points , all of index strictly less than @xmath52 @xcite . they satisfy the sawc as proved by viterbo @xcite . cieliebak @xcite has proved that their floer homology actually vanishes . we can recover this through theorem a by using another of his results @xcite , namely that every such manifold is stein deformation equivalent to a split one @xmath53 . this can be seen as an extension of the classical vanishing result @xmath54 , @xmath55 @xcite . * theorem c ( cieliebak @xcite ) . * _ let @xmath7 be a subcritical stein manifold of finite type . its floer homology vanishes _ @xmath56 the paper is organized as follows . in section [ constructions ] we state the relevant definitions and explain the main properties of the invariant @xmath57 . section [ la preuve de kunneth ] contains the proof of theorem a. the proofs of theorems b and c , together with other applications , are gathered in section [ appli ] . let us point out where the difficulty lies in the proof of theorem a. floer homology is defined on closed manifolds for any hamiltonian satisfying some generic nondegeneracy condition , and this condition is _ stable _ under sums @xmath58 on products @xmath59 . this trivially implies ( with field coefficients ) a knneth formula of the type @xmath60 . on the other hand , floer homology for manifolds with contact type boundary is defined using hamiltonians with a rigid behaviour at infinity and involves an algebraic limit construction . this class of hamiltonians is _ not stable _ under the sum operation @xmath58 on @xmath61 . one may still define floer homology groups @xmath62 , but the resulting homology might well be different , in the limit , from @xmath63 . the whole point of the proof is to show that this is not the case . this paper is the first of a series studying the floer homology of symplectic fibrations with contact type boundary . it treats trivial fibrations with open fiber and base . a spectral sequence of leray - serre type for symplectic fibrations with closed base and open fiber is constructed in @xcite . _ acknowledgements . _ this work is part of my ph.d . thesis , which i completed under the guidance of claude viterbo . without his inspired support this could not have come to being . i am grateful to yasha eliashberg , dietmar salamon , paul seidel , jean - claude sikorav and ivan smith for their help and suggestions . i also thank the referee for pointing out errors in the initial proofs of theorem c and proposition [ prop : symplectic capacities ] . during the various stages of preparation of this work i was supported by the following institutions : laboratoire de mathmatiques , universit paris sud ; centre de mathmatiques de lcole polytechnique ; cole normale suprieure de lyon ; departement mathematik , eth . floer homology has been first defined by a. floer for closed manifolds in a series of papers @xcite which proved arnold s conjecture for a large class of symplectic manifolds including _ symplectically aspherical _ ones . in this situation floer s construction can be summarized as follows . consider a periodic time - dependent hamiltonian @xmath64 with hamiltonian vector field @xmath65 defined by @xmath66 . the associated _ action functional _ @xmath68 is defined on the space of smooth contractible loops @xmath69 the critical points of @xmath70 are precisely the @xmath71-periodic solutions of @xmath72 , and we denote the corresponding set by @xmath73 . we suppose that the elements of @xmath73 are nondegenerate i.e. the time one return map has no eigenvalue equal to @xmath71 . each such periodic orbit @xmath74 has a @xmath75-valued conley - zehnder index @xmath76 ( see @xcite for the definition ) , where @xmath77 is the _ minimal chern number_. the latter is defined by @xmath78 and by the convention @xmath79 if @xmath80 . let also @xmath43 be a compatible almost complex structure . the homological floer complex is defined as @xmath81 @xmath82 @xmath83 here @xmath84 denotes the space of trajectories for the negative @xmath85 gradient of @xmath70 with respect to the metric @xmath86 , running from @xmath74 to @xmath87 : @xmath88 the additive group @xmath89 acts on @xmath90 by translations in the @xmath91 variable , while the symbol @xmath92 stands for an algebraic count of the elements of @xmath93 . we note that it is possible to choose any coefficient ring instead of @xmath94 once the sign assignment procedure is available . the crucial statements of the theory are listed below . the main point is that the symplectic asphericity condition prevents the loss of compactness for floer trajectories by bubbling - off of nonconstant @xmath43-holomorphic spheres : the latter simply can not exist . 1 . under the nondegeneracy assumption on the @xmath71-periodic orbits of @xmath95 and for a generic choice of @xmath43 , the space @xmath96 is a manifold of dimension @xmath97 . if this difference is equal to @xmath71 , the space @xmath98 consists of a finite number of points . moreover , there is a consistent choice of signs for these points with respect to which @xmath99 ; 2 . the floer homology groups @xmath100 are independent of @xmath95 and @xmath43 . more precisely , for any two pairs @xmath101 and @xmath102 there is a generic choice of a smooth homotopy @xmath103 , @xmath104 with @xmath105 , @xmath106 , @xmath107 , @xmath108 defining a map @xmath109 @xmath110 the notation @xmath111 stands for the space of solutions of the equation @xmath112 which run from @xmath74 to @xmath87 . the map @xmath113 induces an isomorphism in cohomology and this isomorphism is independent of the choice of the homotopy . we shall call it in the sequel the `` continuation morphism '' . if @xmath95 is time independent , morse and sufficiently small in some @xmath114 norm , then the @xmath71-periodic orbits of @xmath115 are the critical points of @xmath95 and the morse and conley - zehnder indices satisfy @xmath116 , @xmath117 . here @xmath118 is the conley - zehnder computed with respect to the trivial filling disc . moreover , the floer trajectories running between points with index difference equal to @xmath71 are independent of the @xmath119 variable and the floer complex is equal , modulo a shift in the grading , with the morse complex of @xmath95 . + we infer that for any regular pair @xmath120 we have @xmath121 * remark . * an analogous construction yields a _ cohomological complex _ by considering in ( [ the differential ] ) the space of trajectories @xmath122 . we explain now how the above ideas can be adapted in order to construct a symplectic invariant for manifolds with contact type boundary . we follow @xcite , but similar constructions can be found in @xcite ( see @xcite for a survey ) . a symplectic manifold @xmath1 is said to have a contact type boundary if there is a vector field @xmath123 defined in a neighbourhood of @xmath3 , pointing outwards and transverse to @xmath3 , which satisfies @xmath124 we say that @xmath8 has restricted contact type boundary if @xmath123 is defined globally . we call @xmath123 and @xmath125 the _ liouville vector field _ and the _ liouville form _ respectively , with @xmath126 . we define the reeb vector field @xmath127 as the generator of @xmath128 normalized by @xmath129 . an integral curve of @xmath127 is called a _ characteristic_. we have @xmath130 and this implies that a neighbourhood of @xmath3 is foliated by the hypersurfaces @xmath131 , whose characteristic dynamics are conjugate . the floer homology groups @xmath0 of a manifold with contact type boundary are an invariant that takes into account : * the dynamics on the boundary by `` counting '' characteristics of arbitrary period ; * the interior topology of @xmath8 by `` counting '' interior @xmath71-periodic orbits of hamiltonians . in order to understand its definition , let us explain how one can `` see '' characteristics of arbitrary period by using @xmath71-periodic orbits of special hamiltonians . there is a symplectic diffeomorphism onto a neighbourhood @xmath132 of the boundary @xmath133 , \ , d(s\lambda| ) \big ) \longrightarrow ( \mc v , \ , \omega ) , \quad \delta > 0 { \text{\rm small , } } \ ] ] @xmath134 where @xmath135 denotes the restriction of @xmath136 to @xmath137 ( we actually have @xmath138 ) . we define the _ symplectic completion _ @xmath139 consider now hamiltonians @xmath95 such that @xmath140 for @xmath141 , where @xmath142 \longrightarrow { \mathbb{r}}$ ] is smooth . it is straightforward to see that @xmath143 the @xmath71-periodic orbits of @xmath115 that are located on the level @xmath44 correspond to characteristics on @xmath3 having period @xmath144 under the parameterization given by @xmath145 . the general principle that can be extracted out of this computation is that _ one `` sees '' more and more characteristics as the variation of @xmath146 is bigger and bigger_. we define a hamiltonian to be _ admissible _ if it satisfies @xmath147 on @xmath8 and it is of the form @xmath148 for @xmath44 big enough . it is clear that the a priori @xmath149 bounds require no additional argument than the one in @xcite . the situation is of course different if one enlarges further the admissible class . ] , with @xmath146 convex increasing and such that there exists @xmath150 with @xmath151 constant for @xmath152 . we call such a hamiltonian _ linear at infinity_. moreover , we assume that the slope at infinity of @xmath146 is not the area of a closed characteristic on @xmath3 and that all @xmath71-periodic orbits of @xmath95 are nondegenerate . one method to obtain such hamiltonians is to slightly perturb functions @xmath153 , where @xmath154 is equal to zero in a neighbourhood of @xmath155 , strictly convex on @xmath156 and linear at infinity with slope different from the area of any closed characteristic , by a perturbation localized around the periodic orbits . we point out that there are admissible hamiltonians having arbitrarily large values of the slope at infinity . the _ admissible almost complex structures _ are defined to be those which satisfy the following conditions for _ large enough values of @xmath44 _ : @xmath157 where @xmath158 is an almost complex structure compatible with the restriction of @xmath4 to the contact distribution @xmath159 on @xmath3 . these are precisely the almost complex structures which are invariant under homotheties @xmath160 , @xmath161 for large enough values of @xmath44 . the crucial fact is that the function @xmath162 is plurisubharmonic with respect to this class of almost complex structures . this means that @xmath163 for any nonzero @xmath164 , @xmath165 , @xmath166 big enough , and indeed we have @xmath167 . plurisubharmonicity implies that for any @xmath43-holomorphic curve @xmath168 , @xmath169 big enough one has @xmath170 . in particular the maximum of @xmath171 is achieved on the boundary @xmath172 ( see for example @xcite , theorem 3.1 ) . a similar argument applies to solutions of the floer equation ( [ fl eq ] ) , as well as to solutions of the parameterized floer equation ( [ par fl eq ] ) for _ increasing homotopies _ satisfying @xmath173 @xcite . this implies that solutions are contained in an a priori determined compact set . all compactness results in floer s theory therefore carry over to this situation and so does the construction outlined for closed manifolds . we introduce a partial order on regular pairs @xmath174 as @xmath175 the continuation morphisms ( [ cont morph ] ) form a direct system with respect to this order and we define the floer homology groups as @xmath176 an important refinement of the definition consists in using a truncation by the values of the action . the latter is decreasing along floer trajectories and one builds a @xmath71-parameter family of subcomplexes of @xmath177 , defined as @xmath178-\infty,\ , a[}(h , \ , j ) = \bigoplus_{\scriptsize \begin{array}{c } \gamma \in \mc p ( h ) \\ i_{cz}(\gamma ) = -k \mod \ 2\nu \\ a_h(\gamma ) < a \end{array } } { \mathbb{z}}\langle \gamma \rangle \ .\ ] ] this allows one to define the corresponding quotient complexes @xmath179-\infty,\ , b[}(h , \ , j ) /fc _ * ^{]-\infty,\ , a [ } ( h , \ , j ) , \quad -\infty \le a < b \le \infty\ ] ] and the same direct limit process goes through . we therefore put @xmath180 let us now make a few remarks on the properties of the above invariants . 1 . there is a natural cofinal family of hamiltonians whose values of the action on the @xmath71-periodic orbits is positive or arbitrarily close to @xmath181 ( see the construction of @xmath182 in figure [ hamplateau et rho ] ( 1 ) described in section [ la preuve de kunneth ] ) . this implies that @xmath183 if @xmath184 and @xmath185 does not depend on @xmath186 if the latter is strictly negative . in particular we have @xmath187 2 . the infimum @xmath188 of the areas of closed characteristics on the boundary is always strictly positive and therefore @xmath189 where @xmath190 , @xmath191 , @xmath192 . this follows from the fact that , in the limit , the hamiltonians become @xmath114-small on @xmath193 and the floer complex reduces to a morse complex that computes the relative cohomology , as @xmath194 points inward along @xmath3 . 3 . there are obvious truncation morphisms @xmath195 which induce morphisms @xmath196 , @xmath197 , @xmath198 . if @xmath199 , @xmath200 and @xmath201 we obtain a natural morphism @xmath202 or , written differently , @xmath203 we also note at this point that we have @xmath204 and that the two limits above can be interchanged by general properties of bi - directed systems . * fundamental principle ( viterbo @xcite ) . * _ if the morphism @xmath205 is not bijective , then there is a closed characteristic on @xmath3 . _ indeed , either there is some extra generator in @xmath0 , or some morse homological generator of @xmath206 is killed in @xmath0 . the `` undesired guest '' in the first case or the `` killer '' in the second case necessarily corresponds to a closed characteristic on @xmath3 . the version of floer homology that we defined above has various invariance properties @xcite . the main one that we shall use is the following . [ inv fl h ] the floer homology groups @xmath0 are an invariant of the completion @xmath7 in the following sense : for any open set with smooth boundary @xmath207 such that @xmath208 and the liouville vector field @xmath209 is transverse and outward pointing along @xmath210 , we have @xmath211 _ proof . _ one can realise a differentiable isotopy between @xmath8 and @xmath212 along the liouville vector field . this corresponds to an isotopy of symplectic forms on @xmath8 starting from the initial form @xmath213 and ending with the one induced from @xmath212 , denoted by @xmath214 . during the isotopy the boundary @xmath3 remains of contact type and the symplectic asphericity condition is preserved . an invariance theorem of viterbo @xcite shows that @xmath215 . on the other hand @xmath216 because @xmath217 and @xmath218 are symplectomorphic . before beginning the proof , let us note that the natural class of manifolds for which one can _ define _ floer homology groups for a product is that of manifolds with restricted contact type boundary . the reason is that @xmath219 involves the full manifolds @xmath8 and @xmath18 , not only some neighbourhoods of their boundaries . if @xmath123 and @xmath220 are the conformal vector fields on @xmath8 and @xmath18 respectively and @xmath221 , @xmath222 are the canonical projections , the natural conformal vector field on @xmath61 is @xmath223 . in order for @xmath224 to be defined in a neighbourhood of @xmath225 it is necessary that @xmath123 and @xmath220 be globally defined . we only prove a ) because the proof of b ) is entirely dual . one has to reverse arrows and replace direct limits with inverse limits . the difference in the statement is due to the fact that the inverse limit functor is in general not exact , except when each term of the directed system is a finite dimensional vector space @xcite . for the sake of clarity we shall give the proof under the assumption @xmath226 , @xmath227 , so that the grading on floer homology is defined over @xmath94 . i. we establish the short exact sequence ( [ suite exacte kunneth floer ] ) . here is the sketch of the proof . we consider on @xmath228 a hamiltonian of the form @xmath229 , @xmath230 , @xmath231 , @xmath232 . for an almost complex structure on @xmath233 of the form @xmath234 , with @xmath235 , @xmath236 ( generic ) almost complex structures on @xmath7 and @xmath237 respectively , the floer complex for @xmath238 can be identified , modulo truncation by the action issues , with the tensor product of the floer complexes of @xmath239 and @xmath240 . nevertheless , the hamiltonian @xmath238 is not linear at infinity and hence not admissible . we refer to @xcite for a discussion of the weaker notion of asymptotic linearity and a proof of the fact that @xmath238 does not even belong to this extended admissible class . the main idea of our proof is to construct an admissible pair @xmath241 whose floer complex is roughly the same as the one of @xmath242 . the hamiltonian @xmath41 will have lots of additional @xmath71-periodic orbits compared to @xmath238 , but all these will have negative enough action for them not to be counted in the relevant truncated floer complexes . we define the _ period spectrum _ @xmath243 of a contact type hypersurface @xmath244 in a symplectic manifold as being the set of periods of closed characteristics on @xmath244 , the latter being parameterized by the reeb flow . we assume from now on that the period spectra of @xmath3 and @xmath245 are _ discrete and injective _ i.e. the periods of the closed characteristics form a strictly increasing sequence , every period being associated to a unique characteristic which is transversally nondegenerate . this property is @xmath246-generic among hypersurfaces @xcite , while floer homology does not change under a small @xmath246-perturbation of the boundary ( proposition [ inv fl h ] ) . assuming a discrete and injective period spectrum amounts therefore to no loss of generality . we shall construct cofinal familes of hamiltonians and almost complex structures @xmath247 , @xmath248 , @xmath249 on @xmath7 , @xmath237 and @xmath250 respectively , with the following property . * main property . * _ let @xmath251 be fixed . for any @xmath252 , there is a positive integer @xmath253 such that , for all @xmath254 , the following inclusion of differential complexes holds : _ @xmath255 & fc_k^{[-\delta,\ , b[}(l_\nu,\ , j_\nu ) \ar[dl ] \\ \bigoplus_{r+s = k } fc_r^{[-\delta,\ , 2b[}(h_\nu,\ , j_\nu^1 ) \otimes fc_s^{[-\delta,\ , 2b[}(k_\nu,\ , j_\nu^2 ) \ar[r ] & fc_k^{[-\delta,\ , 4b[}(l_\nu,\ , j_\nu ) } } \ ] ] it is important to note that we require the two composed arrows @xmath256 @xmath257 to be the usual inclusions corresponding to the truncation by the action . in practice we shall construct _ autonomous _ hamiltonians having transversally nondegenerate @xmath71-periodic orbits , but one should think in fact of small local perturbations of these ones , along the technique of @xcite . the latter consists in perturbing an autonomous hamiltonian in the neighbourhood of a transversally nondegenerate @xmath71-periodic orbit @xmath74 , replacing @xmath74 by precisely two nondegenerate @xmath71-periodic orbits corresponding to the two critical points of a morse function on the embedded circle given by @xmath258 . the conley - zehnder indices of the perturbed orbits differ by one . moreover , the perturbation can be chosen arbitrarily small in any @xmath259-norm and the actions of the perturbed orbits can be brought arbitrarily close to the action of @xmath74 . let @xmath260 , @xmath261 be the vertical coordinates on @xmath7 and @xmath237 respectively . let @xmath262 , @xmath263 be _ cofinal _ families of autonomous hamiltonians on @xmath7 and @xmath237 , such that @xmath264 for @xmath265 , @xmath266 for @xmath267 , with @xmath268 , @xmath269 convex and linear of slope @xmath270 outside a small neighbourhood of @xmath71 . we assumed that the period spectra of @xmath3 and @xmath245 are discrete and injective and so we can choose @xmath271 with @xmath272 , @xmath273 . we shall drop the subscript @xmath12 in the sequel by referring to @xmath274 , @xmath275 and @xmath270 as @xmath95 , @xmath276 and @xmath136 . let us denote @xmath277 @xmath278 @xmath279 \b . our starting point is the construction by hermann @xcite of a cofinal family which allows one to identify , in the case of bounded open sets with restricted contact type boundary in @xmath280 , the floer homologies defined by viterbo @xcite and floer and hofer @xcite . we fix @xmath281 and consider the hamiltonian @xmath182 equal to @xmath95 for @xmath282 and constant equal to @xmath283 for @xmath284 , with @xmath283 arbitrarily close to @xmath285 . here @xmath286 is chosen to be small enough and positive . we perform the same construction in order to get a hamiltonian @xmath287 . we suppose ( figure [ hamplateau et rho ] ( 1 ) ) that @xmath182 takes its values in the interval @xmath288 on the interior of @xmath8 where it is also @xmath114-small and that @xmath289 on @xmath290 , with @xmath291 on @xmath292 $ ] and @xmath293 on @xmath294 , where @xmath295 . thus @xmath182 takes values in @xmath296 $ ] for @xmath297 $ ] and in @xmath298 $ ] for @xmath299 $ ] . the hamiltonian @xmath182 has additional @xmath71-periodic orbits compared to @xmath95 . these are either constants on levels @xmath300 with action @xmath301 , or orbits corresponding to characteristics on the boundary , appearing on levels @xmath302 close to @xmath23 . the action of the latter is arbitrarily close to @xmath303 , @xmath304 . the special choice of @xmath23 is motivated by the previous computation . we see in particular that it is crucial to take into account the gap @xmath305 in order to be able to make the action of this kind of orbits tend to @xmath306 . and the truncation function @xmath307 [ hamplateau et rho ] ] \c . we deform now @xmath308 to a hamiltonian that is constant equal to @xmath309 outside the compact set @xmath310 , with @xmath311 this already holds in @xmath312 . we describe the corresponding deformation in @xmath313 and perform the symmetric construction in @xmath314 . let us define @xmath315 @xmath316 with @xmath317 $ ] , @xmath318 on @xmath319 $ ] , @xmath320 for @xmath321 , @xmath307 strictly increasing on @xmath322 $ ] , @xmath323 $ ] on @xmath324 $ ] ( figure [ hamplateau et rho ] ( 2 ) ) . the symplectic form on @xmath325 is @xmath326 , with @xmath327 the contact form on @xmath245 . we get @xmath328 the projection of a periodic orbit of @xmath329 on @xmath330 is a periodic orbit of @xmath331 . in particular , @xmath332 is constant along the projection . moreover , the orbits appear on levels @xmath333 as there is no component @xmath334 in @xmath329 . as a consequence , the coefficients in front of @xmath331 and @xmath335 are constant along one orbit of @xmath329 . a @xmath71-periodic orbit @xmath336 of @xmath329 corresponds therefore to a couple @xmath337 such that * @xmath338 is an orbit of @xmath331 having period @xmath339 ; * @xmath74 is a closed characteristic on the level @xmath340 , having period @xmath341 and having the _ opposite _ orientation than the one given by @xmath335 . we have used the notation @xmath342 . the action of @xmath336 is @xmath343 we have denoted by @xmath344 , @xmath345 the areas of the orbits @xmath74 and @xmath338 respectively . the hamiltonian @xmath287 is constant in the relevant domain and we have directly replaced it by its value @xmath283 . it is useful to notice that @xmath346 . the minus sign comes from the fact that the running orientation on @xmath74 is opposite to the one given by @xmath335 . we have used the symplectic form @xmath347 on the second factor in ( [ action totale ] ) . for @xmath348 , the @xmath349-periodic orbits of @xmath331 belong to one of the following classes . 1 . [ 11 ] constants in the interior of @xmath8 , having zero area ; 2 . [ 22 ] closed characteristics located around the level @xmath350 , whose areas belong to the interval @xmath351 $ ] ; 3 . [ 33 ] if @xmath352 , one has a closed characteristic of area @xmath353 for any @xmath354 $ ] - the interval where @xmath332 is linear of slope @xmath136 ; 4 . [ 44 ] closed characteristics located around the level @xmath355 , whose areas belong to the interval @xmath356 $ ] ; 5 . [ 55 ] constants on levels @xmath357 , having zero area . for @xmath358 fixed we choose the various parameters involved in our constructions such that @xmath359 @xmath360 @xmath361 we now show that the actions of @xmath71-periodic orbits of @xmath329 appearing in the region @xmath362 tend uniformly to @xmath306 when @xmath363 . we estimate the action of the orbits @xmath336 according to the type of their first component @xmath338 and according to the level @xmath364 on which lies @xmath74 . 1 . @xmath338 of type [ 11 ] corresponds to @xmath365 $ ] . 1 . [ premier ] @xmath366 \ \bigcup \ [ b-\epsilon(\lambda),\ , \infty[$ ] . because @xmath367 there is no component of @xmath329 in the @xmath335 direction , orbits @xmath338 appear in degenerate families ( of dimension @xmath368 ) and the action of @xmath336 is @xmath369 . [ deuxieme ] @xmath370 $ ] . orbits @xmath338 come in pairs with closed characteristics @xmath371 of period @xmath372 on @xmath373 . we have @xmath374 the second inequality is valid for sufficiently large @xmath136 , in view of @xmath375 which implies @xmath376 . [ troisieme ] @xmath377 $ ] ( hence @xmath378 $ ] ) . for @xmath379 big enough we have @xmath380 2 . @xmath338 of type [ 22 ] corresponds to @xmath381 $ ] and @xmath382 $ ] . the area of @xmath338 belongs to the interval @xmath383 $ ] . 1 . @xmath384 \ \bigcup \ [ b-\epsilon(\lambda),\ , \infty[$ ] . as in [ premier ] ) we have @xmath385 2 . @xmath386 $ ] . like in [ deuxieme ] ) the total action of @xmath336 is @xmath387 3 . @xmath388 $ ] . following [ troisieme ] ) one has @xmath389 3 . @xmath338 of type [ 33 ] has an action @xmath390 , where @xmath391 . 1 . @xmath384 \ , \bigcup \ , [ b-\epsilon(\lambda),\ , \infty[$ ] : @xmath392 . 2 . @xmath370 $ ] : @xmath393 . [ quatrieme ] @xmath394 $ ] . the technique used in [ troisieme ] ) in order to get the upper bound no longer applies , as @xmath395 can be arbitrarily close to @xmath181 . nevertheless @xmath307 satisfies by definition the inequality @xmath396 . we thus get @xmath397 4 . @xmath338 of type [ 44 ] corresponds to @xmath398 $ ] and @xmath399 . in all three cases a)-c ) we get @xmath400 . @xmath338 of type [ 55 ] corresponds to @xmath401 . like in [ premier ] ) there is no component in the @xmath335 direction for @xmath329 and orbits @xmath338 appear in ( highly ) degenerated families . the total action in all three cases a ) - c ) is @xmath402 . this finishes the proof of the fact that the action of the new orbits of @xmath403 tends uniformly to @xmath306 . the symmetric construction can be carried out for @xmath276 in the region @xmath404 . one gets in the end a hamiltonian @xmath405 which is constant equal to @xmath309 on @xmath406 . we modify now @xmath405 outside the compact set @xmath407 in order to make it linear with respect to the liouville vector field @xmath408 on @xmath409 . let us define the following domains in @xmath410 ( see figure [ param product ] ) : @xmath411 @xmath412 let @xmath413 be a hypersurface which is transversal to @xmath224 such that @xmath414 \ , \ ] ] @xmath415 \ .\ ] ] we parameterize @xmath416 by @xmath417 @xmath418 which is a symplectomorphism if one endows @xmath419 with the symplectic form @xmath420 , where @xmath421 . as an example , for @xmath422 we have @xmath423 . it is easy to see that @xmath424 as a consequence @xmath405 is constant equal to @xmath309 on @xmath425 . we replace it by @xmath426 on @xmath427 , with @xmath428 convex and @xmath429 for @xmath430 . the additional @xmath71-periodic orbits that are created in this way have action @xmath431 . by choosing @xmath432 one ensures @xmath433 , as well as the cofinality of the family of hamiltonians @xmath41 as @xmath434 . indeed , the hamiltonian @xmath41 is bigger than @xmath435 on @xmath436 . note that the choice of @xmath437 equal to @xmath438 and not belonging to the spectrum of @xmath244 is indeed possible if we choose @xmath244 to have a discrete and injective spectrum . [ param product ] ] e. we have constructed a cofinal family of hamiltonians @xmath439 which are linear at infinity and that are associated to the initial hamiltonians @xmath262 and @xmath263 . i claim that the _ main property _ holds for @xmath440 . we assume of course that @xmath441 , @xmath442 and @xmath443 are regular almost complex structures for @xmath274 , @xmath275 , @xmath444 respectively , which are standard for @xmath445 , @xmath446 and @xmath447 . moreover , the almost complex structure @xmath443 is of the form @xmath448 for @xmath449 . the preceding estimates on the action of the @xmath71-periodic orbits show that the sequence ( [ inclusion fondamentale de complexes ] ) of inclusions is certainly valid at the level of modules . this says in particular that the @xmath71-periodic orbits involved in the free modules appearing in ( [ inclusion fondamentale de complexes ] ) are located in a neighbourhood of @xmath8 , @xmath18 and @xmath19 respectively , and we can assume that the latter is contained in @xmath450 for a suitable choice of @xmath244 . classical transversality arguments ( @xcite , prop . 17 ; @xcite , 5.1 and 5.4 ) ensure that we can choose regular almost complex structures of the form described above . the point is now to prove that the inclusions ( [ inclusion fondamentale de complexes ] ) are also valid at the level of _ differential _ complexes . it is enough to show that any trajectory @xmath451 which satisfies @xmath452 , @xmath453 stays in the domain @xmath454 , where the floer equation is split . that will imply that @xmath455 and @xmath456 are floer trajectories in @xmath330 and @xmath457 respectively and will prove ( [ inclusion fondamentale de complexes ] ) at the level of differential complexes . it is of course enough to prove this statement under the assumption that @xmath458 on the whole of @xmath459 : the floer trajectories would then be a - posteriori contained in @xmath460 . moreover , the proof of this fact only makes use of the split structure on the set @xmath461 and this allows one to modify @xmath443 outside @xmath462 in order to formally work with an almost complex structure that is homothety - invariant at infinity . we therefore suppose in the sequel that @xmath463 . arguing by contradiction , assume that the floer trajectory @xmath171 is not contained in @xmath464 . as @xmath171 is anyway contained in a compact set , we infer that the function @xmath465 has a local maximum in @xmath466 , which means that one of the functions @xmath467 or @xmath468 has a local maximum in @xmath469 or @xmath470 respectively . the two cases are symmetric and we can assume without loss of generality that @xmath471 has a local maximum in @xmath472 . in view of the fact that @xmath456 satisfies the floer equation associated to @xmath275 for @xmath473 the maximum principle ensures that the value of @xmath474 at the local maximum is in the interval @xmath4752a , \ , \infty[$ ] . but @xmath476 , @xmath477 with @xmath478 and this implies that @xmath456 crosses the hypersurfaces @xmath479 and @xmath480 . moreover , the piece of @xmath456 contained in @xmath481 is @xmath442-holomorphic because @xmath275 is constant on that strip . we therefore obtain @xmath482 } |w_s|^2_{j_\nu^2 } \ = \ { \text{\rm area}}(w \cap \ { a \le s '' \le 2a \ } ) \ . \nonumber \end{aligned}\ ] ] the last equality holds because @xmath456 is @xmath442-holomorphic in the relevant region . the lemma below will allow us to conclude . it is inspired from hermann s work @xcite , where one can find it stated for @xmath483 . [ petit lemme energetique ] let @xmath484 be a manifold with contact type boundary and @xmath457 its symplectic completion . let @xmath43 be an almost complex structure which is homothety invariant on @xmath485 . there is a constant @xmath486 such that , for any @xmath487 and any @xmath43-holomorphic curve @xmath171 having its boundary components on both @xmath488 and @xmath489 , one has @xmath490 _ proof . _ consider , for @xmath487 , the map @xmath491 by definition the map @xmath492 is @xmath43-holomorphic . on the other hand , by using the explicit form of @xmath43 given in [ constructions ] , one sees that @xmath492 expands the area element by a factor @xmath23 . as a consequence , up to rescaling by @xmath492 it is enough to prove ( [ petite ineg aires ] ) for @xmath493 . we apply gromov s monotonicity lemma @xcite 1.5.b , @xcite 4.3.1 which ensures the existence of @xmath494 and of @xmath495 such that , for any @xmath496 , any @xmath497 $ ] with @xmath498 $ ] and any connected @xmath43-holomorphic curve @xmath44 such that @xmath499 and @xmath500 one has @xmath501 let us now fix @xmath502 small enough so that @xmath503 $ ] for all @xmath504 . as the boundary of @xmath171 rests on both @xmath505 and @xmath506 one can find such a point @xmath507 on the image of @xmath171 . we infer @xmath508 . then @xmath509 is the desired constant . applying lemma [ petit lemme energetique ] to our situation we get a constant @xmath510 that does not depend on @xmath171 and such that @xmath511 . the difference of the actions in ( [ inegalite estimation energie ] ) is at the same time bounded by @xmath512 and we get a contradiction for @xmath136 large enough . for a fixed @xmath252 the floer trajectories corresponding to @xmath444 are therefore contained in @xmath513 for @xmath12 large enough . this proves that the sequence of inclusions ( [ inclusion fondamentale de complexes ] ) is valid at the level of differential complexes . a few more commutative diagrams will now finish the proof . first , as a direct consequence of ( [ inclusion fondamentale de complexes ] ) , one has the commutative diagram @xmath514 \ar[r ] & fh_k^{[-\delta,\ , b[}(l_\nu,\ , j_\nu ) \ar[dl ] \ar@{-->}[d ] \\ h_k \big ( fc_*^{[-\delta,\ , 2b[}(h_\nu,\ , j_\nu ^1 ) \otimes fc_*^{[-\delta,\ , 2b[}(k_\nu,\ , j_\nu ^2 ) \big ) \ar[r ] & fh_k^{[-\delta,\ , 4b[}(l_\nu,\ , j_\nu ) } } \ ] ] by taking the direct limit for @xmath515 and @xmath516 this induces the diagram @xmath517 \ar@{-->}[d ] & fh_k(m\times n ) \ar[dl]_\sim \ar@{-->}[d ] \\ \displaystyle \lim_{\stackrel{\longrightarrow } { b \rightarrow + \infty } } \lim_{\stackrel{\longrightarrow } { \nu { \longrightarrow } + \infty } } h_k \big ( fc_*^{[-\delta , 2b[}(h_\nu , j_\nu ^1 ) \otimes fc_*^{[-\delta , 2b[}(k_\nu , j_\nu ^2 ) \big ) \ar[r ] & fh_k(m \times n ) } } \ ] ] we easily see that _ vertical arrows are isomorphisms _ as the direct limits following @xmath12 and @xmath518 commute with each other ( this is a general property of bidirected systems ) . this implies that _ the diagonal arrow is an isomorphism _ as well . at the same time the algebraic knneth theorem ( * ? ? ? * vi.9.13 ) ensures the existence of a split short exact sequence @xmath519 & \displaystyle \lim_{\begin{array}{c } \scriptsize \longrightarrow \\ b,\ , \nu \rightarrow \infty \end{array } } h_k \big ( fc_*^{[-\delta,\ , 2b[}(h_\nu,\ , j _ \nu ^1 ) \otimes fc_*^{[-\delta,\ , 2b[}(k_\nu,\ , j _ \nu ^2 ) \big ) \ar@{->>}[d ] \\ & \bigoplus_{r+s = k-1 } { \text{\rm tor}}_1^a \big ( fh_r(m ) , \ fh_s(n ) \big ) } } \ ] ] we infer the validity of the short exact sequence ( [ suite exacte kunneth floer ] ) . we note that we use in a crucial way the exactness of the direct limit functor in order to obtain ( [ troisieme diagramme intermediaire ] ) from the knneth exact sequence in truncated homology . we prove now the existence of the morphism from the classical knneth exact sequence to ( [ suite exacte kunneth floer ] ) . we restrict the domain of the action to @xmath520 with @xmath251 small enough . the floer trajectories of a @xmath114-small autonomous hamiltonian which is a morse function on @xmath8 coincide in the symplectically aspherical case with the gradient trajectories in the thom - smale - witten complex . we denote by @xmath521 the morse complexes on the relevant manifolds . by ( [ inclusion fondamentale de complexes ] ) there is a commutative diagram @xmath522 & fc_k^{[-\delta , 4b[}(l_\nu , j_\nu ) \\ \displaystyle{\bigoplus_{r+s = k } c^{{\text{\rm morse}}}_{m+r}(h_\nu , j_\nu^1 ) \otimes c^{{\text{\rm morse}}}_{n+s}(k_\nu , j_\nu^2 ) } \ar[u ] \ar@{=}[r ] & c^{{\text{\rm morse}}}_{m+n+k}(l_\nu , j_\nu^1\oplus j_\nu ^2 ) \ . \ar[u ] } } \ ] ] the relevant morse complexes compute homology relative to the boundary . with an obvious notation the above diagram induces in homology @xmath523^{\qquad \qquad \qquad \qquad \qquad \sim } & fh_k(m\times n ) \\ h_{m+n+k } \big ( c_{m+*}(m , \ , \partial m ) \otimes c_{n+*}(n , \partial n ) \big ) \ar[u]^{\phi } \ar[r]^{\qquad \qquad \sim } & h_{m+n+k}(m\times n , \ , \partial(m\times n ) ) \ar[u]^{c _ * } } } \ ] ] by naturality of the algebraic knneth exact sequence , the map @xmath524 fits into the diagram below , where @xmath525 stands for its domain and target . @xmath526 & { * } \ar@{->>}[r ] & { \displaystyle{\bigoplus_{r+s = k-1 } } } { \text{\rm tor}}_1^a \big ( fh_r(m ) , fh_s(n ) \big ) \\ \hspace{-.1cm}{\displaystyle{\hspace{.03cm}\bigoplus_{r+s = k } } } \hspace{-.1 cm } h_{m+r}(m , \partial m ) \otimes h_{n+s}(n , \partial n ) \ \ \ar@{>->}[r ] \ar[u]_{c_*\otimes c _ * } & { * } \ar@{->>}[r ] \ar[u]_{\phi } & \hspace{-.32cm}{\displaystyle{\bigoplus _ { r+s = k-1 } } } \hspace{-.25 cm } { \text{\rm tor}}_1^a \big(h_{m+r}(m,\partial m ) , h_{n+s}(n,\partial n ) \big ) \ar[u]_{{\text{\rm tor}}_1(c _ * ) } } } \ ] ] diagrams ( [ deuxieme diag pour kunneth geant][troisieme diag pour kunneth geant ] ) establish the desired morphism of exact sequences . [ prop : produc with c ] let @xmath18 be a compact symplectic manifold with restricted contact type boundary and let @xmath237 be its symplectic completion . let @xmath527 , @xmath528 be endowed with the product symplectic form . then @xmath529 we now fix terminology for the proof of theorem c following @xcite . a stein manifold @xmath531 is a triple @xmath532 , where @xmath533 is a complex structure and @xmath534 is an exhausting plurisubharmonic function . we say that @xmath531 is of _ finite type _ if we can choose @xmath534 with all critical points lying in a compact set @xmath276 . the stein domains @xmath535 such that @xmath536 are called _ big stein domains _ of @xmath534 ; they are all isotopic . we call @xmath537 _ subcritical _ if @xmath534 is morse and all its critical points have indices strictly smaller than @xmath538 ( they are anyway at most equal to @xmath538 ) . let @xmath532 be a stein manifold of finite type with @xmath534 morse . following @xcite we define a _ finite type stein deformation _ of @xmath539 as a smooth family of complex structures @xmath540 , @xmath541 $ ] together with exhausting plurisubharmonic functions @xmath542 such that : i ) @xmath543 , @xmath544 ; ii ) the @xmath542 have only morse or birth - death type critical points ; iii ) there exists @xmath545 such that all @xmath546 are regular values for @xmath542 , @xmath541 $ ] . condition ii ) is not actually imposed in @xcite , but we need it for the following theorem , which says that the existence of finite type stein deformations on subcritical manifolds is a topological problem . * theorem ( compare @xcite ) . * _ let @xmath547 and @xmath548 be finite type stein structures on @xmath531 . assume @xmath158 , @xmath549 are homotopic as almost complex structures and @xmath550 , @xmath551 can be connected by a family @xmath542 , @xmath541 $ ] of exhausting smooth functions which satisfy conditions ii)-iii ) above and whose nondegenerate critical points have subcritical index for all @xmath552 $ ] . then @xmath553 , @xmath548 are homotopic by a finite type stein deformation . _ this is the finite type version of theorem 3.4 in @xcite . it holds because the latter is proved by _ h_-cobordism methods within the plurisubharmonic category ( * ? ? ? * lemma 3.6 ) , and these preserve the finite type condition . given a stein manifold of finite type @xmath532 , let us fix @xmath545 such that all @xmath546 are regular values of @xmath534 . let @xmath554 . the big stein domains @xmath535 , @xmath546 are diffeomorphic and endowed with exact symplectic forms @xmath555 for which @xmath556 is of restricted contact type . the floer homology groups @xmath557 are well defined and , by invariance under deformation of the symplectic forms , they are isomorphic @xcite . we define @xmath558 as @xmath557 for @xmath559 large enough . let now @xmath560 , @xmath541 $ ] be a finite type deformation on a stein manifold @xmath531 . given @xmath545 such that all @xmath546 are regular values of @xmath542 for all @xmath541 $ ] , the big stein domains @xmath561 , @xmath541 $ ] , @xmath546 are all diffeomorphic and endowed with exact symplectic forms @xmath562 . the boundaries @xmath563 are of restricted contact type and we can again apply lemma 3.7 in @xcite to conclude that the floer homology groups @xmath564 are naturally isomorphic . in particular @xmath565 . _ proof of theorem c. _ cieliebak proved in @xcite that , given a subcritical stein manifold of finite type @xmath566 , there exists a stein manifold of finite type @xmath567 and a diffeomorphism @xmath568 such that : i ) @xmath569 and @xmath570 are homotopic as almost complex structures ; ii ) @xmath571 and @xmath572 are subcritical morse functions with isotopic big stein domains @xmath573 , respectively @xmath574 . the functions @xmath571 and @xmath572 are associated to handle decompositions @xmath575 , @xmath576 of @xmath573 and @xmath574 such that @xmath577 , @xmath578 and the following property holds . given @xmath579 such that @xmath580 , the attaching maps of @xmath581 , @xmath582 , @xmath583 are isotopic in @xmath584 ( the condition is independent of @xmath585 ) . two such handle decompositions are called _ isotopic_. because the handle decomposition determines the isotopy class of the associated morse function and the two handle decompositions above are isotopic , we infer that the functions @xmath571 and @xmath572 are isotopic , i.e. there exists @xmath586 with @xmath587 . we can therefore apply the previous theorem and conclude that the stein structures @xmath588 and @xmath589 can be connected by a finite type stein deformation . it follows that @xmath590 . by proposition [ prop : produc with c ] , the latter homology group vanishes . on the other hand , by invariance of floer homology under symplectomorphism we have @xmath591 . _ proof of theorem b. _ the statement follows readily from the existence of the commutative diagram given by the second part of theorem a , taking into account the isomorphism @xmath592 , @xmath593 , @xmath594 . the latter is given by the usual knneth formula in singular cohomology with coefficients in a field . * floer , hofer and viterbo @xcite proved the weinstein conjecture in a product @xmath595 , @xmath528 with @xmath596 a closed symplectically aspherical manifold . the weinstein conjecture for a product @xmath233 with @xmath237 subcritical stein and @xmath7 the completion of a restricted contact type manifold has been proved by frauenfelder and schlenk in @xcite . the discussion below makes use of field coefficients . let @xmath597 be small enough . one defines ( see e.g. @xcite ) the capacity of a compact symplectic manifold @xmath8 with contact type boundary as @xmath598-\delta,\ , b]}(m ) \longrightarrow h^{2m}(m , \ , \partial m ) { \text{\rm is zero } } \ } \\ & = & \sup \ { b > 0 \ , : \ , fh^{m}_{]-\delta,\ , b]}(m ) \longrightarrow h^{2m}(m , \ , \partial m ) { \text{\rm is nonzero } } \ } \ .\end{aligned}\ ] ] here @xmath593 . the next result is joint work with a .- biolley , who applies it in her study of symplectic hyperbolicity @xcite . _ proof . _ the main property ( [ inclusion fondamentale de complexes ] ) gives , for @xmath12 large enough and field coefficients , an arrow @xmath600-\delta,\frac b 2]}(h_\nu)\otimes fh^s_{]-\delta,\frac b 2]}(k_\nu ) \longleftarrow fh^{m+n}_{]-\delta , b]}(l_\nu)$ ] . moreover , for fixed @xmath518 and @xmath12 large enough the hamiltonians @xmath274 , @xmath275 and @xmath444 compute the corresponding truncated cohomology groups of @xmath8 , @xmath18 and @xmath19 . like in theorem a.b . we get the commutative diagram @xmath601-\delta , \ , \frac b 2]}(m ) \otimes fh^s_{]-\delta , \ , \frac b 2]}(n ) } \ar[d]^{c_*^{b/2}\otimes c_*^{b/2 } } & & \ar[ll ] fh^{m+n}_{]-\delta , \ , b]}(m \times n ) \ar[d]^{c_*^b } \\ h^{2m}(m,\ , \partial m)\otimes h^{2n}(n,\ , \partial n ) & & \ar[ll]_{\sim } h^{2m+2n}(m\times n,\ , \partial ( m\times n ) ) } } \ ] ] let now @xmath602 . then @xmath603 , hence @xmath604 and therefore @xmath605 . y. eliashberg , symplectic geometry of plurisubharmonic functions . notes by m. abreu . _ gauge theory and symplectic geometry ( montreal , 1995 ) _ , nato adv . , 488 , kluwer , dordrecht , 1997 , 4967 .

* abstract : * we prove that given a three manifold with an arbitrary metric of positive ricci curvature , there exists a sweepout of by surfaces of genus and areas bounded by . we use this result to construct a sweepout of by 1-cycles of length at most , giving a partial answer to a question of l. guth . the sweepout of surfaces is generated from a min - max minimal surface . if further assuming a positive scalar curvature lower bound , we can get a diameter upper bound for the min - max surface . [ section ] [ theorem]proposition [ theorem]corollary [ theorem]remark [ theorem]definition [ theorem]lemma * * 12pt [ runin ] * * .

1510.02896

let @xmath1 be a 3-manifold with positive ricci curvature . in this paper we obtain quantitative results about sweepouts of @xmath1 by 1-cycles and surfaces . [ main_cycle ] every closed 3-manifold @xmath1 of positive ricci curvature admits a map @xmath5 with fibers of length at most @xmath6 for a universal constant @xmath7 . moreover , the family of 1-cycles @xmath8 is continuous in a strong sense . we define this precisely in section 3 . the geometric invariant that we bound from above in theorem [ main_cycle ] is called the k - waist of a riemannian manifold . for an n - dimensional manifold @xmath1 and @xmath9 the k - waist , @xmath10 , is defined as @xmath11 , where the infimum is taken over all proper functions @xmath12 . waists have been defined by gromov and extensively studied in @xcite , @xcite , @xcite , @xcite , @xcite , @xcite . gromov proved deep theorems about waists of manifolds ( see @xcite and @xcite for exposition of some of his work ) , yet many significant questions about waists remain open . in particular , existence of upper bounds for @xmath10 when @xmath13 is a widely open question . in @xcite guth asked if for every riemannian metric @xmath14 on a 3-torus @xmath15 the 1-waist @xmath16 is bounded from above by @xmath17 . more generally , one may ask analogous questions for any riemannian 3-manifold , that is , does there exist a constant @xmath18 , such that for any metric @xmath14 there exists a map @xmath5 with fibers of length at most @xmath19 ? if this is true , it would be a strong generalization of the systolic inequality . in theorem [ main_cycle ] we affirmatively answer this question for 3-manifolds @xmath1 under an additional assumption of @xmath20 . it is to our knowledge the first occasion where such a generalization is proved . theorem [ main_cycle ] generalizes , in the setting of positive ricci curvature , some fundamental theorems about the length of closed geodesics and stationary one - cycles in riemannian manifolds due to gromov and nabutovsky - rotman . as consequence of theorem [ main_cycle ] we obtain [ systolic ] let @xmath1 be a closed 3-manifold of positive ricci curvature . if @xmath1 is homeomorphic to @xmath21 then it contains a non - contractible closed geodesic of length at most @xmath22 . if @xmath23 then it contains a geodesic net of length at most @xmath22 . the first statement of corollary [ systolic ] is a special case of gromov s systolic inequality @xcite . the systole of a riemannian manifold is defined as the length of the shortest non - contractible geodesic loop . in the same paper gromov conjectured that every manifold contains a non - trivial closed geodesic of length at most @xmath24 . nabutovsky and rotman @xcite proved that every riemannian manifold contains a stationary 1-cycle of length at most @xmath25 ( a stationary 1-cycle need not be a closed geodesic ; it may look , for example , like a bouquet of geodesic loops all intersecting at a point with tangent vectors at that point summing up to @xmath26 , see @xcite for more examples ) . if @xmath1 is topologically a sphere , then a min - max argument yields an upper bound for the length of a stationary 1-cycle , giving an alternative proof of a special case of the result of nabutovsky and rotman . here we present a short ( and incomplete ) overview of previously known estimates for sweepouts of manifolds . in @xcite guth proved that every open subset of euclidean space @xmath27 admits a sweepout by relative @xmath28-cycles of volume at most @xmath29 . in general such inequalities do not hold for riemannian manifolds ( see appendix 5 in @xcite and @xcite ) . however , we may control volumes of @xmath30-cycles if we impose an additional requirement on the metric . in @xcite , among other results , it was shown that if @xmath1 is conformally equivalent to a manifold with non - negative ricci curvature then it admits a sweepout by @xmath30-cycles of volume at most @xmath31 ( in @xcite sabourau independently constructed a sweepout of @xmath1 with @xmath32 by @xmath30-cycles of controlled volume ) . + when @xmath20 and @xmath33 we show that we can simultaneously control the area and the genus of surfaces in the sweepout , which will be essential in the proof of theorem [ main_cycle ] . [ main theorem3 ] given a three manifold with an arbitrary metic @xmath0 of positive ricci curvature , i.e. @xmath34 , there exists a minimal surface @xmath35 , such that @xmath36 , for a universal constant @xmath7 . also we have * if @xmath37 is orientable , then the genus @xmath38 of @xmath39 satisfies @xmath40 , and there exists a smooth sweepout @xmath41}$ ] of @xmath0 , such that * * @xmath42 forms a heegaard splitting of @xmath43 , i.e. @xmath44 is an embedded surface of genus @xmath38 , for @xmath45 , and @xmath46 and @xmath47 are graphs ; * * @xmath48 for @xmath49 . * if @xmath37 is non - orientable , then the genus @xmath50 of its double cover @xmath51 satisfies @xmath52 . moreover , by removing @xmath39 from @xmath1 , we get a manifold with boundary @xmath53 with @xmath54 , and there exists a smooth sweepout @xmath55}$ ] of @xmath53 , such that * * @xmath42 forms a heegaard splitting of @xmath53 , i.e. @xmath44 is an embedded surface of genus @xmath38 lying in the interior of @xmath53 , for @xmath56 , and @xmath57 ; * * @xmath58 for @xmath49 . after this paper was submitted for publication ketover , marques and neves @xcite proved that the min - max minimal surface @xmath37 in 3-manifolds with positive ricci curvature must be orientable . this result follows from a clever application of the catenoid estimate they derive . therefore , the second case of theorem 1.3 does not occur . note that this simplifies our construction of a continuous sweepout of @xmath1 by 1-cycles as it rules out case 2 in the proof of theorem [ main_cycles1 ] ( see section 5 ) . we would like to compare our result with that of f. marques and a. neves @xcite . in @xcite , assuming @xmath34 and the scalar curvature lower bound @xmath59 , marques - neves produced a smooth sweepout @xmath55}$ ] , where the genus of @xmath44 is the heegaard genus . ] , and @xmath60 . the advantage of @xcite is that they have better estimates for the genus . however , from the point of view of area estimates ( e.g. for the application to prove theorem [ main_cycle ] ) , our result can be much better than that in @xcite while we still have a relatively good genus estimate . an example illustrating this fact is a long and thin 3-dimensional ellipsoid ; when we normalize the scalar curvature lower bound to be @xmath61 , the width can be very small ( compared to @xmath62 ) . the difference between our method with @xcite is that we use the almgren - pitts min - max theory @xcite for general sweepouts constructed in @xcite , while marques - neves used the colding - de lellis @xcite ( or simon - smith @xcite ) min - max method for smooth sweepouts given by heegaard splittings . we refer to [ open questions ] for more discussion . + the sweepout @xmath42 in theorem [ main theorem3 ] is used to construct a sweepout by 1-cycles of controlled length in theorem [ main_cycle ] . an important open question is whether one can construct a sweepout by _ closed curves _ of controlled length rather than 1-cycle ( see more discussion in [ open questions ] ) . one approach in this direction is to first construct a sweepout of @xmath1 by spheres or tori of controlled area and _ diameter_. for this purpose , we derive the following partial result . in particular , if we further assume a scalar curvature lower bound , we can get a uniform diameter upper bound for the min - max minimal surface . [ main theorem3.1 ] let @xmath0 be as in theorem [ main theorem3 ] ; if the scalar curvature of @xmath0 is bounded from below , i.e. @xmath63 , for some @xmath64 , then the diameter of @xmath35 ( when it is orientable ) or the diameter of its double cover ( when it is non - orientable ) is bounded from above by @xmath65 . the main idea of proving theorem [ main_cycle ] is a dimension reduction type argument . we first construct a nice sweepout by 2-surfaces with controlled area and genus by theorem [ main theorem3 ] . then we continuously sweep out these 2-surfaces by 1-cycles . a large portion of the argument ( in particular , proof of lemma [ main_cycles1 ] ) is devoted to making this family continuous in a strong sense ( cf . section [ intro to 1-cycles ] ) . this continuity is needed to obtain a geodesic net of controlled length as a limit of a min - max sequence of 1-cycles . theorem [ main theorem3 ] is proved by combining several ingredients . we apply the almgren - pitts min - max theory to the sweepout constructed in @xcite and get a min - max minimal surface of controlled area . by using one of the authors morse index bound @xcite , we can get the desired genus bound via schoen - yau genus estimates @xcite . the existence of good heegaard splitting follows from meeks - simon - yau @xcite . the diameter estimates ( theorem [ main theorem3.1 ] ) for the min - max surface comes from schoen - yau diameter estimates @xcite and the morse index estimate . our paper is organized as follows . in [ area and diameter estimates ] , we prove theorem [ main theorem3 ] and theorem [ main theorem3.1 ] . in [ intro to 1-cycles ] we give a precise definition of the sweepout by 1-cycles . in [ parametric sweepouts of surfaces ] , we show how to sweep out a family of surfaces simultaneously by continuous 1-cycles with lengths controlled by the genus and area . in [ proof of main_cycle ] we prove theorem [ main_cycle ] by combining results in [ area and diameter estimates ] and [ parametric sweepouts of surfaces ] . finally , we summarize several interesting open questions in [ open questions ] . : both authors would like to thank larry guth for getting them together and useful comments . y.l . would like to thank alexander nabutovsky and regina rotman for helpful discussions and alexander nabutovksy for pointing out a mistake in the earlier draft . we outline the proof of theorem [ main theorem3 ] . since @xmath66 has positive ricci curvature , we can start with a sweepout constructed in @xcite , and , independently , @xcite , i.e. @xmath67 , such that for a universal constant @xmath7 , @xmath68 now we can adapt such a sweepout to the almgren - pitts theory @xcite as in @xcite . application of the almgren - pitts theory produces a min - max minimal surface @xmath69 , such that an integer multiple @xmath70 , @xmath71 achieves the min - max value @xmath72 , i.e. @xmath73 using ( * ? ? ? * theorem 1.1 ) , there exists a minimal surface @xmath37 , such that @xmath37 has least area among all closed , embedded , minimal hypersurfaces in the following sense . define ( see ( * ? ? ? * ( 1.1 ) ) ) @xmath74 then @xmath75 when it is orientable , or @xmath76 when it is non - orientable . by comparing the area of @xmath69 with that of @xmath37 , we have that : * @xmath77 . moreover , when @xmath37 is orientable , it is proven in ( * ? ? ? * theorem 1.1 ) that * the morse index of @xmath37 is one . when @xmath37 is non - orientable , it is shown by ( * ? ? ? * theorem a ) that * the morse index of the double cover @xmath51 of @xmath37 is one . our definition of @xmath78 is the same as @xmath79 in @xcite ; and when @xmath80 , ( * ? ? ? * theorem a ) reduces to ( * ? ? ? * theorem 1.1 ) , except that ( * ? ? ? * theorem a ) showed that the double cover of a non - orientable min - max surface has morse index 1 . by @xcite ( * ? ? * theorem 2 ) , when @xmath34 , the morse index equal to one implies that the genus @xmath81 of @xmath37 ( when it is orientable ) or the genus @xmath82 of its double cover @xmath51 ( when it is non - orientable ) is bounded by @xmath83 . next we show the existence of good sweepouts generated by @xmath37 . when @xmath37 is orientable , we claim that @xmath37 must be a heegaard splitting ; or equivalently , @xmath84 is a union of two connected component @xmath85 and @xmath86 , such that @xmath85 and @xmath87 are both handlebodies . this is essentially due to @xcite ( see also ( * ? ? ? * lemma 3.2 ) ) . in fact , @xmath37 separates @xmath1 into two connected components @xmath85 and @xmath87 by ( * ? ? ? * proposition 3.5 ) ( see also ( * ? ? ? * lemma 3.2 ) ) . by minimizing area in the isotopy class of @xmath37 inside @xmath85 using ( * ? ? ? * theorem 1 ) , we either get another minimal surface @xmath88 in the interior of @xmath85 , or we get an empty set , i.e @xmath37 can be isotopically changed to a surface of arbitrarily small area . the first case will violate the frankel s theorem @xcite which says that every two closed minimal surfaces must intersect when @xmath34 ; while the second case implies that @xmath85 is a handlebody by ( * ? ? ? * proposition 1 ) . similarly @xmath87 is also a handlebody . by ( * lemma 3.5 ) . ] , we can construct a heegaard splitting @xmath41}$ ] satisfying the requirement of theorem [ main theorem3 ] . when @xmath37 is non - orientable , by removing @xmath37 from @xmath1 , we get a manifold @xmath53 with boundary @xmath89 . the boundary @xmath89 is a minimal surface , and is a double cover of @xmath37 . similar argument as above shows that @xmath53 is a handlebody . again by the same method in ( * ? ? ? * lemma 3.5 ) , we can construct a heegaard splitting @xmath55}$ ] satisfying the requirement . to prove theorem [ main theorem3.1 ] , we will adapt the schoen - yau @xcite diameter estimates via scalar curvature lower bound for stable minimal surfaces as follows . let @xmath39 be a two - sided minimal surface possibly with boundary . here two - sided " means that @xmath39 has a unit normal vector field @xmath90 . given a function @xmath91 , the second variation of area functional along normal deformation in the direction of @xmath92 is given by ( * ? ? ? * chap . 1 , 8) : @xmath93 where @xmath94 is the jacobi operator , and @xmath95 is the second fundamental form of @xmath39 . @xmath39 is stable if @xmath96 for all @xmath91 . [ diameter estimates for stable surface ] given a three manifold @xmath0 , assume that the scalar curvature is bounded from below @xmath97 , @xmath64 . let @xmath39 be a two - sided stable minimal surface with boundary @xmath98 , then the inf - radius @xmath99 of @xmath39 is bounded from below by @xmath100 . the fact that @xmath39 is stable implies that the jacobi operator @xmath101 is non - positive . let @xmath102 be the first dirichlet eigenfunction of @xmath101 , i.e. @xmath103 , @xmath104 . then @xmath105 in the interior of @xmath39 . using @xcite , by rewriting the jacobi operator , we have @xmath106 where @xmath107 and @xmath108 are respectively the scalar curvatures of @xmath1 and @xmath39 . take a point @xmath109 in the interior of @xmath39 , such that the distance of @xmath109 to @xmath110 achieves the inf - radius @xmath99 of @xmath39 . now consider the problem of minimizing the following functional @xmath111 among all curves @xmath112 connecting @xmath109 to @xmath110 . assume that @xmath113 achieves a minimum , then @xmath114 the first variation of @xmath115 at @xmath113 vanishes : @xmath116 here @xmath117 is the riemannian connection of @xmath39 , and @xmath118 is an arbitrary variational vector field along @xmath119 vanishing at the end points of @xmath113 , and @xmath120 in the unit tangent vector field along @xmath119 , and @xmath121 is the length parameter . integrating by parts shows that @xmath122 where @xmath123 is normal component ( with respect to the tangent vector of @xmath113 in @xmath39 ) of @xmath124 . therefore the weighted geodesic equation is @xmath125 the second variation of @xmath115 is non - negative : @xmath126 here @xmath127 is the curvature tensor of @xmath39 . denote @xmath90 by the unit normal vector field along @xmath113 , and let @xmath128 for some function @xmath129 which vanishes at the end points of @xmath113 . using ( [ weighted first variation ] ) and ( [ weighted geodesic equation ] ) we have , @xmath130f^2ds\\ & + \int_{{\gamma}}\varphi({\gamma}(s))\big(|\nabla_v f|^2-k^{{\sigma}}(\nu , v , \nu , v)f^2\big)ds \geq 0 . \end{split}\ ] ] using the fact that @xmath131 and ( [ weighted geodesic equation ] ) , ( [ jacobi operator 1 ] ) can be re - written as @xmath132 combing this with ( [ weighted second variation ] ) and using the fact that @xmath133 and that @xmath97 , we have , @xmath134 now parametrize @xmath113 by the length parameter on @xmath135 $ ] , with @xmath136 , and using integration by part , we get @xmath137 this implies that the following operator @xmath138 is non - negative is the same as that in ( * ? ? ? * p577 ) where the @xmath139 used in @xcite is a constant in our setting . ] : @xmath140 let @xmath141 be the first dirichlet eigen - function of @xmath138 on @xmath135 $ ] , then @xmath142 , and @xmath143 multiply the above inequality with any @xmath144 , @xmath145)$ ] , and use integration by part , then @xmath146 re - arranging , we get @xmath147 by the cauchy - schwartz inequality , @xmath148 so @xmath149 it implies that the operator @xmath150 is non - negative on @xmath135 $ ] , so ode comparison implies that @xmath151 next we prove theorem [ main theorem3.1 ] . let us first assume that @xmath37 is orientable , and hence is two - sided ( c.f . * proposition 3.5 ) ) . pick two points @xmath109 , @xmath152 on @xmath39 such that the distance @xmath153 achieves the diameter . consider the geodesic balls @xmath154 and @xmath155 of @xmath39 . as the morse index of @xmath37 is one , at least one geodesic ball , say @xmath154 is a stable minimal surface with smooth boundary . then the proof of proposition [ diameter estimates for stable surface ] implies that @xmath156 . when @xmath37 is non - orientable , its double cover @xmath51 is then a two - side minimal surface of morse index 1 by theorem [ main theorem3 ] , so we finish the proof . in this section we define what we mean by a family of 1-cycles and a sweepout of a manifold by 1-cycles . following @xcite , @xcite for @xmath157 let @xmath158 denote the space of all @xmath28-tuples @xmath159 of lipschitz maps of @xmath160 $ ] to @xmath1 such that @xmath161 with the following topology . using nash embedding theorem we embed @xmath1 isometrically into a eucledian space and define the distance by the formula @xmath162 . we let @xmath163 . observe that the induced topology on @xmath164 is finer than the flat topology on the space of integer 1-cycles @xcite and that the length functional is continuous on @xmath164 . let @xmath165 denote the space of all constant curves ( points ) . let @xmath166 be an @xmath167polyhedral complex and @xmath168 be a subcomplex of @xmath166 . we say that a family of 1-cycles @xmath169 is a sweepout of @xmath1 if * for each @xmath170 the cycle @xmath171 has zero length * @xmath172 is not contractible in @xmath173 as noted above , constructing a family of cycles that is continuous in @xmath164 is a stronger result than constructing a continuous family of flat cycles . alternatively , we could start with a family that is only continuous in the flat norm and then use almgren - pitts theory of almost minimizing 1-dimensional varifolds as it was done in @xcite ( see @xcite for a much more general and detailed exposition ) . the first step in this construction is to modify this family so that there is no cancellation of mass , that is , so that the family is continuous both in the mass and the flat topology , while controlling the mass of the maximal slice . this is analogous to the procedure that we perform in the proof of lemma [ main_cycles1 ] . both approaches lead to a min - max sequence that converges in the appropriate sense ( in @xmath158 topology in the first case or as varifolds for almgren - pitts theory ) to a stationary geodesic net . the min - max argument for families in @xmath158 has the advantage of being simpler ( see @xcite and appendix of @xcite ) and this is why we follow this approach . it is often of interest to consider families of cycles that arise as fibers of a certain well - behaved mapping from @xmath1 to a space of lower dimension . for example , if @xmath174 is a 2-dimensional closed surface and @xmath175 $ ] is an onto morse function then we consider the family @xmath176}$ ] . for this family we have the following result . [ morse to sweepout ] there exists a sweepout of @xmath174 by 1-cycles @xmath177 } \subset \gamma$ ] , such that for each @xmath178 the image of @xmath171 coincides with @xmath179 except possibly for a finite collection of points . let @xmath28 be the maximum number of connected components of @xmath179 , @xmath180 $ ] . we define @xmath181 by induction on the number of singular points of @xmath129 . since @xmath129 is an onto morse function we have that @xmath182 consists of @xmath183 points @xmath184 . define the first @xmath185 components of @xmath186 to be @xmath187)= p_i$ ] and for @xmath188 set @xmath189)= p_{k_0}$ ] . let @xmath190 be the smallest critical value of @xmath129 . for each @xmath191 and @xmath192 we can define a homotopy @xmath193 , so that @xmath194)$ ] is a connected component of @xmath195 ( we have @xmath196 ) . let @xmath197 be a critical point of @xmath129 and assume that @xmath171 is defined for all @xmath198 . the singularity that occurs at @xmath197 may be a destruction / creation of a connected component of @xmath179 or a splitting / merging of two connected components . in the first case we proceed in the obvious way . consider the case of a splitting . choose a small @xmath199 so that @xmath129 has no critical values in @xmath200 . since the number of connected components of @xmath179 is less than @xmath28 for @xmath201 there exists a constant component @xmath202 of @xmath171 let @xmath203 denote the component that splits into two at time @xmath197 . for @xmath204 we deform homotopically @xmath205 to the point @xmath206 . for @xmath207 we homotop @xmath203 and @xmath205 so that they form two arcs of the same connected component of @xmath179 and their endpoints approach the singular point of @xmath129 at @xmath197 . for @xmath208 we can split the two arcs into two distinct connected components . this ensure continuity of the family of cycles in @xmath209 . we deal with a merging of two components in a similar way . this finishes the construction of a family of 1-cycles @xmath177 } \subset \gamma(\sigma)$ ] corresponding to @xmath129 . to see that this family is a sweepout recall a result of almgren @xcite about homotopy groups of the space of 1-cycles . let @xmath210 denote the space of integer flat cycles in @xmath1 . almgren constructed an isomorphism between homotopy groups of the space of cycles @xmath211 and homology groups @xmath212 of the space @xmath1 . let @xmath213 be the map that sends each cycle in @xmath214 to the corresponding cycle in @xmath210 . we will show that @xmath215 is not contractible in the space of flat cycles and hence , it is not be contractible in @xmath209 . recall the definition of almgren s map . we pick a fine subdivision @xmath216 of @xmath160 $ ] and for each @xmath217 consider a lipschitz chain @xmath218 filling @xmath219 . the almgren s map then sends the family @xmath220 to the homology class of @xmath221 . as long as the area of each filling @xmath218 is sufficiently close to the minimal area , the exact choice of @xmath218 does not matter . hence , by our construction of @xmath220 it is immediately clear that @xmath221 represents the generator of @xmath222 . in this paper we construct two families of cycles in a riemannian 3-manifold of positive ricci curvature : the family of fibers of a mapping @xmath223 and the corresponding family @xmath220 of cycles in @xmath209 , where @xmath171 and @xmath179 coincide except possibly for a finite number of constant curves . because of this correspondence we will often talk about them as if they are the same family . in this section we will prove a parametric version of the following theorem of balacheff and sabourau @xcite . [ surface sweepout ] let @xmath174 be a riemannian surface of genus @xmath224 , area @xmath95 and with a ( possibly empty ) piecewise smooth boundary of length @xmath225 . there exists a morse function @xmath226 $ ] , such that @xmath227 , and the length of @xmath228 is at most @xmath229 for all @xmath230 $ ] and some universal @xmath231 . this version of the theorem is slightly more general and the upper bound for constant @xmath18 is better than in @xcite . these improvements follow from the methods of @xcite , @xcite and @xcite . to each boundary component of @xmath174 we glue a very small disc to obtain a closed surface @xmath232 of area @xmath233 . let @xmath234 denote the unique surface of constant gaussian curvature @xmath235 , @xmath26 or @xmath236 , which lies in the conformal class of @xmath174 and let @xmath237 be a conformal diffeomorphism . for each @xmath238 we will construct a morse function @xmath239 , such that @xmath240 , and the length of @xmath228 is at most @xmath241 for all @xmath230 $ ] . * step 1 . * choose @xmath242 much smaller than the injectivity radius of @xmath232 and suppose @xmath243 . it follows from the systolic inequality on surfaces that the genus of @xmath244 is @xmath26 . let @xmath245 be a small number . by lemma 19 from @xcite if @xmath246 is sufficiently small then there exists a morse function @xmath247 $ ] with @xmath248 and the length of fibers at most @xmath249 . * now we prove that for every open subset @xmath250 there exists a relative cycle @xmath251 ( with @xmath252 ) , subdividing @xmath244 into two subsets of area at least @xmath253 and such that @xmath254 . let @xmath255 be the smallest radius such that there exists a ball @xmath256 ( on the constant curvature conformal representative ) , such that @xmath257 . we consider two cases . suppose first that @xmath258 then it it follows by comparison with a constant curvature space that the annulus @xmath259 can be covered by @xmath260 discs of radius @xmath255 in @xmath234 . let @xmath261 be such that @xmath262 is maximazed . it follows from the choice of @xmath261 and @xmath255 that @xmath263 . using the length - area method ( cf . @xcite ) we find that there exists a cycle in the image of the annulus of length @xmath264 now suppose @xmath265 . in this case we use an idea of @xcite ( cf . proof of lemma 3.3 in @xcite ) of considering systems of balls of radius @xmath236 . let @xmath28 be the smallest integer , such that there exist @xmath266 with @xmath267 . assume that @xmath268 are chosen in such a way that this quantity is maximized . observe that by the choice of @xmath28 we have @xmath269 . for each @xmath270 , let @xmath271 be a collection of @xmath260 balls of radius @xmath236 covering the annulus @xmath272 . by our choice of @xmath273 for each @xmath274 we have @xmath275 . as in the case @xmath276 we use coarea formula and the length - area method to find a relative cycle @xmath251 in the image of @xmath277neighbourhood @xmath278 of length at most @xmath279 . cycle @xmath251 subdivides @xmath244 into two parts each of area at least @xmath280 . for a surface of genus @xmath224 and constant curvature @xmath236 , @xmath26 or @xmath235 we have @xmath281 . we conclude that the length of @xmath251 is bounded above by @xmath282 . * step 3 . * let @xmath283 and @xmath284 be two open subsets of @xmath174 with disjoint interiors and let @xmath285 $ ] , @xmath286 , be morse functions , such that @xmath287 and @xmath288 . by lemma 18 in @xcite there exists a morse function @xmath289 $ ] with @xmath290 and @xmath291 . * step 4 . * to finish the proof we combine three steps above in the following inductive argument . we claim that for each integer @xmath292 and for every @xmath244 with @xmath293 there exists a morse function @xmath239 , such that @xmath240 , and the length of @xmath228 is at most @xmath294 for all @xmath230 $ ] . by step 1 the claim is true for @xmath295 . suppose that every subset of @xmath232 of area at most @xmath296 satisfies the inductive hypothesis . by step 2 we can subdivide @xmath244 into two subsets of area @xmath297 by a cycle of length @xmath298 . by step 3 there exists a desired morse function with length of fibers @xmath299 . @xmath300 can be chosen much smaller than @xmath301 . this finishes the inductive argument . we state a parametric version of this result . [ thm : parametric ] let @xmath174 be a surface of genus @xmath224 and let @xmath302}$ ] be a smooth family of riemannian metrics on @xmath174 , such that the area @xmath303 for some constant @xmath95 . there exists a continuous family of functions @xmath304 , @xmath180 $ ] , such that for each @xmath305 we have that @xmath306 is a @xmath236-cycle in @xmath307 of length at most @xmath308 . all functions @xmath309 are morse except possibly for finitely many with a finite number of ordinary cusp singularities . there exists a corresponding continuous family of sweepouts in @xmath310 ( see section [ intro to 1-cycles ] ) . theorem [ thm : parametric ] easily follows from the following proposition conjectured by a. nabutovsky in a conversation with one of the authors . [ prop : homotopy ] let @xmath174 be a closed riemannian surface and let @xmath311 , @xmath312 , be two morse functions , such that the length of @xmath313 is bounded above by @xmath225 for all @xmath261 . then @xmath314 and @xmath315 are homotopic through functions @xmath309 , @xmath316 , with the length of @xmath306 bounded above by @xmath317 for all @xmath261 and @xmath178 and arbitrarily small @xmath242 . functions @xmath309 are all morse , except for finitely many , which may have a finite number of ordinary cusp singularities . there exists a corresponding continuous family of sweepouts in @xmath209 . note that cusp singularities occur in a generic 1-parameter family of functions , so we can not expect them all to be morse . without any loss of generality we may assume that @xmath318 $ ] for @xmath312 . let @xmath319 , @xmath320 $ ] , be the 1-parametric family of 1-cycles given by the level sets of @xmath321 . since @xmath321 is a morse function we have that the family @xmath322 is a foliation with finitely many singular leaves . the singularities are either constant curves or curves with transverse self - intersections . the set of singular points of @xmath315 and @xmath323 is finite . we can make finitely many local perturbations so that the lengths of preimages changes by less than @xmath246 and the singular points of @xmath315 and @xmath323 are disjoint . moreover , by thom s multijet transversality we can make a perturbation so that each pre - image of @xmath315 has only finitely many tangency points with preimages of @xmath323 . to summarize , without any loss of generality we may assume that families @xmath324 satisfy the following : 1 . if @xmath325 is a singular leaf and @xmath261 is a singular point of @xmath325 then it is disjoint from singular points of @xmath326 for all @xmath327 . 2 . for each @xmath328 all but finitely many @xmath326 intersect @xmath325 transversely ; @xmath326 and @xmath325 have at most one non - transverse touching away from the singular points of @xmath326 and @xmath325 . let @xmath180 $ ] . we define a 1-parameter family @xmath329 , @xmath320 $ ] , as follows . for @xmath330 we set @xmath331 . for @xmath332 we set @xmath333 ) \cup f_0^{-1}\big((-\infty , \frac{s - t}{1-t}])\big)$ ] . the family of cycles @xmath334 is illustrated on figure [ fig : parametric ] . note that for each @xmath327 and @xmath178 the cycle @xmath329 consists of arcs of cycles @xmath335 and @xmath336 . moreover , by properties ( 1 ) and ( 2 ) each @xmath329 is a collection of finitely many piecewise smooth curves with a finite number of corners . we can smooth out the corners by a small perturbation . using properties ( 1 ) and ( 2 ) we can perturb family @xmath329 so that it contains only finitely many singular leaves . hence , we have constructed a continuous family of sweepouts by 1-cycles in @xmath310 . now we would like to construct a continuous family of morse functions with short fibers . note that the above family @xmath337 ( for @xmath178 fixed ) is not a foliation because cycles @xmath338 and @xmath339 may overlap . also , we can not just glue together different fibers of @xmath314 and @xmath315 because they may correspond to different values . we can fix this as follows . choose @xmath340 sufficiently small , so that every ball of radius @xmath341 is @xmath342bilipschitz diffeomorphic to a disc on the plane of the same radius . let @xmath245 be such that cycles @xmath343 and @xmath344 have length less than @xmath341 for all @xmath345 \cup [ 1-\delta,1]$ ] . let @xmath346 $ ] . for @xmath347)$ ] we set @xmath348 . consider a foliation of @xmath349)$ ] by cycles @xmath350 for @xmath351 $ ] . this foliation will be the level sets of @xmath309 on @xmath349)$ ] . we set @xmath352 for @xmath353 . in particular , for @xmath354)$ ] we have @xmath355 . for @xmath356 we have that @xmath357 everywhere except for a small set @xmath358)$ ] . we linearly homotop @xmath359 to @xmath360 . on @xmath361 the function now coincides with @xmath314 . since @xmath244 is contained in a collection of small discs @xmath342bilipschitz to discs in the plane , we can homotop the function to @xmath314 in @xmath244 while controlling the lengths of the fibers . when @xmath362 function @xmath309 coincides with @xmath315 everywhere except for a small set @xmath363)$ ] . similarly , we homotop the function to @xmath315 while controlling the lengths of the fibers . we now prove theorem [ thm : parametric ] . let @xmath18 be the constant from theorem [ surface sweepout ] . fix a small @xmath199 . subdivide @xmath160 $ ] into @xmath364 sufficiently small intervals @xmath365 $ ] , so that for any @xmath366 $ ] we have @xmath367 . for each @xmath217 let @xmath368 be a morse function from theorem [ surface sweepout ] with fibers of length at most @xmath369 . if @xmath370 $ ] then we have that the functions @xmath371 and @xmath372 have fibers of length at most @xmath373 . by proposition [ prop : homotopy ] there exists a family of morse functions @xmath374\}$ ] , such that @xmath375 and @xmath376 . moreover , for any @xmath377 $ ] the fibers of @xmath378 have length at most @xmath379 , when measured with respect to @xmath380 . we then set @xmath381 for @xmath370 $ ] . this finishes the proof of theorem [ thm : parametric ] . in this section we give a proof of theorem [ main_cycle ] . [ main_cycles1 ] given a @xmath83-dimensional riemannian manifold @xmath1 with an arbitrary metric of positive ricci curvature , there exists a smooth map @xmath5 with fibers of length at most @xmath6 . the family of pre - images @xmath382 corresponds to a continuous sweepout in @xmath209 . let @xmath1 be a 3-manifold of positive ricci curvature . by theorem [ main theorem3 ] we have the following two possibilities : case 1 . there exist a smooth function @xmath383 $ ] , such that the fibers of @xmath129 for @xmath384 form a family of smooth diffeomorphic surfaces of genus @xmath385 and area @xmath386 and @xmath387 and @xmath388 are graphs . decomposition into cycles of controlled length then immediately follows by theorem [ thm : parametric ] . the fact that these 1-cycle are continuous in @xmath209 follows from the fact that @xmath389 are continuous in the smooth topology . now we consider case 2 of theorem [ main theorem3 ] . let @xmath390 be a non - orientable min - max minimal surface as in the theorem . let @xmath391 be the genus of the double cover @xmath392 of @xmath234 . let @xmath393 be the set of all points at a distance @xmath178 from @xmath37 . we have that for a sufficiently small @xmath394 and all @xmath395 , the surface @xmath396 is bi - lipschitz diffeomorphic to the double cover @xmath51 of @xmath37 with bi - lipschitz constant @xmath397 . let @xmath398}$ ] denote the tubular neighborhood of @xmath37 . suppose we are interested in constructing a function @xmath5 with fibers forming a family of 1-cycles of controlled length and continuous in the _ flat norm _ , but not necessarily continuous in @xmath209 . then we can argue as follows . let @xmath399 \times \{0\ } \subset \mathbb{r}^2 $ ] be the morse function from theorem [ surface sweepout ] . composing with the covering map we obtain a map from the double cover @xmath400 \times \{0\}$ ] . let @xmath53 denote the manifold with boundary from theorem [ main theorem3 ] , such that interior of @xmath53 is isometric to @xmath401 and @xmath402 . by theorem [ thm : parametric ] we construct a function @xmath403 ^ 2 $ ] , such that the restriction of @xmath404 to @xmath405 is @xmath406 . we then define @xmath407 for all @xmath408 and @xmath409 for @xmath410 . in the remainder of the proof we will modify this construction in order to produce a family , which is continuous in @xmath209 . our argument is similar to constructions of families of cycles with no cancellation of mass in the works of almgren and pitts ( see @xcite ) . by theorem [ main theorem3 ] there exists a sweepout of @xmath411 by surfaces of controlled area . as in the first case we construct a map @xmath412 \times [ 0,1]$ ] with preimages of controlled length , and such that the preimages form a continuous family of @xmath236-cycles . moreover , we can do it in such a way so that @xmath129 restricted to @xmath413 is a morse function and @xmath414}$ ] is a family of @xmath236-cycles sweeping out @xmath413 . next we construct an extension of this map to @xmath244 . there exists a map @xmath415 \times [ -1,0]$ ] , such that the length of @xmath416 , @xmath417\times[-1 , 0]$ ] is at most @xmath418 ; @xmath419 \times 0 $ ] is a morse function ; and the family of cycles @xmath420\times [ -1 , 0]\}$ ] is continuous in @xmath214 . let @xmath421 be the projection map , i.e. @xmath109 is the identity map on @xmath37 and it sends @xmath422 to the unique point @xmath423 with @xmath424 for @xmath425 $ ] . let @xmath426 denote the restriction of @xmath109 to @xmath427 . observe that @xmath426 is locally @xmath428-bi - lipschitz . by theorem [ surface sweepout ] there exists a morse function @xmath429 , such that all preimages of @xmath14 are bounded in length by @xmath430 . we may assume that @xmath431 $ ] . let @xmath432 denote the @xmath236-cycle @xmath433 . for each @xmath320 $ ] we will define a function @xmath434 $ ] with fibers of controlled length . suppose first that @xmath435 is the pre - image of a regular value of @xmath14 . we can write @xmath435 as a union of finitely many disjoint embedded circles @xmath436 in @xmath234 . the set @xmath437 is diffeomorphic to a cylinder or a mobius band . we will construct a foliation of @xmath437 by 1-cycles and use it to define function @xmath438 . choose two distinct points @xmath261 and @xmath439 on @xmath440 and consider a small tubular neighborhood @xmath441 ( resp . @xmath442 ) of @xmath443 ( resp . @xmath444 ) in @xmath437 . we foliate @xmath441 by 1-cycles as depicted on figure [ fig : foliation](a ) . clearly we can define a smooth function from @xmath441 to @xmath160 $ ] , whose fibers are 1-cycles in the foliation and which is non - degenerate everywhere except for a saddle point at @xmath261 . call this type of foliation of @xmath441 saddle foliation_. similarly , figure [ fig : foliation](b ) depicts a foliation of @xmath442 and we define the corresponding function from @xmath442 to @xmath160 $ ] with a singularity of index @xmath445 at @xmath439 ( a maximum point for @xmath446 ) . call this type of foliation of @xmath442 as _ node foliation_. we can extend @xmath441 and @xmath442 so that they cover all of @xmath437 and extend the corresponding foliations and functions in the obvious way . observe that the lengths of the preimages are bounded above by @xmath447 . for @xmath448 we define the function @xmath449 \times [ -1,0]$ ] . hence , we described a construction of @xmath438 on each connected component of @xmath450 in the case when @xmath435 is non - singular . as @xmath327 varies we can vary @xmath261 and @xmath439 and the corresponding foliations continuously and extend the map @xmath438 to @xmath451 , where @xmath452 is the first singular value of @xmath14 after @xmath327 . and @xmath442 by fibers of function @xmath446 . ] let @xmath453 be a singular connected component of @xmath454 . consider the case when @xmath453 is a point . this occurs when we have a creation or a destruction of a connected component . suppose first that @xmath455 has one less connected components than @xmath456 . the case of creation of a connected component is treated similarly . we modify the function @xmath438 on a neighborhood @xmath457 } ) \subset u$ ] as follows . let @xmath458 denote a disc of radius @xmath246 . observe that there exists a diffeomorphism @xmath459 \rightarrow v$ ] , such that for a concentric circle @xmath460 of radius @xmath255 and @xmath461 $ ] we have @xmath462 . let us consider the set @xmath463 . for @xmath464 we define @xmath438 on @xmath463 to be equal to @xmath465 . in particular , if @xmath466 and @xmath467 then @xmath468 consists of two points . for @xmath469 we construct @xmath470 exactly as how we constructed @xmath446 on the pre - image under @xmath109 of a non - singular curve , but we scale the image so that @xmath438 is continuous along the diagonal @xmath471 . we do it as follows . fix @xmath472 and let @xmath473 . construct a foliation of @xmath72 by 1-cycles with two singularities as on figure [ fig : foliation ] . we then define @xmath438 on @xmath72 so that @xmath474 and its preimages are given by 1-cycles in the foliation . suppose now that @xmath475 is a figure-8 curve with a self - intersection at a point @xmath476 . this means that we either have a splitting of one component into two or a merging of two components into one ( for otherwise we could perturb this family of cycles so that no singularity would occur ) . we consider the case of two components merging and the argument for the other case is analogous . for @xmath477 let @xmath478 and @xmath479 be two components that merge into @xmath475 at time @xmath452 . we arrange @xmath480 on @xmath481 ( @xmath482 ) so that @xmath480 takes on its maximum at a point @xmath483 on @xmath481 with @xmath483 converging to the self - intersection point of the figure-8 : @xmath484 as @xmath485 . for @xmath486 we obtain two node foliations ( figure [ fig : foliation](b ) ) glued along @xmath487 . observe that we can arrange the foliations to match properly so that they correspond to preimages of a smooth function on @xmath486 and , moreover , we can extend it to a foliation of @xmath488 by cycles satisfying @xmath489 upper bound on their lengths . this foliation has two node foliations and two saddle foliations . we can arrange for one node foliation and one saddle foliation to collide and annihilate in a saddle - node bifurcation . this can be done without increasing the length bounds by more than @xmath490 . ( thinking of @xmath446 as a height function one can picture saddle - node bifurcation as smoothing out a hill ) . this completes the construction of @xmath438 . it is clear from the construction that the corresponding family of cycles is continuous in @xmath209 . from the construction in the proof it follows that @xmath491 , @xmath320 $ ] , is a family of @xmath236-cycles sweeping out @xmath413 . moreover @xmath438 has only finitely many singularities on @xmath413 , all of them non - degenerate . by proposition [ prop : homotopy ] @xmath491 and @xmath492 can be connected by a family of sweepouts of controlled length . after a small perturbation this produces the desired map from @xmath1 to @xmath493 with fibers of controlled length . we now use theorem [ main_cycle ] to give alternative proofs of results of gromov and nabutovsky - rotman stated in corollary [ systolic ] . first , let @xmath494 . by theorem [ main_cycle ] there exists a function @xmath495 ^ 2 $ ] and a continuous sweepout @xmath496 ^ 2 } \subset \gamma$ ] of @xmath1 by 1-cycles of length at most @xmath6 . corollary [ systolic ] follows from the lemma below . a connected component of some cycle @xmath171 in the sweepout that we constructed is a non - contractible loop in @xmath1 . the idea of the proof of this lemma was suggested to us by nabutovsky and rotman ( see also @xcite , where a version of this lemma is proved for sweepouts of 2-dimensional tori ) . for contradiction we assume that every connected component of @xmath171 is contractible for all @xmath178 . let @xmath497 ^ 2 $ ] be the map constructed in theorem [ main_cycle ] . from @xcite recall the definition of a connected map @xmath498 associated to @xmath129 . the set @xmath499 is defined as the quotient of @xmath1 by the equivalence relation @xmath500 if @xmath261 and @xmath439 are in the same connected component of @xmath179 for some @xmath180 ^ 2 $ ] and @xmath498 is the quotient map . there is a unique map @xmath501 ^ 2 $ ] , such that @xmath502 . observe that @xmath499 must be connected since @xmath1 is . also , by our construction of @xmath129 we can endow @xmath499 with the structure of a polyhedral complex , such that for an interior point @xmath261 of every face we have that every pre - image @xmath503 is a simple closed curve in @xmath1 and for each point @xmath261 contained in the 1-skeleton of @xmath499 the pre - image @xmath503 is a point or a closed curve with a finite number of self - intersections . we claim that @xmath499 is simply - connected . indeed , since @xmath504 is a connected map every loop in @xmath499 lifts to a loop in @xmath1 , so the induced map on the fundamental groups is surjective . on the other hand , the cohomology ring structure of @xmath21 forces the induced map on homology to be trivial . let @xmath505 denote the universal cover of @xmath1 and @xmath506 be the covering map . consider the composition @xmath507 . by our assumption for each @xmath261 we have that @xmath503 is a contractible closed curve in @xmath1 , so @xmath508 consists of @xmath28 disjoint closed curves in @xmath505 . let @xmath509 be the connected map associated to @xmath510 . we obtain that @xmath511 is a @xmath512fold covering space for @xmath499 . this contradicts the fact that @xmath499 is simply - connected . now by applying brikhoff curve - shortening process we obtain a closed geodesic in @xmath1 of length at most @xmath6 proving corollary [ systolic ] for the case when @xmath513 . if @xmath1 is homeomorphic to a 3-sphere then by the min - max argument described in @xcite existence of a sweepout , which is continuous in @xmath209 by short 1-cycles implies existence of a geodesic net with the desired length bound . the first open question is the relation between the almgren - pitts min - max minimal surface and the simon - smith ( also colding - de lellis ) min - max minimal surface in @xmath514 with positive ricci curvature ( or even in any 3-manifold with positive ricci curvature ) . the almgren - pitts minimal surface , which we use in this paper , has area bounded by the @xmath515-power of the volume up to a universal constant , and genus @xmath2 ; while the simon - smith min - max minimal surface has genus @xmath26 ( heegaard genus for @xmath516 ) but no a priori area bound in terms of the volume of the ambient manifold . it is then a natural question to compare them . the second open question is whether we could have diameter bound for the whole min - max family constructed in theorem [ main theorem3 ] when we assume the scalar curvature lower bound instead of just getting a diameter bound for the min - max surface as in theorem [ main theorem3.1 ] . these two questions are related to the problem of finding an upper bound for the length of the shortest non - trivial closed geodesic in manifold @xmath514 . the methods of this paper produce a sweepout of @xmath514 by short 1-cycles which yields a stationary geodesic net in @xmath514 of controlled length , but they do not give any bound for the length of the shortest closed geodesic . for this purpose one would need to consider sweepouts by loops instead of 1-cycles . if a manifold @xmath514 admits a sweepout by 2-spheres or 2-tori of controlled area @xmath95 and diameter @xmath517 then it seems plausible that using methods of @xcite one could bound the length of the shortest closed geodesic in @xmath514 in terms of @xmath95 and @xmath517 . f. almgren , _ the homotopy groups of the integral cycle groups _ , topology 1 ( 1962 ) 257 - 299 . f. almgren , _ the theory of varifolds , mimeographed notes _ , princeton , 1965 . f. balacheff , s. sabourau , _ diastolic and isoperimetric inequalities on surfaces _ , _ ann . sci . ecole norm . _ 43 ( 2010 ) 579 - 605 . a. carlotto , _ generic finiteness of minimal surfaces with bounded morse index _ , b. colbois , d. maerten , _ eigenvalues estimate for the neumann problem of bounded domain _ , j. geom . , 18(4):1022 - 1031 , 2008 . t. colding and w. minicozzi , _ a course in minimal surfaces _ , graduate studies in mathematics volume 121 , american mathematical society , 2011 . e. calabi , j. cao , _ simple closed geodesics on convex surfaces _ , j. diff . 36 ( 1992 ) , 517 - 549 . t. colding and c. de lellis , _ the min - max construction of minimal surfaces _ , surveys in differential geometry , vol . viii ( boston , ma , 2002 ) , 75 - 107 , int . press , somerville , ma , 2003 . t. frankel , _ on the fundamental group of a compact minimal submanifold_. ann . 83 ( 1966 ) 68 - 73 . k. frensel , _ stable complete surfaces with constant mean curvature _ , anais acad . bras . , 60 , 115 - 117 , 1988 . p. glynn - adey and y. liokumovich , _ width , ricci curvature , and minimal hypersurfaces _ , to appear in j. differential geometry , arxiv:1408.3656 . p. glynn - adey and z. zhu , subdividing three dimensional riemannian discs , preprint , arxiv:1508.03746 m. gromov , _ filling riemannian manifolds _ , j. differential geometry , volume 18 , number 1 ( 1983 ) , 1 - 147 . m. gromov , _ dimension , non - linear spectra and width . in geometric aspects of functional analysis _ , pages 132 - 184 . springer , 1988 . m. gromov , _ isoperimetry of waists and concentration of maps _ , geom . ( 2003 ) , no . 1 , 178 - 215 . m. gromov , _ singularities , expanders and topology of maps . i. homology versus volume in the spaces of cycles _ , geom . anal . 19 ( 2009 ) , 3 , 743 - 841 . m. gromov , _ singularities , expanders and topology of maps . part 2 : from combinatorics to topology via algebraic isoperimetry _ , geom . anal . 20 ( 2010 ) , no . 2 , 416 - 526 . m. gromov , _ morse spectra , homology measures , spaces of cycles and parametric packing problems _ , preprint . l. guth , _ the width - volume inequality _ , 17 , 1139 ( 2007 ) . l. guth , _ metaphors in systolic geometry _ , arxiv:1003.4247 . l. guth , _ the waist inequality in gromov s work _ , the abel prize 2008 - 2012 ( h. holden and r. piene , eds . ) , springer , 2014 , pp . 181 - 195 . d. ketover , f.c . marques and a. neves , the catenoid estimate and its geometric applications , arxiv:1601.04514 y. liokumovich , a. nabutovsky , r. rotman , _ contracting the boundary of a riemannian 2-disc _ , to appear in geom . funct . y. liokumovich , _ slicing a 2-sphere _ , journal of topology and analysis , 06(04):573 - 590 , 2014 . y. liokumovich , _ families of short cycles on riemannian surfaces _ , to appear in duke math . , arxiv:1410.8436 . f. marques and a. neves , _ rigidity of min - max minimal spheres in three - manifolds _ , duke math . j. 161 , 14 ( 2012 ) 2725 - 2752 . f. marques and a. neves , _ min - max theory and the willmore conjecture _ , ann . of math . ( 2 ) 179 , ( 2014 ) 683 - 782 . b. meeks , l. simon , and s .- t . yau , _ embedded minimal surfaces , exotic spheres , and manifolds with positive ricci curvature _ , ann . of math . 116 ( 1982 ) , 621 - 659 . y. memarian , _ on gromov s waist of the sphere theorem _ , 3 ( 2011 ) , no . 1 , 7 - 36 . a. nabutovsky , r. rotman , _ volume , diameter and the minimal mass of a stationary 1-cycle _ , geom . 14 ( 2004 ) no . 4 , 748 - 790 . a. neves , _ new applications of min - max theory _ , arxiv:1409.7537v1 . l. mazet and h. rosenberg , _ minimal hypersurfaces of least area _ , arxiv:1503.02938v2 . j. pitts , regularity and singularity of one dimensional stational integral varifolds on manifolds arising from variational methods in the large , symposia mathematics , vol . xiv , roma , italy , 1974 j. pitts , _ existence and regularity of minimal surfaces on riemannian manifold _ , mathematical notes 27 , princeton university press , princeton 1981 . p. papasoglu , e. swensen , _ a sphere hard to cut _ , preprint , arxiv:1509.02307 l. simon , _ lectures on geometric measure theory_. australian national university centre for mathematical analysis , canberra , 1983 . f. smith , on the existence of embedded minimal 2-spheres in the 3-sphere , endowed with an arbitrary riemannian metric , phd thesis 1982 , supervisor : leon simon . stphane sabourau , _ volume of minimal hypersurfaces in manifolds with nonnegative ricci curvature _ , journal fr die reine und angewandte mathematik ( crelle s journal ) , to appear . r. schoen , s.t . yau , _ existence of incompressible minimal surfaces and the topology of three dimensional manifolds with non - negative scalar curvature_. ann . math . 110 ( 1979 ) 127 - 142 . richard schoen and s .- t . yau , _ the existence of a black hole due to condensation of matter _ , commun . phys 90 , 575 - 579 ( 1983 ) . s. t. yau . _ nonlinear analysis in geometry _ , enseign ( 2 ) 33(1987 ) , 109 - 158 . _ min - max minimal hypersurface in @xmath518 with @xmath519 and @xmath520 _ , j. differential geometry , 100 ( 2015 ) 129 - 160 . = yevgeny liokumovich xin zhou + department of mathematics department of mathematics + imperial college london massachusetts institute of technology + south kensington campus 77 massachusetts avenue + london sw7 2az , uk cambridge , ma 02139 - 4307 , usa + email : [email protected] email : [email protected]

motivated by the discovery of young , massive stars in the nuclei of some liner/ `` transition '' nuclei such as ngc 4569 , we have computed photoionization models to determine whether some of these objects may be powered solely by young star clusters rather than by accretion - powered active nuclei . the models were calculated with the photoionization code cloudy , using evolving starburst continua generated by the the starburst99 code of . we find that the models are able to reproduce the emission - line spectra of transition nuclei , but only for instantaneous bursts of solar or higher metallicity , and only for ages of myr , the period when the extreme - ultraviolet continuum is dominated by emission from wolf - rayet stars . for clusters younger than 3 myr or older than 6 myr , and for models with a constant star - formation rate , the softer ionizing continuum results in an emission spectrum more typical of regions . this model predicts that wolf - rayet emission features should appear in the spectra of transition nuclei . while such features have not generally been detected to date , they could be revealed in observations having higher spatial resolution . demographic arguments suggest that this starburst model may not apply to the majority of transition nuclei , particularly those in early - type host galaxies , but it could account for some members of the transition class in hosts of type sa and later . the starburst models during the wolf - rayet - dominated phase can also reproduce the narrow - line spectra of some liners , but only under conditions of above - solar metallicity and only if high - density gas is present ( ) . this scenario could be applicable to some `` type 2 '' liners which do not show any clear signs of nonstellar activity .

astro-ph0003294

emission - line nebulae in galactic nuclei are generally considered to fall into three major categories : star - forming or nuclei , seyfert nuclei , and low - ionization nuclear emission - line regions , or liners . the formal divisions between these classes are somewhat arbitrary , as the observed emission - line ratios of nearby galactic nuclei fall in a continuous distribution between liners and seyfert nuclei and between liners and nuclei ( e.g. , * ? ? ? traditionally , liners have been defined as those nuclei having emission - line flux ratios which satisfy the relations [ ] @xmath33727/ [ ] @xmath35007 @xmath4 and [ ] @xmath36300/ [ ] @xmath35007 @xmath5 @xcite . it is possible to construct alternative but practically equivalent definitions , based on other line ratios , which can be applied to datasets that do not include the wavelengths of the [ ] , [ ] , or [ ] lines ( e.g. , * ? ? ? a sizeable minority of galactic nuclei has emission - line ratios which are intermediate between those of `` pure '' liners and those of typical regions powered by hot stars ; these galaxies would be classified as liners except that their [ ] @xmath36300 line strengths are too small in comparison with other lines to meet the formal liner criteria . objects falling into this category have been dubbed `` transition '' galaxies by @xcite , and although this nomenclature is somewhat ambiguous we adopt it here for consistency with the spectroscopic survey of @xcite . that survey defined the transition class in terms of the following flux ratios : [ ] @xmath35007/h@xmath6 @xmath7 3 , + 0.08 @xmath8 [ ] @xmath36300/h@xmath9 @xmath7 0.17 , + [ ] @xmath36583/h@xmath9 @xmath10 0.6 , + [ ] @xmath116716 , 6731 /h@xmath9 @xmath10 0.4 . @xcite have used the term `` weak- [ ] liners '' to refer to galaxies having [ ] @xmath36583/h@xmath9 @xmath12 ( typical of liners ) but which have [ ] @xmath36300/h@xmath9 @xmath13 . this category is essentially identical to the transition class of @xcite , and we will refer to these galaxies as transition objects in this paper . according to the survey results of @xcite , this transition class accounts for 13% of all nearby galaxies , making them about as numerous as seyfert nuclei . the hubble type distribution of transition galaxies is intermediate between that of liners , which are most common in e / s0/sa galaxies , and that of nuclei , which occur most often in hubble types later than sb @xcite . roughly @xmath14 of galaxies with hubble types ranging from s0 to sbc belong to the transition class . there is not a consensus , however , as to whether these transition objects should be regarded as star - forming nuclei , as accretion - powered active nuclei , or as composite objects powered by an agn and by hot stars in roughly equal proportion . there is a large body of literature on the subject of the excitation mechanism of liners which is relevant to the similar transition class . a variety of physical mechanisms has been proposed to explain the emission spectra of liners , including shocks , photoionization by a nonstellar ultraviolet ( uv ) and x - ray continuum , and photoionization by hot stars . ( see filippenko 1996 for a review . ) the possibility that liners ( and seyfert nuclei as well ) might be photoionized by starlight was raised by @xcite , who suggested that very hot ( @xmath15 k ) wolf - rayet ( w - r ) stars in a metal - rich starburst could give rise to an ionizing continuum with a nearly power - law shape in the extreme - uv . more recent atmosphere models have indicated substantially lower temperatures for w - r stars , however , casting doubt on the warmer hypothesis @xcite . subsequent photoionization models have attempted to explain liner and transition - type spectra as resulting from massive main - sequence stars . @xcite found that the spectra of weak- [ ] liners could be explained in terms of photoionization by o3o4 stars having effective temperatures of @xmath16 k , at ionization parameters of @xmath17 to @xmath18 . @xcite carried this line of argument farther , proposing that genuine liner spectra could be generated by early o stars with @xmath19 k , provided that a high - density component ( @xmath20 @xmath2 ) is present in the nlr ; the high densities are needed to boost the strengths of high critical - density emission lines , most notably [ ] @xmath36300 . similar conclusions were reached by @xcite , who explored the effects of absorption by ionized gas as a means to harden the effective ionizing spectrum . recent observations , particularly in the uv and x - ray bands , have provided convincing evidence that many liners are in fact agns , particularly the `` type 1 '' liners which have a broad component to the h@xmath9 emission line ( for a recent review see * ? ? ? the possibility has remained , however , that some liners and transition nuclei are powered entirely by bursts of star formation . an important shortcoming of the model calculations performed by @xcite and @xcite is that the ionizing continua used as input were those of single o - type stars ; these studies did not address the question of whether a liner or transition - type spectrum could result from the the _ integrated _ ionizing continuum of a young stellar cluster . compared with these single - star models , the contribution of late - o and b stars will soften the ionizing spectrum , making the emission - line ratios tend toward those of normal regions . w - r stars , on the other hand , will harden the ionizing spectrum during the period when these stars are present , roughly @xmath21 myr after the burst . another drawback of the o - star models is that they require the presence of stars with effective temperatures higher than are thought to occur in regions of solar or above - solar metallicity , in order to produce a liner or transition - type spectrum rather than an region spectrum . their applicability to galactic nuclei is therefore somewhat unclear . other mechanisms have been proposed for generating liner or transition - type spectra . shock excitation by supernova remnants in an aging starburst may give rise to some transition objects ; the nucleus of ngc 253 is a likely candidate for such an object @xcite . also , post - agb stars and planetary nebula nuclei will produce a diffuse ionizing radiation field which could be responsible for the very faint liner emission ( with h@xmath9 equivalent widths of @xmath22 ) observed in some ellipticals and spiral bulges @xcite . an alternate possibility is that the transition galaxies may simply be composite systems consisting of an active nucleus surrounded by star - forming regions . for a galaxy at a distance of 10 mpc , for example , a 2-wide spectroscopic aperture will include regions within 50 pc of the nucleus . galaxies having emission lines both from a liner nucleus and from surrounding star - forming regions , in roughly equal proportions , will appear to have a transition - type spectrum . this interpretation was advocated by @xcite as the most likely explanation for the majority of transition galaxies , and is consistent with the observed hubble type distribution for the transition class . other authors have similarly contended that transition galaxies are agn/ region composites , based on optical line - profile decompositions @xcite and near - infrared spectra @xcite . two of the 65 transition nuclei observed in the @xcite survey have a broad component to the h@xmath9 emission line , indicating the likely presence of an agn , and it is probable that many more transition nuclei contain obscured agns which were not detected in the optical spectra . on the other hand , radio observations do not appear to support the composite agn / starburst interpretation . in a vla survey of nearby galactic nuclei , nagar et al . ( 1999 ) find compact , flat - spectrum radio cores in more than 50% of liner nuclei , but in only 6% ( 1 of 18 ) of transition objects . this discrepancy suggests that the simple picture of an ordinary liner surrounded by star - forming regions may not apply to the majority of transition objects . recent results from the _ hubble space telescope _ ( _ hst _ ) have shed new light on the question of the excitation mechanism of transition nuclei . as shown by @xcite , the uv spectrum of the well - known transition nucleus in ngc 4569 over 1200 - 1600 is virtually identical to that of a w - r knot in the starburst galaxy ngc 1741 , indicating that o stars with ages of a few myr dominate the uv continuum . @xcite find that the nuclear star cluster in ngc 4569 is producing sufficient uv photons to ionize the surrounding narrow - line region , a key conclusion which provides fresh motivation to study stellar photoionization models . the brightness of the ngc 4569 nucleus , and the consequently high s / n observations that have been obtained , make it one of the best objects with which to study the transition phenomenon . the recent availability of the starburst99 model set @xcite has prompted us to reexamine the issue of ionization by hot stars in liners and transition nuclei . these models give predictions for the spectrum and luminosity of a young star cluster , for a range of values of cluster age , metal abundance , and stellar initial mass function ( imf ) properties . using the photoionization code cloudy @xcite in combination with the starburst99 model continua , we have calculated the expected emission - line spectrum of an region illuminated by a young star cluster , to test the hypothesis that some liners and transition nuclei may be powered by starlight . similar calculations have been performed by @xcite , but for the physically distinct case of metal - poor objects representing galaxies . other examples of photoionization calculations for regions using evolving starburst continua are presented by @xcite , @xcite , and @xcite . before describing the photoionization modeling , we review the properties of ngc 4569 , as it is among the best - known examples of the transition class . ngc 4569 is a virgo cluster spiral of type sab , with a heliocentric velocity of @xmath23 km s@xmath24 , and we assume a distance of 16.8 mpc for consistency with the catalog of @xcite . its nucleus is remarkably bright for a non - seyfert , and so compact in the optical that @xcite suspected it to be a foreground galactic star . it is also an unusually bright uv source , with the highest 2200 luminosity of the liners and transition objects observed by @xcite and @xcite . _ hst _ faint object spectrograph ( fos ) spectra show that the uv continuum is dominated by massive stars , with prominent p cygni profiles of @xmath31549 , @xmath31400 , and @xmath31240 @xcite . the uv spectrum is nearly an exact match to the spectrum of one of the starburst knots in the w - r galaxy ngc 1741 , an object with a likely age in the range 36 myr @xcite . the optical spectrum of the ngc 4569 nucleus is dominated by the light of a - type supergiants , providing additional evidence for recent star formation @xcite . one key result of the @xcite study was the conclusion that the nuclear starburst in ngc 4569 is producing sufficient numbers of ionizing photons to power the narrow - line region , assuming that the surrounding nebula is ionization - bounded , _ even without correcting for the effects of internal extinction on the uv continuum flux_. in fact , there appears to be substantial extinction within ngc 4569 , as demonstrated by the uv continuum slope as well as the presence of deep interstellar absorption features @xcite . @xcite derive an internal reddening of @xmath25 mag from the h@xmath9/h@xmath6 ratio , while @xcite estimate a uv extinction of @xmath26 mag at 1300 by comparison of the observed uv slope with the expected spectral shape of an unreddened starburst . despite the fact that ngc 4569 is often referred to as a liner , and in some cases presumed to contain an agn on the basis of that classification , there is no single piece of evidence which conclusively demonstrates that an agn is in fact present at all . the _ hst _ images and spectra are all consistent with the nucleus being a young , luminous , and compact starburst region . no broad - line component is detected on the h@xmath9 emission line @xcite , and no narrow or broad emission lines are visible at all in the uv spectrum other than the p cygni features that are generated in o - star winds @xcite . only the optical emission - line ratios point to a possible agn classification . in a thorough study of optical and _ iue _ uv spectra , @xcite concluded that there was at best weak evidence for the presence of an agn in ngc 4569 , and that any agn continuum component , if present , must have an unusually steep spectrum . furthermore , while the nucleus of ngc 4569 is certainly extremely compact , the uv and optical _ hst _ images show that the nucleus is not dominated by a central point source . at 2200 , the nucleus appears extended in wfpc2 images with fwhm sizes of 13 and 9 pc along its major and minor axes @xcite . optical wfpc2 images have been discussed recently by @xcite , who state that the nucleus is unresolved by _ we have obtained these same images from the _ hst_archive . while the nucleus is certainly compact , we find that it is clearly extended even at the smallest radii . a 12-second , cr - split exposure in the f547 m ( @xmath27-band ) filter is unsaturated and allows a radial profile measurement . we find a fwhm size of 14 pc by 8 pc along the major and minor axes , consistent with the size of the nuclear cluster measured at 2200 . @xcite estimated that at most 23% of the nuclear uv flux could come from a central point source . from the equivalent widths of stellar - wind features in the uv spectrum , @xcite give a similar upper limit of @xmath28 to the possible contribution of a truly featureless continuum to the observed uv flux . x - ray observations of ngc 4569 with _ rosat _ have revealed a source coincident with the nucleus which is unresolved at the 2 resolution of the hri camera @xcite . this does not necessarily indicate that an agn is present , however , as the optical / uv size of the starburst core is an order of magnitude smaller than the hri resolution . asca _ observations show that the x - ray emission is extended over arcminute scales in both the hard ( 27 kev ) and soft ( 0.52 kev ) bands @xcite . interestingly , the compact source seen in the _ rosat _ image is detected only in the soft _ asca _ band , while there is no detectable contribution from a compact , hard x - ray source . the spectral shape of the compact soft x - ray component is consistent with an origin either in an agn or in x - ray binaries @xcite , but the lack of a compact hard x - ray source argues against the agn interpretation . if an agn is present , it must be highly obscured even at hard x - ray energies , with an obscuring column of @xmath29 @xmath30 @xcite . in radio emission , vla observations show that the ngc 4569 nucleus is an extended source with a size of 4 and no apparent core @xcite , in contrast with the compact , agn - like cores found in some liners @xcite . shock excitation has often been considered as a mechanism to power the narrow emission lines in liners . however , the lack of narrow emission features in the uv spectrum of ngc 4569 argues against shock - heating models for this object , as existing shock models generally predict strong uv line emission ( e.g. , * ? ? ? shock - excited filaments in supernova remnants show strong emission in high - excitation uv lines such as @xmath31549 and @xmath31640 ( e.g. , * ? ? ? * ; * ? ? ? * ) which are altogether absent from the ngc 4569 spectrum . similarly , the shock - excited nuclear disk of m87 @xcite has a high - excitation uv line spectrum which bears no resemblance to the ngc 4569 spectrum . from an analysis of infrared spectra , @xcite proposed that the ngc 4569 nucleus is powered by an 811 myr - old starburst , by a combination of stellar photoionization and shock heating from supernova remnants . while this hypothesis may be applicable to some liners and transition galaxies , the uv spectrum of ngc 4569 shown by @xcite is inconsistent with a burst of such advanced age , as the p cygni features of , , and would have disappeared from a single - burst population after about 6 myr . the overall picture emerging from these observations is that the ngc 4569 nucleus is a compact , luminous , and young starburst . the only reason to invoke the presence of an agn at all would be to explain the higher strengths of the low - ionization forbidden lines in comparison with values observed in normal nuclei . if it were indeed possible for a young starburst to produce transition or liner - type emission lines in the surrounding gas , then there would be no reason to consider agn models for ngc 4569 . as discussed by @xcite and @xcite , the key ingredient necessary for generating a liner or transition - type emission - line spectrum is an ionizing continuum which is harder than that produced by typical clusters of ob stars . a harder continuum will produce a more extended partially - ionized zone in the surrounding region , boosting the strength of the low - ionization lines which are typical of liner spectra : [ ] @xmath36300 , [ ] @xmath33727 , [ ] @xmath3@xmath36548,6583 , and [ ] @xmath3@xmath36716,6731 . to represent the ionizing continuum of a young starburst , we have chosen the starburst99 model set ; we refer the reader to @xcite for the details of the methods used to construct these models . briefly , the starburst99 code employs the geneva stellar evolution models of @xcite , with enhanced mass - loss rates , for high - mass stars . atmospheres are represented by the models compiled by @xcite and @xcite . figures 112 of @xcite display the spectral energy distributions of the starburst99 model clusters for a range of burst ages and for a variety of initial conditions . from the figures , some important trends are readily apparent . during the first 2 myr after an instantaneous burst , the continuum is dominated by the hottest o stars , and there is essentially no emission below 228 , corresponding to the ionization energy of he@xmath31 . the appearance of w - r stars during the period 35 myr after the burst results in a dramatic change in the uv continuum , as these stars emit strongly in the he@xmath32 continuum below 228 . from 6 myr onwards , the w - r stars disappear and the uv continuum rapidly fades and softens as the burst ages . only the models with an upper mass limit of @xmath33 = 100@xmath34 generate the hard , w - r - dominated uv continuum ; the model sequences with @xmath33 = 30@xmath34 do not generate significant numbers of photons below 228 for any ages because the progenitors of w - r stars are not present in the initial burst . constant star - formation rate models with @xmath33 = 100@xmath34 form w - r stars continuously after 3 myr , but the overall shape of the uv continuum is softer than in the instantaneous burst models , because of the continuous formation of luminous o stars . these results provide a useful starting point for the photoionization calculations . if it is possible for the region surrounding a young cluster to resemble a liner or transition object , then this is most likely to occur when the ionizing continuum is hardest , when w - r stars are present during @xmath35 myr after a burst . very massive stars ( in the range 30100 @xmath34 or greater ) must be present in the burst or else the requisite w - r stars will not appear . the formation of w - r stars is enhanced at high metallicity , so the ability to generate a liner or transition - type spectrum may be a strong function of metal abundance as well as age . to create grids of photoionization models , we fed the uv continua generated by the starburst99 models into the photoionization code cloudy ( version 90.04 ; * ? ? ? for each time step , a grid of models was calculated by varying the nebular density and the ionization parameter , which is defined as the ratio of ionizing photon density to the gas density at the ionized face of a cloud . real liners and transition nuclei are likely to contain clouds with a range of values of density and ionization parameter , and more general models incorporating density and ionization stratification can be constructed as linear combinations of these simple single - zone models . all models were run with the following range of parameters : burst age from 1 to 10 myr at increments of 1 myr , with log @xmath36 ranging from @xmath37 to @xmath38 at increments of 0.5 , and a constant density ranging from log ( @xmath39/@xmath2 ) = 2 to 6 at increments of 1 . as a starting point , we computed a grid for an instantaneous burst with an imf having a power - law slope of @xmath40 , @xmath33 = 100 @xmath34 , solar metallicity in stars and gas , and a single plane - parallel slab of gas with no dust ; we will refer to this as model grid a. the solar abundance set was taken from @xcite and @xcite . other grids were computed as variations on this basic parameter set , with the following modifications made in different model runs : a constant star - formation rate ; metallicity 0.2 , 0.4 , or 2@xmath41 in both stars and gas ; and spherical geometry for the nebula . to assess the effects of the highest - mass stars , we also ran custom model grids , via the starburst99web site , with @xmath42 and 120 @xmath34 . the depletion of heavy elements onto grains can result in marked changes to the emergent emission - line spectrum of an region , both by removing gas - phase coolants from the nebula and by grain absorption of ionizing photons , which will modify the effective shape of the ionizing continuum . in metal - rich regions , these effects will tend to boost the strengths of the low - ionization emission lines relative to the dust - free case @xcite . to assess the effects of dust in transition nuclei , we calculated additional model grids which included dust grains with a galactic ism dust - to - gas ratio along with the corresponding gas - phase depletions . the dusty models were all calculated using the solar abundance set for the undepleted gas . dust grains were assumed to have the optical properties of galactic ism grains , as described by @xcite , @xcite , and @xcite . from the cloudy output files , we tabulated the strengths relative to h@xmath6 of the major emission lines which are prominent in liners . the calculations were performed under the assumption that the region is ionization bounded . for this case , the outer extension of the cloud was set to be the radius at which @xmath43 falls to 4000 k , beyond which essentially no emission is generated in the optical or uv lines . as a test , we ran a grid of models with the stopping temperature set to 1000 k , and we verified that the emission - line ratios were essentially identical to the default case of 4000 k stopping temperature . we also verified that the important diagnostic line ratios differed by @xmath44 dex between the spherical and plane - parallel cases when all other input parameters were unmodified , and all results discussed in this paper refer to the plane - parallel models . in the calculations , the longest timescales for atomic species to reach equilibrium were of order @xmath45 years , much shorter than the evolution timescale of the stellar cluster , justifying the assumption that each time step of the cluster evolution could be used independently to calculate the nebular conditions . table [ table1 ] gives a summary of the model parameters , for the model grids which appear in the following discussion . the model results are displayed in figures [ oratio][ohimetal ] . to compare the model outputs with the observed properties of a variety of galaxy types , we have used the emission - line data compiled by @xcite . this catalog has the advantages of a homogeneous classification system , small measurement aperture ( @xmath46 ) , and careful starlight subtraction to ensure accurate emission - line data . in order to reduce confusion and to keep the sample of comparison objects to a reasonable number , we included only objects with unambiguous classifications as , liner , transition , or seyfert . objects with borderline or ambiguous classifications , such as `` liner / seyfert , '' were excluded for clarity . the comparison sample was further reduced by excluding galaxies in which any of the emission lines h@xmath9 , h@xmath6 , [ ] @xmath35007 , [ ] @xmath36300 , [ ] @xmath36583 , or [ ] @xmath3@xmath36716,6731 was undetected or was flagged as having a large uncertainty in flux ( `` b '' or `` c '' quality flags ) . the measured line ratios are corrected for both galactic and internal reddening . figure [ oratio ] plots the ratio [ ] @xmath35007/h@xmath6against [ ] @xmath36300/h@xmath9 at a burst age of 4 myr , for the solar - metallicity model grids a , b , c , and d. the model results are plotted for a density of @xmath47 @xmath2 , as an approximate match to the density of @xmath48 @xmath2 measured for ngc 4569 @xcite . the diagram shows that the instantaneous burst models ( a and b ) are a good match to the line ratios of the transition nuclei , for log @xmath49 . the dusty models from grid b fall more centrally within the region defined to contain transition objects , but the models without dust still closely match transition nuclei having lower [ ] @xmath36300/h@xmath9 ratios . figures [ nratio ] and [ sratio ] show the corresponding diagrams for [ ] @xmath36583/h@xmath9 and for [ ] @xmath116716 , 6731/h@xmath9 , respectively . in both cases we find that the single - burst models span the region occupied by transition nuclei in the diagnostic diagrams . in the constant star - formation rate models ( c and d ) , the uv continuum remains softer than in the w - r - dominated phase of the instantaneous burst models , because of the ongoing formation of luminous o stars . as a result , the low - ionization emission lines are significantly weaker than in the instantaneous burst models at 4 myr . these constant star - formation rate sequences are a reasonable match to the region of nuclei in the diagram , and for [ ] /h@xmath9and [ ] /h@xmath9 the agreement with nuclei is improved at lower densities of @xmath50 @xmath2 , a value more typical of nuclei . these constant star - formation rate models are probably appropriate for galaxies having spatially extended , ongoing star formation in their nuclei . we note that the model curves shown in the figures should not be expected to follow the locus of nuclei in each plot , despite the fact that the models are generated with a starburst continuum . the range of line ratios observed in regions is primarily a sequence in metal abundance @xcite , while our models are shown as sequences in @xmath36 for a given metallicity and density . another point to note about the diagrams is that some of the transition galaxies fall outside the region nominally defined for transition objects , particularly in the [ ] /h@xmath9 ratio ( figure [ oratio ] ) . these galaxies were classified as transition objects by @xcite on the basis of meeting the majority of the classification criteria . similarly , some overlap can be seen in the diagrams between the regions occupied by liners and seyfert nuclei ; this again reflects the fact that galaxies span a continuous range in the values of these emission - line ratios . the variation of [ ] line strength as a function of density is shown in figure [ odensity ] for model grid a at @xmath51 myr . for the range of densities considered ( @xmath52 to @xmath53 @xmath2 ) , the models closely overlap the transition region in the diagram at log @xmath49 . introducing ism depletion and dust grains to the nebula primarily increases the [ ] /h@xmath9 ratio at low density . as expected , the [ ] /h@xmath9 ratio increases with @xmath39 up to densities of @xmath54 @xmath2 , while at densities approaching the critical density of the @xmath36300 transition ( @xmath55 @xmath2 ) this ratio saturates and begins to turn over . figure [ oage ] shows the [ ] /h@xmath9 ratio as a function of burst age ( at @xmath56 @xmath2 ) for model grid a , for ages of 2 to 6 myr . this diagram highlights the dramatic changes that w - r stars generate in the surrounding nebula . from 3 to 5 myr after the burst , when w - r stars dominate the uv continuum , the harder ionizing continuum boosts the strength of [ ] by an order of magnitude and the model sequences appear adjacent to the transition region , with relatively little evolution in the line ratios during this period . as the burst ages beyond 6 myr , the [ ] /h@xmath9 ratio continues to fall , and the emission - line strengths drop rapidly as the ionizing continuum softens and its luminosity decreases . at @xmath57 , the wr - dominated phase occurs slightly later , during time steps 4 , 5 , and 6 myr . the [ ] /h@xmath9 and [ ] /h@xmath9 ratios have a similar dependence on burst age , and are displayed in figures [ nage ] and [ sage ] , respectively ; these results are quite similar to the calculations presented by @xcite to illustrate the effects of the w - r continuum on the [ ] /h@xmath9 ratio . while the [ ] /h@xmath9 ratios in the models at @xmath58 are too low by @xmath59 dex to fit within the nominal transition region , they still closely match those transition nuclei having relatively low values for [ ] /h@xmath9 , and the low - ionization line strengths can be further enhanced by the inclusion of dust and depletion ( as in figure [ oratio ] ) . one puzzling aspect of figure [ oage ] is that for ages outside the range 35 myr , the models predict [ ] /h@xmath9 ratios too low to match the majority of nuclei . @xcite and @xcite discuss this same problem in the context of low - metallicity starburst galaxies . they propose that shocks generated by supernovae and stellar winds provide the additional [ ] emission , without making a significant contribution to the [ ] or [ ] line strengths . shocks could play a similar role in transition galaxies , as in the model of @xcite , although the lack of high - excitation uv line emission in transition nuclei is problematic for the shock hypothesis . a higher upper mass cutoff alleviates this problem to some extent , at least for very young bursts . increasing @xmath33 to 120 @xmath34 boosts [ ] /h@xmath9 by @xmath60 dex for ages of @xmath61 myr . due to the very short lifetimes of the highest - mass stars , however , the @xmath33 = 120 , 100 , and 70 @xmath34 model grids result in identical emission - line spectra from @xmath62 myr onward . metal abundance is an additional parameter which must be considered . the models displayed up to this point were all calculated for a solar abundance set , while the nuclei of early - type spirals are likely to have enhanced heavy - element abundances . as discussed by @xcite , the continuum shortwards of 228 is strongest in the high - metallicity models , because the increased mass - loss rates lead preferentially to the formation of w - r stars at high metal abundance . figure [ olometal ] plots the [ ] /h@xmath9 ratio for abundances of @xmath63 = 0.2 , 0.4 , 1 , and 2 @xmath41 , at a density of @xmath47 @xmath2 and @xmath51 myr . from this diagram , it is clear that solar or higher abundances are necessary to match the [ ] strengths of the transition nuclei ; at lower abundances the line ratios are a better match to those of the high - excitation ( low - metallicity ) nuclei . figure [ ohimetal ] displays the density dependence of the [ ] /h@xmath9 ratio for the @xmath64 model grid ; by comparison with the solar - metallicity model grid in figure [ odensity ] , the higher abundances result in a lower - excitation spectrum with enhanced [ ] emission , due to the harder extreme - uv continuum . it would be advantageous to compare the model results with a wider variety of emission lines . unfortunately , measurements of other optical emission lines are scarce for transition nuclei . the ho et al . survey did not include the [ ] @xmath33727 line , and there is no other homogeneous catalog of [ ] measurements for transition galaxies . to be consistent with a liner or transition - type classification , a model calculation must result in the flux ratio [ ] @xmath33727 / [ ] @xmath35007 @xmath65 . in fact , all of the models with log @xmath66 do satisfy this criterion . thus , any of our models which is consistent with the ho et al . liner or transition classification criteria is also consistent with the original @xcite criterion for the [ ] / [ ] ratio in liners . the relative strengths of uv lines such as ] @xmath32326 , ] @xmath31909 , and @xmath31549 can provide further diagnostics , but none of these lines is detected in ngc 4569 @xcite . the only other transition nucleus having _ hst _ uv spectra available is ngc 5055 , and its spectrum appears to be devoid of uv emission lines as well @xcite . we ran one additional model grid to test whether different model atmospheres for o stars would lead to different results . the starburst99continua were calculated using stellar atmosphere models compiled by @xcite , which are based on the @xcite model set for the massive stellar component . the recent costar model grid of @xcite , which includes non - lte effects , stellar winds , and line blanketing for o stars , makes dramatically different predictions for the ionizing spectra . as shown by @xcite , the costar models yield a luminosity in the he@xmath32 continuum which is four orders of magnitude greater than that predicted by the kurucz models , for the most massive o stars which dominate the uv luminosity at burst ages of @xmath67 myr . in the costar - based models the photon output of the cluster below 228 is essentially constant from 0 to 5 myr . to investigate the effects of this harder o - star continuum on the emission - line spectra , we ran a grid of models using the evolving starburst continua computed by @xcite with the costar atmospheres . model parameters were the same as for model grid a except that an upper mass limit of 120 @xmath34 was used . we find that using the costar atmospheres has a relatively minor effect on our results . in comparison with the starburst99-based models having @xmath33 = 120@xmath34 , the costar model grid yields an increase in the [ ] /h@xmath9 ratio of @xmath68 dex for @xmath69 myr , while [ ] /h@xmath9 and [ ] /h@xmath9 are essentially unaffected . during the period @xmath70 myr , the costar - based models result in an region spectrum , demonstrating that w - r stars are still required in order to generate liner or transition - type line ratios . the strength of the [ ] emission lines at 7291 and 7324 is often used as a diagnostic of dust and depletion , because in the absence of depletion these lines are predicted to be strong in photoionized gas ( e.g. , * ? ? ? * ; * ? ? ? ( the @xmath37291 line is a cleaner diagnostic since @xmath37324 is blended with [ ] @xmath37325 . ) however , for a 4 myr - old burst with nebular conditions of @xmath56 @xmath2 , @xmath58 , and an undepleted solar abundance set , our calculations yield a maximum prediction of only 0.2 for the ratio of [ ] @xmath3@xmath37291 , 7324 to h@xmath6 . only at very low ionization parameters ( @xmath71 ) does the [ ] emission become stronger than h@xmath6 . since h@xmath6 is only barely visible in the spectra of many transition objects ( prior to careful starlight subtraction , at least ) , typical observations may not have sufficient sensitivity to detect faint [ ] lines in these objects . high - quality spectra of liners do not show [ ] emission @xcite , indicating that ca is likely to be depleted onto grains in these objects , but similar data are not generally available for transition nuclei . if the [ ] lines are found to indicate a high level of depletion onto dust grains in transition nuclei , this would also provide a further argument against shock - heating models , as shocks will tend to destroy grains ( e.g. , * ? ? ? the results shown in the preceding figures demonstrate that the starburst models are in fact able to reproduce the major diagnostic emission - line ratios of transition nuclei with reasonable accuracy , during the period @xmath72 = 35 myr when w - r stars are present . for a density of @xmath45 @xmath2 and an age of 4 myr , the solar - metallicity models with and without depletion bracket the range of values observed in real transition nuclei for the line ratios [ ] /h@xmath9 , [ ] /h@xmath9 , and [ ] /h@xmath9 . we do not attempt to fine - tune a model to produce an exact match with the spectrum of ngc 4569 , but the basic solar - metallicity dust - free model at @xmath51 myr with @xmath73 @xmath2 and log @xmath74 closely fits the observed [ ] /h@xmath9 ratio , while overpredicting [ ] /h@xmath9 by @xmath60 dex and underpredicting [ ] /h@xmath9 by @xmath75 dex . we emphasize that the starburst models are only able to produce transition - type spectra for the case of an instantaneous burst ; that is , when the burst duration is shorter than the timescale for evolution of the most massive stars . multiple - burst populations can only yield a transition spectrum if the dominant population is @xmath0 myr old and the older or younger bursts do not contribute significantly to the ionizing photon budget . models with a constant star - formation rate produce region spectra at all ages , as the softer ionizing continua do not produce sufficient [ ] @xmath36300 emission in the surrounding region to match transition - type spectra . the parameter which is most important for determining the hardness of the ionizing continuum is the number ratio of w - r stars to o stars , which exceeds @xmath76 during the w - r - dominated phase in the starburst99 models at solar metallicity , and approaches or exceeds unity at @xmath57 . in the constant star - formation rate models at solar metallicity , the w - r / o ratio levels off at @xmath77 after about 4 myr . the compact size of the ngc 4569 nucleus is consistent with the requirement that the burst duration must be brief ( @xmath78 myr ) in order to generate a transition - type spectrum . the fwhm size of the starburst core in ngc 4569 is only @xmath79 pc . for such a burst to occur in @xmath78 myr would require a propagation speed for star formation of only @xmath79 km s@xmath24 . in fact , the typical velocities in the ngc 4569 nucleus are much greater than 10 km s@xmath24 : the [ ] @xmath36583 line has a velocity width of 340 km s@xmath24 @xcite . thus , the ngc 4569 nucleus could represent the result of a single , rapid burst of star formation . although our results suggest that transition galaxy spectra may be attributed to a starburst with a high w - r / o - star ratio , the demographics of transition nuclei and nuclei indicate that many transition galaxies are probably not formed by this mechanism . in the starburst99 models , the w - r - dominated phase in an instantaneous burst lasts for @xmath80 myr ( i.e. , 3 time steps in the calculations ) . an region surrounding an instantaneous burst will be visible for @xmath81 myr , after which the emission lines will fade rapidly ( e.g. , * ? ? ? * ) . thus , for an instantaneous burst population , the transition phase and the nucleus phase will have approximately equal lifetimes . if all nuclei consisted of instantaneous burst stellar populations with nebular conditions conducive to the formation of transition - type spectra , then nuclei and transition nuclei should be roughly equal in number . in reality , it is likely that a large fraction of star - forming nuclei contain multiple bursts of star formation and/or conditions of low density or low metallicity , so all star - forming nuclei should not be expected to evolve through a transition - type phase . although it is difficult to make specific predictions , it is probably safe to conclude that for a given hubble type , transition nuclei generated solely by starbursts should be considerably less numerous than ordinary nuclei . the statistics compiled by @xcite provide a basis for comparison . in early - type galaxies ( e and s0 ) , transition nuclei outnumber nuclei by a 3-to-1 margin . only for hubble types sb and later do nuclei begin to outnumber transition nuclei by a factor of 2 or more . the most straightforward interpretation of this trend is that in early - type host galaxies , the majority of transition nuclei are actually agn/ region composites , as proposed by @xcite and others . at intermediate and late hubble types , the population of transition nuclei may consist of both composite objects and `` pure '' starbursts evolving through the w - r - dominated phase . the presence of transition nuclei in a small fraction ( @xmath82 ) of elliptical galaxies @xcite presents a particularly intriguing problem . the ho et al . survey detected five transition nuclei in ellipticals but not a single case of an elliptical galaxy hosting an nucleus . given that the models which have been considered for transition nuclei involve star formation , either alone or in combination with an agn , this observation is rather puzzling . perhaps faint agns in elliptical nuclei can produce transition - type spectra without substantial star formation activity . four of the five transition nuclei found in ellipticals by @xcite have borderline or ambiguous spectroscopic classifications , however , so `` pure '' transition nuclei in ellipticals are evidently quite rare . given these results , one might expect to see transition - type emission spectra in some fraction of disk regions in spiral galaxies , but in fact such spectra are never found . single - burst models for disk region spectra are only compatible with observed line ratios for model ages of @xmath70 myr @xcite , as the harder ionizing spectrum after 3 myr makes the models overpredict the strengths of the low - ionization lines . bresolin et al . suggest that either current stellar evolution models are at fault , or that disk regions are disrupted before reaching an age of 3 myr , in which case the w - r phase would not be observed in the nebular gas . an alternate ( and perhaps more attractive ) possibility is that the majority of nuclei , as well as disk regions , are better described by the models with constant star - formation rate , or contain multiple bursts of star formation with an age spread of a few myr , which would result in a spectrum similar to the constant star - formation rate models . for understanding the physical nature of transition objects , the observational challenge is to search for any unambiguous signs of nonstellar activity . detection of broad h@xmath9 emission , or a compact source of hard x - ray emission with a power - law spectrum , would provide evidence for an agn component . high - resolution optical spectra ( from _ hst _ ) could provide a means to spatially resolve a central agn - dominated narrow - line region from the surrounding starburst - dominated component . since direct evidence for accretion - powered nuclear activity in transition nuclei is generally lacking , it should not be assumed that any given transition object actually contains an agn unless observations specifically support that interpretation . one further effect that should be considered in starburst models in the future is photoionization by the x - rays generated by the starburst . x - ray binaries and supernova remnants will provide high - energy ionizing photons , resulting in a spatially extended source of soft x - ray emission as observed in the nucleus of ngc 4569 @xcite , for example . ( the massive main sequence stars will contribute only a negligible amount to the total x - ray luminosity of a starburst ; see helfand & moran 1999 . ) photoionization by x - rays will naturally lead to an enhancement of the low - ionization forbidden lines , and this could contribute to the excitation of some transition galaxies . the strength of [ ] @xmath36300 is the key distinguishing factor between liners and transition nuclei , and matching the observed strength of this line is the major challenge of starburst models for liners . our calculations show that liner spectra can only be generated by the starburst99 clusters under a very specific and limited range of circumstances . model grids a and b , while matching the [ ] / h@xmath9 ratio of transition nuclei quite well , do not overlap at all with the main cluster of liners in figure [ oratio ] , even at high densities and even when depletion and dust grains are included . only grid g with @xmath57 is able to replicate the high [ ] / h@xmath9 ratios of most liners , and only during @xmath83 46 myr and at densities of @xmath39 @xmath84 @xmath2 . in agreement with previous models , we find that values of log @xmath85 @xmath86 to @xmath87 reproduce the observed [ ] / h@xmath9ratios of liners . however , at such high densities the models underpredict the strengths of [ ] and [ ] relative to h@xmath9 . single - zone models require @xmath88 @xmath2 to match the [ ] /h@xmath9 ratios of liners and @xmath89 @xmath2 for [ ] /h@xmath9 . agreement with liner spectra can be achieved with a simple two - zone model , in which high - density and low - density components are present , similar to the scenario proposed by @xcite . as an example , a two - component model constructed from grid a containing gas at ( @xmath90 @xmath2 , @xmath91 ) and at ( @xmath92 @xmath2 , @xmath93 ) produces emission - line ratios which are consistent with all the liner classification criteria of both the @xcite and @xcite systems , if the two density componenets are scaled so as to contribute equally to the total h@xmath6 luminosity . as a local comparison , observations of near - infrared emission indicate the presence of clouds having @xmath94 @xmath2 in the galactic center region @xcite , so it is plausible that other galactic nuclei may contain ionized gas at similarly high densities even in the absence of an observable agn . a starburst origin for some liner 2 nuclei would provide a natural explanation for the lower values of the x - ray / h@xmath9 flux ratio seen in these objects , in comparison with agn - like liner 1 nuclei @xcite . it seems unlikely , however , that many liners are generated by this starburst mechanism . about 15% of liners are known to have a broad component of the h@xmath9 emission line , indicating a probable agn @xcite . by analogy with the seyfert population , a much larger fraction of liners is likely to have broad - line regions which are either obscured along our line of sight , or are simply too faint to be detected in ground - based spectra against a bright background of starlight . many liners show signs of nuclear activity that can not be explained by stellar processes : compact flat - spectrum radio sources or jets , compact x - ray sources with hard power - law spectra , or double - peaked broad balmer - line emission , for example . as an increasing body of observational work supports the idea that many liners are in fact agns , there is less incentive to consider purely stellar models for their excitation . demographic arguments , similar to those given above for transition nuclei , can be applied for the liner population . since the liner phase only occurs for instantaneous bursts at high density and high metallicity , the starburst scenario implies that nuclei should be considerably more numerous than starburst - generated liners for a given hubble type . while liners are common in early - type hosts , nuclei are not found in elliptical hosts and are seen in fewer than 10% of s0 galaxies this disparity is a strong argument against a starburst origin for those liners in early - type galaxies . in later hubble types the situation is less clear , however . nuclei occur in @xmath95 of spirals of type sc and later , while liners occur in just 5% of these galaxy types @xcite . it is conceivable that some of the liners in intermediate to late - type hosts could have a starburst origin , and this issue could be resolved by further uv and x - ray observations in the future . interestingly , the ho et al . survey did not find any examples of broad h@xmath9 emission in liners or transition objects with hosts of type sc or later ; perhaps star formation plays a more prominent role than accretion - powered activity in these objects . while a few liners show spectral features of young stars in the uv @xcite , the quality of the observational data is poor in comparison with the ngc 4569 uv spectrum , and it is difficult to set meaningful constraints on the age of the young stellar population . ngc 404 is a possible candidate for a starburst - generated liner , but in its uv spectrum the p cygni features are weak in comparison with ngc 4569 , indicating either an older burst population or dilution by a featureless agn continuum @xcite . the liners having uv spectral features from massive stars may also host obscured agns which can be detected in other wavebands . for example , the uv continuum of the liner ngc 6500 appears to have its origin in hot stars @xcite , but observations of a parsec - scale radio jet unambiguously demonstrate that nonstellar activity is occurring as well @xcite . the starburst models presented here run into two obvious problems . first , w - r galaxies are almost never known to have liner or transition - type spectra . second , liners and transition nuclei almost never show w - r features in their spectra . is there any way to reconcile the starburst models with these facts ? w - r galaxies are identified by the appearance of the 4650 blend in their spectra ( e.g. , * ? ? ? since the formation of w - r stars is enhanced at high @xmath63 , the strength of this feature relative to h@xmath6 increases dramatically with metallicity , from @xmath44 at @xmath96 to @xmath97 at @xmath98 @xcite . however , in the nuclei of early - type spirals where high metallicities are expected , a nuclear starburst will be surrounded by the old stellar population of the galactic bulge , making the detection of the w - r bump extremely difficult @xcite . most of the currently known w - r galaxies are late - type spirals or irregular galaxies @xcite in which the w - r bump is visible against the nearly featureless starburst continuum . when the w - r bump is detected in galaxies , its amplitude above the continuum level is generally far smaller than that of h@xmath6 or even h@xmath99(e.g . . in most of the liner and transition galaxy spectra in the catalog of @xcite , however , h@xmath6 barely appears and h@xmath99 is too weak to be visible at all prior to continuum subtraction . even in a high - metallicity environment , where the total intensity of the w - r bump can be comparable to that of h@xmath6 , the amplitude of the w - r bump above the continuum will be much lower than that of h@xmath6 because the flux in the w - r feature is spread over @xmath100 . thus , the detection of w - r emission in galactic nuclei is strongly biased toward late - type , bulgeless galaxies . in late - type or dwarf irregular galaxies where the w - r bump is visible , the w - r / o - star ratio is expected to be much smaller owing to the lower metallicity , and the resulting softer ionizing spectrum will tend to produce an region spectrum rather than a transition object . the gas density as a function of hubble type may play a role as well ; in a study of nuclei , @xcite find a weak trend toward lower nebular densities in later - type host galaxies . observational detection of w - r features in transition nuclei is perhaps the clearest test of the starburst models , if sufficiently sensitive observations can be obtained . the uv spectrum of ngc 4569 is consistent with an age of @xmath101 myr , an age at which w - r stars are expected to be present . previous optical spectra have not revealed the 4650 w - r bump in ngc 4569 , but further observations with high s / n and small apertures would be worthwhile . the lack of @xmath31640 emission in the uv spectrum of ngc 4569 is potentially a more serious problem since the burst population should dominate at short wavelengths . the models of @xcite predict an equivalent width of at least 2 in the w - r - generated @xmath31640 line during the period 36 myr for an instantaneous burst of solar or higher metallicity , while the observed upper limit of @xmath102 erg s@xmath24 @xmath30 @xcite corresponds to an equivalent width limit of @xmath103 . it should be noted , however , that the @xmath31640 line lies at the extremely noisy blue end of the fos g190h grating setting , in a region where detection of emission or absorption features is difficult . two w - r galaxies may provide useful points of reference for the starburst models . ngc 3367 is classified by @xcite as an nucleus on the basis of its [ ] /h@xmath9 and [ ] /h@xmath9 ratios , although its [ ] /h@xmath9 ratio of 0.83 is more consistent with a liner or transition - type classification and its emission lines are markedly broader than those of typical nuclei . @xcite describe ngc 3367 as a starburst - dominated transition object ( see also * ? ? ? the 4650 w - r bump was noted by @xcite , who also suggested a liner/ classification and a composite source of ionization . as a borderline nucleus / transition object with clear evidence for w - r stars , this object deserves further study , to determine whether there is indeed an agn or whether the enhanced low - ionization emission may be the result of ionization by the w - r population . the electron density of 835 @xmath2 measured from the [ ] doublet @xcite is also noteworthy , as this is among the highest densities found for an nucleus in the ho et al . survey . another intriguing object is the nucleus of ngc 6764 , which has been classified variously as a seyfert , a liner , and a starburst by different authors ( see * ? ? ? this galaxy exhibits prominent emission in the 4650 w - r blend @xcite . a recent study by @xcite demonstrates that the narrow emission lines are consistent with a liner classification , but there are no unambiguous signs of nonstellar activity in the nucleus . @xcite find that the nucleus contains @xmath104 w - r stars , and that the overall properties of the object are consistent with ionization by the starburst alone , rather than by a starburst / agn composite . if this conclusion is confirmed by further observations , ngc 6764 could be considered the best candidate for a liner photoionized by a starburst during its w - r - dominated phase . @xcite derived an age of 910 myr for the starburst in ngc 6764 based on near - infrared emission - line diagnostics , but this age is inconsistent with the presence of w - r stars , at least for the case of an instantaneous burst . compared with typical liners and transition nuclei , conditions for detection of w - r spectral features in these two objects are perhaps more favorable . both host galaxies are of late hubble types ( sbc for ngc 3367 and sbbc for ngc 6764 ) in comparison with the majority of liners and transition nuclei , so the level of contamination by the surrounding old stellar population is relatively low . furthermore ( and partly as a result of this ) , the emission - line equivalent widths in these two nuclei are relatively high for liners or transition nuclei . if the emission - line spectra of ngc 3367 or ngc 6764 had been superposed on a luminous early - type spiral bulge , the w - r emission might never have been noticed . the detection of w - r emission in a liner does not automatically imply a purely starburst origin for the emission lines , of course , since starbursts and agns are often known to coexist . mrk 477 is a well - known example of a seyfert 2 galaxy having a large population of w - r stars in its nucleus @xcite . in the context of the starburst model , however , the crucial test is to search for additional examples of liners or transition nuclei which exhibit a high ratio of w - r to o stars but no signs of accretion - powered activity . the most important limitation of these calculations comes from the accuracy of the input continua . the conclusion that the starburst models are able to reproduce transition or liner spectra under some circumstances depends crucially on the presence of w - r stars to provide a hard and luminous ionizing continuum . unfortunately , the continuum shape and luminosity of w - r stars in the extreme - uv band are quite uncertain , particularly in the he@xmath32 continuum where stellar winds have a dramatic effect . several up - to - date reviews of the numerous difficulties involved in modeling w - r spectra can be found in the volume edited by @xcite . the starburst99 model grid uses the w - r atmospheres of @xcite , which are calculated for a pure helium composition , but more recent atmosphere models are beginning to include the effects of line - blanketing , as well as clumping and departures from spherical symmetry . as discussed by @xcite , the w - r / o - star ratio is also extremely model dependent , and may be revised in future generations of models . this would have a direct effect on the strength of the extreme - uv continuum and consequences for the nebular emission lines . furthermore , the starburst99 models neglect binary evolution , although this is more likely to affect the w - r / o ratio at low metallicity . since the results of the transition - object models are highly dependent on the most uncertain portion of the w - r spectrum , new photoionization calculations should be computed to assess the impact of different w - r evolution and atmosphere models in the future . the shape of the imf in starburst regions is a subject of some debate , and the possible variation of the imf with metallicity is of particular importance for galactic nuclei , which are likely to have @xmath105 . @xcite and @xcite suggested that @xmath33 should be lower in regions of higher metal abundance , but this issue has not been settled definitively . star - count observations demonstrate that the imf slope and @xmath33 do not appear to vary with metallicity @xcite , at least for @xmath106 . while nebular diagnostics in galaxies are generally consistent with a salpeter imf with @xmath107 @xmath34 at subsolar metallicity ( e.g. , * ? ? ? * ) , at high metallicity the observational situation is somewhat ambiguous . @xcite find that the mean stellar temperature in regions decreases significantly with increasing @xmath63 , and that the @xmath35876 / h@xmath6 ratios of regions at @xmath108 are more consistent with @xmath33 = 30 @xmath34 than with @xmath33 = 100 @xmath34 . such a low value for @xmath33 would pose serious difficulties for any starburst models of liners and transition nuclei , as the massive progenitors of w - r stars would not be present . counterbalancing this trend , the strong tidal forces , turbulence , and magnetic field strengths in galactic nuclei may act to raise the jeans mass and favor the formation of more massive stars @xcite . in the galactic center , there are stars with initial masses of @xmath109 @xmath34 @xcite , and one galactic center object ( the pistol star ) may have @xmath110 as high as 200250 @xmath34 @xcite . thus , the proposed trend toward lower values of @xmath33 at high metallicity in disk regions may not apply to galactic nuclei . detailed comparison of the uv spectra of galaxies such as ngc 4569 with starburst population synthesis models can provide useful constraints on the population of high - mass stars in nuclear starbursts . despite these uncertainties , these photoionization models have a major advantage compared with previous generations of w - r or o - star models for liners and transition nuclei , in that the starburst99 models with standard parameters are constructed to represent the actual stellar populations in starbursts , to the best of current knowledge . previous o - star models @xcite required the presence of hypothetical , unusually hot stars in order to explain liner or transition spectra , and they did not address the evolution of the young stellar population at all . the starburst models presented here provide a more plausible mechanism to generate a transition - type spectrum , even if this model may apply only to a relatively small fraction of the population of transition galaxies . our primary conclusion is that for standard starburst parameters and for nebular conditions which may be typical of galactic nuclei , the starburst models are able to reproduce the important diagnostic emission - line ratios for liner/ transition galaxies , otherwise known as weak- [ ] liners . the key ingredient needed to generate a transition - type spectrum is a uv continuum dominated by w - r stars , a condition which occurs during @xmath111 35 myr after an instantaneous burst . a transition - type emission spectrum may thus be a phase in the evolution of some nuclear regions in which the ionizing continuum is generated by a single - burst stellar population . the models are also able to produce an [ ] / h@xmath9 ratio high enough to match liner spectra , but only for conditions of above - solar metallicity combined with the presence of high - density ( @xmath112 @xmath2 ) clouds . a sensitive search for w - r spectral features in transition nuclei would provide a test of this starburst scenario . this model may apply only to a small fraction of liners and transition nuclei ; many liners and some transition objects show clear signs of nonstellar activity , and the starburst models may not apply at all to objects in early - type host galaxies . further multiwavelength observations of transition nuclei will be of great utility for determining what fraction of them contain genuine active nuclei , and what fraction appear to be purely the result of stellar phenomena . research by a.j.b . is supported by a postdoctoral fellowship from the harvard - smithsonian center for astrophysics . this research was also supported financially by grant ar-07988.02 - 96a , awarded to j.c.s . by stsci , which is operated by aura for nasa under contract nas5 - 26555 . this work would not have been possible without the excellent software created and distributed by gary ferland and the cloudy team , and by claus leitherer and the starburst99 team . we also thank gary ferland for providing a helpful referee s report , claus leitherer for additional helpful comments on the manuscript , and daniel schaerer for supplying model starburst spectra in electronic form . line - ratio diagram of [ ] @xmath35007 / h@xmath6 against [ ] @xmath36300 / h@xmath9 , for model grids a ( solid line ) , b ( long - dashed line ) , c ( short - dashed line ) , and d ( dot - dashed line ) , at a burst age of 4 myr and @xmath56 @xmath2 . the input continuum has solar metallicity , imf power - law slope @xmath40 , and @xmath33 = 100 @xmath34 . the following description applies to this and all subsequent plots : the small squares along each model line correspond to the model grid points at log @xmath36 = @xmath113 , and @xmath37 , with @xmath36 increasing upward along the line . the points plotted represent galaxies from the ho et al . catalog , as follows : _ small circles _ : nuclei . _ squares : _ liner/ transition objects . ngc 4569 is represented by an open square . _ triangles : _ `` pure '' liners . _ crosses : _ seyfert nuclei . the dotted line encloses the region defined for transition nuclei according to the criteria of @xcite .

difference between bulk material and deposited film is shown to have an appreciable influence on the casimir force . analysis of the optical data on gold films unambiguously demonstrates the sample dependence : the dielectric functions of the films deposited in different conditions are different on the level that can not be ignored in high precision prediction of the force . it is argued that the precise values of the drude parameters are crucial for accurate evaluation of the force . the dielectric function of perfect single crystalline gold is discussed . it is used to establish the upper limit on the absolute value of the force . it is demonstrated that the force between films is smaller than that between bulk samples mainly due to the presence of voids and electron scattering on the grain boundaries in the films . the minimal reduction of the force is estimated as 2% for the smallest separations investigated in the most precise experiments . the other sample effects can reduce the force further but the correction is expected to be smaller than 1% .

cond-mat0412123

the casimir force @xcite between uncharged metallic plates attracts considerable attention as a macroscopic manifestation of the quantum vacuum @xcite . with the development of microtechnologies that routinely allow control of the separation between bodies smaller than 1 @xmath0 the force became a subject of systematic experimental investigation . modern high precision experiments were made using different techniques such as torsion pendulum @xcite , atomic force microscope ( afm ) @xcite , microelectromechanical systems ( mems ) @xcite and different geometrical configurations : sphere - plate @xcite , plate - plate @xcite , crossed cylinders @xcite . in most cases the bodies were covered with gold evaporated or sputter deposited to the thickness of 100 - 200 nm . in the only plate - plate configuration experiment @xcite the bodies were covered with chromium . one of the bodies was covered with copper in the mems experiment @xcite . the root mean square ( rms ) roughness of deposited metal films was varied from 30 - 40 nm as in the mems experiments @xcite to nearly atomically flat surfaces as in the crossed cylinders experiment @xcite . relatively low precision , 15% , in the force measurement was reached for the plate - plate configuration @xcite because of the parallelism problem . in the torsion pendulum experiment @xcite the force was measured with the accuracy of 5% . in the experiments @xcite it was found with 1% precision . in the most precise up to date experiment @xcite the improvement was achieved due to the use of the dynamical method @xcite . additionally the change in the resonance frequency of the mechanical oscillator was measured using the phase jump instead of the resonance behavior of the amplitude . in this way the force between gold and copper covered bodies was found with a relative accuracy of 0.25% . to draw any conclusion from the experiments one has to predict the force theoretically with the precision comparable with the experimental errors . it is a real challenge to the theory because the force is material dependent ; it is very difficult to fix the material properties since different uncontrolled factors are involved . in its original form , the casimir force @xcite @xmath1 was calculated between the ideal metals . it depends only on the fundamental constants and the distance between the plates @xmath2 . the force between real materials was found for the first time by lifshitz @xcite . the material properties enter in the lifshitz formula via the dielectric function @xmath3 at imaginary frequencies @xmath4 . correction to the expression ( [ fc ] ) is significant at the separations between bodies smaller than 1 @xmath0 . to calculate the force the lifshitz formula is used with the optical data taken from the handbooks @xcite . the data are available only up to some low - frequency cutoff @xmath5 . for good metals such as @xmath6 @xmath7 @xmath8 the data can be extrapolated to the lower frequencies @xmath9 with the drude dielectric function @xmath10 which includes two parameters : the plasma frequency @xmath11 and the relaxation frequency @xmath12 . these parameters can be extracted from the optical data at the lowest accessible frequencies . the exact values of the drude parameters are very important for the precise evaluation of the force . when two plates are separated by the distance @xmath2 , one can introduce a characteristic imaginary frequency @xmath13 of electromagnetic field fluctuations in the gap . the important fluctuations are those that have frequency @xmath14 . for @xmath15 the characteristic frequency is in the near infrared region . this qualitative consideration is often continued to the real frequency domain @xcite with the real characteristic frequency @xmath16 . because @xmath17 is in the near infrared it is stated that the low - frequency behavior of the dielectric function is not significant for the casimir force . it is not difficult to see that this consideration is wrong @xcite . the force depends on the dielectric function at imaginary frequencies @xmath3 . with the help of the dispersion relation @xmath18 can be expressed via the observable function @xmath19 : @xmath20 because for metals @xmath21 is large at low frequencies , the main contribution in the integral in eq . ( [ k - k ] ) is given by the low frequencies even if @xmath22 is in the visible range . the characteristic frequency is well defined on the imaginary axis @xmath23 but it is wrong to introduce it in the real frequency domain . this aspect is carefully discussed in this paper . two different procedures to get the drude parameters were discussed in the literature . in the first approach @xcite the plasma frequency was fixed independently on the optical data assuming that each atom gives one conduction electron ( for @xmath24 ) with the effective mass equal to the mass of free electron . the optical data at the lowest frequencies were used then to find @xmath25 with the help of eq . ( [ drude ] ) . in this way the parameters @xmath26 and @xmath27 have been found . in the second approach @xcite no assumptions were made and both of the parameters were extracted from the low - frequency optical data by fitting them with eq . ( [ drude ] ) . when the best data from ref . @xcite were used the result @xcite was close to that found by the first approach , but using different sources for the optical data collected in ref . @xcite an appreciable difference was found @xcite . this difference was attributed to the defects in the metallic films which appear as the result of the deposition process . it was indicated that the density of the deposited films is typically smaller and the resistivity is larger than the corresponding values for the bulk material . however , it was difficult to make definite conclusions without analysis of significant amount of the data on the deposited metallic films . in this paper this analysis is performed and it is proven that there is a genuine sample dependence of optical properties of gold . first of all it means that the gold films prepared in different ways will have different drude parameters . the change in the optical properties will lead to the sample dependence of the casimir force , which is carefully investigated in this paper . the main ideas were outlined before @xcite with the conclusion that the absolute value of the casimir force between gold films is of about 2% smaller than between bulk material at the smallest separations investigated in the experiments @xcite . here we come to the same conclusion but present much more details and facts . the paper is organized as follows . in sec . [ sec2 ] it is explained qualitatively and quantitatively why the exact values of the drude parameters are crucial for the precise calculation of the casimir force . existing optical data for gold are reviewed and analyzed in sec . it is unambiguously demonstrated that there is the sample dependence of the optical properties of gold films . in sec . [ sec4 ] the dielectric function of the perfect gold single crystal is discussed and the upper limit on the casimir force is deduced . in sec . [ sec5 ] a simple two parameter model is proposed allowing us to take into account the main reasons for the sample dependence of the casimir force : voids and grains in the films . the volume fraction of voids and the mean grain size are bounded from the available optical data and the smallest sample correction to the force is estimated . in sec . [ sec6 ] necessary experimental information on the films that has to be known for the precise evaluation of the force is briefly discussed . comparison between the afm experiment and prediction of this paper is given . our conclusions are summarized in the last section . let us consider the casimir force between similar metallic plates separated by a distance @xmath2 at a temperature @xmath28 . it depends on the material optical properties via the dielectric function @xmath29 at imaginary frequency @xmath30 @xcite : @xmath31 , \label{fpp}\ ] ] where @xmath32 is the wave vector along the plates ( @xmath33 ) . this formula includes the reflection coefficients for two polarization states @xmath34 and @xmath35 which are defined as @xmath36 here @xmath37 and @xmath38 are the normal components of the wave vectors in vacuum and metal , respectively : @xmath39 the imaginary frequencies here are the matsubara frequencies @xmath40 for most of the experimental configurations the force has to be calculated between sphere and plate . while the sphere radius @xmath41 is much larger than the minimal separation @xmath2 , the force can be approximately calculated with the help of the proximity force approximation @xcite that gives the following result : @xmath42 . \label{fsp}\ ] ] this approximation is good with the precision @xmath43 that is typically better or on the level of 0.1% . the material dielectric function @xmath44 is the main source of uncertainty for precise prediction of the casimir force . this function can not be measured directly but it can be expressed via the measurable function @xmath45 with the help of the well known dispersion relation ( [ k - k ] ) . the equation ( [ k - k ] ) shows that the casimir force is intimately connected with the dissipation in the material given by @xmath46 that is , indeed , the result of the fluctuation - dissipation theorem ( see discussion in ref . @xcite ) . the exponent @xmath47 in eqs . ( [ fpp ] ) , ( [ fsp ] ) shows that the main contribution to the force comes from the region @xmath48 . for this reason one can say that the field fluctuations with frequencies nearby the characteristic imaginary frequency @xmath49 give the main contribution to the force . this frequency is in the infrared region for typical separations @xmath50 investigated in the experiments . therefore , for precise calculation of the force we have to know the dielectric function at imaginary frequencies @xmath51 with the highest possible accuracy . this qualitative analysis is often extended to the real frequencies @xmath52 @xcite that is , of course , not true . as follows from the dispersion relation ( [ k - k ] ) real frequencies @xmath52 essentially different from @xmath53 dominate in @xmath54 . suppose that the material can be described by the drude dielectric function ( [ drude ] ) . then the integrand in eq . ( [ k - k ] ) has the form @xmath55 . this function is significant in the range @xmath56 . since @xmath12 corresponds to the far infrared region , we have to know the function @xmath57 with the highest precision in the far infrared but not in the infrared . let us consider this important conclusion in detail . throughout this paper the energy units will be used to measure the frequency . the following transition coefficients are helpful to transform it in the wavelength or in @xmath58 : @xmath59 for precise calculation with eq . ( [ fpp ] ) or ( [ fsp ] ) the dielectric function @xmath46 is usually taken from the handbooks . [ fig1 ] comprises the handbook data for gold @xcite . the solid line shows the actual data taken from two original sources : the points to the right of the arrow are those by thye @xcite and to the left by dold and mecke @xcite . for frequencies smaller than the cutoff frequency @xmath5 ( @xmath60 for these data ) the data are unavailable and @xmath57 has to be extrapolated into the region @xmath61 this is a delicate procedure because the contribution of the low frequencies dominates in @xmath62 . in the field of the casimir effect the following procedure has been adopted @xcite . the plasma frequency is calculated using the relation @xcite ( solid line ) . the arrow separates the data from the original sources @xcite ( on the left ) and @xcite ( on the right ) . the dotted line shows the extrapolation to low frequencies with the drude parameters indicated in the upper right corner . the cutoff frequency @xmath63 and the edge of the interband transition @xmath64 separate three frequency domains marked as _ 1 _ , _ 2 _ , and _ 3_.,width=325 ] @xmath65 where @xmath66 is the number of conduction electrons per unit volume , @xmath67 is the charge and @xmath68 is the effective mass of electron . assuming that each atom gives one conduction electron , that the effective mass coincides with the mass of free electron and using the bulk density of @xmath24 it was found @xmath26 . with this value of the plasma frequency the best fit of available low - frequency data with the drude function ( [ drude ] ) gave for the relaxation frequency @xmath27 @xcite . the dotted line in fig . [ fig1 ] shows the drude extrapolation with these parameters . one can separate three frequency regions in fig . [ fig1 ] marked as _ 1 _ , _ 2 _ , and _ 3_. in the high energy domain @xmath69 , where @xmath70 @xcite is the edge of interband absorption , the casimir force only weakly depends on the behavior of the dielectric function . however , as we will see in the next section , analysis of the interband absorption can give valuable information about voids in the metallic films . the region _ 2 _ extends from the cutoff frequency @xmath5 to the absorption edge @xmath64 . the curve in this region is also sensitive to the sample quality . it gives perceptible contribution to @xmath3 but what is more important is that this domain completely defines the drude parameters . these parameters will be used to extrapolate the curve to the low - frequency region _ 1 _ , @xmath71 . the low - frequency region gives the main contribution to @xmath3 and for this reason the precise values of the drude parameters are very important . let us write the dielectric function at imaginary frequencies in the following form : @xmath72 where the indexes 1,2 and 3 indicate the integration range in eq . ( [ k - k ] ) that are @xmath73 , @xmath74 , and @xmath75 , respectively . @xmath76 can be calculated explicitly using the drude function ( [ drude ] ) . it gives @xmath77 . \label{eps1}\ ] ] two other functions @xmath78 and @xmath79 have to be calculated numerically . the result for all functions including @xmath62 is shown in fig . one can clearly see that @xmath80 dominates in the dielectric function at imaginary frequencies up to @xmath81 . it means that for all experimentally investigated situations the region _ 1 , _ where the extrapolated optical data are used , gives the main contribution to @xmath82 . for example , for the separation between bodies @xmath83 the characteristic imaginary frequency is @xmath84 . at this frequency the contributions of different domains to @xmath85 are @xmath86 , @xmath87 , and @xmath88 . . the low - frequency ( @xmath9 ) contribution @xmath89 clearly dominates while @xmath90.,width=325 ] in the following section it will be demonstrated that the dielectric function of the films used for the force measurement can deviate considerably from the handbook data . the main reason for this is genuine sample dependence of the optical properties . here let us consider the question how sensitive is the casimir force to this deviation . to this end consider small variations @xmath91 of the dielectric function @xmath92 . the relative change in the force can be written via the relative variation of @xmath93 as @xmath94 where @xmath95 is the @xmath96-th term in the sum ( [ fpp ] ) or ( [ fsp ] ) and @xmath97 is the total force . to see how important is the change of the dielectric function at real frequencies , it is more convenient to introduce instead of @xmath98 the relative variations for each frequency domain _ 1 _ , _ 2 _ , or _ 3 _ : @xmath99 with @xmath100 here @xmath101 is defined as @xmath102 , @xmath103 is the measured deviation at real frequencies and @xmath104 is a particular frequency interval coinciding with the domain @xmath105 . with these definitions the relative change in the force can be presented as a linear combination of @xmath106 : @xmath107 . \label{df}\ ] ] the coefficients in front of @xmath108 can be called the weight functions . they describe the weight with which domain @xmath105 contributes to the force . these weight functions are shown in fig . [ fig3 ] for the separation @xmath83 . one can clearly see that the weight for the domain _ 1 _ is the largest . when @xmath2 increases the weight functions become more narrow and the domains _ 2 _ and _ 3 _ become less important . in the next section the relative variation of @xmath109 for real gold films will be discussed where more detailed information on @xmath106 will be presented . ) for different frequency domains . the point on the ordinate axis shows the weight of the @xmath110 term ( see eq . ( [ zterm ] ) ) , which exists only for the plasma model prescription @xcite.,width=325 ] the zero term @xmath110 in eqs . ( [ df])-([df ] ) has to be described separately . in the lifshitz formula ( [ fpp ] ) or ( [ fsp ] ) the @xmath110 term rose controversy in the literature @xcite . it is still an unresolved problem in the field of the casimir effect . the point of disagreement between different authors is the value of the reflection coefficient @xmath111 in the limit @xmath112 . bostrm and sernelius @xcite found @xmath113 applying the drude behavior to @xmath3 at @xmath112 . svetovoy and lokhanin @xcite got @xmath114 demanding continuous transition to the ideal metal at @xmath112 . bordag et al . @xcite have used the dielectric function of the plasma model in the limit @xmath112 and found @xmath115 there is no problem with @xmath116 coefficient which is always @xmath117 in the @xmath112 limit . in the first two approaches @xmath111 and @xmath116 do not depend on the material and the zero term should be omitted in eqs . ( [ df])- ( [ df ] ) . in the plasma model approach @xcite @xmath111 depends on the plasma frequency and the @xmath110 term will change when @xmath11 varies . the contribution of the zero term to eq . ( [ df ] ) will be @xmath118 in fig . [ fig3 ] the coefficient in front of @xmath119 is shown by the dot on the ordinate axis . the main conclusion that can be drawn from the analysis above is that the deviation of the drude parameters @xmath11 and @xmath12 from their handbook values gives the largest contribution to the change of the casimir force . in this section we are going to show that the optical properties of gold films depend on the details of the deposition method and following treatment to the extent that can not be ignored in the calculation of the casimir force . optical properties of gold were extensively investigated in 50 - 70th . today most of the researches are interested in gold nanoclusters but these data are inappropriate for our purposes . there are only a few exceptions . reflectivity of evaporated gold films was investigated @xcite in the wavelength range @xmath120 @xmath0 to analyze its dependence on the mean grain size . since only normal reflectance has been measured , one can not get the information on both real and imaginary part of the dielectric function but these data are very helpful . the reflectivity of gold in submillimeter range , @xmath121 , was measured with high precision @xcite . in ref . @xcite the films evaporated on glass with two different methods were characterized ellipsometrically in the photon energy range @xmath122 . this work allows one to compare the interband absorption with the older measurements . in addition , the optical properties of films deposited with the magnetron dc sputtering and with the filtered arc process @xcite were measured at the photon energies @xmath123 and compared with each other . one can indicate also two recent studies @xcite where new ellipsometric equipment was tested but the preparation of the films was poorly described . we will use the recent papers where it is possible but most of the data will be taken from the old works . in many old works the importance of sample preparation methods was recognized and carefully discussed . complete bibliography of the old publications up to 1981 one can find in ref . @xcite . let us start from the interband absorption or domain _ 3_. this range is not very significant for the casimir force but it provides us with important information how the sample preparation procedure influences the data . there is significant amount of data in this range obtained by combined reflectance and transmittance @xcite , ellipsometric spectroscopies @xcite on unannealed @xcite or annealed @xcite thin film @xcite or bulk @xcite samples measured in air @xcite or ultrahigh vacuum @xcite . a representative picture of typical variations of @xmath57 is shown in fig . the data for this figure were chosen to represent very dense ( bulk - like ) film @xcite , the films similar to those used for the casimir force experiments @xcite , and some intermediate case @xcite . thye @xcite described her films very carefully . the samples were semitransparent @xmath24 films with the thickness @xmath124evaporated in ultrahigh vacuum on supersmooth fused silica . the deposition rate was a few @xmath125 and the substrate was kept in most cases at room temperature . after the deposition the films were annealed under the same vacuum at @xmath126 , the process being monitored by simultaneous dc resistance measurements . the structure of the films was investigated by x - ray and electron - microscopy methods . good films were continuous formed of joined regularly shaped crystallites , the average lateral size of which ranged from 300 to 500 nm . the main defects were grain boundaries . it was demonstrated that flatness and smoothness of the film surfaces were very good . the dc resistivity of the films were found to be very sensitive to the conditions of preparation . the errors in the optical characteristics of the films were estimated on the level of a few percents . johnson and christy @xcite evaporated their films onto fused quartz substrates at room temperature . the evaporation rate was relatively high , 60 / s , and the thickness was in the range 250 - 500 . the presented data @xcite were found for unannealed samples . the structural characterization of the films was not made . errors in the optical data were estimated on the level of 2% . wang et al . @xcite evaporated the films on fused silica or optical glasses with two different methods . for the thermal evaporation the rate was 50 - 100 / s and for the electron - beam evaporation it was 5 / s . the substrates were kept at room temperature and there was no annealing after the deposition . roughness of the films or structural details were not reported . opaque @xmath127 thick films were characterized ellipsometrically in the photon energy range @xmath128 . errors were not specified but typical precision of the method is of about 1% . as one can see in fig . [ fig4 ] the spectra differ mainly on the scale factor . the experimental errors are too small to explain significant variation of the data . one can conclude that there is genuine sample dependance of optical characteristics of gold films . differences among these spectra have been attributed to various effects such as roughness @xcite , surface films @xcite , unspecified internal electron - scattering defects @xcite , strain @xcite , grain - boundary material @xcite , and voids @xcite . special investigation directed to quantitative understanding the discrepancies on a systematic basis was undertaken by aspnes et al . @xcite ( see also @xcite and @xcite ) . the authors analyzed the literature data and their own gold films evaporated by electron beam or sputter - deposited . the best films were obtained on cleaved @xmath129 substrates . glass , @xmath130 , and sapphire substrates were also investigated . air cleaved @xmath129 substrates were superior because gold start to grow on them epitaxially . scaling behavior of the spectra was attributed to different volumes of voids in the films prepared by different methods . it was also stressed that grains in a film imply voids , because it is impossible to produce grain boundaries in fcc or hexagonal close - packed lattices without losing density . to describe the contribution of voids in the dielectric function , the bruggeman effective medium approximation @xcite was used @xcite . nowadays this approximation is successfully applied for interpretation of the spectra of different nanocomposite materials ( see , for example , @xcite ) . if @xmath131 is the dielectric function of @xmath24 in its homogeneous form , then the effective dielectric function , @xmath132 , of the material containing a volume fraction of voids @xmath133 can be found from the equation @xmath134 where @xmath135 is the dielectric function of the host material . in the bruggman approximation it is assumed that the host material coincides with the effective medium , @xmath136 , so it treats both void and material phases on an equal . if the void fraction is small , @xmath137 , then the solution is @xmath138 aspnes et al . @xcite analyzed available data for @xmath139 in the interband spectral range fitting them with eq . ( [ ema ] ) . it was found out that in their e - beam evaporated unannealed sample the volume fraction of voids was 4% larger than in thye annealed thin films @xcite but 5% smaller than @xmath133 in johnson and christy @xcite unannealed thin films . one of the films was sputter - deposited on @xmath129 substrate that was held at liquid nitrogen temperature . this sample was full of voids that were clearly seen with the transmission electron microscopy . the volume fraction was estimated as @xmath140 in respect to thye @xcite film . one should mention that roughness of the film also will reduce @xmath139 in the interband region but this is a minor effect @xcite because the films in refs . @xcite were quite smooth . however , significant roughness could be the reason why wang et al . @xcite thermally evaporated film shows low absorptivity . concluding this discussion one can say that the sample preparation procedure significantly influences the optical properties of the films in the interband spectral range . at first sight it seems unimportant for the casimir force since variation of the dielectric function in this range gives only minor correction to the force . however , the volume fraction of voids , @xmath133 , that can be estimated from these data defines the exact value of the effective plasma frequency which is very important for the precise prediction of the casimir force . to see it clearly consider eq . ( [ solema ] ) at low frequencies where @xmath141 . in this limit the effective dielectric function will be simply proportional to that of the homogeneous material @xmath142 in the drude region where @xmath143 is given by eq . ( [ drude ] ) this relation is equivalent to the existence of the effective plasma frequency @xmath144 because voids reduce the number of conduction electrons per unit volume . correction to the force due to variation of @xmath145 in the interband spectral range is expected to be small . nevertheless , let us estimate it . for numerical calculation it was assumed that the bulk material can be described by thye data @xcite but the film material is given by johnson and christy data @xcite . besides , it was assumed that the difference between these data is negligible at the photon energies larger than @xmath146 . for the third term in eq . ( [ df ] ) and for the sphere - plate geometry it was found @xmath147 for the separations @xmath148 , respectively . indeed , the correction is negative since any deviation from the perfect material will reduce the reflection coefficients and , therefore , the absolute value of the force will be smaller . at separations larger than @xmath127 the correction to the force is completely negligible on the level of 0.1% . when the photon energy is below the interband threshold , @xmath149 , only free carriers should contribute to the dielectric response . however , the drude function ( [ drude ] ) obviously fails to describe the data nearby the transition ( see points below @xmath64 in fig . [ fig4 ] ) . steep rise of @xmath150 near @xmath151 originates from @xmath152 bands to fermi - surface transition near @xmath153 point @xcite but there is a long tail which extends well below @xmath154 . this tail can not be explained by the temperature broadening of the threshold . it was demonstrated @xcite that an additional transition at @xmath155 near @xmath156 point in the brillouin zone is involved . this transition has a smooth edge due to topological reason . in addition , the surface roughness and the grains can contribute to the tail broadening @xcite . this tail is not very important for the casimir force but it marks the region where the drude parameters can not be determined precisely . in contrast with the interband region there are only a few sources where the dielectric function was measured in the infrared . the available data for @xmath139 in the range @xmath157 are presented in fig . the dots represent the points from the handbook @xcite comprising dold and mecke data @xcite for @xmath158 and thye data @xcite for higher frequencies . dold and mecke did not describe carefully the sample preparation . it was reported only @xcite that the films were evaporated onto a polished glass substrate and measured in air by using an ellipsometric technique . annealing of the samples was not reported . these data are typically referred in connection with the casimir force calculation . the squares are the data by weaver et al . @xcite that were found for the electropolished bulk @xmath159 sample . originally the reflectance was measured in a broad interval @xmath160 and then the dielectric function was determined by the kramers - kronig analysis . due to indirect determination of @xmath143 the recommended accuracy of these data is only 10% . motulevich and shubin @xcite result for @xmath24 films is marked with the open circles . in this paper the films were carefully described . gold was evaporated on polished glass at pressure @xmath161 . the investigated films were @xmath162 thick . the samples were annealed in the same vacuum at @xmath163 fore more than 3 hours . the optical constants @xmath96 and @xmath164 ( @xmath165 ) were measured by the polarization methods in the spectral range @xmath166 . the errors in @xmath96 and @xmath164 were estimated as 2 - 3% and 0.5 - 1% , respectively . moreover , the density and conductivity of the films were measured . the crosses represent padalka and shklarevskii data @xcite for evaporated unannealed @xmath24 films . the solid line shows the only modern data @xcite for @xmath24 films sputtered on @xmath167 substrate . no additional details about the film preparation were reported . the precision of optical data was claimed to be better than 1% in the entire spectral range . for evaporated unannealed films @xcite and thye data @xcite for higher frequencies . the squares are the data @xcite for electropolished bulk @xmath159 . the open circles are the data for carefully prepared well annealed evaporated films @xcite . the crosses represent the data @xcite for unannealed evaporated films . the only modern data @xcite are given by the solid line.,width=325 ] different researches stressed that the conduction electrons are much more sensitive to slight changes in the material structure @xcite . as one can see in fig . [ fig5 ] it is , indeed , the case . variety of the data in any case can not be explained by the experimental errors . the observed scattering of the data is the result of different preparation procedures and reflects genuine difference between samples . bennett and ashley @xcite observed considerable reduction in the reflectance of @xmath24 films when instead of ultrahigh vacuum ( @xmath168 ) the films were evaporated in the ordinary vacuum ( @xmath169 ) . it was noted also that it was more difficult to prepare films with reproducible reflectance in the ordinary vacuum . even more important are the deposition method ( there is a whole spectrum of evaporation and sputtering methods ) , type of the substrate , its temperature and quality , and deposition rate . when we are speaking about the casimir force measurement with the precision of 1% or better there is no any more such a material as gold in general . there is only a material prepared in definite conditions . the data in fig . [ fig5 ] below @xmath64 should be described quite well with the drude function @xmath170 while @xmath171 the data on the log - log plot should look like a straight line with the slope @xmath172 shifted along the ordinate axis mostly due to variation of @xmath173 for different samples . the data in fig . [ fig5 ] are in general agreement with the expectations excluding the solid line . the latter can not be described with the drude behavior , probably , due to poor preparation of the film . padalka and shklarevskii data ( crosses ) demonstrate a long drude tail that can be observed up to @xmath174 . to all appearance it is again the result of poor sample preparation . the drude parameters can be found fitting both @xmath175 and @xmath176 with the function ( [ drude ] ) . this procedure will be discussed later . now let us estimate the change in the casimir force due to variation of @xmath57 in the infrared . for this purpose we calculate @xmath177 in eq . ( [ deli ] ) using as @xmath178 the difference between the squares @xcite and points @xcite in fig . [ fig5 ] in the interval @xmath179 . with this @xmath180 the second term in eq . ( [ df ] ) is calculated that gives the relative change in the casimir force due to variation of @xmath181 in the indicated frequency interval . for the sphere - plate geometry it was found that @xmath182 for the separations @xmath148 , respectively . the change in the force is negative and quite large . the main reason for this is the difference between the relaxation frequencies of the bulk material ( squares ) and poorly evaporated film ( dots ) . as one can see in fig . [ fig5 ] below @xmath183 the dielectric function @xmath184 for the film is roughly two times larger than that for the bulk material ( squares ) so @xmath177 is approximately @xmath185 for all important @xmath96 . there is no contradiction with the expectation that the force must be larger for the bulk material . we took into account here only the frequency interval @xmath186 . when the contribution of the low - frequency domain @xmath187 will be added the total force will be larger for bulk gold . the dielectric function in the low - frequency domain , @xmath61 , where direct optical data are inaccessible , is defined by the dielectric function in the infrared domain , @xmath179 , because one has to extrapolate the optical data from this domain to inaccessible region . the real and imaginary parts of @xmath143 follow from the drude function ( [ drude ] ) with an additional term @xmath188 in @xmath189 : @xmath190 the polarization term @xmath188 has appeared here due to the following reason . the total dielectric function @xmath191 includes contributions due to conduction electrons @xmath192 and the interband transitions @xmath193 . the polarization term consists of the atomic polarizability and polarization due to the interband transitions @xmath194 @xmath195 where @xmath196 is the atomic polarizability and @xmath197 is the concentration of atoms . if there is no the drude tail the last term can be considered as a constant . this is because the interband transitions have the threshold behavior with the onset @xmath64 and the kramers - kronig relation allows one to express @xmath198 as @xmath199 for @xmath200 this integral does not depend on @xmath52 . in reality the situation is more complicated because the transition is not sharp and many factors can influence the structure of the drude tail . it will be assumed here that @xmath188 is a constant but the fitting procedure will be shifted to frequencies where the drude tail is not important . in practice ( [ druderi ] ) can be applied for @xmath158 . as we have seen above the plasma frequency @xmath11 of a sample under investigation can deviate from that of perfect bulk material due to presence of voids ( see eq . ( [ ompeff ] ) ) . the casimir force is measured between films which are far from perfect . the best optical properties demonstrate well annealed films prepared in ultrahigh vacuum @xcite . the annealed films were not used for the force measurement and it is hardly reasonable to do in any future experiments . for high precision calculation of the force the procedure proposed in ref . @xcite is not good any more . one can not use the bulk plasma frequency to extrapolate the dielectric function to experimentally inaccessible frequencies . the second drude parameter , the relaxation frequency @xmath12 , is even more sensitive to the sample preparation procedure . therefore , the drude parameters have to be extracted from the optical characteristics of the same sample which is used for the force measurement . our purpose here is to establish the magnitude of the force change due to reasonable variation of the optical properties . to this end the available low - frequency data for @xmath201 shown in fig . [ fig5 ] and @xmath202 ( not shown ) were fitted with eq . ( [ druderi ] ) . the results together with the expected errors are collected in table [ tab1 ] . .the drude parameters found by fitting the available infrared data for @xmath203 and @xmath57 with eq . ( [ druderi ] ) . the first error is the statistic one and the second error ( if present ) is the systematic one . [ cols="<,<,<,<,<",options="header " , ] the values presented in table [ tab2 ] one can compare with the result by lambrecht and reynaud @xcite who gave for @xmath83 the values @xmath204 and @xmath205 . they are very close to that presented in table [ tab2 ] . this coincidence is because the increase of the force due to the infrared region was compensated by its decrease in the far infrared since larger value @xmath27 was used . therefore , the result of ref . @xcite represents rather the upper limit on the force than the real force corresponding to the handbook data . the latter force should be calculated with the drude parameters given in table [ tab1 ] ( row 1 ) . for the sphere - plate geometry at @xmath206 it is 6.3% smaller as was demonstrated in sec . [ sec3.3 ] . for small separations presented in table [ tab2 ] @xmath207 gives the relative temperature correction . it is well known that this correction is negative for the bostrm and sernelius approach @xcite making the absolute value of the force smaller ; it is positive when the @xmath110 term is assumed to coincide with that for the ideal metal @xcite . in the latter case the relative correction is nearly constant at small separations and is of about 2% independently on the geometry . in the former case it decreases from -0.5% to -4.2% for the plate - plate and from -1.1% to -6.7% for sphere - plate geometries . we did not include in the upper limit the correction due to surface roughness @xcite for obvious reason : it should be specified for a definite experiment . this correction always increases the absolute value of the force . additionally , the nonlocal effects in the interaction of electromagnetic field with a solid were not included in the upper limit on the force . it was demonstrated @xcite that the nonlocal screening effect due to excitation of plasma oscillations in metals will reduce the casimir force on a few percents at small separations . the influence of the anomalous skin effect on the force was also analyzed recently @xcite . it was found that this effect gives only minor correction which reduces the force smaller that 0.5% . as we have seen above the main uncertainty of the casimir force comes from the uncertainty of the drude parameters . the reason for variation of the effective plasma frequency , @xmath11 , is the presence of voids in the deposited films as was discussed in sec . [ sec3.1 ] . the relaxation frequency , @xmath12 , is much more sensitive to the details of the film preparation @xcite . it is well known that the resistivity @xmath208 of deposited films is larger than that for the single crystalline material . the resistivity of unannealed medium thick ( @xmath209 ) gold films @xcite was ranged from 4 to @xmath210 in dependence on the method ( electron beam , sputtering ) and rate ( 1 / s , 20 / s ) of deposition on @xmath167-substrate . for well crystallized annealed film deposited in ultrahigh vacuum thye @xcite has found @xmath211 but for unannealed film this value was as high as @xmath212 . these values should be compared with the resistivity of bulk gold @xmath213 . the difference between film and bulk values was attributed to the presence of grain boundaries in the films @xcite . in this section a simple two parameter model will be presented allowing one to estimate the force between films via the volume fraction of voids @xmath133 and the mean grain size @xmath214 . mayadas and shatzkes @xcite proposed the first model for calculation of the dc conductivity , @xmath215 , of films taking into account the electron scattering on the grain boundaries . this model successfully describes films and nanowires @xcite . in the model the films were treated as joined grains with bulk material properties ; the grain boundaries contributed to the electron transport . a semiclassical approach based on the boltzmann transport equation was used with the collision term associated to the grain boundaries . the background scattering was taken into account via the bulk relaxation time . the grains were considered to have columnar shape to be laterally bounded by planes . the grain boundaries were represented by two uncorrelated sets of @xmath216-potentials along each lateral direction . the following expression was found for the conductivity of films @xmath217 \right ) , \label{cond}\ ] ] where @xmath218 is the conductivity of single crystalline material and the parameter @xmath219 is defined as @xmath220 here @xmath221 is the fermi velocity in the homogeneous material , @xmath214 is the mean grain size , and @xmath222 is the reflection coefficient of electrons on the grain boundary . it is important to stress that @xmath223 depends only on the mean grain size ; dependence on the standard deviation from the mean value can be neglected @xcite . variation of the dc conductivity one can connect with the effective relaxation frequency , @xmath224 , by the relation @xmath225 comparing it with eq . ( [ cond ] ) one can find @xmath226 . however , this simple concept does not work at frequencies where the conductivity demonstrates the dispersion . recently the same approach was generalized to the case of optical conductivity @xmath227 @xcite , which is converging to the dc conductivity in the low - frequency limit , and the empirical parameter @xmath222 was extracted from comparison of the model with the measured reflectivity of gold films of different grain size . to find the optical conductivity the authors @xcite solved a linearized boltzmann transport equation taking into account the anomalous skin effect as well . so , in general , the conductivity is nonlocal . it means that it depends not only on frequency but also on the wave vector . it was found that the anomalous skin effect gives only very small correction to the reflectance of the films . to simplify the model we will neglect the nonlocal effect . recently the anomalous skin effect was discussed in connection with the casimir force @xcite . it was found that the effect gives only a minor contribution to the force and for this reason it could be estimated separately for a homogeneous material neglecting the grain boundaries . in the local limit for the optical conductivity the expression similar to eq . ( [ cond ] ) was found @xcite : @xmath228 \right ) , \label{optcond}\ ] ] where the conductivity in the drude model @xmath229 and the function @xmath230 are defined as @xmath231 the longitudinal component of the electric field , which is perpendicular to the film surface , does not feel the grain boundaries because the grains are columnar . therefore , eq . ( [ optcond ] ) is true only for the transverse field component . for this reason one has to distinguish two dielectric functions in the model : the longitudinal function @xmath232 , which coincides with the dielectric function of single crystalline gold , and the transverse function @xmath233 . the latter one can be presented in the following form @xcite : @xmath234 . \label{epst}\ ] ] using these dielectric functions sotelo et al . @xcite predicted the film reflectivity that was compared with the measured reflectivity of the films with different mean grain size @xmath214 . the experimental data were fitted in the visible and near infrared range to find the only parameter @xmath235 . this parameter was estimated as @xmath236 . the results for different grain sizes all fall within the indicated errors . prediction of the model was compared with the reflectivity in the infrared and excellent agreement was found . to estimate influence of the film preparation procedure on the casimir force , we use two parameter model taking into account the main reasons for the deviation of the optical properties from the bulk material . voids in the film are described with the effective medium approximation ( see eq . ( [ ema ] ) ) where the volume fraction of voids , @xmath133 , is considered as a free parameter . the other source of significant variation of the optical properties is the electron scattering on the grain boundaries . as was shown above this effect is completely defined by the mean grain size @xmath214 if the reflection coefficient of electron , @xmath222 , is fixed . the grain size @xmath237 is the second free parameter which can be easy measured with afm for any given film . because the contribution of each effect , voids or grains , to the casimir force is estimated on the level of a few percents , they can be considered separately . to make the calculations we have to fix the dielectric function of single crystalline gold . instead of linear interpolation between @xmath238 and @xmath239 in the drude tail region , as was done for the upper limit evaluation , more realistic quadratic interpolation on the linear plot is used to fit the minimum @xmath46 that happens at @xmath240 for all the data . the real part of dielectric function , @xmath241 , does not show any special features in the drude tail region . that is why for @xmath242 the drude function is used below @xmath243 and the handbook data @xcite above @xmath244 . defined in this way function @xmath245 shows much smaller discontinuity at @xmath246 than in the handbook between dold and mecke data @xcite and thye data @xcite . it should be stressed that these details in the dielectric function behavior can influence the force only on the level of 0.1% . it is relatively easy to calculate the change in the force due to variation of the void fraction in the film . it is assumed that the void fraction is small and one can use eq . ( [ solema ] ) for the effective dielectric function @xmath247 . this function is substituted in the dispersion relation ( [ k - k ] ) to find @xmath248 . the effective dielectric function at imaginary frequencies is used then in the lifshitz formula to find the casimir force . the relative correction to the force , @xmath249 , as a function of distance between bodies for @xmath250 is shown in fig . [ fig8 ] for the sphere - plate geometry ( dashed line ) and plate - plate geometry ( solid line ) . it was found that the dependence on the void fraction can be considered as linear at least up to @xmath251 . hence , the graph can be used to estimate the correction for any reasonable value of @xmath133 . the correction is given for the case when the temperature effect is negligible ( plasma model prescription for the @xmath110 term ) . the thermal correction can be estimated using table [ tab2 ] . . the dependence of the correction on @xmath252 is approximately linear.,width=325 ] as was discussed in sec . [ sec3.1 ] all unannealed samples demonstrate presence of voids . the void fraction for the best unannealed sample evaporated with e - beam on @xmath129 substrate was estimated as @xmath253 @xcite assuming that annealed thye films had no voids . for johnson and christy films @xcite @xmath133 was estimated as 0.09 and the same value is realized for e - beam evaporated films in wang et al . experiment @xcite . we expect that @xmath254 is the smallest void fraction in unannealed samples because this value is realized for the films growing epitaxially @xcite . in the afm experiment @xcite at the minimal separation @xmath255 the correction to the force for the best unannealed sample is estimated then as @xmath256 . in the mems experiment @xcite , where the smallest separation was @xmath257 , for the correction it is found @xmath258 . these values practically coincide with that presented in ref . @xcite , where the calculation was performed using as the perfect material the handbook data extrapolated to low frequencies according to ref . this coincidence shows once more that the force is much more sensitive to the variation of the drude parameters than to any other details of the dielectric function . here and in ref . @xcite the same value of the plasma frequency , @xmath259 , was used for the perfect material . it is not so straightforward to find the correction to the force due to variation of the grain size . the reason of complication is that the effect is described by two different dielectric functions @xmath260 and @xmath233 . in this case the reflection coefficients @xmath111 and @xmath116 have to be expressed via the surface impedances @xmath261 and @xmath262 @xcite . discussion of the impedance approach to the casimir force evaluation rose controversy in the literature @xcite . the problem was connected with the use of the approximate leontovich impedance @xcite for the force calculation . this approach was criticized @xcite on the basis that , in general , one has to use two different impedances one for each polarization and it was demonstrated that the right local impedances reproduced the lifshitz formula . recently the problem was settled finally in ref . @xcite where the classical theory of nonlocal impedances @xcite was applied to the casimir problem . it was demonstrated that the leontovich impedance is a good approximation for the propagating fields but it fails to describe the evanescent fluctuations that give significant contribution to the force . calculating the correction to the force we follow the method developed in ref . @xcite . the casimir force can be calculated using the same lifshitz formulas ( [ fpp ] ) and ( [ fsp ] ) where the reflection coefficients are expressed via the impedances : @xmath263 in general , the impedances are functions of frequency @xmath22 and wave vector component along the plate @xmath264 . they can be written via the dielectric functions @xcite and analytically continued to the imaginary frequencies @xcite : @xmath265 @xmath266 , \label{zp}\ ] ] where @xmath267 is the absolute value of the wave vector . because in our case the dielectric functions are local ( does not depend on @xmath164 ) , the integrals here can be calculated explicitly . it gives @xmath268 @xmath269 the dielectric functions that enter @xmath270 and @xmath271 can be written at imaginary frequencies with the help of eq . ( [ epst ] ) as @xmath272 @xmath273 , \label{epsim}\ ] ] where @xmath3 is the dielectric function of perfect material and @xmath274 to calculate the dependence of the casimir force on the grain size , the impedances ( [ zsex ] ) and ( [ zpex ] ) were substituted in the reflection coefficients ( [ rimp ] ) which were used then in the lifshitz formulas ( [ fpp ] ) , ( [ fsp ] ) . correction to the force as a function of the grain size is shown in fig . the results are given for the closest separations @xmath255 in the afm experiment @xcite ( dashed line ) and @xmath257 in the last mems experiment @xcite ( solid line ) . the correction found here is somewhat larger than in ref . for the grain size @xmath275 it is estimated as -0.9% for the afm and -1.2% for the mems experiments in contrast with the corresponding values -0.5% and -0.9% reported previously @xcite . the difference is explained by the larger value of the relaxation frequency for the perfect material , @xmath27 , used in ref . @xcite . it was already stressed that , in general , the effect of finite grain size is not reduced to the effective relaxation frequency which follows from eqs . ( [ cond ] ) , ( [ cond1 ] ) @xmath276 ^{- 1}. \label{omteff}\ ] ] however , the calculation showed that the correction to the force with a good precision is given by the relation @xmath277 this relation holds true at least in the investigated region @xmath278 and @xmath279 . the function @xmath280 is shown in fig . [ fig10 ] for the sphere - plate geometry ( dashed line ) and for the plate - plate geometry ( solid line ) . therefore , the concept of the effective relaxation frequency is happened to be good in practice . is connected with the correction by the relation ( [ adef]).,width=325 ] the grain size is difficult to predict because it depends on many details of the deposition procedure . definitely one can say only that for evaporated films the grain size increases with the film thickness @xmath281 as @xmath282 @xcite . in general , it is not true that the grain size is close to the film thickness . for example , the grain size for thermally evaporated film @xcite with the thickness of 205 nm can be estimated from the presented image as @xmath283 . the image of 120 nm thick evaporated film @xcite gives @xmath284 . the discussion above shows that instead of the physical parameters defining the film optical properties such as the volume fraction of voids @xmath133 and the grain size @xmath214 one can use the effective drude parameters connected with @xmath133 and @xmath214 by eqs . ( [ ompeff ] ) and ( [ omteff ] ) . these effective parameters can be extracted directly from the optical data in the infrared as was described in sec . [ sec3.3 ] . the drude parameters are the most important information about the film that one has to know to make high precision prediction of the casimir force . the details of the film dielectric response in the near infrared and visible range can change the force only on the level of 0.5% or smaller . voids and grain boundaries are the main but not the only factors affecting the electron system in the metallic film . the effective mass of electron , @xmath285 , can be modified by the strains which exist in the film and change slightly the crystallographic parameters . it would modify the fermi surface but the changes are not sufficient to account for the observed variation of @xmath286 @xcite . some minor contribution , however , is not excluded . internal electron - scattering defects like impurities or local defects can affect the relaxation time . this mechanism becomes dominant over the grain boundaries in alloys . for monoatomic material like @xmath287 one can hardly believe that the local defects ( 0-dimensional ) can play more significant role than the grain boundaries ( 2-dimensional ) . the drude tail is very sensitive to the details of the film preparation @xcite . the mechanisms of this dependence are not well understood @xcite but we have seen that the drude tail can change the force less that 0.5% at the smallest separation . a number of arguments were provided in this paper to show that the optical properties of deposited gold films differ from the properties of bulk material on the level which can not be ignored in the precise evaluation of the casimir force . the best properties demonstrate carefully prepared annealed films but all the experiments were made with unannealed films . the main physical reasons for the deviation from the bulk properties are the voids in the films reducing the density of conduction electrons and the electron scattering on the grain boundaries significantly increasing the effective relaxation frequency . it was demonstrated also that the casimir force is much more sensitive to the material optical properties in the infrared and far infrared range than to the dielectric response in the near infrared and visible range . for this reason the precise values of the effective drude parameters are the most important information that has to be extracted from the optical measurements for successful prediction of the force . a reliable way to get this information is the direct optical characterization of the films used for the casimir force measurement . ordinary ellipsometry in the visible range , @xmath128 , can be helpful to see the quality of the film but can not be used to get the drude parameters . both real and imaginary parts of the dielectric function in the wavelength range @xmath288 can be routinely measured with the infrared spectroscopic ellipsometers ( for recent publications using these apparatus see @xcite ) . the data can be fitted with the drude dielectric function to find the effective plasma frequency @xmath11 and the relaxation frequency @xmath25 characterizing the film . special attention has to be paid to the sample preparation procedure to reduce the drude tail and roughness of the film . up to date the best known method for metal deposition is the filtered arc deposition process . the gold films deposited with this method demonstrated the bulk - like behavior in the near infrared @xcite without annealing . the method is superior to evaporation or sputtering due to significant fraction , 70 - 80% , of @xmath24 ions with the energy @xmath289 in the plasma . for high precision experiments the sample preparation and characterization is as important as the measurement of the force itself . while there is no direct optical data for the films used in the casimir force experiments , one can establish the upper limit on the force using the dielectric function of single crystalline gold . this upper limit was discussed in sec . it was already noticed @xcite that the afm experiment @xcite gives at smallest separations the force which is larger than the upper limit . much more detailed analysis of the dielectric function of perfect single crystalline gold carried out in this paper supports this conclusion . let us be more precise . for illustration we consider only one experimental point of the closest approach in the afm experiment , but the same conclusions will be true for all the points with @xmath290 . the first point presented in fig . @xcite corresponds @xmath255 and @xmath291 . this and the other points of close approach were found from the original postscript figure presented in the preprint quant - ph/0005088 with the digitizing errors ( 0.5 pixel ) @xmath292 in the separation and @xmath293 in the force . one can compare @xmath294 with the upper limit on the force which was discussed in sec . [ sec4 ] . for the reduction factor it was found @xmath295 that is equivalent to the force @xmath296 . the difference @xmath297 should be compared with the experimental error that was estimated as @xmath298 @xcite . due to error in the absolute separation @xmath299 @xcite the first experimental point could correspond to @xmath300 . in this case the upper limit is @xmath301 and @xmath302 . it should be stressed that the theoretical uncertainty in the force because of @xmath303 is @xmath304 /f\left ( 63\,nm\right ) = 4.1\%$ ] . it is considerably larger than the claimed 1% agreement between the theory and experiment @xcite . this problem has been already emphasized before @xcite . even for @xmath300 the upper limit is still @xmath305 smaller than the measured force . in reality the theoretical force should be smaller than the upper limit due to deviation of the film optical properties from the properties of bulk material . the smallest volume fraction of voids was estimates in sec . [ sec5.2 ] as 4% . it reduces the absolute value of the force at @xmath255 on 1.8% . the electron scattering on the grain boundaries reduces the force additionally . as the largest possible value of the mean grain size for the unannealed @xmath306-thick gold film we take @xmath307 . according to eq . ( [ cond ] ) it corresponds to the resistivity @xmath308 or equivalently to the effective relaxation frequency @xmath309 . the minimal reduction of the force due to finite grain size is estimated then as 0.5% . therefore , in the very best situation the theoretical force is expected to be smaller than @xmath310 or @xmath311 for @xmath300 . the difference with the experimental value is striking . recently the result of the afm experiment @xcite has been reanalyzed @xcite . three of 30 scans were rejected due to excessive noise while the rest 27 scans were used to calculate the force and the standard deviation . smaller value was found for the standard deviation @xmath312 . the only point for the force was reported : at @xmath300 the mean force was @xmath313 . both position and the force are smaller than that in ref . @xcite but the authors did not make any comment in this respect . the problem with the uncertainty of absolute separation @xmath303 @xcite was dismissed on the basis that the experiment is in a good agreement with the theory . the theoretical result for the force was found as in ref . @xcite using the handbook data and the drude parameters @xmath26 , @xmath27 to extrapolate to the low frequencies . we have seen in sec . [ sec4 ] that this procedure gives the force very close to the upper limit . the films used in the experiments should have different drude parameters . in this respect the authors @xcite stated that @xmath11 is determined by the properties of the elementary cell which can not be influenced by the sample quality . this statement is wrong because , as shows eq . ( [ omp ] ) , @xmath11 depends on the concentration of the conduction electrons which , indeed , depends on the sample quality . the voids will reduce the electron concentration in the sample . even more confusing is the discussion of the sample dependence of @xmath12 . the authors introduce not imaginary but real characteristic frequency @xmath314 and assume that this frequency gives the main contribution to the force . as was discussed in detail in sec . [ sec2 ] low frequencies @xmath315 give the main contribution to the dispersion integral ( [ k - k ] ) for @xmath316 . this fact was already stressed in a number of publications @xcite . because of this mistake the authors @xcite took into account some minor effects in the near infrared like electron - electron scattering that gave quadratic frequency dependence of the relaxation frequency @xcite but missed very important and well established effect like significant increase of the film resistivity in comparison with the bulk material . the same analysis can not be done for the latest mems experiment @xcite mainly because of huge roughness correction . in contrast with the afm experiment , where the roughness correction was smaller than 0.5% , in this experiment the roughness correction is as large as 25% . one can hardly believe that the calculations based on the local separation @xcite can be true for the shape dependent casimir force with the precision on the level of 1% . in the experiment @xcite one of the plates was covered with copper but in the present paper an extensive analysis of gold was made . for this reason one can say only that the sample effect should reduce the force on the level of 2% . it is much larger than the declared experimental errors . the roughness correction and sample effects do not allow us to support the conclusion @xcite on the possibility to distinguish between different approaches to the temperature correction . in this paper we analyzed the sample dependence of the casimir force . it was already noted @xcite that the optical properties of metallic films used to measure the force can deviate significantly from the bulk material but here , for the first time , serious analysis of the effect based on the facts about gold films was undertaken . it was stressed that the behavior of the dielectric function of a metal in the far infrared , where direct optical data are inaccessible , is crucial for the precise evaluation of the force . the casimir force depends on the dielectric function @xmath82 connected with the observable function @xmath57 via the dispersion relation ( [ k - k ] ) . because @xmath145 is large at low frequencies , the far infrared range gives the most significant contribution in @xmath3 and , therefore , in the force . for example , at @xmath317 the contribution of the real frequency domain @xmath318 to @xmath319 was estimated as 75% . sample dependence of the optical properties of gold films was discussed in sec . the data for 40 years of investigations have been collected and analyzed . these data clearly showed that significant difference between samples can not be explained by the experimental errors and had to be attributed to genuine sample dependence of the optical properties . the mechanisms of this dependence have been clarified in the literature . variation of @xmath57 in the visible range was connected with the volume fraction of voids @xmath133 in the films . the void fraction is especially significant in unannealed films . for the best unannealed films which start to grow epitaxially on @xmath129 substrate @xmath133 was estimated as 4% but in the worst cases it could be more than 20% . the fraction of voids is very important for the casimir force because it defines the precise value of the plasma frequency . the concentration of conduction electrons in the sample decreases with the increase of @xmath133 and , as the result , @xmath11 decreases . analysis of the data available in the infrared range revealed the same tendency of the dependence on the sample preparation procedure . it was demonstrated that at @xmath158 the data for real and imaginary parts of @xmath320 could be reasonably well described with the drude model . the fitting procedure allowed one to find the effective drude parameters for the investigated samples ( see table [ tab1 ] ) . surprisingly , the handbook data @xcite in the range @xmath158 taken from the old paper by dold and mecke @xcite happen to be bad . the best fit gave @xmath321 , @xmath322 . it means that the films were poorly prepared and were full of voids and defects for the electron scattering . on the other hand bulk @xmath159 and well annealed gold films demonstrated much better characteristics close to that typically used for the calculation of the casimir force @xcite : @xmath323 , @xmath27 . it was stressed that the force was measured between unannealed films for which one can hardly expect that the effective drude parameters coincide with that for the bulk material . however , using the `` ideal '' dielectric function of gold one can establish the upper limit on the casimir force . this dielectric function was carefully described in sec . the upper limit happened to be very close to the force evaluated in ref . the use of nearly the same @xmath324 and @xmath325 for `` ideal '' gold in this paper and in ref . @xcite explains the coincidence of the forces . thus , the result of ref . @xcite should be considered rather as the upper limit on the force than as the force between real films . the expected deviations of the force from the case of perfect single crystalline gold were analyzed in sec . [ sec5 ] . an appreciable correction to the casimir force was found due to the presence of voids in the film . even for the smallest fraction of voids , @xmath253 , the relative correction to the absolute value of the force in the afm experiment @xcite was found to be -1.8% at the smallest separation @xmath255 . for the mems experiment @xcite at @xmath257 the correction was -1.0% . the other effect that was estimated was influence of the grain boundaries in the film on the electron scattering . this effect is the main reason for the increase of the film resistivity or the effective relaxation frequency . for the mean grain size of 50 nm ( typical for 100 nm thick film ) the corrections due to the finite grain size were estimated as -0.9% for the afm experiment and -1.2% for the mems experiment at the smallest separations in each case . both corrections can not be ignored in the experiment where the force was measured with the precision on the level or better than 1% . a number of additional factors can affect the dielectric function . not all of them are well understood but one can indicate strains in the film or some internal electron - scattering defects . all additional effects are expected to reduce the absolute value of the force less than 1% . the best way to get the precise values of the drude parameters is the direct measurement of the dielectric functions of the film used in a casimir force experiment . the most important frequency range where this function should be known is the infrared range @xmath158 . it can be done with the infrared spectroscopic ellipsometers . special attention has to be payed to the film preparation procedure .

restrictions imposed by gauge invariance in noncommutative spaces together with the effects of ultraviolet / infrared mixing lead to strong constraints on possible candidates for a noncommutative extension of the standard model . we study a general class of noncommutative models consistent with these restrictions . specifically we consider models based upon a gauge theory with the gauge group coupled to matter fields transforming in the ( anti)-fundamental , bi - fundamental and adjoint representations . we pay particular attention to overall trace - u(1 ) factors of the gauge group which are affected by the ultraviolet / infrared mixing . typically , these trace - u(1 ) gauge fields do not decouple sufficiently fast in the infrared , and lead to sizable lorentz symmetry violating effects in the low - energy effective theory . in a 4-dimensional theory on a continuous space - time making these effects unobservable would require making the effects of noncommutativity tiny , . this severely limits the phenomenological prospects of such models . however , adding additional universal extra dimensions the trace - u(1 ) factors decouple with a power law and the constraint on the noncommutativity scale is weakened considerably . finally , we briefly mention some interesting properties of the photon that could arise if the noncommutative theory is modified at a high energy scale . desy 06 - 087 + ippp/06/34 + dcpt/06/68

hep-ph0606106

gauge theories on spaces with noncommuting coordinates , @xmath2=i\,\theta^{\mu\nu } \ , \ ] ] provide a very interesting new class of quantum field theories with intriguing and sometimes unexpected features . these noncommutative models can arise naturally as low - energy effective theories from string theory and d - branes . as field theories they must satisfy a number of restrictive constraints detailed below , and this makes them particularly interesting and challenging for purposes of particle physics model building . for general reviews of noncommutative gauge theories the reader can consult e.g. refs . @xcite . there are two distinct approaches used in the recent literature for constructing quantum field theories on noncommutative spaces . the first approach uses the weyl - moyal star - products to introduce noncommutativity . in this case , noncommutative field theories are defined by replacing the ordinary products of all fields in the lagrangians of their commutative counterparts by the star - products @xmath3 noncommutative theories in the weyl - moyal formalism can be viewed as field theories on ordinary commutative spacetime . for example , the noncommutative pure gauge theory action is @xmath4 where the commutator in the field strength also contains the star - product . the important feature of this approach is the fact that phase factors in the star - products are not expanded in powers of @xmath5 and the @xmath5 dependence in the lagrangian is captured entirely . this ability to work to all orders in @xmath5 famously gives rise to the ultraviolet / infrared ( uv / ir ) mixing @xcite in the noncommutative quantum field theory which we will review below . the second indirect approach to noncommutativity does not employ star - products . it instead relies @xcite on the seiberg - witten map which represents noncommutative fields as a function of @xmath5 and ordinary commutative fields . this approach essentially reduces noncommutativity to an introduction of an infinite set of higher - dimensional ( irrelevant ) operators , each suppressed by the corresponding power of @xmath5 , into the action . there are two main differences compared to the weyl - moyal approach . first , in practice one always works with the first few terms in the power series in @xmath5 and in this setting the uv / ir mixing can not be captured . second , the seiberg - witten map is a non - linear field transformation . therefore , one expects a non - trivial jacobian and possibly a quantum theory different from the one obtained in the weyl - moyal approach . in this paper we will use the original direct formulation of the theory on a noncommutative space in terms of the weyl - moyal star product . in the context of weyl - moyal noncommutative standard model building , a number of features of noncommutative gauge theories have to be taken into account which are believed to be generic @xcite : 1 . the mixing of ultraviolet and infrared effects @xcite and the asymptotic decoupling of u(1 ) degrees of freedom @xcite in the infrared ; 2 . the gauge groups are restricted to u(@xmath6 ) groups @xcite or products of thereof ; 3 . fields can transform only in ( anti-)fundamental , bi - fundamental and adjoint representations @xcite ; 4 . the charges of matter fields are restricted @xcite to @xmath7 and @xmath8 , thus requiring extra care in order to give fractional electric charges to the quarks . building upon an earlier proposal by chaichian _ @xcite , the authors of ref . @xcite constructed an example of a noncommutative embedding of the standard model with the purpose to satisfy all the requirements listed above . the model of @xcite is based on the gauge group @xmath9 with matter fields transforming in noncommutatively allowed representations . higgs fields break the noncommutative gauge group down to a low - energy commutative gauge theory which includes the standard model group @xmath10 . the @xmath11 group here corresponds to ordinary qed , or more precisely to the hypercharge @xmath12 abelian gauge theory . the generator of @xmath11 was constructed from a linear combination of _ traceless _ diagonal generators of the microscopic theory @xmath13 because of this , the uv / ir effects which can affect only the overall trace-@xmath14 subgroup of each @xmath15 were not contributing to the hypercharge @xmath16 however some of the overall trace-@xmath17 degrees of freedom can survive the higgs mechanism and thus contribute to the low - energy effective theory , in addition to the standard model fields . these additional trace-@xmath14 gauge fields logarithmically decouple from the low - energy effective theory and were neglected in the analysis of ref . @xcite . here , we take these effects into account . we will find that the noncommutative model building constraints , and , specifically , the uv / ir mixing effects in the trace - u(1 ) factors in the item 1 above , lead to an unacceptable defective behavior of the low - energy theory , when we try to construct a model having the photon as the only massless colourless u(1 ) gauge boson . our findings pose extremely severe constraints on such models effectively ruling them out . one way out is to modify some of the assumptions . we will discuss the introduction of universal extradimensions and modifications of the noncommutative field theory at very high energy scales . the uv / ir mixing in noncommutative theories arises from the fact that certain classes of feynman diagrams acquire factors of the form @xmath18 ( where @xmath19 is an external momentum and @xmath20 is a loop momentum ) compared to their commutative counter - parts . these factors directly follow from the use of the weyl - moyal star - product ( [ stardef ] ) . at large values of the loop momentum @xmath20 , the oscillations of @xmath18 improve the convergence of the loop integrals . however , as the external momentum vanishes , @xmath21 the divergence reappears and what would have been a uv divergence is now reinterpreted as an ir divergence instead . this phenomenon of uv / ir mixing is specific to noncommutative theories and does not occur in the commutative settings where the physics of high energy degrees of freedom does not affect the physics at low energies . there are two important points concerning the uv / ir mixing @xcite which we want to stress here . first , the uv / ir mixing occurs only in the trace - u(1 ) components of the noncommutative @xmath22 theory , leaving the @xmath23 degrees of freedom unaffected . second , there are two separate sources of the uv / ir mixing contributing to the dispersion relation of the trace - u(1 ) gauge fields : the @xmath24 effects and the @xmath25 effects , as will be explained momentarily . a study of the wilsonian effective action , obtained by integrating out the high - energy degrees of freedom using the background field method , and keeping track of the uv / ir mixing effects , has given strong hints in favour of a non - universality in the infrared @xcite . in particular , the polarisation tensor of the gauge bosons in a noncommutative @xmath22 gauge theory takes a form @xcite @xmath26 here @xmath27 are adjoint labels of @xmath22 gauge fields , @xmath28 , such that @xmath29 correspond to the overall @xmath14 subgroup , i.e. to the trace - u(1 ) factor . the term in ( [ poltensor ] ) proportional to @xmath30 would not appear in ordinary commutative theories . it is transverse , but not lorentz invariant , as it explicitly depends on @xmath31 nevertheless it is perfectly allowed in noncommutative theories . it is known that @xmath25 vanishes for supersymmetric noncommutative gauge theories with unbroken supersymmetry , as was first discussed in @xcite . in general , both @xmath24 and @xmath25 terms in ( [ poltensor ] ) are affected by the uv / ir mixing . more precisely , as already mentioned earlier , the uv / ir mixing affects specifically the @xmath32 components and generates the @xmath33 components in ( [ poltensor ] ) . the uv / ir mixing in @xmath32 affects the running of the trace - u(1 ) coupling constant in the infrared . for a pure noncommutative gauge theory in 4 continuous dimensions one finds , @xmath34 leading to a logarithmic decoupling of the trace - u(1 ) gauge fields from the @xmath23 low - energy theory , see refs . @xcite for more detail . for nonsupersymmetric theories , @xmath33 can present more serious problems . in theories without supersymmetry , @xmath35 at small momenta , and this leads to unacceptable quadratic ir singularities @xcite . in theories with softly broken supersymmetry ( i.e. with matching number of bosonic and fermionic degrees of freedom ) the quadratic singularities in @xmath33 cancel @xcite . however , the subleading contribution @xmath36 survives @xcite unless the supersymmetry is exact . for the rest of the paper we will concentrate on noncommutative standard model candidates with softly broken supersymmetry , in order to avoid quadratic ir divergencies . in this case , @xmath37 is a measure of susy breaking . ] as explained in @xcite . the presence of such @xmath25 effects will lead to unacceptable pathologies such as lorentz - noninvariant dispersion relations giving mass to only one of the polarisations of the trace - u(1 ) gauge field , leaving the other polarisation massless . the presence of the uv / ir effects in the trace - u(1 ) factors makes it pretty clear that a simple noncommutative u(1 ) theory taken on its own has nothing to do with ordinary qed . the low - energy theory emerging from the noncommutative u(1 ) theory will become free at @xmath38 ( rather than just weakly coupled ) and in addition will have other pathologies @xcite . however , one would expect that it is conceivable to embed a commutative @xmath23 theory , such as e.g. qcd or the weak sector of the standard model into a supersymmetric noncommutative theory in the uv , but some extra care should be taken with the qed u(1 ) sector @xcite . we will show that the only realistic way to embed qed into noncommutative settings is to recover the electromagnetic u(1 ) from a _ traceless _ diagonal generator of some higher @xmath22 gauge theory . so it seems that in order to embed qed into a noncommutative theory one should learn how to embed the whole standard model @xcite . we will see , however , that the additional trace - u(1 ) factors remaining from the noncommutative @xmath22 groups will make the resulting low - energy theories unviable ( for the 4 dimensional models considered in the first half of this paper ) . in order to proceed we would like to disentangle the mass - effects due to the higgs mechanism from the mass - effects due to non - vanishing @xmath39 hence we first set @xmath40 ( this can be achieved by starting with an exactly supersymmetric theory ) . it is then straightforward to show ( see @xcite ) that the higgs mechanism alone can not remove all of the trace - u(1 ) factors from the massless theory . more precisely , the following statement is true : _ consider a scenario where a set of fundamental , bifundamental and adjoint higgs fields breaks @xmath41 such that @xmath42 is non - trivial . then there is at least one generator of the unbroken subgroup @xmath42 with _ non - vanishing trace_. this generator can be chosen such that it generates a u(1 ) subgroup . _ we can now count all the massless u(1 ) factors in a generic noncommutative theory with @xmath40 and after the higgs symmetry breaking . in general we can have the following scenarios for massless u(1 ) degrees of freedom in @xmath42 : 1 . [ possa]@xmath43 is traceless and in addition there is one or more factors of trace - u(1 ) in @xmath42 . [ possb ] @xmath43 arises from a mixture of traceless and trace - u(1 ) generators of the noncommutative product group @xmath44 3 . [ possc ] @xmath43 has an admixture of trace - u(1 ) generators as in ( [ possb ] ) plus there are additional massless trace - u(1 ) factors in @xmath42 . in the following sections we will see that none of these options lead to an acceptable low - energy theory once we have switched on @xmath45 , i.e. once we have introduced mass differences between superpartners . it is well - known @xcite that @xmath45 leads to strong lorentz symmetry violating effects in the dispersion relation of the corresponding trace - u(1 ) vector bosons , and in particular , to mass - difference of their helicity components . if option ( [ possa ] ) was realised in nature , it would lead ( in addition to the standard photon ) to a new colourless vector field with one polarisation being massless , and one massive due to @xmath39 the options ( b ) and ( c ) are also not viable since an admixture of the trace - u(1 ) generators to the photon would also perversely affect photon polarisations and make some of them massive . in the rest of this note we will explain these observations in more detail . we end this section with some general comments on noncommutative standard modelling . in an earlier analysis @xcite the trace - u(1 ) factors were assumed to be completely decoupled in the extreme infrared and , hence , were neglected . however , it is important to keep in mind that the decoupling of the trace - u(1 ) s is logarithmic and hence slow . for a 4 dimensional continuum theory one finds that even in presence of a huge hierarchy between the noncommutative mass scale @xmath46 , say of the order of the planck scale @xmath47 , and the scale @xmath48 ( electroweak and qcd scale , respectively ) , where the su(@xmath6 ) subgroup becomes strong , the ratio @xmath49 is not negligible . in particular , the above inequality holds for any @xmath50 . hence , the complete decoupling of the trace - u(1 ) degrees of freedom at small non - zero momenta does not appear to be fully justified and the trace - u(1 ) would leave its traces in scattering experiments at accessible momentum scales @xmath51 ( see sect . 2 for more detail ) . however , eq . ( [ ration1 ] ) already gives us a hint how one can avoid that the trace - u(1 ) s leave observable traces . the logarithms in eq . ( [ ration1 ] ) are a typical property of the 4 dimensional theory . adding universal extra dimensions ( where gauge fields can propagate into the extra dimensions ) one expects that one gets a much faster power like decoupling . we will explore this possibility in in sect . [ powerlaw ] . finally , starting from the original motivation from string theory another possibility to avoid the conclusions stated above presents itself . viewed as originating from string theory , the noncommutative field theory is only a low energy limit . at very high scales the noncommutative field theory is not necessarily a good description anymore . we discuss a simple ( but not too unreasonable ) modification and study its consequences in sect . [ birefringence ] . uv / ir mixing manifests itself only in the trace - u(1 ) part of the full noncommutative u(@xmath6 ) . for this part it strongly affects @xmath52 and is responsible for the generation of nonvanishing @xmath53 ( if susy is not exact ) . in this section we will briefly review how the uv / ir mixing arises in the trace - u(1 ) sector and how this leads us to rule out options ( a ) and ( c ) discussed in sect . 1 . following refs . @xcite , we will consider a u(@xmath6 ) noncommutative theory with matter fields transforming in the adjoint and fundamental representations of the gauge group . we use the background field method , decomposing the gauge field @xmath54 into a background field @xmath55 and a fluctuating quantum field @xmath56 , and the appropriate background version of feynman gauge , to determine the effective action @xmath57 by functionally integrating over the fluctuating fields . to determine the effective gauge coupling in the background field method , it suffices to study the terms quadratic in the background field . in the effective action these take the following form ( capital letters denote full u(@xmath6 ) indices and run from @xmath7 to @xmath58 ) , . ] @xmath59 at tree level , @xmath60 is the standard transverse tensor originating from the gauge kinetic term . in a commutative theory , gauge and lorentz invariance restrict the lorentz structure to be identical to the one of the tree level term . in noncommutative theories , lorentz invariance is violated by @xmath5 . the most general allowed structure is then given by eq . ( [ poltensor ] ) . the second term may lead to the strong lorentz violation mentioned in the introduction . this term is absent in supersymmetric theories @xcite . let us start with a discussion of the effects noncommutativity has on @xmath52 and the running of the gauge coupling . that is , for the moment , we postpone the study of @xmath25-effects by considering a model with unbroken supersymmetry valid also in the non - supersymmetric case . ] . as usual , we define the running gauge coupling as @xmath61 where @xmath62 is the microscopic coupling ( i.e. the tree level contribution ) and @xmath63 includes only the contributions from loop diagrams . henceforth , we will drop the loop subscript . to evaluate @xmath64 at one loop order one has to evaluate the appropriate feynman diagrams . the effects of noncommutativity appear via additional phase factors @xmath65 in the loop - integrals . using trigonometric relations one can group the integrals into terms where these factors combine to unity , the so called planar parts , and those where they yield @xmath66 , the so called non - planar parts . for fields in the fundamental representation , the phase factors cancel exactly and only the planar part is non - vanishing . fundamental fields therefore contribute as in the commutative theory @xcite . in all loop integrals involving adjoint fields one finds the following factor @xcite , @xmath67 using trigonometric and group theoretic relations this collapses to @xmath68 we can now easily see that all effects from uv / ir mixing , marked by the presence of the @xmath69 , appear only in the trace - u(1 ) part of the gauge group . the planar parts , however , are equal for the u(1 ) and su(@xmath6 ) parts . summing everything up we find the planar contribution ( the coefficients @xmath70 are given in table [ coefficients ] and @xmath71 is the casimir operator in the representation @xmath72 ) @xmath73\bigg),\end{aligned}\ ] ] where @xmath74 is the mass of a spin @xmath75 particle belonging to the representation @xmath76 of the gauge group , @xmath77 and @xmath78 appears via dimensional transmutation similar to @xmath79 in qcd . we have chosen the renormalisation scheme , i.e. the finite constants , such that @xmath80 vanishes at @xmath81 . .coefficients appearing in the evaluation of the loop diagrams . [ cols="^,^,^,^,^",options="header " , ] for the trace - u(1 ) part the nonplanar parts do not vanish and we find @xmath82 with @xmath83 \int_{0}^{1}dx \ k_{0 } ( \sqrt{a } |\tilde{k}|)\right\ } \ \ , \\ \label{pi12 } \tilde{\pi}\!\!\ ! & = & \!\!\!{4c({\bf g})\over ( 4\pi)^2}\left\ { { d_j \over \tilde{k}^2}- \left(c_j k^2 - d_j { \partial^2 \over \partial^2 |\tilde{k}| } \right ) \int_{0}^{1}dx \ k_0 ( \sqrt{a } |\tilde{k}|)\right\ } , \end{aligned}\ ] ] where @xmath84 is the casimir operator in the adjoint representation . [ 0.95 ] ( 190,180)(40,0 ) ( solid ) and @xmath85 ( dashed ) for a u(2 ) theory with two matter multiplets and all particles of equal mass @xmath86 , from top to bottom ( left side , solid ) , as a function of the momentum @xmath19 , for a choice of @xmath87 , with @xmath88 . , title="fig:",width=359 ] ( -40,-15)[c][1.2]@xmath89 ( -280,150)[c][1.7]@xmath90 for illustration , we plot in fig . [ u1gaugecoupling ] the coupling ( [ defcoupling ] ) for a toy model which is a supersymmetric u(2 ) gauge theory with two matter multiplets and all masses ( of all fields ) taken to be equal . we observe that even for large masses the running of the u(1 ) part ( solid lines ) does not stop in the infrared . for masses smaller than the noncommutative mass scale @xmath91 the trace - u(1 ) gauge coupling has a sharp bend at @xmath46 where the nonplanar parts start to contribute . for larger masses the running stops at the mass scale @xmath92 only to resume running at a scale @xmath93 which is , of course , again due to the nonplanar parts . the dashed lines in fig . [ u1gaugecoupling ] give the running of the su(2 ) part which receives no nonplanar contributions and behaves like in an ordinary commutative theory . for @xmath94 the su(2 ) gauge coupling reaches a landau pole at @xmath81 , for all non vanishing masses the running stops at the mass scale . we observe that the ratio between the su(2 ) coupling and the trace - u(1 ) coupling is not negligibly small over a wide range of scales , in support of our assertion ( [ ration1 ] ) in sec further support comes from looking at the following approximate form for the running of the gauge coupling . we assume the hierarchy @xmath95 , @xmath96,\,\quad\,\,\,\ , \rm{for}\quad k^2 \ll m^2.\end{aligned}\ ] ] here , we have simplified the discussion by writing @xmath97 is the noncommutativity mass - scale . heuristically , @xmath98 but it may depend on the direction . e.g. , for @xmath99 in the canonical basis , @xmath100 only when @xmath101 one has @xmath102 otherwise the scale @xmath103 depends on @xmath104 @xmath105 it is nevertheless a useful scale . the gauge coupling for the su(@xmath6 ) subgroup @xmath106 is obtained by setting @xmath107 . for simplicity let us now consider a situation where we have only fields in the adjoint representation . one finds @xcite that @xmath108 , and @xmath109 to reach @xmath110 we need @xmath111 and in turn @xmath46 to be large . as a generic example let us use @xmath112 ( the scale where the ordinary electroweak su(2 ) would become strong , in absence of electroweak symmetry breaking ) and @xmath113 . in this regime our formulas ( [ run1 ] ) and ( [ run2 ] ) approximate the full result to a very high precision since threshold effects are negligible . ] . we find @xmath114 taking electroweak symmetry breaking into account we have to replace @xmath115 by @xmath116 with @xmath117 in ( [ abschaetz ] ) . we find @xmath118 let us increase the coupling strength of the su(@xmath6 ) by using @xmath119 . @xmath113 is now quite close to the strong coupling scale of the su(@xmath6 ) . without symmetry breaking we find @xmath120 we might be able to reduce this number by some orders of magnitude but without using an extreme field content it remains always extraordinarily large indeed , one can typically find a scale @xmath19 which is not too close to the strong coupling scale of the su(@xmath6 ) which strengthens the bounds dramatically . therefore , as a conservative estimate we propose and hence @xmath121 for @xmath122 ( we have omitted the lorentz indices for simplicity ) . ] @xmath123 let us note that this strong constraint is based on the assumption that the 4 dimensional noncommutative field theory is a valid description up to arbitrarily high momentum scales . this assumption is not necessarily fulfilled if the noncommutative theory is embedded into a more fundamental theory , e.g. string theory . in the later sects . [ powerlaw ] and [ birefringence ] we will investigate situations where this assumption is not valid anymore and the constraints can be weakened . to conclude this subsection , let us point out that , in a scattering experiment ( as depicted in fig . [ scattering ] ) , @xmath19 is really the scale of the internal momentum , and therefore , non - vanishing . @xmath124 , too , is non - vanishing in appropriate ( remember that we have lorentz symmery violation ) directions of @xmath125-channel scattering . [ 0.9 ] ( 190,110)(0,0 ) ( 3,10 ) ( 40,60)(0,100 ) ( 0,20)(40,60 ) ( 40,60)(100,60)2 ( 140,100)(100,60 ) ( 100,60)(140,20 ) ( 40,60)2 ( 20,60)[c][1.0]@xmath126 ( 120,60)[c][1.0]@xmath126 ( 100,60)2 ( 70,75)[c][1.1]@xmath19 ( 70,65)[c][1.2]@xmath127 in the previous subsection we made @xmath53 vanish by working in a supersymmetric theory . let us now study , what happens , when supersymmetry is ( softly ) broken . looking only at the trace - u(1 ) degrees of freedom of a generic noncommutative theory we have @xmath128.\ ] ] one easily checks that @xmath129 if susy is unbroken , all masses are equal . using supersymmetric matching between bosonic and fermionic degrees of freedom , @xmath130 we reproduce the vanishing of @xmath53 . if susy is softly broken this cancellation is not complete anymore ( in fact ( [ susycanc ] ) still holds and this removes the leading power - like ir divergence in @xmath53 , however , the subleading effects in @xmath53 survive ) . @xmath53 gets a contribution @xcite @xmath131+o(k^2 ) \\\nonumber \!\ ! & = & \!\ ! c \delta m^2_{\rm{susy}}+c'\sum_{j}\alpha_{j}d_j m^{2}_{j}\log(m^2_{j}\tilde{k}^2)+\cdots,\end{aligned}\ ] ] with known constants @xmath132 , @xmath133 and @xmath134 . this has dire consequences for the gauge boson . let us look at the equations of motion resulting from this additional lorentz symmetry violating contribution to the polarisation tensor . in presence of a higgs field which generates a mass term @xmath92 and using unitary gauge the field equations in presence of non vanishing @xmath25 read @xmath135 using that unitary gauge implies lorentz gauge , @xmath136 , we can simplify @xmath137 to proceed further it is useful to specify a direction for the momentum and the noncommutativity parameters . the photon flies in 3-direction and we have @xmath138 what is the corresponding value of @xmath139 ? since @xmath99 breaks lorentz invariance , we need to specify @xmath99 in a particular frame . for the latter , a natural one is the system where the cosmic microwave background is at rest . in this frame , we assume that the only non - vanishing components of @xmath99 are @xmath140 this yields , @xmath141 , @xmath142 in this direction , ( [ fieldnc2 ] ) yields @xmath143 in the other transverse direction , @xmath144 we find @xmath145 finally we have the third polarisation ( which can be gauged away if and only if @xmath94 ) , @xmath146 which results in @xmath147 we note that the different polarisation states do not mix due to the presence of @xmath53 . the second and the third polarisation state behave more or less like in the ordinary commutative case . however , the first has a modified equation of motion , ( [ direction1 ] ) , in presence of a non - vanishing @xmath53 ) one has to use the rescaled equation ( we set @xmath94 for simplicity ) @xmath148 . for @xmath149 , the second term vanishes since @xmath52 diverges in this limit . therefore , we find an additional solution . however , this solution is rather strange . it does not correspond to a pole in the propagator ( it goes like a @xmath150 ) moreover , if one calculates the cross section @xmath53 still upsets the angular dependence quite severely compared to the ordinary commutative case . ] . this is another strong argument against a trace - u(1 ) being the photon @xcite . if the gauge symmetry is unbroken and @xmath94 we usually have two massless polarisations . however , a non vanishing @xmath53 reduces this to one . the other one gets an additional mass @xmath151 . since only one polarisation is affected this is a strong lorentz symmetry violating effect . moreover , a negative @xmath53 would lead to tachyons while a positive mass is phenomenologically ruled out by the constraint @xcite @xmath152 on the photon mass ) to zero is not an option . since we have only a finite number of masses this is at best possible for a finite number of values of @xmath153 and we will surely find values of @xmath153 where @xmath53 is nonzero . ] . if we take the trace - u(1 ) as an additional ( to the photon ) gauge boson from the unbroken subgroup @xmath42 , we would still get strong lorentz symmetry violation since the trace - u(1 ) is not completely decoupled . in summary , we found in this section that additional trace - u(1 ) subgroups are not completely decoupled and should lead to observable effects . in particular , if susy is not exact we have non - vanishing @xmath53 which gives rise to strong lorentz symmetry violation which has not been observed . this rules out possibilities ( [ possa ] ) and ( [ possc ] ) of sec . 1 . moreover , we confirmed that a trace - u(1 ) is not suitable as a photon candidate . from the previous section we concluded that the trace - u(1 ) groups are unviable as candidates for the sm photon . therefore , it has been suggested to construct the photon from traceless u(1 ) subgroups @xcite . it turns out , however , that typically trace and traceless parts mix and the trace parts contribute their lorentz symmetry violating properties to the mixed particle . for u(2 ) broken by a fundamental higgs , the standard higgs mechanism yields the symmetry breaking @xmath154 . however , the remaining u(1 ) is a mixture of trace and traceless parts . if susy is broken , the trace - u(1 ) has a @xmath53 part in the polarisation tensor . taking this into account we find the following matrix for the equations of motion @xmath155 where the adjoint u(2 ) and polarisation indices are @xmath156 . we omitted the values @xmath157 and @xmath158 for the adjoint u(2 ) indices which do not mix with the trace - u(1 ) and are not qualitatively different from the commutative case . the matrix is block diagonal and the second and third polarisation ( lower right corner ) behave more or less like their commutative counterparts . we can concentrate on the upper left @xmath159 matrix corresponding to the transverse polarisations affected by @xmath53 . this @xmath159 matrix admits two solutions for the equations of motion . expanding for small @xmath53 we find , @xmath160 in analogy to ( [ direction1 ] ) . in absence of @xmath53 the first solution in eq . ( [ u2solution ] ) is a massless one corresponding to the massless combination of gauge bosons ( think of it as the photon ) . the second is a massive combination ( similar to the @xmath161 boson ) . the presence of non - vanishing @xmath53 again leads to a mass @xmath162 for the first solution and rules out the `` massless '' combination as a reasonable photon candidate . this example demonstrates that the disastrous effects of @xmath53 are also present in any combination which has an admixture of trace - u(1 ) degrees of freedom . hence , this rules out possibilities ( [ possb ] ) and ( [ possc ] ) from the introduction . in the introduction we already mentioned that a possible way out of the dilemma with the trace - u(1 ) s is the introduction of universal extra dimensions . let us now investigate this option . in most of the following discussion we will adopt a four - dimensional point of view in describing extra - dimensional theories . that is , because we are interested in renormalisation group effects associated with the 4-dimensional momentum , it makes more sense to include the effects of extra dimensions by considering the effect of a simple kaluza - klein tower of states . ( in the uv - complete string models there are other effects which , at one - loop order and in compact dimensions significantly larger than the string length , will be secondary . ) intuitively it is obvious that the main factor affecting the running of the gauge couplings will be the noncommutativity parameter @xmath124 , and in particular how it mixes the additional ( compact ) dimensions with the ordinary four large dimensions . we will now give a somewhat heuristic presentation of how @xmath124 affects the running of the gauge couplings . a more precise and general calculation is given in @xcite and we will just quote the results from there in the last part of this section . let us start by briefly reviewing power law running in the uv at scales well above the compactification scale . in the uv regime the planar diagrams dominate the two point function and so there is no difference to the ordinary commutative case ( see @xcite ) . because of this it is sufficient to use an intuitive approach based on thresholds . consider first the most simple case of one compact extra dimension of size @xmath163 . neglecting threshold effects the one loop running of the gauge coupling in four dimensions typically follows ( @xmath164 ) @xmath165 where the @xmath166 are coefficients depending on the spin and representation of the particle @xmath167 . in the sum only particles with mass @xmath168 smaller than the momentum scale @xmath169 contribute ( in any suitable massive renormalisation scheme ) . this leads to the typical decoupling of massive modes . for simplicity , let us now consider a situation where all particles have ( approximately ) the same mass @xmath170 . we find @xmath171 where we have chosen the sign of the constant @xmath172 such that it is positive when the theory is asymptotically free . ( for example , in @xmath173 supersymmetric pure gauge theory @xmath174 in this notation . ) above the compactification scale , more precisely at @xmath175 , the first kaluza - klein mode gives an identical contribution to the @xmath176-function , and in general one finds @xmath177 where @xmath178 is the number of kaluza - klein modes ( including the zero mode ) contributing at the scale @xmath19 . since the mass of the @xmath179th kaluza - klein mode is given by @xmath180 one easily finds the approximate formula @xmath181 where we have introduced the constant @xmath182 to account for the details of the compactification and threshold effects . this already suggests power law running . more precisely , one easily checks that for @xmath183 and appropriate initial conditions the solution approaches @xmath184 which is indeed a power law . expressions ( [ number ] ) and ( [ powersolution ] ) are easily generalized to arbitrary dimension @xmath185 ( @xmath183 ) @xmath186 where again the constant @xmath187 depends on the details of the compactification . the flow equation ( [ flow ] ) for the running coupling can be also discussed using the more natural effective coupling @xmath188 of the @xmath134-dimensional theory , @xmath189 from the lower - dimensional viewpoint ( [ rescaling ] ) can be understood by remembering that the amplitudes of all kaluza - klein modes add up and therefore increase the effective coupling by a factor @xmath190 . inserting ( [ rescaling ] ) into ( [ flow ] ) yields the flow equation for @xmath191 , @xmath192 if we start at small values for @xmath188 the coupling increases toward the infrared until it reaches a fixed point at @xmath193 . the corresponding coupling of the 4-dimensional theory is then @xmath194 in agreement with the last equation in ( [ flow ] ) . this discussion implies that power - law running in extra dimensions originates from a fixed point in the effective higher - dimensional coupling constant @xmath188 . this implies that the power - law running of @xmath195 is a strong coupling phenomenon in terms of @xmath188 and one should exercise caution since eqs . ( [ flow ] ) and ( [ rescaled ] ) are one - loop results . in particular a large number of extra - dimensions increases the value of the fixed point coupling and the approximation may break down . the issues of existence of a fixed point of @xmath188 were investigated in literature on extra - dimensional gauge theories , see e.g. @xcite . from now on we will continue assuming that ( ordinary commutative ) extra - dimensional gauge theories do provide a power - law running of the coupling in the extreme ultraviolet ( i.e. at energies well above the compactification scale ) . we will then show that in noncommutative settings the mixing between ultraviolet and infrared degrees of freedom will induce in the extreme infrared a power - law decoupling of the trace - u(1 ) degrees of freedom . as specified in eq . ( [ poltensor ] ) @xmath52 and therefore the gauge coupling depends on the additional scale @xmath124 ( cf . @xcite ) @xmath196 . in fact , the coupling depends only on the absolute values @xmath153 as well as @xmath197 , as can be seen from eqs . ( [ pi11 ] ) and ( [ pi12 ] ) . since we are mostly interested in low - energy physics ( compared to the compactification scale ) the effects of extra dimensions can contribute only through loops in perturbation theory . thus the external momenta @xmath198 are taken to be 4-dimensional , i.e. external particles will not include excited kaluza - klein modes , while internal loop momenta @xmath199 ( in feynman diagrams ) are kept general . in this section we consider a scenario where only the four infinite dimensions are noncommutative , @xmath200 where @xmath201 and @xmath202 from eq . ( [ defcoupling ] ) together with ( [ pinonplanar ] ) one easily finds that in a 4-dimensional noncommutative gauge theory with all particles of equal non - zero mass @xmath203 , the trace - u(1 ) couplings runs according to @xmath204 here @xmath205 is a positive number which specifies the non - planar contribution to the running gauge coupling . from eq . ( [ irrun ] ) one can see that in general noncommutative theory when we lower momentum - scale @xmath169 sufficiently , even very massive modes start to contribute . this holds for kaluza - klein modes , too , as long as we have noncommutativity only in the four infinite dimensions according to eq . ( [ fourdcase ] ) . in analogy to ( [ flow ] ) we find ( @xmath206 ) @xmath207 the right hand side of the ir flow equation in ( [ flowir ] ) has the opposite sign to that of the uv flow equation ( [ flow ] ) . this implies that the trace - u(1 ) coupling @xmath195 becomes small in the ir and the uv regimes . the enhancement by the @xmath208 factor gives the _ power - like _ decoupling of these unwanted degrees of freedom from the su(@xmath6 ) theory ( which is unaffected by the uv / ir mixing effects ) . if the matrix @xmath99 has nonvanishing entries that mix the ordinary four dimensions with the extra dimensions we may have a non - vanishing @xmath209 in the calculation of the polarisation tensor this leads to phase factors in the sum over the kaluza - klein modes , @xmath210 ( in addition to the usual @xmath5-dependent phases in non - planar contributions ) . in this situation it is advantageous to directly perform the sum over kaluza - klein modes in the polarisation tensor . we have done this explicitly in @xcite . here we will quote the result ( for an @xmath211 supersymmetric u(@xmath6 ) theory without adjoint matter fields ) , @xmath212 where @xmath213 are the compactification radii and @xmath124 is now the total noncommutative momentum @xmath214 ( @xmath215 ) . this equation is valid for @xmath216 with @xmath46 still defined as @xmath217 . the fact that the actual running is now given by replacing the 4-dimensional components of @xmath124 with the total @xmath218 is not too surprising since the infrared running comes from very ultraviolet modes , i.e. it involves momenta much higher than the compactification scale where the theory is effectively higher - dimensional . at these scales there is no distinction between the ordinary four dimensions and the extra dimensions . ( [ resultshort ] ) has the additional advantage that it already corresponds to the integrated result . it directly gives @xmath126 without the need to solve a differential equation ( @xmath219 ) , @xmath220 here we have fixed , @xmath221 where we still consider the @xmath211 case and none of the matter fields are in the adjoint representation . @xmath222 is a renormalisation constant determined from the bare coupling and the planar diagrams only . therefore in the regime ( [ regimelow ] ) this constant is connected to the gauge coupling of the su(@xmath6)-part ( up to logarithmic corrections which we neglected in our approximation ) @xmath223 in noncommutative field theories the gauge coupling is not the only part of the polarisation tensor that is affected by power law running . recall that in noncommutative field theories the ( 4-dimensional ) polarisation tensor has an additional lorentz symmetry violating part @xcite , which is called @xmath53 in eq . ( [ poltensor ] ) . for softly broken supersymmetry only the ir - singular ( pole ) contribution to @xmath25 vanishes , but a constant term @xmath224 remains . in ( [ nonsusy ] ) the sums run over _ all _ bosons and fermions . therefore , if we have compactified extra dimensions , we must include the kaluza - klein modes , effectively multiplying the four - dimensional @xmath225 by the number of kaluza - klein modes . the number of contributing kaluza - klein modes is , again , given roughly by @xmath226 of eq . ( [ flowir ] ) . hence , we find @xmath227 repeating the analysis of sect . [ eom ] one finds , again , one ordinary massless polarisation state and one with a lorentz symmetry breaking mass , @xmath228 which is roughly constant although both @xmath52 and @xmath53 scale with a power law . yet , these power laws cancel since they are the same for @xmath52 and @xmath53 . we found in sect . [ massec ] that the lorentz violating mass term for the trace - u(1 ) factors remains roughly constant . hence trace - u(1 ) s are still unsuitable as photon candidates . with a similar argument as in sect . [ u2example ] one finds that this also holds for mixtures of trace and traceless parts . therefore a suitable photon candidate must be constructed ( as in four dimensions ) from an unbroken combination of traceless generators . in @xcite we found that such a combination can only exist together with additional unbroken u(1 ) s which have nonvanishing trace . here the results of sect . [ powerlaw ] help us , since they allow for a fast decoupling of trace - u(1 ) degrees of freedom . this is in contrast to the four - dimensional case , where the ( only ) logarithmic decoupling necessitated incredibly large noncommutativity scales @xmath1 . with additional ( compactified ) space dimensions we have power law running according to ( [ flowir ] ) . this decouples the unwanted trace - u(1 ) s much faster in the ir thereby weakening the constraints dramatically . let us now estimate the new constraints obtained from power law running . as already mentioned earlier , current experiments probe the regime well below @xmath229 . to apply eq . ( [ resultu1 ] ) we also need @xmath230 , @xmath231 this is also assured , since the discussion of sect . [ simple ] shows that for @xmath232 the trace - u(1 ) and the su(@xmath6 ) have gauge couplings which are of the same order . ( until @xmath233 both gauge couplings are approximately equal and power law running sets in only below @xmath234 . ) neglecting the slow logarithmic running of the su(@xmath6 ) couplings we find from eqs . ( [ resultu1 ] ) and ( [ resultsun ] ) , @xmath235 where the @xmath236 in the second line holds for a pure noncommutative u(@xmath6 ) . to have @xmath237 we need @xmath238 as an illustration we have plotted the excluded region in fig . [ exclusion ] . [ 0.90 ] ( 190,180)(40,0 ) -plane ( in tev ) . the blue region is excluded because the trace - u(1 ) still has nonnegligible coupling . we have chosen @xmath239 , @xmath240 , @xmath241 , @xmath242 , and @xmath243.,title="fig:",width=359 ] ( -30,-10)[c][1.4]@xmath244 ( -300,150)[c][2.0]@xmath245 this shows that when we allow for a 5% uncertainty in the electromagnetic coupling at 100 gev , the allowed region of @xmath246 starts already at a few tev , depending on the compactification scale . in the last sect . [ powerlaw ] we have already seen that a modification of the theory at high energy scales ( there it was the introduction of extra dimensions ) can alter the behavior of the noncommutative field theory at infrared scales . therefore it makes sense to investigate the consequences of a modification at a high energy scale @xmath247 even for a 4 dimensional theory . a simple and natural possibility to model a non - local uv - finite microscopic theory like , e.g. , string theory , is to simply cut off all fluctuations with momenta larger than @xmath78 ( for a more detailed discussion of this choice see @xcite ) . as we will see this cutoff softens the problem of the unwanted mass term for the photon considerably . instead of a mass term one has vacuum birefringence at low momentum scales . if @xmath46 is close enough to the cutoff scale @xmath248 this vacuum birefringence can be pushed beyond the current experimental limits . thereby a window for @xmath46 opens where noncommutativity is still allowed . as experimental and observational sensitivity is likely to improve in the near future this provides an interesting probe for scales @xmath46 very close to the planck scale . in the following we will restrict ourselves to the case of a pure u(1 ) noncommutative gauge theory . the discussion of the previous sections shows how this can be generalised to more realistic situations where the photon gets an admixture of a trace - u(1 ) . let us now cut off the fluctuations with momenta larger than @xmath78 by introducing a factor of @xmath249 in the integral over the schwinger time @xmath125 . one obtains ( s. @xcite ) , @xmath250 \left[k_{0}\left({\sqrt{\delta_{j}}\over \lambda}\right ) -k_{0}\left({\sqrt{\delta_{j}}\over \lambda_{\rm eff}}\right)\right ] \nonumber\\ & + & { 1\over ( \pi)^2}\tilde{p}_{\mu}\tilde{p}_{\nu}\,\lambda_{\rm eff}^2\sum_{j}\alpha_{j}d(j ) \int_{0}^{1}dx\,\delta_{j}k_{2}\left({\sqrt{\delta_{j}}\over \lambda_{\rm eff}}\right ) \nonumber \\ & + & \delta_{\mu\nu}\left[\mbox { gauge non - invariant term } \right ] . \label{polarization}\end{aligned}\ ] ] we will neglect the gauge non - invariant terms in the following . they could be treated and eliminated by using modified ward - takahashi identities @xcite . the employed regularisation cuts off the modes @xmath251 in the loop integral in a smooth way . of course there are lots of different possibilities to do this . since universality does not hold , different regularisations will in principle lead to different results . however , as long as we leave the qualitative feature `` all momenta @xmath252 are cut off '' holds , we expect that the qualitative results we obtain remain true . let us first concentrate on @xmath52 , i.e. the running gauge coupling . in fig . [ gauge ] we plot the running gauge coupling for various values of the cutoff @xmath78 . as expected the running stops at the uv scale @xmath78 . in an ordinary commutative theory we would expect no further changes . here , however , we observe that the running stops , again , at an infrared scale @xmath253 . therefore the running for @xmath254 is essentially the same as that of a commutative u(1 ) gauge theory . [ 0.95 ] ( 190,180)(40,0 ) , respectively . we have fixed the maximal gauge coupling to be @xmath255 . one can clearly see that for finite values of the cutoff the running stops at @xmath256 in the uv and at @xmath253 in the ir.,title="fig:",width=359 ] ( -40,-10)[c][1.4]@xmath257 ( -285,150)[c][2.0]@xmath90 let us now turn to the @xmath53 part of the polarisation tensor . it , too , is affected by the presence of a finite uv cutoff . for softly broken broken we can easily derive the following approximate expressions @xmath258 where @xmath259 are known constants . following the arguments of sect . [ eom ] we can now solve the equations of motion for the two transverse photon polarisations ( noncommutativity matrix as given in eq . ( [ thetamatrix ] ) ) , @xmath260 let us now concentrate on the polarisation state @xmath261 which is affected by the presence of @xmath53 . inserting the approximate expressions ( [ approximate ] ) we can now study the dispersion relation , @xmath262 eq . ( [ limit1 ] ) yields the lorentz symmetry violating mass term of the order of @xmath225 already discussed in detail in sect . [ eom ] . without cutoff , i.e. in the limit @xmath263 this mass term persists down to @xmath264 . thereby excluding any chance that this can be the photon observed in nature . in presence of the cutoff eq . ( [ limit1 ] ) is only applicable for @xmath265 . masslessness of the photon is well tested up to at least @xmath266 . using @xmath267 this gives us a conservative lower bound of @xmath268 . nevertheless , this opens a rather large window of opportunity compared to the @xmath263 case where there was no allowed range of @xmath269 . for small photon momentum eq . ( [ limit2 ] ) applies . to understand ( [ limit2 ] ) better , let us restore the light speed @xmath270 in our equations and use @xmath271 for the frequency of the wave , @xmath272 with @xmath273 where we have used @xmath274 and @xmath275 . from eq . ( [ speed ] ) we can see that the photon @xmath261 propagates with a speed @xmath276 . since the @xmath261 photon propagates with @xmath270 we observe birefringence , i.e. different polarisations propagate with different speed . although @xmath277 seems to be quite small we should compare this to the current experimental sensitivity . in @xcite a study of all possible dimension four lorentz violating operators in electrodynamics was conducted and constraints derived . the most general dimensions four lagrangian which is gauge and cpt invariant but violates lorentz symmetry is , @xmath278 comparing the propagator derived from ( [ general ] ) with eq . ( [ poltensor ] ) we find @xmath279 in @xcite the coefficients of @xmath280 have been constrained using various methods . for laboratory measurements their estimate translates to , @xmath281 depending on the pattern of the noncommutativity . astrophysical obervations already provide a much tighter bound of @xmath282 while the strongest constraints come from observations of objects at cosmological distances ( see also @xcite ) @xmath283 noncommutative gauge symmetry in the weyl - moyal approach leads to two main features which have to be taken into account for sensible model building . first , there are strong constraints on the dynamics and the field content . the only allowed gauge groups are u(@xmath6 ) . in addition , the matter fields are restricted to transform as fundamental , bifundamental and adjoint representations of the gauge group . second , there are the effects of ultraviolet / infrared mixing . those lead to asymptotic infrared freedom of the trace - u(1 ) subgroup and , if the model does not have unbroken supersymmetry , to lorentz symmetry violating terms in the polarisation tensor for this trace - u(1 ) subgroup . for a 4 dimensional continuum theory we have demonstrated that , although the trace - u(1 ) decouples in the limit @xmath264 , the coupling is not negligibly small at finite momentum scales @xmath19 , as they appear , for example , in scattering experiments . therefore , observations rule out additional unbroken ( massless ) trace - u(1 ) subgroups . noncommutativity explicitly breaks lorentz invariance . therefore an additional lorentz symmetry violating structure is allowed in the polarisation tensor . this structure is absent only in supersymmetric models . if supersymmetry is ( softly ) broken , this additional structure is present in the polarisation tensor of the trace - u(1 ) . it leads to an additional mass @xmath284 for one of the transverse polarisation states @xcite . the tight constraints on the photon mass therefore exclude trace - u(1 ) s as a candidate for the photon . it turns out that even a small admixture of a trace part to a traceless part ( unaffected by these problems ) is fatal . the only way out seems to be the construction of the photon from a completely traceless generator . a group theoretic argument shows , that this is impossible whithout having additional unbroken u(1 ) subgroups . however , those are already excluded from the arguments given above . this result severely restricts the possibilities to construct a noncommutative standard model extension . if all of the constraints given at the beginning are fulfilled the noncommutativity scale is pushed to scales far beyond @xmath285 . in general there is no reason to assume that the simple noncommutative model used here describes correctly the physics at energies ranging from a few ev up to the planck mass . in fact , due to the ultraviolet / infrared mixing , a different ultraviolet embedding of the theory would modify the theory not only in the ultraviolet , but also in the infrared which can drastically alter these conclusions . e.g. , a powerlike decoupling of the trace - u(1 ) can effectively hide them from observation . we have demonstrated that in a noncommutative u(@xmath6 ) gauge theory with compact extra dimensions , the ultraviolet / infrared mixing effects lead to such a fast power - like decoupling of the trace - u(1 ) degrees of freedom . in such a setting the bounds are weakened considerably if the compactification scale is small enough . as an alternative to extra dimensions we have discussed a modification obtained by simply cutting off all fluctuations with momenta larger than a cutoff @xmath248 . the presence of an ultraviolet cutoff @xmath78 induces an effective infrared scale @xmath286 below which the theory behaves essentially like a commutative gauge theory and the theory shows strong effects of noncommutativity at all scales . ] . in particular , up to threshold effects the running is that of a commutative field theory . if supersymmetry is broken , we have a lorentz symmetry violating mass term at scales @xmath287 in accord with @xcite . however , below @xmath288 the mass term turns into a modification of the phase velocity of plane wave solutions , leading to birefringence . nevertheless , if such a trace - u(1 ) gauge boson is to be interpreted as ( part of ) a photon a mass is not acceptable and birefringence must be smaller than the experimental limits . using the most stringent limits from cosmological observations one obtains a rather strong limit of @xmath289 . if we use the more conservative astrophysical or laboratory limits the same argument yields only @xmath290 . in this setting high precision measurements of the properties of light are a wonderful tool to test ( nearly ) planck scale physics . 9 j. jaeckel , v. v. khoze and a. ringwald , `` telltale traces of u(1 ) fields in noncommutative standard model extensions , '' jhep * 0602 * ( 2006 ) 028 [ hep - ph/0508075 ] . s. a. abel , j. jaeckel , v. v. khoze and a. ringwald , `` noncommutativity , extra dimensions , and power law running in the infrared , '' jhep * 0601 * ( 2006 ) 105 [ hep - ph/0511197 ] . s. a. abel , j. jaeckel , v. v. khoze and a. ringwald , `` vacuum birefringence as a probe of planck scale noncommutativity , '' in preparation . n. seiberg and e. witten , `` string theory and noncommutative geometry , '' jhep * 9909 * ( 1999 ) 032 [ hep - th/9908142 ] . m. r. douglas and n. a. nekrasov , `` noncommutative field theory , '' rev . mod . phys . * 73 * ( 2001 ) 977 [ hep - th/0106048 ] . r. j. szabo , `` quantum field theory on noncommutative spaces , '' phys . * 378 * ( 2003 ) 207 [ hep - th/0109162 ] . s. minwalla , m. van raamsdonk and n. seiberg , `` noncommutative perturbative dynamics , '' jhep * 0002 * , 020 ( 2000 ) [ hep - th/9912072 ] . a. matusis , l. susskind and n. toumbas , `` the ir / uv connection in the non - commutative gauge theories , '' jhep * 0012 * ( 2000 ) 002 [ hep - th/0002075 ] . j. madore , s. schraml , p. schupp and j. wess , `` gauge theory on noncommutative spaces , '' eur . j. c * 16 * ( 2000 ) 161 [ hep - th/0001203 ] . x. calmet , b. jurco , p. schupp , j. wess and m. wohlgenannt , `` the standard model on non - commutative space - time , '' eur . j. c * 23 * ( 2002 ) 363 [ hep - ph/0111115 ] . v. v. khoze and j. levell , `` noncommutative standard modelling , '' jhep * 0409 * ( 2004 ) 019 [ hep - th/0406178 ] . v. v. khoze and g. travaglini , `` wilsonian effective actions and the ir / uv mixing in noncommutative gauge theories , '' jhep * 0101 * ( 2001 ) 026 [ hep - th/0011218 ] . t. j. hollowood , v. v. khoze and g. travaglini , `` exact results in noncommutative n = 2 supersymmetric gauge theories , '' jhep * 0105 * ( 2001 ) 051 [ hep - th/0102045 ] . k. matsubara , `` restrictions on gauge groups in noncommutative gauge theory , '' phys . lett . b * 482 * ( 2000 ) 417 [ hep - th/0003294 ] . a. armoni , `` comments on perturbative dynamics of non - commutative yang - mills theory , '' nucl . phys . b * 593 * ( 2001 ) 229 [ hep - th/0005208 ] . j. m. gracia - bondia and c. p. martin , `` chiral gauge anomalies on noncommutative r**4 , '' phys . b * 479 * ( 2000 ) 321 [ hep - th/0002171 ] . s. terashima , `` a note on superfields and noncommutative geometry , '' phys . b * 482 * ( 2000 ) 276 [ hep - th/0002119 ] . m. chaichian , p. presnajder , m. m. sheikh - jabbari and a. tureanu , `` noncommutative gauge field theories : a no - go theorem , '' phys . b * 526 * ( 2002 ) 132 [ hep - th/0107037 ] . m. hayakawa , `` perturbative analysis on infrared aspects of noncommutative qed on r**4 , '' phys . lett . b * 478 * ( 2000 ) 394 hep - th/9912094 ; `` perturbative analysis on infrared and ultraviolet aspects of noncommutative qed on r**4 , '' hep - th/9912167 . m. chaichian , p. presnajder , m. m. sheikh - jabbari and a. tureanu , `` noncommutative standard model : model building , '' eur . j. c * 29 * ( 2003 ) 413 hep - th/0107055 . l. alvarez - gaume and m. a. vazquez - mozo , `` general properties of noncommutative field theories , '' nucl . b * 668 * ( 2003 ) 293 [ hep - th/0305093 ] ; + `` comments on noncommutative field theories , '' hep - th/0311244 . s. eidelman _ et al . _ [ particle data group ] , `` review of particle physics , '' phys . b * 592 * ( 2004 ) 1 . p. aschieri , t. grammatikopoulos , h. steinacker and g. zoupanos , `` dynamical generation of fuzzy extra dimensions , dimensional reduction and symmetry breaking , '' hep - th/0606021 . k. r. dienes , e. dudas and t. gherghetta , `` grand unification at intermediate mass scales through extra dimensions , '' nucl . b * 537 * ( 1999 ) 47 [ hep - ph/9806292 ] . d. m. ghilencea , `` regularisation techniques for the radiative corrections of the kaluza - klein states , '' phys . d * 70 * , 045011 ( 2004 ) [ hep - th/0311187 ] ; `` compact dimensions and their radiative mixing , '' phys . d * 70 * , 045018 ( 2004 ) [ hep - ph/0311264 ] ; `` wilson lines corrections to gauge couplings from a field theory approach , '' nucl . b * 670 * , 183 ( 2003 ) [ hep - th/0305085 ] ; `` higher derivative operators as loop counterterms in one - dimensional field theory orbifolds , '' jhep * 0503 * , 009 ( 2005 ) [ hep - ph/0409214 ] . h. gies , `` renormalizability of gauge theories in extra dimensions , '' phys . d * 68 * ( 2003 ) 085015 [ hep - th/0305208 ] . m. reuter and c. wetterich , `` effective average action for gauge theories and exact evolution equations , '' nucl . b * 417 * , 181 ( 1994 ) . u. ellwanger , `` flow equations and brs invariance for yang - mills theories , '' phys . b * 335 * , 364 ( 1994 ) [ hep - th/9402077 ] . f. freire , d. f. litim and j. m. pawlowski , `` gauge invariance and background field formalism in the exact renormalisation group , '' phys . b * 495 * , 256 ( 2000 ) [ hep - th/0009110 ] . v. a. kostelecky and m. mewes , `` signals for lorentz violation in electrodynamics , '' phys . d * 66 * ( 2002 ) 056005 [ hep - ph/0205211 ] . v. a. kostelecky and m. mewes , `` cosmological constraints on lorentz violation in electrodynamics , '' phys . rev . lett . * 87 * ( 2001 ) 251304 [ hep - ph/0111026 ] .

we report ultraviolet spectra of the high - redshift ( ) quasar , he 2347 , taken by the cosmic origins spectrograph ( cos ) on the _ hubble space telescope _ ( _ hst _ ) . spectra in the g130 m ( medium - resolution , 11351440 ) and g140l ( low - resolution , 10302000 ) gratings exhibit patchy gunn - peterson absorption in the 303.78 ly line of he ii between 2.392.87 ( g140l ) and 2.742.90 ( g130 m ) . with cos , we obtain better spectral resolution , higher - s / n , and better determined backgrounds than previous studies , with sensitivity to abundance fractions in filaments of the cosmic web . the optical depths from cos are higher than those with the _ far ultraviolet spectroscopic explorer _ ( _ fuse _ ) and range from to , with a slow recovery in mean optical depth to at . the / optical - depth ratio varies ( 10100 for , and 5500 for ) on scales ( 10.8 mpc in comoving radial distance at ) , with numerous flux - transmission windows between 11351186 . the absorption extends to 1186.26 ( ) , including associated absorbers with and minimal proximity effect " of flux transmission at the edge . we propose a qso systemic redshift , some higher than that derived from emission . three long troughs ( 410 or 2560 mpc comoving distance ) of strong absorption between are uncharacteristic of the intergalactic medium if reionized at . contrary to recent indirect estimates ( ) from optical depths , the epoch of reionization may extend to .

1008.2957

the epoch of reionization in hydrogen has become a topic of considerable interest ( barkana & loeb 2001 ; fan , carilli , & keating 2006 ; meiksin 2009 ; furlanetto et al . 2009 ) as a probe of the transition from neutral to ionized hydrogen in the intergalactic medium ( igm ) . this transition occurred somewhere between redshifts @xmath3 612 , marking the exit from the cosmic dark ages " , beginning at the time when the first stars and galaxies formed at redshifts @xmath24 ( tegmark et al . 1997 ; ricotti , gnedin , & shull 2002 ; trenti & shull 2010 ) . helium underwent similar reionization from to ( that is , from he@xmath25 to he@xmath26 ) at @xmath27 ( reimers et al . 1997 ; shull et al . 2004 ) , most likely mediated by the harder ( @xmath28 ev ) radiation from quasars and other active galactic nuclei ( agn ) . with a 4 ryd ionization potential , he@xmath29 is harder to ionize than h@xmath30 , and he@xmath26 recombines 56 times faster than h@xmath29 ( osterbrock & ferland 2006 ; fardal , giroux , & shull 1998 ) . for these reasons , and the fact that most hot stars lack strong 4 ryd continua , it is believed that agn are the primary agents of reionization . ionization models find that , owing to its resilience , he@xmath29 is much more abundant than h@xmath30 , with predicted column - density ratios @xmath31 n()/n ( ) @xmath32 50100 ( miralda - escud et al . 1996 ; fardal , giroux , & shull 1998 ) . determining when and how the universe was ionized has been an important question in cosmology for decades ( gunn & peterson 1965 ; sunyaev 1977 ) . although recent progress has led to quantitative constraints , we still do not know whether galaxies are the sole agents of hydrogen reionization , and the epoch of reionization remains uncertain . we _ do _ know that hydrogen reionization in the igm is complete by @xmath33 , based on the rapid evolution between redshifts @xmath34 and @xmath35 of ly@xmath2 gunn - peterson ( gp ) absorption from neutral hydrogen along lines of sight to qsos ( fan et al . 2002 ; gnedin 2004 ) . the optical depth of the cosmic microwave background ( komatsu et al . 2010 ) sets a @xmath36 constraint on instantaneous reionization at @xmath37 and a @xmath38 confidence interval of @xmath39 . a limit @xmath40 is set by the detection of ly@xmath2 emitters at @xmath41 ( major evolution in the observed luminosity function would be expected if neutral gas were present around the emitters ) . these observations are in line with @xmath42cdm simulations ( trac & cen 2007 ; cen 2003 ; gnedin 2004 ) and ionization - front ( i - front ) models ( venkatesan , tumlinson & shull 2003 ; bolton & haehnelt 2007 ; shull & venkatesan 2008 ) . the current data leave open several hydrogen ionization scenarios , some involving simple reionization at @xmath43 and others with more complex ionization histories that depend on the star - formation rate at @xmath44 and the transition from pop iii to pop ii stars . numerical models of the reionization process often follow the expansion and overlap of i - fronts , but the details are complicated by geometric uncertainties in the ionizing sources , higher recombination rates in denser gas , and radiative transfer of lyman continuum ( lyc ) photons . in particular , the gas is highly structured in both density and temperature . the reionization epoch ( @xmath45 ) and ionization processes are also poorly characterized . the thermal evolution of the igm includes many subtle effects , with degeneracies arising from fluctuations in density , temperature , and ionization fraction , as well as spectral hardness of the ionizing sources . previous estimates of @xmath21 were based on indirect measurements of temperature , ionization , and average optical depth of ly@xmath2 lines of , and on apparent shifts in / absorption - line ratios arising from changes in the 4-ryd continuum radiation field as was ionized to . early estimates of the igm temperature evolution and equation of state from line widths suggested a change at @xmath46 ( ricotti , gnedin , & shull 2000 ; schaye et al . theuns et al . ( 2002 ) analyzed the average optical depth of ( ) ly@xmath2 lines in 1061 quasar spectra , using low - resolution data from the sloan digital sky survey ( bernardi et al . 2003 ) . from a weak dip in @xmath47 at @xmath48 , these authors suggested that reionization started at redshift @xmath49 and lasted until @xmath50 . several recent studies with higher - resolution spectroscopic data have not confirmed the @xmath51 feature ( mcdonald et al . 2005 ; kim et al . faucher - giguere et al . ( 2008 ) found a feature in @xmath51 ( ) near @xmath52 , in a sample of 86 high - resolution , high - s / n quasar spectra taken with keck ( esi , hires ) and magellan ( mike ) . however , they question the reionization explanation of the depth and narrow redshift extent . they also note that at least three physical effects could produce a feature in in @xmath47 : changes in igm temperature , electron number density , and ionizing background . recent numerical simulations ( bolton et al . 2009 ; mcquinn et al . 2009 ; lidz et al . 2010 ) also question whether reionization is responsible for the proposed feature in @xmath47 at @xmath53 . the most direct estimates of reionization come from far - ultraviolet spectra of lyman-@xmath2 absorption toward high - redshift agn . because of strong galactic interstellar absorption in the lyman continuum , the @xmath54 line is only observable in the far ultraviolet , at redshifts @xmath55 when the line shifts longward of 912 . a small number of bright agn at @xmath3 2.93.3 have been observed with the _ hubble space telescope _ ( _ hst _ ) and the _ far ultraviolet spectroscopic explorer _ ( _ fuse _ ) . these data show moderate optical depths in the line , ranging up to @xmath56 , with a gradual recovery of transmission ( @xmath57 ) at @xmath8 . the best absorption data have been acquired toward three high - redshift agn : q0302@xmath1003 ( jakobsen et al . 1994 ; hogan , anderson , & rugers 1997 ; heap et al . 2000 ) ; and he 2347@xmath14342 ( reimers et al . 1997 ; kriss et al . 2001 ; smette et al . 2002 ; shull et al . 2004 ; zheng et al . 2004 ) ; and hs 1700@xmath586416 ( davidsen , kriss , & zheng 1996 ; fechner et al . 2006 ) . from an analysis of the optical - depth evolution in moderate - resolution fuse data toward he 2347@xmath14342 , shull et al . ( 2004 ) suggested that the reionization epoch of he@xmath29 occurred at @xmath59 . observations of with _ hst _ and _ fuse _ also constrain the thermodynamic state of the igm and the spectra of ionizing sources . for example , the high abundance ratio , n(he ii)/n(h i ) @xmath32 50100 , observed toward he 2347@xmath14342 and hs 1700@xmath586416 ( kriss et al . 2001 ; smette et al . 2002 ; shull et al . 2004 ; zheng et al . 2004 ; fechner et al . 2006 ) requires a soft ionizing radiation field , @xmath60 , with mean spectral index @xmath61 ( fardal et al . as observed in the qso rest - frame , the euv ( 12 ryd ) spectral indices of quasars exhibit a wide range ( telfer et al . 2002 ; scott et al . 2004 ) . more intriguing are the spectroscopic observations of and taken by _ fuse _ , the keck telescope , and the very large telescope ( vlt ) . kriss et al . ( 2001 ) and shull et al . ( 2004 ) report variations in @xmath62 over 210 mpc scales ( @xmath63 ) which imply spatial fluctuations in the radiation field at the ionization edges of ( 1 ryd ) and ( 4 ryd ) . most of these variations probably arise in the 4 ryd continuum , whose photons have greater penetrating power . these flux differences at 1 ryd and 4 ryd are then amplified by absorption and reprocessing by the igm ( haardt & madau 1996 ; fardal et al . 1998 ) . in this paper , we present new , high - quality , far - uv spectroscopic observations of intergalactic absorption toward he 2347@xmath14342 , using the cosmic origins spectrograph ( cos ) on _ hst _ ( green et al . 2010 ; osterman et al . this target is of considerable interest as a bright ( @xmath64 , @xmath65 ) continuum source , previously observed by the _ fuse _ , keck , and vlt spectrographs to study intergalactic ly@xmath2 absorption in both and . he 2347@xmath14342 is the first of three agn targets scheduled for cos guaranteed - time observations of the reionization epoch . the data were acquired with both the moderate - resolution g130 m grating ( @xmath66 , @xmath67 ) and the low - resolution g140l grating ( @xmath68 , @xmath69 ) . in the 303.78 line of , these wavelength bands allow us to probe redshifts down to @xmath70 ( g130 m ) and @xmath71 ( g140l ) . the cos data enable many improvements compared to previous studies . first , the high far - uv throughput of the cos / g130 m grating provides higher signal - to - noise ( s / n ) and better photometric accuracy . second , the low background of the cos detectors allows us to detect very weak flux - transmission through the igm , characterize low flux levels in absorption troughs , and probe regions of high optical depth , @xmath72 . finally , low - resolution g140l spectra ( 1000 - 2000 ) allow us to study the recovery of the optical depth at @xmath8 and characterize the continuum longward of the edge . in 2 we discuss the observations and data reduction techniques for both g130 m and g140l gratings . in 3 we display the fitted quasar continuum , @xmath73 , which we use to extrapolate below the edge at @xmath74 . from this continuum and the transmitted fluxes , we derive optical depths in the ly@xmath2 forest and analyze the fluctuating absorption . we observe minimal effects of local photoionization expected from proximity to this very luminous qso . in the ly@xmath2 forest at @xmath75 , we see narrow ( 25 , 1230 mpc of comoving radial distance ) windows of partial flux transmission , which we interpret as patchy reionization " , interspersed with broad absorption troughs ( 410 intervals , 2560 mpc ) with high optical depths , @xmath76 . at @xmath77 ( @xmath78 ) the mean opacity drops to @xmath79 , consistent with the overlap of the i - fronts . because is much more abundant than , its opacity through the cosmic web remains high . in 4 we summarize our observations and their implications for the reionization epoch . with reasonable assumptions about the physical state of the igm at @xmath80 , borne out by studies of and ly@xmath2-forest absorption and cosmological simulations , we can probe the reionization epoch down to fractional ionization @xmath81 in underdense regions ( @xmath82 ) relative to the mean baryon density of the cosmic web . this ionization level provides much greater sensitivity to conditions at @xmath83 than possible with at @xmath33 ( fan et al . coupled with galaxy surveys , cosmological simulations , and photoionization modeling , these uv spectra will help us understand the propagation and overlap of the helium ionization fronts . he2347@xmath14342 was observed with the short - wavelength , medium - resolution far - uv mode of hst - cos ( g130 m ) on 2009 november 05 for a total of 28,459 s. descriptions of the cos instrument and on - orbit performance characteristics are in preparation ( green et al . 2010 ; osterman et al . 2010 ) . in order to achieve continuous spectral coverage across the g130 m bandpass ( 1135 @xmath85 1440 ) and to minimize fixed - pattern noise , we made observations at two central wavelength settings ( 1291 and 1300 ) with four focal - plane offset locations in each grating setting ( fp - pos = 1 , 2 , 3 , 4 ) . this combination of grating settings ensures the highest signal - to - noise observations at the shortest wavelengths available to the g130 m mode ( @xmath86 ) at a resolving power of @xmath87 ( @xmath88 km s@xmath89 ) . in addition , cos observed he2347@xmath14342 in the low - resolution g140l mode for 11,558 s on 2009 november 06 . the g140l segment - a data provide some overlap with the g130 m observations between 12401440 , and they extend our low - resolution spectral coverage ( @xmath68 ) out to @xmath90 . from g140l segment - b observations and custom data - processing steps described in section 2.2 , we extend our reliable spectral coverage over part of the g140l bandpass , between 10301175 . table 1 lists the relevant cos observational parameters for the data sets used in this work . cccccc 2009 nov 05 & g130 m & 1291 & 1,2,3,4 & 9 & 16535 + 2009 nov 05 & g130 m & 1300 & 1,2,3,4 & 9 & 11924 + 2009 nov 06 & g140l & 1230 & 1,2,3,4 & 4 & 11558 all observations were centered on the qso target , he2347@xmath14342 at j2000 coordinates ( r.a . @xmath91 , decl . = @xmath92 ) . cos performed nuv imaging target acquisitions with the mirrora / psa mode , yielding good centering in the cos aperture to maximize throughput and resolving power . the g130 m data were processed with the cos calibration pipeline , calcos v2.11b , and combined with a custom idl coaddition procedure . the analog nature of the cos micro - channel plate makes it susceptible to temperature changes . electronically injected pulses ( stims ) at opposite corners of the detector allow for the tracking and correction of any drift ( wavelength zero - point ) and/or stretch ( dispersion solution ) of the recorded data location as a function of temperature . at the beginning of the observations of he 2347@xmath14342 , the detector proved to be substantially colder than the nominal operating temperature of the micro - channel plate . this allowed one of the electronic stim - pulses to accumulate at the bottom of the detector , compromising the wavelength solution for the first four g130 m exposures . the detector warmed to the normal operating range for the remaining g130 m observations . we reprocessed the earlier exposures using extrapolated stim positions to ensure a consistent wavelength solution over the course of the observations . the g140l data were unaffected . flat - fielding , alignment , and coaddition of the processed cos / g130 m exposures were carried out using idl routines developed by the cos gto team specifically for cos fuv datadanforth / costools.html for our coaddition and flat - fielding algorithm and additional discussion . ] . first , the data were corrected for fixed - pattern instrumental features . although attempts at a true `` flat - fielding '' of cos data show promise , the technique is not yet sufficiently robust to improve data of moderate s / n . the quantum efficiency is improved by the presence of a series of wires , called the quantum - efficiency grid , placed above the detector . these wires create shadows in the spectrum that are removed during data reduction . an ion - repellor grid reduces the background rate by preventing low - energy thermal ions from entering the open - faced detector . we are able to correct the narrow shadows of @xmath9315% opacity arising from the grid wires . a one - dimensional map of grid - wire opacity for each detector was shifted from detector coordinates into wavelength space and divided from the flux and error vectors . exposure time in the location of grid wires was decreased by @xmath94 , giving these pixels less weight in the final coaddition . we also modified the error vector and local exposure time at the edges of the detector segments to underweight the flux contributions from these regions . with a total of eight different wavelength settings in the final co - addition , any residual instrumental artifacts from grid - wire shadows and detector segment boundaries should have little effect on the final spectrum . we performed two sets of wavelength - scale shifts . first , the g130 m exposures were aligned with each other and interpolated onto a common wavelength scale . then , we transferred the g130 m wavelengths to a ground - based reference frame of the data , taken by d. reimers with the vlt / uves ( ultraviolet visual echelle spectrograph ) in eso program 071.a-0066 . these data were used in the analysis paper of _ fuse _ data ( shull et al . one exposure in each cos detector was picked as a wavelength reference , and the remaining exposures were cross - correlated with it . the wavelength region of cross - correlation for each case was picked to include a strong interstellar absorption feature , although shifts were typically a resolution element or less . residual errors on the target centering within the cos aperture caused a systematic shift between the strong interstellar lines and their expected lsr velocities . the cos wavelength solution is relatively accurate , but it may suffer from linear offsets from an absolute frame . since our later analysis involves comparing the rapidly - varying ( optical ) and ( uv ) profiles , wavelength alignment is critical . the only absorption features in the normalized vlt data clearly not associated with the ly@xmath2 forest are a pair of narrow absorbers at 4787.12 and 4795.08 , consistent in relative strength and separation with the doublet at 1548.195 and 1550.770 , redshifted to @xmath95 . at this redshift , cos / g130 m covers 384466 . the @xmath96 line at 1194.1 in the cos data corresponds to the doublet in the optical data , with similar line profiles . we therefore shift the cos data by @xmath97 ( approximately one g130 m resolution element ) to align with the optical spectra . the flux at each position was taken to be the exposure - time weighted mean of flux . since exposure time was reduced in certain wavelength locations , as noted above , pixels near detector edges and where grid - wire shadows were removed received less weight than those in less suspect locations . the combined g130 m data show @xmath98 per 7-pixel resolution element , sufficient to detect narrow absorption features near 1200 down to equivalent widths @xmath99 m at @xmath100 significance . the g140l observations were originally processed with calcos 2.11b . however , the @xmath101 data from this processing is not appropriate for scientific analysis . we describe our custom processing in section 2.2 . figure 1 shows the g130m / g140l continuum from 1200 to 2000 , which we have fitted and extrapolated below the edge at 1186.2 . the g140l wavelengths were shifted to match g130 m , with no corrections needed to match g130 m and g140l fluxes . these curves were de - reddened using color excess , @xmath102 , corresponding to column density @xmath103 derived by fitting interstellar ly@xmath2 absorption . we adopted a standard interstellar ratio , n@xmath104 @xmath105 mag@xmath106 , from shull & van steenberg ( 1985 ) and diplas & savage ( 1994 ) and used the uv selective extinction curve of fitzpatrick & massa ( 1999 ) . our derived extinction , @xmath102 , lies intermediate between previous values of @xmath107 ( reimers et al . 1997 ) and @xmath108 ( smette et al . figure 2 shows the g130 m spectrum ( @xmath109 ) with prominent absorption troughs , numerous flux - transmission windows shortward of the edge , and the two proposed qso systemic redshifts , @xmath110 ( reimers et al . 1997 ) and @xmath111 ( this paper ) . this issue is discussed further in section 3.1 . the strong features are listed in table 2 , with the optical depths calculated from the depth of flux transmission , @xmath112 , relative to the extrapolated continuum @xmath113 . uncertainties in @xmath114 were computed as root - mean - square averages over redshift bins , @xmath115 , ranging from 0.004 to 0.02 at wavelength centroids @xmath116 . we conservatively rounded errors to the nearest 0.25 in @xmath114 , using uncertainties arising from converting errors in flux transmission to errors in optical depth . in regions of lower s / n in the g140l grating ( @xmath117 ) , we smoothed over two resolution elements ( @xmath118 ) to reduce the noise before the optical depth and error were evaluated . the g140l - b spectra shown in figure 1 were smoothed over three resolution elements . cccccc 2.90151 & 0.00400 & 1185.20 & @xmath119 5 & g130 m & @xmath120 + 2.89765 & 0.00800 & 1184.03 & @xmath119 5 & g130 m & @xmath120 + 2.88721 & 0.00823 & 1180.86 & @xmath119 5 & g130 m & @xmath120 + 2.87445 & 0.01317 & 1176.98 & @xmath119 5 & g130 m & z04 - 3a + 2.84700 & 0.03456 & 1168.64 & @xmath119 5 & g130 m & z04 - 3cde + 2.77471 & 0.05267 & 1146.68 & @xmath119 5 & g130 m & z04 - 3hij + 2.71324 & 0.00823 & 1128.01 & @xmath121 & g140l & @xmath120 + 2.69082 & 0.01646 & 1121.20 & @xmath122 & g140l & @xmath120 + 2.66414 & 0.00494 & 1113.09 & @xmath123 & g140l & @xmath120 + 2.65061 & 0.00494 & 1108.98 & @xmath124 & g140l & @xmath120 + 2.63785 & 0.01317 & 1105.11 & @xmath125 & g140l & @xmath120 + 2.61814 & 0.00988 & 1099.12 & @xmath126 & g140l & @xmath120 + 2.59804 & 0.00988 & 1093.01 & @xmath127 & g140l & @xmath120 + 2.57059 & 0.01317 & 1084.67 & @xmath128 & g140l & @xmath120 + 2.55000 & 0.02008 & 1078.42 & @xmath129 & g140l & @xmath120 + 2.52999 & 0.02008 & 1072.34 & @xmath130 & g140l & @xmath120 + 2.49327 & 0.01975 & 1061.19 & @xmath131 & g140l & @xmath120 + 2.47001 & 0.02008 & 1054.12 & @xmath132 & g140l & @xmath120 + 2.43683 & 0.01811 & 1044.04 & @xmath133 & g140l & @xmath120 during the servicing mission-4 observatory verification ( smov-4 ) period , it was discovered that the mgf@xmath134/al mirrors of @xmath84 have retained approximately 80% of their pre - flight reflectivity at @xmath135 ( mccandliss et al . 2010 ) . this result has opened the door for use of the short - wavelength response of cos , in particular the g140l mode , to perform spectroscopic observations at wavelengths inaccessible to previous _ hst _ instruments ( @xmath77 ) . see appendix a of france et al . ( 2010 ) for preliminary work with this mode . the calcos pipeline processing of g140l segment b ( 400 @xmath85 1175 ) is not sufficiently mature to produce one - dimensional spectra appropriate for scientific analysis . we therefore processed the data with a newer version of the calcos pipeline ( v2.12 ) , employing specially created reference files for flux and wavelength calibrations in the short - wavelength segment - b setting . the custom reference files that were created to supplement the calcos 2.12 release were a first - order dispersion solution for @xmath136 and a flux calibration curve . the wavelength calibration was created by fitting line centers to the pressure - broadened lyman series lines observed in cos calibration spectra of white dwarfs ( mccandliss et al . while g140l segment a uses a second - order polynomial wavelength solution , the relatively small amount of physical detector area used by g140l segment b allows a reasonable fit with a linear solution . we find a dispersion of @xmath137 m pixel@xmath89 , and a long - wavelength zero - point of 1245.26 . to align with g130 m data , we shifted g140l - b wavelengths by @xmath138 . for example , 1245.26 ( old g140l ) becomes 1249.16 ( new g140l ) . the g140l - a wavelengths were shifted by only @xmath139 ( @xmath140 resolution element ) . the calibration curve , translating count rate ( counts s@xmath89 ) into flux ( erg @xmath105 s@xmath89 @xmath141 was created using the g140l segment - b effective area curve derived by mccandliss et al . ( we caution that there may still be significant uncertainty in this flux calibration . at @xmath142 , we should be within @xmath143% of the correct value . the accuracy is likely @xmath932025% from 10251110 and @xmath931015% from 1120 longward . continued testing of g140l - b observations will refine this calibration . this custom reduction produces reprocessed one- and two - dimensional spectral data products , which can be coadded following the procedure described above and in danforth et al . the coadded data were used to create a low - resolution ( @xmath144 , @xmath145 ) spectrum of he 2347@xmath14342 at @xmath146 . in principle , the cos / g140l wavelength coverage extends to very low wavelengths ( @xmath147 ) . however , the combination of low instrumental effective area , @xmath148 @xmath149 at @xmath150 ( mccandliss et al . 2010 ) and interstellar absorption by atomic and molecular hydrogen limit the working wavelength coverage . for the absorption toward he 2347@xmath14342 , we present g140l data from 1030 @xmath151 1174 , corresponding to @xmath152 . we cross - calibrated the g140l segment - b spectrum with the short - wavelength data from g130 m , finding that the flux calibration is better than 10% in the overlap region . the data were then used to extend our investigation of the ionization state in the igm down to @xmath153 ( 1030 ) . figure 3 provides a comparison among three spectrographs : cos / g130 m , cos / g140l , and _ fuse _ ( lif1 and lif2 channels ) . there is a fair amount of agreement in fluxes among these instruments , although the _ fuse _ backgrounds at 11601187 are obviously incorrect . this comparison gives us confidence in using g140l data down to 1030 . the systemic redshift of he 2347@xmath14342 is of interest , both to fix the expected location of the edge and to infer the metagalactic radiation field from the qso proximity effect ( dallaglio , wisotzki , & worseck 2008a , b ) . a significant puzzle for this sight line is the 6 offset between the extent of strong absorption ( out to 1186 ) and the expected edge ( 1180 ) based on an uncertain qso emission - line redshift . reimers et al . ( 1997 ) found @xmath154 , based on the weak @xmath19 emission line , while scannapieco et al . ( 2006 ) quoted @xmath155 with no error bar . emission - line redshifts were also measured with uves at the vlt / ut2 by dallaglio et al . ( 2008b ) , who quoted @xmath156 based on the + blend and assuming a rest wavelength @xmath157 for the blend . given the solar abundance ratio ( o / si)@xmath158 , this blend is probably dominated by three lines of , with a statistically weighted centroid at 1303.49 ( morton 2003 ) . in total , the blend consists of three lines decaying to the ground - state fine - structure levels of [ @xmath159 $ ] at 1302.17 , 1304.86 , 1306.03 and two lines of @xmath160 $ ] at 1304.37 and 1309.28 . dallaglio et al . ( 2008a ) measured the redshifts of the strongest emission lines to be @xmath161 ( ly@xmath2 ) and @xmath162 ( ) . however , there are well - known shifts in emission - line redshifts between the qso systemic redshift and the higher ionization lines " ( e.g. , ly@xmath2 , @xmath163977 , @xmath1631549 ) . for instance , 1000 - 3000 km s@xmath89 offsets are commonly seen in the emission - line spectra of agn ( espey et al . 1989 ; corbin 1990 ) . these issues provide systematic uncertainties in determining @xmath164 , either from the weak / blend or from higher ionization lines . thus , the precise qso systemic redshift remains uncertain . because of these problems with emission - line redshifts , we propose to use igm absorption to determine the qso redshift . we suggest that the qso systemic redshift is @xmath111 ( figure 4 ) based on the extent of the and absorption . this redshift is chosen as the midpoint between the centroid of the strong absorption feature seen in both and and the red edge of absorption . figure 4 provides an overview of the cos / g130 m spectrum from 11001190 , together with a blowup of the region from 1173 to the edge near 1186 . we see absorption extending well beyond the wavelengths ( 1180.21180.5 ) corresponding to the qso emission - line redshifts , @xmath165 and @xmath166 , suggested by reimers et al . ( 1997 ) and dallaglio et al . ( 2008b ) respectively . the absorption between 11811186 suggests that the qso actually lies at @xmath167 . interestingly , we see no ionization effects from proximity to the qso , a topic discussed further in section 3.6 . the absorption profile shows an absorption feature at 1186.5 , with optical depth @xmath168 , equivalent width 95 m , and fwhm = 0.14 . this may be proximate absorption or absorption from an intervening system . the spectrum near 1186.5 shows a clear flux minimum and recovery toward shorter wavelengths typical of weak absorption lines . a distinct absorber would appear as a broader shelf " on the side of the @xmath169 trough . we do not make a specific line identification for the feature . it does not seem to be low - redshift ly@xmath170 nor any of the usual igm metal - line systems ( danforth & shull 2008 ) . reimers et al . ( 1997 ) attribute the and absorption at @xmath171 to a multicomponent system of associated absorbers between 11811186 ( redshifts @xmath172 to @xmath173 ) . many of these systems have strong metal lines of , , and , and a few have anomalous ratios of and ly@xmath2 absorption . these observations suggest that the gas might be exposed to high fluxes of ionizing radiation and affected by nucleosynthetic anomalies in nuclear outflows . smette et al.(2002 ) and fechner et al . ( 2004 ) both noted evidence for line - locking " in and ( although the latter group found no statistical evidence compared to simulated line lists ) . line - locking occurs when two absorbers are separated by a velocity equal to the separation of the two doublet lines , resulting from radiation pressure on material physically near the qso ( weymann , carswell , & smith 1981 ) . these radiative forces will drive gas outward from the qso , whereas the redshifted associated absorbers would have infall velocity @xmath174 km s@xmath89 relative to the qso if @xmath175 . thus , our proposed redshift , @xmath176 , provides a more plausible explanation . the uncertainty in qso systemic redshift illustrates the need for infrared observations . unfortunately , the redshifted hydrogen balmer lines fall in difficult spectral bands ( h@xmath2 at 2.563 @xmath177 m , h@xmath170 at 1.898 @xmath177 m ) , as does the [ o ii ] @xmath178 doublet at 1.455 and 1.456 @xmath177 m . the forbidden lines of [ o iii ] @xmath179 occur at more promising wavelengths , 1.955 @xmath177 m and 1.936 @xmath177 m . thus , infrared spectroscopic searches for the [ o iii ] forbidden lines may offer the best opportunity to resolve the issue of @xmath180 . the standard gp optical - depth formula ( fan et al . 2006 ) for species ( @xmath181 = or ) is @xmath182 where @xmath183 @xmath184 hz is the integrated classical oscillator cross section , @xmath185 is the absorption oscillator strength for ly@xmath2 transitions of and at rest wavelengths @xmath186 and 303.78 , and @xmath187 is their number density ( @xmath188 ) . the hubble expansion parameter is @xmath189^{1/2}$ ] in a flat friedman cosmology ( @xmath190 ) . all cosmological parameters and the resulting coefficients are based on wmap-7 concordance values of cmb modeling from komatsu et al . ( 2010 ) , @xmath191 , @xmath192 , and @xmath193 , and hubble constant @xmath194 with @xmath195 . to set the ion densities @xmath187 , we define @xmath196 and @xmath197 as the abundance fractions of and , respectively . ( the quantity @xmath196 is often called the neutral fraction . ) the factors @xmath198 are the overdensities of h ( at @xmath33 ) and he ( at @xmath46 ) relative to the mean co - moving baryon density , @xmath199 g @xmath188 . with a primordial helium abundance @xmath200 by mass ( peimbert et al . 2007 ) and @xmath201 by number , this baryon density corresponds to and number densities , @xmath202 ^3 \ ; , \\ n_{\rm heii } & = & ( 1.01 \times 10^{-6}~{\rm cm}^{-3 } ) \ ; x_{\rm heii } \ ; \delta_{\rm he } \left [ \frac { ( 1+z)}{4 } \right ] ^3 \ ; . \end{aligned}\ ] ] in the high-@xmath203 limit , @xmath204 , we can express the hubble parameter as @xmath205^{3/2}$ ] scaled to the epoch @xmath46 relevant for the beginning of reionization . thus , the gunn - peterson optical depths for h i ( scaled to the @xmath206 epoch ) and he ii ( scaled to the @xmath207 epoch ) are : @xmath208 ^{3/2 } \ ; x_{\rm hi } \ ; \delta_{\rm h } \ ; , \\ \tau_{\rm heii } ( z ) & = & ( 3.55 \times 10 ^ 3 ) \left [ \frac { ( 1+z)}{4 } \right ] ^{3/2 } \ ; x_{\rm heii } \ ; \delta_{\rm heii } \ ; .\end{aligned}\ ] ] very small abundance fractions , @xmath209 and @xmath210 , will saturate the gunn - peterson ly@xmath2 absorption in and , with optical depths @xmath211 , even in under - dense regions with @xmath212 . it is primarily for this reason that the hydrogen reionization epoch is so difficult to measure in ly@xmath2 absorption at @xmath213 ( fan et al . . however , absorption at @xmath214 can probe 50100 times higher abundance fractions in @xmath197 . for the maximum observable values of optical depth ( @xmath215 ) and with typical densities in low - density filaments and voids , the intervals of high absorption at @xmath216 correspond to fractions , @xmath217 \left [ \frac { ( 1+z ) } { 3.8 } \right ] ^{-3/2 } \left [ \frac { \delta_{\rm he } } { 0.1 } \right ] ^{-1 } \ ; .\ ] ] these overdensities , @xmath218 , are appropriate for the low - column - density ( @xmath219 ) regions of the ly@xmath2 forest . figure 5 shows the optical depths for @xmath75 , using cos data from both g140l and g130 m and _ fuse_. to construct this figure , we reconstructed horizontal error bars " representing the redshift bins ( @xmath115 ) used in these papers . we used unpublished @xmath115 and uncertainties in optical depths @xmath114 from shull et al . ( 2004 ) . from zheng et al . ( 2004 ) , we used values of @xmath115 and @xmath114 from their figure 3 , with 25% uncertainties in @xmath114 . we find reasonable agreement with the _ fuse _ optical depths ( shull et al . 2004 ) down to @xmath220 , although some of the g140l fluxes are higher than those in the _ fuse _ extension to @xmath221 ( zheng et al . the smaller _ fuse _ optical depths may arise from uncertain background subtractions . defining the redshift ranges for these data points is difficult for optical depths that vary rapidly over @xmath222 . future cos spectra of fainter -gp targets ( often @xmath223 magnitude agn ) will have low s / n ratios and will probably not show the fluctuations in optical depths found here . at best , they will measure broad - band optical depths , @xmath224 . for he 2347@xmath14342 , we list below the and optical depths averaged over @xmath225 windows , together with the standard deviations of their distributions : * ( @xmath3 2.42.5 ) @xmath226 , @xmath227 * ( @xmath3 2.52.6 ) @xmath228 , @xmath229 * ( @xmath3 2.62.7 ) @xmath230 , @xmath231 there may be a slight trend to higher @xmath232 at lower redshifts , as we discuss in section 3.4 . considering the ratios of the broad - band averages , we find that @xmath233 rises from @xmath234 at @xmath235 to values @xmath236 at @xmath237 and @xmath238 at @xmath239 . despite the large spreads of the optical - depth distributions , the means are fairly well determined . figure 6 shows examples of the and ionization structures found in our cosmological simulations ( smith et al . 2010 ) , computed with ` enzo ` , an n - body , hydrodynamical code modified to compute time - dependent h and he ionization . one can see the projected topology of the cosmic web , the baryon overdensity , the ionization fractions of and at @xmath80 , and their ratio , @xmath240 . observationally , we measure the ratio of column densities , @xmath11 n()/n ( ) , which we approximate as @xmath241 . the factor of 4 arises from the larger frequency bandwidth , with the absorption - line cross section , @xmath242 , scaling as oscillator strength times wavelength . our igm simulations at @xmath243 show a mean value of the ionization ratio , @xmath244 , corresponding to @xmath245 for primordial he / h = 0.0823 by number . we have confirmed that a substantial amount of observable , with @xmath246 0.0010.01 , comes from ionized gas with @xmath247 ( much of it at @xmath248 ) and @xmath249 to @xmath250 . this gas is primarily confined to photoionized filaments in the cosmic web , with some contribution from collisional ionization ( at @xmath251 k ) . further analysis of these simulations will appear in a later paper . the and ly@xmath2-forest absorbers have traditionally been assumed to arise in photoionized gas , irradiated by metagalactic ionizing ( euv ) radiation of specific intensity @xmath252 , where @xmath181 refers to ( @xmath253 ev ) or ( @xmath254 ev ) . for these hydrogenic species , whose photoionization cross sections above threshold @xmath255 scale as @xmath256 , the photoionization rates @xmath257 . the parameters @xmath258 are local spectral indices of the ionizing background at 1 ryd and 4 ryd , respectively , and provide minor corrections to the photoionization rates . for typical metagalactic fluxes and igm densities during reionization , hydrogen and helium are mostly ionized , with fractions @xmath259 and @xmath260 . because the first ionization potentials of ( 13.6 ev ) and ( 24.58 ev ) are not greatly disparate , we expect the and ionization fronts to coincide for appropriate values of ionizing spectral index . for qso ionizing spectra of power - law form , the and ionization fronts , driven by the photon fluxes at 1 ryd and 4 ryd , respectively , will propagate at the same ( photon flux - limited ) velocity for a critical spectral index , @xmath261 corresponding to our adopted he / h abundance , @xmath262 . the agreement with the _ hst_-observed mean spectral index of agn at 12 ryd , @xmath263 ( telfer et al . 2002 ) , suggests that and ionization fronts normally propagate together . in photoionization equilibrium in regions of high ionization , @xmath264 and @xmath265 , the / abundance ratio is ( fardal et al . 1998 ; shull et al . 2004 ) , @xmath266 the numerical coefficient has been increased from 1.70 to 1.77 , reflecting an updated value of the primordial he / h abundance , @xmath262 , which we use as an approximation to @xmath267 . in this equation , @xmath268 , @xmath269 , @xmath270 , and @xmath271 are the case - a recombination rate coefficients and photoionization rates for and , and @xmath272 and @xmath273 are the specific intensities of the ionizing radiation field at 1 ryd and 4 ryd . the parameter @xmath274 is used to scale the temperature dependence of the case - a recombination rates over the range @xmath275 k to 20,000 k ( osterbrock & ferland 2006 ) : @xmath276 and @xmath277 . equation ( 8) is an approximation , which will break down when @xmath197 becomes significant . it also does not include the effects of collisional ionization of , which can become important in the shocked filaments of shocked gas at @xmath251 k , the so - called warm - hot intergalactic medium ( whim ) . as noted in figure 6 , our simulations suggest that some of the in shocked igm filaments may be collisionally ionized gas at @xmath251 k. however , the bulk of the is photoionized gas at @xmath278 k. ccccc mean ( @xmath232 ) & @xmath279 & @xmath280 & @xmath281 & @xmath282 + median ( @xmath232 ) & @xmath283 & @xmath284 & @xmath285&@xmath286 + mean ( @xmath287 ) & @xmath288 & @xmath289 & @xmath290 & @xmath291 + median ( @xmath287 ) & @xmath292 & @xmath293 & @xmath294 & @xmath293 + as shown in figures 2 and 4 , we have confirmed the presence of three long troughs with strong absorption ( @xmath72 ) and no detectable flux . these troughs have been noted previously ( reimers et al . 1997 ; smette et al . 2002 ; shull et al . 2004 ) , but the very low cos backgrounds allow us to better characterize their optical depth . we also detect numerous windows of flux transmission ( e.g. , at 1138 , 1139 , 1144 , 1154 , 1160 , 1174 , and 1179 see figure 2 ) . the g130 m data provide accurate boundaries in redshift . from those , we can determine their comoving radial sizes , using wmap-7 cosmological parameters : 1 . _ ( trough 1 ) _ @xmath295 ( 61 mpc ) , 2 . _ ( trough 2 ) _ @xmath296 ( 39 mpc ) , 3 . _ ( trough 3 ) _ @xmath297 ( 25 mpc ) . these dimensions , all much larger than 10 mpc , are incompatible with zones of photoionization around typical qsos . they are also unlikely to exist at the epoch following the overlap of expanding cosmological i - fronts . recent theoretical modeling has studied the statistics of troughs ( mcquinn et al . 2009 ; mcquinn 2009 ; furlanetto 2009a , b ; dixon & furlanetto 2009 ) . these studies examine large - scale fluctuations in the ly@xmath2 forest transmission , during and after reionization . a robust result is that long troughs of large optical depth , @xmath298 , can not be explained in models with a smoothly evolving ionizing radiation field ( furlanetto & dixon 2010 ) . furthermore , the observed 2560 mpc troughs of high opacity between @xmath299 can not be accommodated in standard reionization models unless reionization is delayed until @xmath8 . the absorption is therefore expected to be patchy at @xmath300 , as we have observed . it would be useful to conduct surveys ( worseck et al . 2007 ) of galaxies and qsos near the sight line at these trough redshifts to identify potential sources of ionization ( or lack thereof ) . as discussed in section 3.2 , the opacity greatly exceeds that of ; the ratio @xmath301 for typical conditions in an igm photoionized by qsos of low space density . thus , the propagation and overlap of i - fronts driven by ionizing continua will likely differ from those of hydrogen . the greater opacity means that filaments of the cosmic web still have @xmath302 at @xmath303 ( see figures 5 and 6 ) . thus , the topology of overlapping ionized zones probably differs from that of merging regions at @xmath206 . these theoretical results are explored in a later paper ( smith et al . 2010 ) using adaptive - mesh cosmological simulations with the ` enzo ` code . because reionization involves the stochastic propagation and overlap of ionization fronts ( i - fronts ) , one naturally expects variations in the abundances of and . the metagalactic flux of high - energy ( 4 ryd continuum ) radiation varies primarily because of the low space density of the dominant ionizing sources ( agn ) and varying attenuation through the igm . the variations in opacity can be especially large during reionization , because the 4 ryd continuum is strong near agn ( inside the i - fronts ) but heavily attenuated in the troughs , far from the agn and within the gaseous filaments of the cosmic web . figure 4 shows both and absorption , overlaid in redshift , illustrating the correspondence of flux - transmission windows with weak spots ( low absorption ) in the ly@xmath2 forest . the non - gaussian wings in the cos line - spread function ( lsf ) is most significant for weak absorbers , but we believe this is completely accounted for in the convolution of the data by the cos lsf . the resolving power ( @xmath304 ) of the cos lsf is more appropriately the effective gaussian resolving power ( green et al . the main difference in the convolved and unconvolved data in figure 4 is simply a degradation in resolution by a factor of 2.5 . the cos lsf contributes to a redistribution of flux in the narrowest lines , but this is accounted for with the non - gaussian lsf . the / ratio , @xmath241 , tracks the ratio of ionizing fluxes at 1 and 4 ryd ( see equation [ 8 ] ) . table 3 presents the mean and median values of @xmath232 in the g130 m and g140l data , and over various redshifts . in different columns , we present these statistics for the g140l and g130 m spectral regions , and for the region ( @xmath305 ) where the two gratings overlap . to calculate @xmath306 , we convolve the high - resolution vlt data with either the medium - resolution ( g130 m ) or low - resolution ( g140l ) cos line - spread functions , depending which dataset is representing at a particular redshift . the cos data are then smoothed by seven pixels ( @xmath307 resolution element ) and up - sampled to the finer vlt wavelength scale . optical depths @xmath308 and @xmath309 are calculated on a pixel - by - pixel basis . figures 7 and 8 show @xmath310 and @xmath311 and demonstrate that optical depth saturates in certain regions . over the range @xmath75 , 32% of pixels are saturated , but this fraction depends strongly on redshift and spectral resolution . in the low - resolution data at @xmath312 , we find that @xmath114 is almost always unsaturated . at @xmath313 , we see long absorption troughs in which 4045% of the pixels have @xmath314 . we computed mean and median values of @xmath232 using both saturated and unsaturated pixels . as the saturated pixel values for are entirely in the numerator , our average @xmath232 values are lower limits . using only unsaturated pixels reduces @xmath233 systematically by 10 - 20% at @xmath315 . correcting for larger true values of @xmath316 would increase the mean statistics . there is no change at @xmath317 . we confirm the fluctuations seen by _ fuse _ between @xmath318 ( kriss et al . 2001 ; shull et al . 2004 ) on small scales , @xmath319 . figures 7 and 8 show that @xmath232 fluctuates on scales down to @xmath222 at g140l resolution and @xmath320 at g130 m resolution . for reference , the redshift interval @xmath321 corresponds to a comoving radial distance of 10.8 mpc at @xmath14 . regions of hot gas ( @xmath251 k ) might appear as high-@xmath232 systems , in which is far more collisionally ionized than . however , the cos data also exhibit several instances of low-@xmath232 absorbers with @xmath322 . since @xmath241 , the low-@xmath232 systems appear primarily where the optical depth is large and optical depth small . if photoionized , these absorbers may be irradiated by very hard radiation fields , with large ratios of @xmath323 ( equation 8) , perhaps near an agn with a particularly flat euv spectral index , @xmath324 . several good examples of low@xmath232 absorbers appear at @xmath325 , 2.785 , 2.830 , 2.865 , and 2.904 . most of these systems have high @xmath326 , the exception being the absorber at @xmath327 in which both @xmath114 and @xmath326 are low , and their ratio poorly determined . for high-@xmath232 systems , the g130 m data exhibit regions with @xmath328 , typically in gas with large @xmath114 but low @xmath326 . extreme excursions in @xmath232 may arise from noise , and their detectable range depends on spectral resolution . we therefore take care to assess the range over which we can measure reliable optical depths . for ratios within the range @xmath11 10100 ( g140l ) and @xmath11 5500 ( g130 m ) , we believe nearly all the fluctuations are real . unfortunately , strong troughs of absorption at @xmath329 hinder our ability to measure @xmath114 for individual systems with the high - resolution g130 m . many of the @xmath232-fluctuations appear at @xmath8 , observed with the lower - resolution g140l . these small - scale @xmath232-variations have been the topic of numerous theoretical papers ( zuo 1992 ; fardal & shull 1993 ; fardal et al . 1998 ; bolton et al . 2005 , 2006 ) . the key focus of these papers is on fluctuations in the small numbers of agn within the characteristic attenuation length , corresponding to the @xmath330 sphere " at 4 ryd photon energy . because the opacity is so much larger than that of , one expects large variations in the ionizing flux , @xmath273 , on comoving scales of 2040 mpc at @xmath331 ( fardal et al . 1998 ; furlanetto 2009 ; dixon & furlanetto 2009 ) . a second important effect is that the observed low - redshift agn exhibit wide variations in their euv spectral indices ( telfer et al . 2002 ; scott et al . 2004 ) , which produce strong source variations in the flux ratio @xmath332 . finally , radiative transfer of the metagalactic ionizing radiation reprocesses the ionizing spectra , producing further variations ( fardal et al . what is now firmly established by the new cos data are the variations in @xmath232 on @xmath333 mpc comoving radial distance scales ( @xmath222 ) . it is surprising to note the sharpness of the edge at 1186.26 and the lack of any proximity effect near the qso . if ionizing radiation were escaping from the agn , the surrounding igm should be ionized within 510 mpc of the qso . the absorption extends @xmath334 beyond the previous qso redshift estimate , @xmath335 ( reimers et al . as discussed in section 3.1 , this dilemma is easily resolved by moving the quasar redshift to @xmath173 , an offset ( 1460 km s@xmath89 at @xmath336 ) that corresponds to 19.5 mpc comoving proper distance . however , the two proposed systemic redshifts , @xmath175 and 2.904 , raise several questions . does the qso lie inside an and cavity produced by its photoionizing radiation ? or , is the qso ionizing radiation absorbed by circumnuclear gas ? is there infalling gas near the agn that produces associated absorption with @xmath337 ? it is worth investigating these possibilities . for ( 1 ryd ) and ( 4-ryd ) ionizing - photon luminosities , @xmath338 and @xmath339 , what strmgren radii will this bright qso carve out ? these distances are obviously limiting cases , in ionization equilibrium , whereas one expects the ionization zones around the qso to involve time - dependent propagation of ionization fronts . indeed , the lack of a proximity effect could arise from this source just starting to burn its way out of the galactic nucleus . given the strength of the @xmath336 absorber , we will assume higher gas densities associated with the qso and scale to an overdensity ( @xmath340 ) at @xmath336 . the h and he densities are then : @xmath341 , @xmath342 , and @xmath343 . at @xmath344 k , @xmath345 and @xmath346 . a typical @xmath347 quasar produces a total of @xmath348 lyc photons s@xmath89 , but this qso is much brighter ( @xmath64 ) . extrapolation from 1200 down to the lyman limit ( 889.5 at @xmath176 ) shows that he 2347 - 4342 has an observed specific flux , @xmath349 erg @xmath105 s@xmath89 @xmath89 or @xmath350 erg @xmath105 s@xmath89 hz@xmath89 . from this flux and a luminosity distance , @xmath351 cm at @xmath173 , we estimate an ionizing photon luminosity of @xmath352 photons s@xmath89 in the 4-ryd continuum . scaling to @xmath353 , we estimate a strmgren radius , @xmath354 ^{1/3 } \nonumber \\ & \approx & ( 4~{\rm mpc } ) \left [ \frac { s_0^{{\rm ( he ) } } } { 10^{57}~{\rm s}^{-1 } } \right]^{1/3 } ( \delta_h/100)^{-2/3 } t_{4.3}^{0.255 } \ ; . \nonumber\end{aligned}\ ] ] a similar calculation , using the observed flux , @xmath355 erg @xmath105 s@xmath89 @xmath89 , at the redshifted ( 3561 ) edge , gives @xmath356 photons s@xmath89 and an strmgren radius of @xmath357 ^{1/3 } \nonumber \\ & \approx & ( 6~{\rm mpc } ) \left [ \frac { s_0^{{\rm ( h ) } } } { 5\times10^{57}~{\rm s}^{-1 } } \right]^{1/3 } ( \delta_h/100)^{-2/3 } t_{4.3}^{0.286 } \ ; . \nonumber\end{aligned}\ ] ] we note that the ratio of the extrapolated fluxes at the respective ionization limits is fairly small . in the qso rest - frame , @xmath358 , indicating a fairly hard ( flat ) power - law spectrum @xmath359 . these estimated strmgren radii are a substantial fraction of the proper distance of 19.5 mpc between the two proposed qso redshifts , @xmath173 and @xmath360 . even though the local ionization zone may not extend as far as these equilibrium strmgren radii , the expected proximity effect is minimally visible , except for a narrow feature with @xmath361% flux transmission at @xmath362 , just below the edge ( figure 4 ) . we conclude that ionizing radiation from this qso is strongly absorbed in the proximity of the nucleus , possibly by circumnuclear gas or by the strong absorber at @xmath65 . the photoionization cross sections for and above threshold are approximately @xmath363 and @xmath364 . thus , to absorb a substantial amount of the ionizing continua requires column densities @xmath365 ^3 \ ; \tau(e ) \nonumber \\ { \rm n}_{\rm heii } & = & ( 6.35 \times 10^{17}~{\rm cm}^{-2 } ) \left [ e/54.4~{\rm ev}\right ] ^3 \ ; \tau(e ) \end{aligned}\ ] ] for continuum optical depth @xmath366 at photon energy @xmath367 . for this very bright qso , the absence of a proximity effect indicates a significant attenuation of the 4 ryd continuum , with optical depth @xmath368 at mean photon energies 1.5 times the 54.4 ev threshold , or n@xmath369 . what are the observational consequences of such large column densities of ionized gas ? it has been suggested ( reimers et al . 1997 ) that resonance lines at 303.415 and 305.596 might be detectable , but we find no obvious candidates . it would be useful to search for signs of splittings of these lines , which are separated by 2.182 @xmath370 at @xmath65 . some of the associated absorbers have anomalously stronger than , suggesting that the gas is close to the qso with helium mostly fully ionized ( ) . more clues may come from the strong absorbers at @xmath371 ( 1184.71186.2 ) which would be closest to the qso if @xmath176 . this system has high optical depth in both and ( see figure 4 ) . very strong metal - line absorption is seen in with weaker absorption in , , and ( reimers et al . 1997 ; fechner et al . 2004 ) . by fitting higher lyman - series absorbers , fechner et al . ( 2004 ) found log n@xmath372 , while the metal - line absorbers gave log n@xmath373 , log n@xmath374 , and log n@xmath375 . they proposed that this absorber is exposed to the strongest and hardest radiation from the qso , and therefore may shield the other associated absorbers . however , their photoionization models gave log n@xmath376 , which is 100 times too small to provide the necessary absorption ( eq . [ 9 ] ) . the column is 10 times too small to shield the 1 ryd continuum . fox , bergeron , & petitjean ( 2008 ) obtained newer ( high s / n ) spectra on these metal - line absorbers at @xmath377 , 2.8972 , and 2.9041 . they found somewhat higher column densities of and and a slightly higher column density , @xmath378 in the @xmath379 absorber . this column is still insufficient to shield the qso s ionizing continuum . we conclude that no satisfactory explanation exists for the absence of a proximity effect around this very luminous qso , other than the possibility that it has only recently turned on ( within the last myr ) . the high throughput and low background of the cosmic origins spectrograph allow us to probe the reionization epoch toward he 2347@xmath14342 at redshifts @xmath203 = 2.42.9 . because of photoionization and recombination rates , is a far more abundant species than , with typical ratios @xmath380 varying from 10200 in the filaments of the ly@xmath2forest . with cos , we find that reionization of to is patchy at @xmath381 , with 510 troughs of high optical depth ( @xmath316 ) punctuated by narrow windows of flux transmission . previous studies of this qso sight line ( kriss et al . 2001 ; shull et al . 2004 ; zheng et al . 2004 ) found a gradual decrease in optical depth at @xmath8 , suggesting a reionization epoch @xmath59 . with our new cos data , we find a more complicated picture , in which is slowly reionized at @xmath382 . 1 . we have found clear evidence that the epoch of reionization is delayed below @xmath207 , as we see patchy ionization down to @xmath383 . for this bright agn , high - s / n , low - background observations from cos allow us to measure absorption , the uv continuum of the qso , and troughs of high-@xmath114 . the high spectral resolution of the g130 m grating probes absorption fluctuations and the ionization edge , and the g140l grating allows us to study reionization down to @xmath384 . 2 . comparison of cos / g130 m and g140l data with previous _ hst _ and _ fuse _ data confirms the far - fuv continuum , absorption troughs , transmission windows , and optical - depth below the edge at 1186.26 . the cos / g140l fluxes agree fairly well with _ fuse _ data at @xmath313 ( shull et al . 2004 ) , but below 1130 ( @xmath8 ) considerable differences appear . the cos fluxes are lower than those of fuse inferred by zheng et al . ( 2004 ) , and our derived optical depths are therefore somewhat higher at @xmath385 . the broad - band optical depths ( averaged over @xmath225 ) only drop below @xmath386 at @xmath387 . we observe optical - depth variations in and that imply fluctuations in the / abundance ratio on scales @xmath319 , corresponding to 410 mpc in comoving distance . such variations likely arise from agn source variations ( 14 ryd spectral indices ) and igm radiative transfer effects . the number of low-@xmath232 systems ( mostly low @xmath114 and high @xmath326 ) seen with cos is lower than claimed in fuse observations . we see intervals of patchy flux - transmission , as well as three long ( 510 ) intervals of strong absorption ( 2560 mpc proper distance ) between @xmath295 , @xmath296 , and @xmath297 . these troughs probably arise from dense filaments in the baryon distribution and the scarcity of strong photoionizing sources ( agn ) within 3050 mpc . 5 . at @xmath3 2.72.9 , probes of fraction , @xmath388 are now possible at optical depths @xmath76 . the absorption is 50100 times more sensitive to trace abundances than at @xmath206 , where a neutral fraction @xmath389 saturates the ly@xmath2 gunn - peterson absorption . what future work is needed to better understand the reionization ? first , we need to assess whether the he 2347@xmath14342 sight line is typical . the cos - gto team will be observing two more agn targets , hs1700 + 6416 and q0302 - 003 , which may clarify whether the patchy reionization at @xmath390 is common to most sight lines . second , along the current sight line , we have identified many regions of partial flux transmission and three long troughs of absorption , where surveys of galaxies and agn will help clarify the sources of ionization . third , infrared spectra of the [ ] @xmath1635007 , 4959 narrow emission lines ( at 1.955 @xmath177 m and 1.936 @xmath177 m ) could confirm the qso systemic redshift . pinning down @xmath180 is crucial for understanding the nature of the ionization state of the gas near the qso , which shows no evidence of a proximity effect . on the theoretical front , cosmological simulations with time - dependent h and he chemistry and i - front radiative transfer should elucidate the structure topology of the filaments and voids that we observe in the ly@xmath2-forest and troughs . serious studies of reionization have been hindered by the lack of many agn targets sufficiently bright to obtain far - uv spectra . the spectrographs on _ fuse _ and _ hst _ have made good progress on three agn sight lines , but newly discovered gp targets ( syphers et al . 2009a , b ) are considerably fainter than he 2347@xmath14342 or hs1700@xmath586416 . new observations of these fainter targets will probably measure only broad - band optical depths , for which the high - resolution cos spectra may provide guidance in interpretation . the long - term study of the reionization process , from @xmath391 , must await the construction of a future large - aperture , far - uv telescope in space ( shull et al . 1999 ; postman et al . 2009 ) . it would be especially useful if this mission had high sensitivity and moderate spectral resolution ( @xmath392 ) extending from 1130 down to the 912 hydrogen lyman limit . this would enable the study of patchy reionization in @xmath54 down to redshifts @xmath393 . it is our pleasure to acknowledge the thousands of people who made hst servicing mission 4 a huge success . we thank brian keeney , steve penton , stphane bland , and the rest of the cos / gto team for their work on the calibration and verification of the early cos data , and cristina oliviera for her helpful input on g140l segment - b calcos processing . we thank dieter reimers for discussions regarding the qso systemic redshift and for providing vlt data on the ly@xmath2 forest . mark giroux , blair savage , david syphers , and the referee provided numerous comments that improved the arguments made in this paper . this work is based on observations made with the nasa / esa _ hubble space telescope _ , obtained from the data archive at the space telescope science institute . stsci is operated by the association of universities for research in astronomy , inc . under nasa contract nas5 - 26555 . our work was also supported by nasa grants nnx08ac146 and nas5 - 98043 and the astrophysical theory program ( nnx07-ag77 g from nasa and ast07 - 07474 from nsf ) at the university of colorado at boulder .

we describe a method to identify inclined water maser disks orbiting massive black holes and to potentially use them to measure black hole masses . due to the geometry of maser amplification pathways , the minority of water maser disks are observable : only those viewed nearly edge - on have been identified , suggesting that an order of magnitude additional maser disks exist . we suggest that inward - propagating masers will be gravitationally deflected by the central black hole , thereby scattering water maser emission out of the disk plane and enabling detection . the signature of an inclined water maser disk would be narrow masers near the systemic velocity that appear to emit from the black hole position , as identified by the radio continuum core . to explore this possibility , we present high resolution ( 0.070.17 ) very large array line and continuum observations of 13 galaxies with narrow water maser emission and show that three are good inclined disk candidates ( five remain ambiguous ) . in the best case , for cgcg 120 , we show that the maser and continuum emission are coincident to within pc ( milliarcsec ) . subsequent very long baseline interferometric maps can confirm candidate inclined disks and have the potential to show maser rings or arcs that provide a direct measurement of black hole mass , although the mass precision will rely on knowledge of the size of the maser disk .

1702.06545

water masers arising from thin disks around massive black holes provide high brightness temperature non - thermal dynamical tracers of gas in keplerian orbits . as such , water maser disks viewed edge - on provide tracers of the keplerian potential and enable measurement of the black hole mass , provided a distance is known in order to translate the apparent angular disk size into a physical size ( e.g. , * ? ? ? maser accelerations and proper motions can also be observed , and because the circular velocity is known from doppler shifts , the geometric distance to the black holes ( and host galaxies ) can be determined ( e.g. , * ? ? ? geometric distances obtained from water masers provide a crucial independent measurement of the hubble constant and can be used to calibrate other distance indicators such as the period - luminosity relation of cepheids @xcite . these measurements require maser disks that are viewed within a few degrees of edge - on : otherwise , maser beaming directs emission away from the observer because masers propagate along velocity - coherent paths through the disk . for thin disks , this propagation occurs along the radial path along the line of sight toward the black hole and along the disk tangent points . for warped disks , such as that found in ngc 4258 , the picture is more nuanced because a warped disk provides numerous sightlines and inclinations that intersect velocity - coherent parts of the disk @xcite . nonetheless , inclined maser disks are generally not seen in water maser surveys . or are they ? it seems likely that inclined water maser disks have already been detected by single dish surveys , but they have been discarded because they show no high velocity lines that have canonically been used to identify maser disks . in this paper , we propose a mechanism to produce detectable maser emission from inclined disks that may also be used to obtain black hole masses ( section [ sec : thought ] ) , we present a method to detect inclined maser disks based on extant surveys ( section [ sec : candidates ] ) , we present karl g. jansky very large array ( vla ) observations of candidate inclined disks ( sections [ sec : obs ] and [ sec : results ] ) , and we provide a list of inclined disk candidates for further study ( sections [ sec : analysis ] and [ sec : discussion ] ) . in what follows , we assume a flat cosmology with parameters @xmath3 km s@xmath4 mpc@xmath4 , @xmath5 , and @xmath6 , and we calculate distances from redshifts in the cmb rest frame . consider a typical 22 ghz water maser thin disk in keplerian orbit around a massive black hole . this thin disk shows systemic velocity masers at the center of the observed disk , along a radial amplification path , and it shows high - velocity red- and blue - shifted masers at the tangent points of the disk , along azimuthal paths , amplifying spontaneous emission and background host galaxy continuum ( figure [ fig : schematic ] , left ) . if a thin disk is inclined by more than a few degrees , the masers are no longer beamed toward the observer and the disk would not be detected . the known water maser disks show an inclination typically within @xmath75@xmath8 of edge - on ( e.g. , * ? ? ? * ) , suggesting that an order of magnitude additional maser disks exist , and they are simply beamed in directions we can not observe . note that the famous water maser disk in ngc 4258 ( e.g. , * ? ? ? * ; * ? ? ? * ) is atypical , with a disk inclination of 72@xmath8 , but the disk warp provides sightlines that are nearly edge - on , and this is where the masers are seen @xcite . for the following discussion we will use fiducial parameters similar to those of many of the known maser disks @xcite . we assume a distance of 10 mpc , a black hole mass of @xmath9 ( and therefore a schwarzschild radius of @xmath10pc ) , and a circular rotation speed of 1000 km s@xmath4 in the maser - active part of the disk located 0.5 pc from the black hole and spanning 0.2 pc . we assume a disk thickness equal to that of ngc 4258 : @xmath11 cm or 0.18 mpc ( 1@xmath12 ; * ? ? ? consider how this system appears to an observer at the location of the black hole : the disk describes a plane in the sky . water masers reaching this observer must be purely radial ( propagating inward ) but will have a coherent amplification path in all radial directions ( figure [ fig : schematic ] , center ) . all keplerian doppler shifts will be across the line of sight , so all maser emission reaching this location will have zero velocity in this reference frame ( ignoring the transverse doppler shift , which is negligible at @xmath72 km s@xmath4 ) . the black hole - frame observer will thus see a ring of maser light or maser spots with low velocity spread . the maser - active part of the disk would subtend @xmath72.5 arcminutes ( 1@xmath12 ) . the expected maser opening angle would be similar , @xmath73 arcminutes , assuming a maser path - to - size ratio of @xmath70.2 pc/0.2 mpc @xmath13 , implying maser light spanning @xmath7500 @xmath14 at the black hole . the black hole kerr metric would thus be bathed in maser light with a continuous distribution of impact parameters , implying gravitational deflection of maser light into nearly all angles ( the deflection at 100 @xmath14 is 1.1@xmath8 and grows inversely with impact parameter @xmath15 as @xmath16 ; @xcite ) . one would therefore expect incoming maser light to be scattered into many angles , representing a gravitational lensing de - focusing of a given incoming beam . but portions of multiple maser beams from around the disk may be deflected into a given direction , making the net flux as a function of angular direction uncertain and in need of numerical modeling . ngc 4258 has a warped disk ( e.g. , * ? ? ? * ) , which means that the in - going radial masers can be misaligned and thus be beamed above or below the dynamical center of the disk , and that the continuous distribution of disk inclination angles will fully populate the black hole metric volume with beams from multiple angles , further enhancing the sampling of impact parameters and deflection angles . if warped disks are common , then maser disks with high velocity lines will be detected more often than one would assume based on random inclination alone ( in other words , parts of many disks will have inclinations appropriate for detectable maser amplification ) , and the black hole will be illuminated by in - going masers from above and below the nominal disk plane , producing a range of incoming maser impact parameters and angles with respect to the disk , thereby enhancing the gravitational deflection probability toward any given observer . regardless of whether disks are warped or not , maser light will be scattered away from the maser disk plane , making inclined disks potentially detectable , but likely faint compared to edge - on disk masers . masers amplify seed photons , and these seed photons can be the portion of a radio continuum that matches the maser line frequency , appropriately doppler - shifted , or they can be spontaneously emitted maser line photons . in known maser disks , the systemic radial masers typically amplify the agn radio continuum , and the high - velocity tangential masers amplify spontaneous emission or possibly radio continuum from the host galaxy itself . in - going radial masers do not have access to agn seed photons and must amplify either continuum from the host galaxy or inward - directed spontaneous emission from the outer parts of the disk . in either case , it is reasonable to expect in - going masers to be weaker than the out - going systemic masers from edge - on disks . it is unclear how the amplification of in - going masers would compare to the amplification of the high - velocity masers in edge - on disks because the tangential amplification pathway may or may not have a physically longer or higher column density velocity - coherent path for amplification . if one were to observe a water maser disk from a direction other than edge - on , would water maser emission be seen ? the thought experiment described here suggests that it would , and the inclination of the disk with respect to the observer would select the deflection angle ( or equivalently the impact parameter ) of the observed maser light . for example , deflection by 10@xmath8 would allow observation of masers from the back side of a disk with an inclination of 80@xmath8 , requiring an impact parameter of 12 @xmath14 . other parts of the disk could be viewed if light reaches smaller impact parameters . if , as expected , there is a continuous range of impact parameters , then a continuous distribution of maser light from an extended portion of the disk will contribute to a spectrally narrow maser line complex seen by the observer at nearly any inclination ( but perhaps with higher likelihood and intensity at higher inclinations ) . the observational signatures of an inclined water maser disk would be : + 1 . a narrow line or line complex + 2 . at the systemic velocity + 3 . at the apparent black hole location . + the black hole location would be indicated by the radio continuum core , ideally observed at the same frequency as the water maser . the observational signatures of an inclined water maser disk , however , may also arise from other mechanisms . water masers can be produced in radio jet - molecular cloud interactions ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) , in star - forming regions ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) , and in outflows ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * but note that some objects in the latter survey may be candidates for inclined disk masers ) . these are likely to be the main contaminant among an inclined disk survey sample . vlbi identification of the maser with an agn via spatial coincidence with the core radio continuum identified by the spectral index can resolve the ambiguity ( section [ sec : results ] ) . in contrast to edge - on disk systemic masers , it is unclear whether inclined disk maser lines will show proper motion or acceleration . this may depend on the clumpiness of maser - emitting regions , on the maser beam sizes , on the physical extent of the disk that is sampled by the observed line , and on the deflection angle to the observer . one might expect lensed masers to show little or no time variability , but the substructure seen in known water maser disks and the natural variability of water masers suggests that this may be a bad assumption . many extragalactic water masers detected in previous surveys meet some or all of the above observational criteria : inclined water maser disks may have already been detected ! in most cases , when a single systemic velocity line is detected in water maser surveys , there is no interferometric follow - up because distance or black hole mass measurements ( traditionally ) require edge - on disks , the signature of which are the high velocity lines emitted from the tangent points of the disk . high resolution observations may also be frustrated by weak or variable water masers . in section [ sec : obs ] we present interferometric mapping of a sample of narrow - line systemic velocity water masers that appear to be inclined disk maser candidates . if observations show that the masers remain unresolved and are centered at the location of the central massive black hole ( as identified by simultaneous radio continuum observations at 20 ghz ) , then they remain candidates and should be mapped with vlbi . the einstein radius of a strong gravitational lens is @xmath17 where @xmath18 , @xmath19 , and @xmath20 are the angular diameter distances to the lens , to the source , and between the lens and the source , respectively , and @xmath21 for the black hole - maser disk configuration ( after * ? ? ? for the fiducial parameters listed above ( @xmath22 @xmath23pc , @xmath24 pc , and @xmath25 mpc ) , @xmath26 mas . since the angular resolution ( hpbw ) of the very long baseline array ( vlba ) is 0.3 milliarcseconds at 22.2 ghz ( which should be compared to twice the einstein radius ) , the einstein radius for the fiducial maser disk and black hole would require space - based vlbi to resolve . on the other hand , terrestrial vlbi could resolve a water maser einstein ring for a more massive black hole with a physically larger maser disk : for @xmath27 @xmath23pc ( @xmath28 ) , @xmath29 pc , and @xmath25 mpc , @xmath30 mas . unfortunately , water maser disks have yet to be identified orbiting @xmath31 black holes , but it is unclear whether is this a selection effect or a consequence of physics @xcite . the most massive black hole measured using a water maser disk to date is in ngc 1194 with @xmath32 @xcite . were an einstein ring observable from a back side in - going maser in an edge - on disk or from an in - going maser from an edge - on portion of an inclined warped disk , then one can infer a black hole mass from the angular size of the ring . one does need independent measurements of the luminosity distance to the black hole ( @xmath18 ) and the size of the maser - emitting disk ( @xmath20 ) . @xmath18 can be obtained from an assumed cosmology and the cosmological redshift , but @xmath20 may be more difficult to measure . the precision of black hole masses obtained from this method may be limited by our ability to measure , model , or estimate the radius of the maser - emitting part of the disk . einstein rings require linear alignment between the source , the lens , and the observer , which is not germane to the inclined maser disk geometry ( for unwarped disks ) . instead , one would navely expect to see multiple images of the same maser , which can also be related to the black hole mass . this expectation , which is correct for isotropic emitters , is probably incorrect for masers . an important difference between maser emission and the standard treatment of gravitational lensing is maser beaming : while there may be sightlines in a gravitational lens geometry that land on the emitter , emission may not be seen if light is not beamed along that sightline . for a general pointlike isotropic light source offset from the lens by angle @xmath33 , there are two solutions to the lens equation : @xmath34 @xmath35 represents the angle between the lens and the source image appearing outside @xmath36 , and @xmath37 represents the angle of the image appearing inside @xmath36 ( * ? ? ? * equation 24 , figures 5 and 7 ) . for an inclined maser disk configuration , no emission is directed significantly out of the disk plane , so no maser emission would be seen in the @xmath35 direction . on the other hand , @xmath37 may be small enough that this sightline is included in the maser beam passing very close to the central black hole . in this case , the source - observer deflection angle @xmath38 nearly matches the complement of the disk inclination : @xmath39 ( figure [ fig : schematic ] , right ) . in contrast to canonical lensing , we expect that the maser beaming will produce only one maser spot image , where light is deflected in the manner shown schematically in figure [ fig : schematic ] ( right ) . this maser image will lie inside the einstein radius . for isotropic extended emitters , this image would also be demagnified , but masers are not isotropic emitters and can be exceptionally compact ( equivalently , they demonstrate high brightness temperatures ) . the degree of demagnification is therefore unclear and requires numerical ray - tracing to assess . for the fiducial parameters above , and assuming an inclination of @xmath40 , @xmath41 mas , @xmath42 mas , and therefore @xmath43 @xmath23as , which is equivalent to 11 @xmath14 . the maser beam spans this distance from the black hole , so the maser emission can be lensed toward the observer in this case . for the @xmath31 black hole with a larger maser disk described above , @xmath44 mas , @xmath45 mas , and therefore @xmath46 @xmath23as , which is again equivalent to 11 @xmath14 ( the impact parameter determines the deflection angle , which is determined by the inclination ) . one would therefore expect lensed masers from inclined disks to be faint and appear to arise from the black hole location ( @xmath47 mas ) . this treatment assumes a single pointlike maser rather than an extended continuous disk of masers or a set of distributed maser spots . in this more realistic scenario , numerical ray - tracing is required to connect the observable lensed maser image to the black hole mass and maser disk configuration . lensed masers from inclined disks may appear to be pointlike or they may describe arcs . in - going masers from the far side of an edge - on disk or in - going masers from an inclined but warped disk may produce einstein rings . single - epoch vlbi maps can therefore provide black hole masses , but space - based vlbi may be required for the typical @xmath48 black hole associated with water maser disks . the mass measurement precision will be limited by uncertainty about the maser disk size rather than by the distance to the object ( the maser will provide the systemic redshift , which can be converted into a distance , and the distance uncertainty will be dominated by peculiar velocity departures from the hubble flow ) . if there are in - going radial water masers in maser disks ( and there is no compelling reason to think otherwise ) , then there should be systemic water masers seen in edge - on disks from the back side of the disk , both lensed and unlensed by the black hole . are such masers in extant data ? if there are also front - side systemic masers in an edge - on disk , then they will be orders of magnitude brighter than back - side masers , even in the presence of strong lensing , simply due to amplification considerations . the front - side masers can amplify agn radio continuum , whereas the back - side masers will either be driven by stimulated emission ( but with an amplification pathway equal to the front - side masers ) or by host galaxy continuum , which is substantially weaker than agn continuum at 22 ghz in many cases . the back - side maser contribution may therefore be confused by the front - side emission , particularly since both types of maser emission will occur at the systemic velocity . in rare cases , a back - side maser might be distinguishable from the front - side emission either by a position or a velocity offset ( there is some spread to systemic velocity masers , e.g. @xcite ) . the observational signature of a back - side maser would be an acceleration in the opposite sense of the front - side systemic maser acceleration ( i.e. , negative acceleration under the convention that positive velocities are redshifted ) . this is not seen in published systemic maser acceleration measurements ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) , but such a signal could be lost amid the brighter and numerous front - side masers . if a back - side maser is lensed into an arc or einstein ring by the central black hole , then it might be extended in vlbi maps . resolved systemic maser emission would therefore be an additional observable signature of back - side masers . if inclined water maser disks can be detected via gravitational lensing or deflection of in - going masers by massive black holes , then they have likely already been detected in surveys for maser disks . but they were rejected as disk candidates because they lacked high - velocity emission . inclined maser disks will appear to have maser emission only at the systemic velocity of the galaxy or agn . we therefore use extant water maser surveys to select inclined disk candidates . using extant water maser surveys , most of which favor seyfert 2 agn , we examined single - dish spectra compiled by the megamaser cosmology projecthttps://safe.nrao.edu / wiki / bin / view / main / privatewatermaserlist ] to select objects showing a narrow systemic velocity maser or maser complex . we also imposed a 30 mjy line flux limit and excluded objects south of @xmath49 declination . this process identified 16 inclined maser disk candidates ( table [ tab : obs ] ) , and most candidates ( 14 ) have only been observed with a single dish . those that do have interferometric maps are ngc 3556 , which was mapped using the vla in cnb and dna configurations @xcite with no 22 ghz continuum detected ( 1@xmath12 rms noise of @xmath70.5 mjy beam@xmath4 ) , and ngc 3735,which was mapped using a - array @xcite , but no 22 ghz continuum was detected . since the inclinations of known maser disks are nearly edge - on , the number of inclined maser disks must be large , roughly an order of magnitude larger than the number of detected maser disks . however , the gravitational lensing or deflection of detectable in - going radial masers adds substantial uncertainty to the detection expectations . we do not know how common observable lensed inclined maser disks are in the universe because we do not know the opening angle of the masers , the size of the maser spots with respect to the black hole s schwarzschild radius , the brightness of in - going masers , which will depend on the 22 ghz seed photons from the host galaxy , or the degree of gravitational lensing demagnification . we observed the 22.23508 ghz @xmath50 ortho water maser line and 20 ghz radio continuum toward 16 candidate inclined maser disks ( see section [ sec : candidates ] and table [ tab : obs ] ) using the vla in a configuration ( the highest angular resolution configuration ) . observations of program 15a-297 spanned june 19 2015 through september 26 2015 in five sessions . the fifth session occurred during the a - array to d - array reconfiguration . each session included visits to flux and bandpass calibrators , and observations of each target object were interleaved with nearby complex gain calibrators with a @xmath74 minute switching cadence . spectral line and continuum observations were simultaneous . the spectral line observations were centered on the redshift of the host galaxy , had 1.11.9 km s@xmath4 spectral resolution , used 1536 channels to span 128 mhz ( 17002900 km s@xmath4 ) , and used dual circular polarization and 8-bit sampling . continuum observations spanned 4 ghz , 1822 ghz , using 32 spectral windows spanning 128 mhz each using 128 channels in dual circular polarization and 3-bit sampling . table [ tab : obs ] lists the details of the observations and the rms noise in the line cubes and continuum maps . typical beam sizes were 80100 milliarcseconds . noise was about 3 mjy beam@xmath4 in 1.2 km s@xmath4 channels in the spectral line cubes and about 1520 @xmath23jy beam@xmath4 in the continuum . the exception was the @xmath51 water maser j0804 + 3607 @xcite that was redshifted to 13.4 ghz , in ku band , which necessarily had lower angular and spectral resolution and lower rms noise in the line but higher rms continuum noise . in this case , the continuum was centered on 14 ghz and spanned 1216 ghz . the observing session on september 15 2015 had poor 22 ghz weather for a configuration and could not be calibrated or imaged . ngc 3359 , ngc 3556 , and ngc 3735 are therefore only listed in table [ tab : obs ] and are not discussed further or included in any analysis of the remaining 13 objects . all data reduction and analysis was performed using the common astronomy software applications package ( casa ; * ? ? ? calibration and flagging used a modified casa pipeline plus additional manual flagging . imaging used briggs weighting with robustness 0.5 . spectra were extracted from spectral line cubes using a maser - centered beam , and integrated line maps were restricted to line - emitting channels . all spectra use the optical velocity definition in the barycentric reference frame . ccccrcccc ngc 291 & 2015 - 06 - 21 & 23.3 & 12884 & 20.4 & 5033 & 2.6 & 1.2 & 17 + ngc 520b & 2015 - 06 - 21 & 23.3 & 10289 & 27.8 & 1412 & 3.1 & 1.1 & 20 + j0350@xmath00127 & 2015 - 06 - 21 & 23.3 & 10087 & 20.5 & 8170 & 3.1 & 1.2 & 15 + ic 485 & 2015 - 09 - 26 & 23.4 & 8482 & 19.6 & 4746 & 2.2 & 1.2 & 18 + j0804 + 3607 & 2015 - 09 - 26 & 23.3 & 151130 & @xmath081.3 & 1065917 & 0.9 & 1.9 & 47 + cgcg 120@xmath0039 & 2015 - 09 - 26 & 23.3 & 8581 & 33.7 & 4442 & 2.1 & 1.2 & 20 + j0912 + 2304 & 2015 - 09 - 06 & 23.3 & 9281 & 65.6 & 6759 & 2.9 & 1.2 & 16 + j1011@xmath01926 & 2015 - 09 - 06 & 23.4 & 17386 & @xmath020.8 & 9849 & 5.3 & 1.2 & 19 + ngc 3359 & 2015 - 09 - 15 & 23.3 & 11272 & 48.6 & 96 & & 1.1 & + ngc 3556 & 2015 - 09 - 15 & 23.3 & 10472 & 45.0 & 64 & & 1.1 & + ngc 3735 & 2015 - 09 - 15 & 23.3 & 11268 & 36.8 & 2113 & & 1.2 & + ugc 7016 & 2015 - 09 - 06 & 23.3 & 11491 & @xmath074.8 & 5443 & 3.8 & 1.2 & 16 + ngc 5256 & 2015 - 06 - 19 & 23.3 & 9684 & @xmath073.8 & 5548 & 3.0 & 1.2 & 17 + ngc 5691 & 2015 - 06 - 19 & 23.4 & 13185 & 39.3 & 159 & 3.7 & 1.1 & 17 + cgcg 168@xmath0018 & 2015 - 06 - 19 & 23.3 & 8682 & 50.0 & 6360 & 2.7 & 1.2 & 15 + j1939@xmath00124 & 2015 - 06 - 19 & 25.0 & 9687 & 37.7 & 4138 & 3.0 & 1.2 & 16 + water masers were detected in 9 out of 13 objects , and 20 ghz continuum emission was detected in 7 out of 13 objects . only five objects show both maser and continuum emission , and two objects were detected in neither line nor continuum . figures [ fig : ngc291][fig:2masx1939 ] show a 1 square field of view of the first moment maser maps and continuum contours and they show spectra of the detected objects ( in line , continuum , or both ) . table [ tab : positions ] lists the maser and continuum centroids based on two - dimensional gaussian fits . for the spectral lines , these fits were made to the integrated line maps . the maser emission was universally unresolved , but the continuum emission was formally resolved when deconvolved from the beam in all but two objects , ic 485 and cgcg 168@xmath0018 . table [ tab : positions ] also lists the maser - continuum offsets in angular and physical units . offsets for four of the five objects detected in both maser and continuum are non - significant with @xmath52 uncertainties ranging from about 1 to 40 pc . the only object showing a significant offset between the maser and continuum centroids is cgcg 168@xmath0018 , with a @xmath53 pc offset ( figure [ fig : cgcg168 - 018 ] ) . table [ tab : masers ] lists the measured and derived water maser properties : peak and integrated flux densities , luminosity distance , isotropic line luminosity , the range of velocities spanned by the line emission 3@xmath12 above the noise , the velocity of peak emission , and the adopted systemic velocity . for j0804 + 3607 , we list redshifts rather than velocities . the detected maser velocities are consistent with previous observations , although the masers can be substantially offset from the systemic velocities , which are obtained from optical and hi 21 cm lines . it is unclear whether these velocity offsets are physical ( i.e. , due to different line - emitting regions genuinely having different velocities , as is seen in shock - induced maser emission ) , due to obscuration ( optical vs. radio lines ) , or due to measurement error , particularly in optical redshifts . we therefore do not rely on the velocity offset between the maser emission and the adopted systemic velocity as a criterion for assessing the likelihood of a maser arising from an inclined disk . isotropic maser luminosities range from kilomaser values ( @xmath54 in ngc 520b ) to the exceptionally luminous , @xmath55 in j0804 + 3607 ( section [ sec : discussion ] ) . table [ tab : continuum ] shows the 20 ghz radio continuum properties of the seven detected objects . we include the peak flux density , the integrated flux density , the spectral index derived solely from the 1822 ghz bandpass , and the deconvolved angular size . among the six continuum sources with enough signal - to - noise to derive a significant spectral index , four are steep spectrum ( @xmath56 to @xmath57 ) and two are flat ( @xmath58 ) . one of the latter , cgcg 120@xmath0039 , is only marginally resolved.ic 485 , which does not have a spectral index measurement , but which shows both maser and continuum emission , does not have a resolved continuum . cllllcc ngc 291 & & & 00:53:29.9101(11 ) & @xmath008.46.03.740(14 ) & & + ngc 520b & 01:24:34.91412(14 ) & + 03.47.29.7864(22 ) & 01:24:34.9099(28 ) & + 03.47.29.7783(92 ) & 64(42 ) & 8.6(5.6 ) + j0350@xmath00127 & 03:50:00.352168(29 ) & @xmath001.27.57.39574(53 ) & & & & + ic 485 & 08:00:19.752486(74 ) & + 26.42.05.0526(10 ) & 08:00:19.7515(14 ) & + 26.42.05.050(14 ) & 13(19 ) & 7.6(10.5 ) + j0804 + 3607 & 08:04:31.01144(40 ) & + 36.07.18.1937(52 ) & 08:04:31.01138(12 ) & + 36.07.18.19927(91 ) & 5.6(5.3 ) & 39.6(37.2 ) + cgcg 120@xmath0039 & 08:49:14.07078(16 ) & + 23.22.48.9408(20 ) & 08:49:14.07097(13 ) & + 23.22.48.9346(17 ) & 6.7(2.7 ) & 3.5(1.4 ) + j0912 + 2304 & 09:12:46.36659(33 ) & + 23.04.27.2421(28 ) & & & & + j1011@xmath01926 & 10:11:50.56731(17 ) & @xmath019.26.43.9645(59 ) & & & & + ugc 7016 & & & & & & + ngc 5256 & & & 13:38:17.79219(15 ) & + 48.16.41.1389(19 ) & & + & & & 13:38:17.24843(76 ) & + 48.16.32.2095(92 ) & & + ngc 5691 & & & & & & + cgcg 168@xmath0018 & 16:30:40.90329(19 ) & + 30.29.19.7066(29 ) & 16:30:40.90134(21 ) & + 30.29.19.7216(20 ) & 29.3(3.6 ) & 21.6(2.7 ) + j1939@xmath00124 & 19:39:38.91545(38 ) & @xmath001.24.33.2553(39 ) & & & & + crrrrcrrc ngc 520b & 35(3 ) & 166(14 ) & 28 & 3.05(26 ) & 22702273 & 2272(1 ) & 2288(8 ) & 1 + j0350@xmath00127 & 347(4 ) & 7200(150 ) & 180 & 5181(108 ) & 1233612393 & 12369(1 ) & 12322(18 ) & 2 + ic 485 & 78(2 ) & 2470(130 ) & 125 & 868(46 ) & 83078387 & 8356(1 ) & 8338(10 ) & 3 + j0804 + 3607 & 9(1 ) & 80(6 ) & 3997 & @xmath59 & 0.660380.66051 & 0.66045(1 ) & 0.65654(37 ) & 4 + cgcg 120@xmath0039 & 82(2 ) & 438(46 ) & 116 & 133(14 ) & 75597565 & 7565(1 ) & 7684(26 ) & 5 + j0912 + 2304 & 16(3 ) & 252(44 ) & 164 & 151(26 ) & 1085510878 & 10855(1 ) & 10861(26 ) & 6 + j1011@xmath01926 & 44(4 ) & 624(82 ) & 123 & 213(28 ) & 80438065 & 8048(1 ) & 8065(31 ) & 7 + cgcg 168@xmath0018 & 25(2 ) & 279(36 ) & 162 & 163(21 ) & 1113411164 & 11140(1 ) & 11015(29 ) & 5 + j1939@xmath00124 & 30(2 ) & 519(80 ) & 93 & 102(16 ) & 61706206 & 6198(1 ) & 6226(20 ) & 8 + ccccccc ngc 291 & 263(24 ) & 1.93(20 ) & @xmath02.0(3 ) & 451(52)146(19 ) & 51.8(3.4 ) & 177(20)57(7 ) + ngc 520b & 272(31 ) & 4.31(52 ) & @xmath00.23(19 ) & 775(100)164(24 ) & 95.6(2.3 ) & 105(14)22(3 ) + ic 485 & 77(15 ) & 0.180(48 ) & & < 9388 & @xmath069.5 & < 5249 + j0804 + 3607 & 4247(83 ) & 4.71(16 ) & @xmath01.5(3 ) & 62(12)23(13 ) & 71(17 ) & 437(85)162(92 ) + cgcg 120@xmath0039 & 482(20 ) & 0.555(39 ) & @xmath00.27(16 ) & 44(8)23(14 ) & 133(33 ) & 23(4)12(7 ) + ngc 5256 & 633(28 ) & 1.543(68 ) & @xmath02.0(2 ) & 127(6)96(8 ) & 148.7(9.6 ) & 72(3)55(5 ) + & 78(17 ) & 0.48(11 ) & & 433(23)68(58 ) & 163.1(3.1 ) & 247(13)39(33 ) + cgcg 168@xmath0018 & 267(16 ) & 0.308(30 ) & @xmath00.95(17 ) & < 8785 & 77.8 & < 6463 + figure [ fig : offsets ] shows the projected physical offset between the water maser line and continuum centroids for the five objects detected in both line and continuum ( table [ tab : positions ] ) . only cgcg 168@xmath0018 shows a significant offset of @xmath60 milliarcseconds or @xmath53 pc , but based on its 1822 ghz continuum spectral index of @xmath61 ( table [ tab : continuum ] ) , the continuum could be jet emission and not the core of the agn . all other maser - continuum offsets are not significant and are therefore consistent with inclined maser disk expectations . 018 shows a significant offset . error bars are plotted symmetrically . [ fig : offsets],scaledwidth=45.0% ] among the 13 water maser hosts , we did not detect water masers in four objects : ngc 291 , ugc 7016 , ngc 5256 , and ngc 5691 . the water maser in ngc 520b is likely associated with star formation and is rejected as an inclined maser disk candidate ( see [ subsubsec : ngc520b ] ) . objects with unresolved masers but not detected in the 20 ghz continuum remain ambiguous ( these are j0350@xmath00127 , j0912 + 2304 , j1011@xmath01926 , and j1939@xmath00124 ; note that none of these have other sub - arcsecond radio continuum observations ) , as does cgcg 168@xmath0018 , the only object to show a significant maser - continuum offset ( see [ subsubsec : cgcg168 - 018 ] ) . ic 485 , j0804 + 3607 , and cgcg 120@xmath0039 remain inclined maser disk candidates : they show no significant maser - continuum offset and no high velocity lines ( by selection ) . while the maser emission from ic 485 is broad and multi - component , this type of structure spanning @xmath7100 km s@xmath4 is seen in the systemic masers in many disk systems ( e.g. , * ? ? ? j0808 + 3607 and cgcg 120@xmath0039 have narrow maser lines and are particularly good inclined disk candidates . we discuss the previous water maser observations , general characteristics , and our results for each individual object below . ngc 291 is a barred spiral galaxy with a seyfert 2 nucleus @xcite . the maser was detected as a narrow @xmath760 mjy line in 2006 by gbt program gbt06a-009,^[footnote : maser_url]^ but we did not detect it ( 2.6 mjy beam@xmath4 rms per 1.2 km s@xmath4 channel ; table [ tab : obs ] and figure [ fig : ngc291 ] ) nor did @xcite in 2002 ( 15 mjy rms per 1.3 km s@xmath4 channel ) . we detect extended 20 ghz continuum with 1822 ghz spectral index @xmath62 ( table [ tab : continuum ] ) . ngc 520b is one galaxy in the colliding pair in ngc 520 ( e.g. , * ? ? ? @xcite detected a maser in 2005 , measuring a @xmath63 mjy peak with the vla and @xmath64 . our vla detection shows a @xmath65 mjy peak and @xmath66 ( figure [ fig : ngc520b ] and table [ tab : masers ] ) . our vla map shows extended 20 ghz radio continuum emission ( figure [ fig : ngc520b ] , bottom ) with a flat 1822 ghz spectral index of @xmath67 ( table [ tab : continuum ] ) and a similar east - west morphology to the @xcite 14.9 ghz map . in our map the maser is slightly north of the @xcite maser position . there is a significant maser - continuum peak offset , although the single - component centroid fits listed in table [ tab : positions ] are consistent because the radio emission is extended . @xcite favor a star formation origin for the water maser in ngc 520b , but can not rule out a low luminosity agn . the preponderance of evidence suggests that ngc 520b is a poor candidate for an inclined maser disk . j0350@xmath00127 is an almost otherwise unknown spiral or possibly irregular galaxy included in the 2mass redshift survey @xcite and detected in water maser emission by the gbt programs gbt09 - 051^[footnote : maser_url]^ in 2010 and gbt10c-019^[footnote : maser_url]^ in 2011 . the peak flux density was @xmath7350 mjy in 2011 , in good agreement with our vla measurement of @xmath68 mjy ( table [ tab : masers ] and figure [ fig:2masx0350 ] ) . this is a broad and luminous maser ( @xmath69 ) with no associated 20 ghz continuum , down to an rms noise of 15 @xmath23jy beam@xmath4 ( table [ tab : obs ] ) . without a continuum detection , the nature of this maser remains ambiguous , although it is almost certainly associated with an agn . ic 485 is a spiral galaxy detected in the water maser line in 2006 in gbt program gbt06c-035.^[footnote : maser_url]^ the maser has a @xmath780 mjy peak and shows a broad profile . @xcite lists this object as a maser nondetection . our vla observations show a broad , multi - component maser with a similar peak ( @xmath70 mjy ) and a high luminosity , @xmath71 ( figure [ fig : ic485 ] and table [ tab : masers ] ) . the 20 ghz continuum is detected but unresolved and faint ( @xmath72 @xmath23jy beam@xmath4 peak flux density ; table [ tab : continuum ] ) . this galaxy was also detected at 1.4 ghz ( 4.4 mjy ) , and its dominant radio energy source was classified as star formation by @xcite . this classification does not exclude the presence of an agn : @xcite classify the optical nucleus of ic 485 as a liner . the maser - continuum offset is not significant less than 10.5 pc ( 1@xmath12 ; table [ tab : positions ] ) so this maser remains an inclined disk candidate . this is a type 2 quasar showing luminous water maser emission at @xmath51 @xcite . the isotropic maser luminosity was @xmath73 in 2005 ( * ? ? ? * error from the quoted 20% calibration uncertainty ) and @xmath74 in 2015 ( this work ) , consistent with no variation . the maser and 20 ghz continuum emission are coincident to within 40 pc ( 1@xmath12 ; table [ tab : positions ] and figure [ fig : sdssj0804 ] ) , which is less precise than the other objects due to the much larger distance and lower observing frequency . the 1216 ghz spectral index is @xmath75 ( table [ tab : continuum ] ) . this object remains an inclined maser disk candidate . the water maser in this little - studied galaxy was detected in 2013 in gbt program gbt13a-236.^[footnote : maser_url]^ it had a peak flux density of @xmath7210 mjy , was blueshifted from the systemic velocity , and showed several components . we detect the maser now at @xmath76 mjy peak , but still showing multiple components ( figure [ fig : cgcg120 - 039 ] ) and the 20 ghz continuum at @xmath77 mjy , which are spatially coincident to within a remarkably small @xmath1 pc ( table [ tab : positions ] ) . the 1822 ghz spectral index is flat : @xmath78 ( table [ tab : continuum ] ) . this object is a good inclined maser disk candidate . the water maser in the galaxy j0912 + 2304 was detected in 2008 by gbt program gbt07a-034,^[footnote : maser_url]^ showing a @xmath730 mjy peak . @xcite list this as a water maser nondetection . the vla observations show a @xmath79 mjy peak , but no 20 ghz continuum down to an rms noise of 16 @xmath23jy beam@xmath4 ( figure [ fig:2masx0912 ] and tables [ tab : obs ] and [ tab : masers ] ) . the provenance of the maser remains ambiguous . the maser in this almost unknown galaxy was detected in gbt program gbt07a-066^[footnote : maser_url]^ in 2008 . @xcite list this object as a water maser nondetection . the gbt detection shows a broad line with a @xmath780 mjy peak . the vla detection shows a broad multi - component maser with a @xmath80 mjy peak in good agreement with the systemic velocity ( figure [ fig:2masx1011 ] and table [ tab : masers ] ) . we do not detect the 20 ghz continuum . the nature of this maser therefore remains ambiguous . the water maser in this barred spiral @xcite was detected in 2007 by gbt program gbt07a-034,^[footnote : maser_url]^ but @xcite list this as a nondetection . the gbt detection spectrum had a peak flux density of @xmath755 mjy , but we did not detect the maser or the 20 ghz continuum emission . vla rms noise values were 3.8 mjy beam@xmath4 per 1.2 km s@xmath4 channel in the spectral line cube and 16 @xmath23jy beam@xmath4 in the continuum map ( table [ tab : obs ] ) . @xcite discovered the water maser in this merging luminous infrared galaxy in 2003 . it had a peak flux density of 99 mjy ( @xmath81 ) and was redshifted from the systemic velocity by @xmath7300 km s@xmath4 . @xcite claim that all maser emission originates from the southern nucleus ( a sey 2 nucleus according to , e.g. , * ? ? ? we detect two 20 ghz continuum sources that are consistent with the positions of the two nuclei ( tables [ tab : positions ] and [ tab : continuum ] ; * ? ? ? * ) , but neither location ( nor the larger region that includes the overlap region between the two nuclei ) shows maser emission down to an rms noise of 3.0 mjy beam@xmath4 in 1.2 km s@xmath4 channels ( table [ tab : obs ] and figure [ fig : ngc5256 ] ) . the spectral index of the northern nuclear continuum is @xmath82 in the 1822 ghz band ( table [ tab : continuum ] ) . the water maser in ngc 5691 , a barred non - seyfert spiral galaxy @xcite , was detected in 2009 by gbt program agbt08c-035^[footnote : maser_url]^ with a @xmath745 mjy peak flux density . it was listed by @xcite as a maser nondetection . we did not detect the maser or any 20 ghz continuum in this galaxy , with rms noise levels of 3.7 mjy beam@xmath4 per 1.1 km s@xmath4 channel and 17 @xmath23jy beam@xmath4 , respectively ( table [ tab : obs ] ) . cgcg 168@xmath0018 is a little - studied galaxy classified as an agn by @xcite and listed as a water maser nondetection by @xcite . water maser emission was detected by gbt program gbt07a-066^[footnote : maser_url]^ in 2008 . the gbt water maser spectrum shows a @xmath750 mjy peak , detected by this work at @xmath83 mjy ( table [ tab : masers ] ) . the @xmath84 mjy continuum shows a 18 - 22 ghz spectral index of @xmath61 ( table [ tab : continuum ] ) . the maser emission and 20 ghz continuum are unresolved but show a significant relative offset of @xmath53 pc ( table [ tab : positions ] and figure [ fig : cgcg168 - 018 ] ) . this is the only object in the sample that shows a significant offset between the maser and continuum emission ( figure [ fig : offsets ] ) . while the offset suggests that the maser emission is not deflected from an inclined maser disk , the continuum spectral index suggests that the continuum may arise from a jet , and the radio core is not detected . the provenance of the maser therefore remains ambiguous . @xcite detected the water maser in this spiral galaxy hosting a sey 2 nucleus in 2002 using the tidbinbilla antenna . the maser showed a peak flux density of @xmath728 mjy , and @xcite report @xmath85 . no 22 ghz continuum was detected by @xcite using the vla ( @xmath86 mjy ) . we likewise detect no continuum , with rms noise of 16 @xmath23jy beam@xmath4 ( table [ tab : obs ] ) , but we do detect the maser emission with peak @xmath87 mjy ( @xmath88 ) although with a substantially different maser profile ( figure [ fig:2masx1939 ] and table [ tab : masers ] ) . the 6 and 20 cm continua were detected at @xmath89 mjy and @xmath90 mjy , respectively , by @xcite , and the 6 cm continuum position agrees with the maser position to within @xmath71(@xmath7430 pc ) . the nature of this maser remains ambiguous . this paper presents a physical mechanism that may enable detection of inclined water maser disks orbiting massive black holes via the lensing / deflection of in - going systemic masers . the observational signature of an inclined disk is a maser line or line complex with limited doppler extent that appears to arise at the location of the black hole , as identified by its radio continuum core . with enough angular resolution , it may be possible to measure the black hole mass if the maser emission forms a lensing arc or einstein ring , but the mass precision will be limited by one s ability to measure or estimate the size of the maser - emitting portion of the disk . we suggest that if inclined maser disks can be detected at all , then they have probably already been detected in single - dish surveys but discarded for interferometric follow - up because they did not show high - velocity lines . we present original 0.070.17 ( 4100 pc ) resolution vla observations of inclined maser disk candidates with the goal of identifying systems where the maser emission is unresolved and is coincident with the 20 ghz continuum emission . of the 16 masers observed with the vla , we obtained useful data for 13 , and among these , five were detected in both 22 ghz maser line emission and in 20 ghz continuum . of these five , one maser is most likely associated with star formation ( ngc 520b ) , and one shows a significant spatial offset between the maser emission and the continuum ( cgcg 168@xmath0018 , but it could still host an inclined maser disk this case is ambiguous ) . three objects are good inclined maser disk candidates that merit further study with vlbi : ic 485 , j0804 + 3607 , and cgcg 120@xmath0039 . five maser hosts remain ambiguous , based either on non - detected or offset continua : j0350@xmath00127 , j0912 + 2304 , j1011@xmath01926 , cgcg 168@xmath0018 , and j1939@xmath00124 . more straightforward methods for measuring black hole masses from molecular lines may be in the offing . for example , @xcite and @xcite have used carbon monoxide kinematics in thin disks that approach or are within the black hole gravitational sphere of influence to obtain black hole mass measurements . @xcite , in particular , demonstrate the ability of alma to measure black hole masses with @xmath710% uncertainty . although they lack the intrinsic brightness of masers that enables vlbi mapping , thermal molecular lines have the advantage of being observable at any disk inclination . we thank a. hamilton and m. eracleous for helpful discussion and b. butler for assistance with metadata repair of bespoke observing configurations . we also thank the anonymous referee for important feedback . this research has made use of nasa s astrophysics data system bibliographic services and the nasa / ipac extragalactic database ( ned ) , which is operated by the jet propulsion laboratory , california institute of technology , under contract with the national aeronautics and space administration . abazajian , k. , adelman - mccarthy , j. k. , ageros , m. a. , et al . 2004 , , 128 , 502 adelman - mccarthy , j. k. , ageros , m. a. , allam , s. s. , et al . 2007 , , 172 , 634 adelman - mccarthy , j. k. , ageros , m. a. , allam , s. s. , et al . 2008 , , 175 , 297 amiri , n. , & darling , j. 2016 , , 826 , 136 argon , a. l. , greenhill , l. j. , reid , m. j. , moran , j. m. , & humphreys , e. m. l. 2007 , , 659 , 1040 barth , a. j. ; boizelle , b. d. , darling , j. , baker , a. j. , buote , d. a. , ho , l. c. , & walsh , j. l. 2016 , , 822 , l28 barth , a. j. , darling , j. , baker , a. j. , boizelle , b. d. , buote , d. a. , ho , l. c. , & walsh , j. l. 2016 , , 823 , 51 barvainis , r. & antonucci , r. 2005 , , 628 , l89 van den bosch , r. c. e. , greene , j. e. , braatz , j. a. , constantin , a. , & kuo , c .- y . 2016 , , 819 , 11 braatz , j. a. , henkel , c. , greenhill , l. j. , moran , j. m. , & wilson , a. s. 2004 , , 617 , l29 braatz , j. a. , reid , m. j. , humphreys , e. m. l. , henkel , c. , condon , j. j. , & lo , k. y. 2010 , , 718 , 657 brogan , c. , johnson , k. , & darling , j. 2010 , , 716 , l51 brown , m. j. i. , moustakas , j. , smith , j .- d . t. , da cunha , e. , jarrett , t. h. , imanishi , m. , armus , l. , brandl , b. r. , & peek , j. e. g. 2014 , , 212 , 18 castangia , p. , tarchi , a. , henkel , c. , & menten , k. m. 2008 , , 479 , 111 claussen , m. j. , diamond , p. j. , braatz , j. a. , wilson , a. s. , & henkel , c. 1998 , , 500 , l129 condon , j. j. , cotton , w. d. , & broderick , j. j. 2002 , , 124 , 675 darling , j. , brogan , c. , & johnson , k. 2008 , , 685 , l39 darling , j. 2011 , , 732 , l2 davis , t. a. , bureau , m. , cappellari , m. , sarzi , m. , & blitz , l. 2013 , nature , 494 , 328 einstein , a. 1915 , preussische akademie der wissenschaften , sitzungsberichte , ( part 2 ) , 831 einstein , a. 1936 , science , 84 , 506 gallimore , j. f. , baum , s. a. , odea , c. p. , brinks , e. , & pedlar , a. 1996 , , 462 , 740 gao , f. , braatz , j. a. , reid , m. j. , lo , k. y. , condon , j. j. , henkel , c. , kuo , c. y. , impellizzeri , c. m. v. , pesce , d. w. , & zhao , w. 2016 , , 817 , 128 greene , j. e. , seth , a. , kim , m. , lsker , r. , goulding , a. d. , gao , f. , braatz , j. a. , henkel , c. , condon , j. , lo , f. k. y. , & zhao , w. 2016 , in press , ( arxiv:1606.00018 ) greenhill , l. j. , henkel , c. , becker , r. , wilson , t. l. , & wouterloot , j. g. a. 1995 , , 304 , 21 greenhill , l. j. , herrnstein , j. r. , moran , j. m. , menten , k. m. , & velusamy , t. 1997 , , 486 , l15 greenhill , l. j. , kondratko , p. t. , lovell , j. e. j. , kuiper , t. b. h. , moran , j. m. , jauncey , d. l. , & baines , g. p. 2003 , , 582 , l11 greenhill , l. j. , booth , r. s. , ellingsen , s. p. , herrnstein , j. r. , jauncey , d. l. , mcculloch , p. m. , moran , j. m. , norris , r. p. , reynolds , j. e. , & tzioumis , a. k. 2003 , , 590 , 162 henkel , c. , peck , a. b. , tarchi , a. , nagar , n. m. , braatz , j. a. , castangia , p. , & moscadelli , l. 2005 , , 436 , 75 herrnstein , j. r. , greenhill , l. j. , moran , j. m. , diamond , p. j. , inoue , m. , nakai , n. , & miyoshi , m. 1998 , , 497 , l69 hewett , p. c. & wild , v. 2010 , , 405 , 2302 hofner , p. , baan , w. a. , & takano , s. 2006 , , 131 , 2074 huchra , j. p. , macri , l. m. , masters , k. l. , et al . 2012 , , 199 , 26 humphreys , e. m. l. , reid , m. j. , moran , j. m. , greenhill , l. j. , & argon , a. l. 2013 , , 775 , 13 jones , d. h. , read , m. a. , saunders , w. , et al . 2009 , , 399 , 683 kewley , l. j. , heisler , c. a. , dopita , m. a. , & lumsden , s. 2001 , , 132 , 37 kondratko , p. t. , greenhill , l. j. , & moran , j. m. 2005 , , 618 , 618 kondratko , p. t. , greenhill , l. j. , moran , j. m. , et al . 2006 , , 638 , 100 kuo , c. y. , braatz , j. a. , condon , j. j. , impellizzeri , c. m. v. , lo , k. y. , zaw , i. , schenker , m. , henkel , c. , reid , m. j. , & greene , j. e. 2011 , , 727 , 20 kuo , c. y. , braatz , j. a. , lo , k. y. , reid , m. j. , suyu , s. h. , pesce , d. w. , condon , j. j. , henkel , c. , & impellizzeri , c. m. v. 2015 , , 800 , 26 liu , x. , shen , y. , strauss , m. a. , & hao , l. 2011 , , 737 , 101 mazzarella , j. m. & boroson , t. a. 1993 , , 85 , 27 mcmullin , j. p. , waters , b. , schiebel , d. , young , w. , & golap , k. 2007 , astronomical data analysis software and systems xvi ( asp conf . 376 ) , ed . r. a. shaw , f. hill , & d. j. bell ( san francisco , ca : asp ) , 127 miyoshi , m. , moran , j. , herrnstein , j. , greenhill , l. , nakai , n. , diamond , p. , & inoue , m. 1995 , nature , 373 , 127 mulchaey , j. s. , regan , m. w. , & kundu , a. 1997 , , 110 , 299 nair , p. b. & abraham , r. g. 2010 , , 186 , 427 nakai , n. , inoue , m. , miyazawa , k. , miyoshi , m. , & hall , p. 1995 , , 47 , 771 narayan , r. & bartelmann , m. 1995 , proceedings of the 1995 jerusalem winter school , formation of structure in the universe , ed . a. dekel & j. p. ostriker ( cambridge university press ) peck , a. b. , henkel , c. , ulvestad , j. s. , brunthaler , a. , falcke , h. , elitzur , m. , menten , k. m. , & gallimore , j. f. 2003 , , 590 , 149 riess , a. g. , macri , l. m. , hoffmann , s. l. , scolnic , d. , casertano , s. , filippenko , a. v. , tucker , b. e. , reid , m. j. , jones , d. o. , silverman , j. m. , chornock , r. , challis , p. , yuan , w. , brown , p. j. , & foley , r. j. 2016 , , 826 , 56 schawinski , k. , urry , c. m. , virani , s. , et al . 2010 , , 711 , 284 springob , c. m. , haynes , m. p. , giovanelli , r. , & kent , b. r. 2005 , , 160 , 149 stanford , s. a. & balcells , m. 1990 , , 355 , 59 tarchi , a. , henkel , c. , peck , a. b. , & menten , k. m. 2002 , , 389 , l39 tarchi , a. , henkel , c. , peck , a. b. , & menten , k. m. 2002 , , 385 , 1049 tarchi , a. , castangia , p. , columbano , a. , panessa , f. , & braatz , j. a. 2011 , , 532 , a125 tarchi , a. , castangia , p. , henkel , c. , surcis , g. , & menten , k. m. 2011 , , 525 , a91 theureau , g. , hanski , m. o. , coudreau , n. , hallet , n. , & martin , j .- 2007 , , 465 , 71 vader , j. p. , frogel , j. a. , terndrup , d. m. , & heisler , c. a. 1993 , , 106 , 1743 de vaucouleurs , g. , de vaucouleurs , a. , corwin , h. g. , buta , r. j. , paturel , g. , & fouque , p. 1991 , third reference catalogue of bright galaxies ( rc3 ) , springer - verlag : new york zakamska , n. l. , strauss , m. a. , krolik , j. h. , et al . 2003 , , 126 , 2125 zhu , g. , zaw , i. , blanton , m. r. , & greenhill , l. j. 2011 , , 742 , 73

distilling highly entangled quantum states from weaker ones is a process that is crucial for efficient and long - distance quantum communication , and has implications for several other quantum information protocols . we introduce the notion of distillation under limited resources , and specifically focus on the energy constraint . the corresponding protocol , which we call the canonical distillation of entanglement , naturally leads to the set of canonically distillable states . we show that for non - interacting hamiltonians , almost no states are canonically distillable , while the situation can be drastically different for interacting ones . several paradigmatic hamiltonians are considered for bipartite as well as multipartite canonical distillability . the results have potential applications for practical quantum communication devices .

1411.7936

over the last twenty five years or so , entangled quantum states shared between distant parties have been proved to be essential for several quantum protocols @xcite . however , unavoidable destruction of quantum coherence due to noisy quantum channels diminishes the quality of the shared quantum state , thereby posing a challenge to the implementation of such protocols . invention of distillation protocols @xcite to purify highly entangled states from collection of states with relatively low entanglement has been proven crucial in order to overcome such difficulties in device independent quantum cryptography @xcite , quantum dense coding @xcite , and quantum teleportation @xcite - the three pillars of quantum communication . entanglement distillation is also indispensable in quantum repeater models @xcite , used to overcome the exponential scaling of the error probabilities with the length of the noisy quantum channel connecting distant parties sharing the quantum state . existence of bound entangled ( be ) states @xcite - entangled states from which no pure entangled state can be obtained using local operations and classical communications ( locc ) - further highlights the importance of identifying distillable states . entanglement distillation protocols have also been used in problems related to topological quantum memory @xcite . laboratory realization of single copy distillation has been performed and possible experimental proposal of multicopy distillation has been given @xcite . there is a close correspondence between entanglement and energy @xcite . moreover , consideration of statistical ensembles of quantum states of a system in terms of various constraints on its energy and number of particles is crucial in several areas of physics , including in quantum communication . an important example is the classical capacity of a noiseless quantum channel @xcite for transmitting classical information using quantum states . the classical capacity is quantified by the von neumann entropy of the maximally mixed quantum state that can be sent through the noiseless quantum channel . the holevo bound@xcite dictates that at most @xmath0 bits of classical information can be transmitted using @xmath0 distinguishable qubits , thereby predicting an infinite capacity for infinite dimensional systems , such as the bosonic channels @xcite . since the energy required to achieve infinite capacity is also infinite , such non - physicality can be taken care of by calculating the capacity under appropriate energy constraints . constraints on available energy can also be active in other quantum information protocols including infinite- as well as finite - dimensional systems and in particular may give rise to a novel understanding of the interplay between entanglement and energy . for example , to implement ideas like quantum repeaters for long - distance quantum state distribution , an energy - constrained protocol for the distillation of entanglement may be necessary . evidently , in that case , the energy of the states involved in the distillation process must follow constraints according to the physical situation in hand . the logical choice of such constraints may include bounds on average energy , or maximum available energy of the quantum states . in this paper , we consider the process of distillation of highly entangled quantum states of shared systems from weakly entangled ones within the realm of limited resources . specifically , we propose that a distillation protocol have to be carried out under an energy constraint , and refer to it as canonicaldistillation . we prove that non - interacting hamiltonians lead to situations where canonically distillable states form a set of measure zero . the situation , however , drastically changes with the inclusion of interaction terms . we consider several paradigmatic interacting hamiltonians of spin-@xmath1 systems , viz . the transverse - field @xmath2 model @xcite , the longitudinal - field @xmath2 model , and the @xmath3 model in an applied field @xcite , and the concept of canonical distillation is probed in each case . the interrelation between canonical distillability and the temperature in thermal states is also investigated . the findings are generic in the sense that they hold also in higher dimensions and for higher number of parties . the energy constraint in these cases is introduced by respectively considering a bilinear - biquadratic hamiltonian @xcite of two spin-@xmath4 particles and a multisite transverse @xmath2 model . the paper is organized as follows . in sec . [ definition ] , we define the canonical distillability of bipartite as well as multipartite quantum states . [ cd - biparty ] contains the results on application of the canonical distillation protocol in bipartite systems described by well - known spin hamiltonians . the canonical distillability of a number of mixed states with respect to these hamiltonians are also tested . in sec . [ cd - multiparty ] , we discuss the canonical distillability of multipartite states , focusing on three - qubit pure states belonging to the greenberger - horne - zeilinger ( ghz ) @xcite and the w @xcite classes . [ conclusion ] contains the concluding remarks . ) , while the boundary between the nppt be states and the nppt distillable states is non - convex under an assumption @xcite . separable and ppt states are convex states . ] we begin by providing a formal definition of _ canonical _ distillation of entanglement for two - qubit systems in the asymptotic limit . generalization to higher dimensions and higher number of parties are considered later . in usual entanglement distillation @xcite , one intends to produce the largest number , @xmath5 , of copies of the maximally entangled bell pair , @xmath6 , starting from @xmath0 @xmath7 copies of an entangled two - qubit state , @xmath8 , using only locc . let us assume that an locc on @xmath0 copies of the state @xmath8 creates the state @xmath9 which is close to @xmath5 copies of @xmath10 , or its local unitary equivalent , @xmath11 , so that @xmath12 where @xmath13 . here , @xmath14 , with @xmath15 and @xmath16 being unitary operators on the qubit hilbert space . the distillable entanglement is given by @xmath17 , where the maximum is over all locc protocols satisfying eq . ( [ fidelity ] ) . to introduce appropriate energy constraint , suppose that the two - qubit quantum system in the state @xmath8 is described by the hamiltonian @xmath18 . we assume that the system is in contact with a heat bath such that the average energies of the input and output states of the distillation protocol are equal . this average energy conservation leads to the constraint @xmath19 here we assume that @xmath0 is sufficiently large , so that @xmath20 can be approximated by @xmath21 . it can be shown , by virtue of eq . ( [ fidelity ] ) , that the approximation is an equality for @xmath22 . in that limit , we have @xmath23 we are assuming that the @xmath24 bipartite systems that are traced out and any additional ancillary systems that are used and then discarded out during the locc protocol for the canonical distillation do not contribute to the energy equation . such dissipative channels can be incorporated into the definition , but leads to further intractability in the analysis . also , the average energy constraint can be replaced by a maximal available energy constraint , wherein we expect the broad qualitative features , of the case where the average energy is considered , to be retained . the canonically distillable entanglement , @xmath25 , is the maximum value of @xmath26 that satisfies eq . ( [ cd_actual ] ) for some @xmath27 , and is consistent with eq . ( [ fidelity ] ) . the states with a non - zero @xmath25 are called canonically distillable ( cd ) . one must note that for the two - qubit systems , @xmath28 just like @xmath29 , determination of @xmath25 under the canonical energy constraint is a difficult problem . however , significant insight on cd states can be obtained by considering a weaker version of the energy constraint , viz . @xmath30 we refer to this as the weak canonical energy constraint ( wcec ) . replacing eq . ( [ cd_actual ] ) by eq . ( [ cdeq ] ) leads us to the concept of special cd ( scd ) states , which is clearly a superset of the set of cd states , while being still within the set of distillable states . the definition of canonical distillability can be extended to bipartite states of arbitrary dimensions , where the individual parties have dimensions @xmath31 . in that case , the states @xmath10 and @xmath32 are replaced by @xmath33 and @xmath34 , respectively , with @xmath35 being a maximally entangled pure state in @xmath36 , and @xmath37 . here , @xmath38 , with @xmath39 being an unitary operator on @xmath40 for @xmath41 , and @xmath42 forms the computational basis in @xmath40 . since we are trying to create maximally entangled states in @xmath36 , the relations in ( [ distillability : qubits ] ) are still valid . this is the case where the state @xmath33 is considered to have unit entanglement . if @xmath33 is considered to possess @xmath43 ebits of entanglement , all the expressions in ( [ distillability : qubits ] ) need to be multiplied by @xmath43 . a schematic diagram of the major divisions in the state space with respect to distillability is depicted in fig . [ distil : phase ] . model . ( a ) the probability distribution , @xmath44 , of average energy , @xmath45 , of haar uniformly chosen random two - qubit pure states for different values of @xmath46 at @xmath47 . the distributions becomes sharply peaked at @xmath48 for low values of @xmath46 . ( b ) the variation of the probability @xmath49 that a two - qubit pure state is scd as a function of @xmath46 for different values of @xmath50 ( see eq . ( [ p : general ] ) ) . in this and in all the following figures in this paper , the calculation of @xmath49 does not use the assumption of independence . all quantities employed are dimensionless . ] we conclude this section by pointing out that multipartite extension of canonical distillation can be achieved by considering @xmath0 copies of an @xmath51-party state , @xmath52 , from which @xmath5 copies of @xmath53 , a certain pure state , or its local unitary equivalent @xmath54 , can be created using locc under the canonical constraint . the first result that we prove on canonical distillability of bipartite quantum states is for the case of a general non - interacting hamiltonian @xmath55 , defined on a system of two qudits , each having dimension @xmath56 . the hamiltonian is given by @xmath57 here , @xmath58 with @xmath59 ( @xmath60 , @xmath61 , @xmath62 ) being the @xmath56-dimensional spin operators of a quantum spin-@xmath63 particle @xmath64 , @xmath65 and @xmath66 are unit vectors , @xmath67 denotes the identity operator in @xmath68 , and the subscripts @xmath4 and @xmath69 denote the two qudits . * proposition i. * _ for a system of two qudits described by a non - interacting hamiltonian , almost no states are scd . _ * proof . * the maximally entangled state , @xmath33 ( eq . ( [ qudit_state ] ) ) , and its local unitary equivalents have zero magnetizations in their single - party local density matrices . therefore , @xmath70 @xmath71 , @xmath61 , @xmath62 . hence , for two - qudit hamiltonians of the form ( [ loc - ham ] ) , the wcec reduces to the form @xmath72 the probability that a state ( pure or mixed ) chosen randomly from the entire state space to lie on this surface ( eq . ( [ cd - hyperplane ] ) ) is vanishingly small . hence , almost no two - qudit states are scd if the system is described by a non - interacting hamiltonian of the form @xmath55 . @xmath73 since the cd states form a subset of the scd states , proposition i immediately leads to the following corollary . + * corollary 1 . * _ for a system of two qudits described by a non - interacting hamiltonian , almost no states are cd . _ + * note . * proposition i and corollary 1 are true also in the general case when the local parts of the hamiltonian @xmath55 are expressed as linear combinations of generators of @xmath74 . to investigate whether introduction of interaction terms in the hamiltonian has any effect on canonical distillability of the states of the system , we consider a hamiltonian of the form @xmath75 . here , @xmath76 is the interacting part of @xmath18 whereas @xmath55 is the local part having a generic form as given in eq . ( [ loc - ham ] ) , and @xmath77 and @xmath61 are appropriate system parameters . without any loss of generality , one can scale the system by the parameter @xmath77 so that the hamiltonian of the system takes the form @xmath78 with @xmath79 . let us consider the minimalistic interacting hamiltonian for the system of two qudits given by @xmath80 where @xmath81 and @xmath82 are unit vectors . we refer to @xmath83 as the participation ratio of the interaction part , @xmath76 , in the hamiltonian @xmath18 , with respect to the local part , @xmath55 . * proposition ii . * _ introduction of the minimalistic interacting part in the hamiltonian , with arbitrarily small participation ratio with respect to the local part , results in a non - zero probability of a randomly chosen state to be scd . _ * for a system of two qudits described by the hamiltonian @xmath78 , where @xmath76 and @xmath55 are given by eqs . ( [ smallinteraction ] ) and ( [ loc - ham ] ) , respectively , the wcec reduces to @xmath84 where we have used eq . ( [ loc - zero ] ) . suppose that the limits of variation of the quantity , @xmath85 , of the wcec for a specific value of @xmath86 are @xmath87 and @xmath88 , i.e. , @xmath89 $ ] . note that @xmath87 and @xmath88 respectively are the minimum and the maximum eigenvalues of @xmath18 for a specific value of @xmath86 , while let @xmath90 and @xmath91 be the same of @xmath76 . the range of the right hand side of the wcec is given by @xmath92 $ ] , where @xmath93 with @xmath94 . evidently , the probability of a randomly chosen two - qudit state to be scd is non - zero iff _ @xmath95 , and _ ( ii ) _ @xmath96 $ ] and @xmath97 $ ] have a non - zero overlap . considering the form ( [ smallinteraction ] ) of @xmath76 , @xmath98 is the component of the spin-@xmath63 operator , @xmath99 , along the direction @xmath100 , and let its eigenvectors be @xmath101 . let @xmath102 be the eigenvectors of @xmath103 with the eigenvalues @xmath104 . in this basis , the state given in eq . ( [ qudit_state ] ) can be replaced by @xmath105 a convenient choice of unitary operators is where @xmath106 and @xmath107 , which results in @xmath108 similarly , for a different choice of unitary operators , viz . @xmath109 and @xmath110 , one can obtain @xmath111 . from eq . ( [ int_range_actual ] ) , @xmath112 and @xmath113 , thereby proving @xmath95 . to prove condition _ ( ii ) _ , we point out that the hamiltonian @xmath18 is traceless , which implies @xmath114 and @xmath115 for a specific value of @xmath86 . from the above discussion , it is proved that @xmath116 and @xmath117 , which is possible only when the ranges @xmath96 $ ] and @xmath97 $ ] have a finite overlap . hence , the proof . corollary 2 follows directly from proposition ii , with an additional feature . * corollary 2 . * _ for a system of two qubits described by a hamiltonian of the form @xmath78 , a non - zero probability for a randomly chosen two - qubit state to be scd is guaranteed by a finite overlap of the ranges @xmath97 $ ] and @xmath118 $ ] @xmath119 @xmath96 $ ] . _ * proof . * for a system of two qudits , considering the form of @xmath76 given in eq . ( [ smallinteraction ] ) , @xmath120 . for a two - qubit system ( @xmath121 ) , @xmath122 , implying @xmath123 and @xmath124 . therefore , varying the unitary operators , one can exhaust the full range of the right hand side of wcec . having @xmath118 $ ] @xmath119 @xmath96 $ ] is the additional feature in corollary 2 with respect to proposition ii . @xmath73 next , we wish to estimate the probability that a given quantum state , @xmath8 , is scd with respect to the hamiltonian @xmath18 for a specific value of @xmath86 . if the states @xmath8 are uniformly distributed in the energy range @xmath97 $ ] , the required probability would just be the ratio of the lengths of the two energy ranges , @xmath97 $ ] and @xmath96 $ ] . this , however , is not the case , and the states , @xmath8 , for any given rank , @xmath125 , are typically distributed on the energy range @xmath97 $ ] with a bell - shape . let @xmath126 denotes the probability that an arbitrary two - qudit state @xmath8 of rank @xmath125 has average energy between @xmath45 and @xmath127 . to keep the notations uncluttered , the symbol @xmath125 is not included in the probability density function @xmath44 . let @xmath128 be the probability density that the state @xmath8 ( of rank @xmath125 ) is distillable ( in the usual , non - canonical sense ) given that its average energy is @xmath45 . therefore , the probability , @xmath49 , that a given state , @xmath8 , of rank @xmath125 , is scd with respect to the hamiltonian @xmath18 is given by @xmath129 there does not , as yet , exist an efficient method to estimate the quantity @xmath128 in arbitrary dimensions . however , in @xmath130 systems , distillability is equivalent to being non - positive under partial transposition ( nppt ) @xcite . for such systems , we can perform numerical simulations to estimate this quantity . interestingly , in all the systems that we have considered , numerical evidence indicates that @xmath128 is independent of @xmath45 . we will refer to this as the assumption of independence , and denote @xmath128 by @xmath131 when the assumption is valid . we refer to @xmath131 as the distillability factor ( df ) . we have , therefore , the following proposition under the assumption of independence . * proposition iii . * _ the probability of an arbitrary entangled state , pure or mixed , of a two - qudit system defined by the hamiltonian @xmath78 for a specific value of @xmath86 to be scd is given by _ @xmath132 note here that none of the figures presented in this paper uses this assumption for the numerical calculations . however , we expect that the formula in proposition ii will be useful in cases where numerical simulation is more challenging . model . ( a ) the probability distribution , @xmath44 , of average energy , @xmath45 , for haar uniformly chosen random mixed states of different ranks with @xmath133 , @xmath47 . the distribution becomes sharply peaked at @xmath48 when the rank of the states increase . ( b ) the variation of the probability @xmath49 as a function of @xmath46 at @xmath47 for states of different ranks . other considerations are the same as in fig . [ xypure ] . ] in the case of @xmath134 , the spin operators are the pauli matrices , @xmath135 , acting on the qubit @xmath136 . following proposition i , there is a vanishing probability that a randomly chosen two - qubit state of a system described by a non - interacting two - qubit hamiltonian of the form @xmath137 with @xmath138 is scd . now we consider some examples of well - known two - qubit interacting hamiltonians and investigate the canonical distillability of entangled states of two - qubit systems described by these hamiltonians . we consider three well - known spin hamiltonians , namely , ( 1 ) the @xmath2 model in a transverse field @xcite , ( 2 ) the @xmath2 model in a longitudinal field , and ( 3 ) the @xmath3 model in an external field @xcite . in the case of @xmath2 model in a transverse field . the white line depicts the boundary between thermal states that do and do not satisfy the wcec on the @xmath139 plane . below the boundary , the constraint is satisfied while above the boundary , it is not . the different shades in the figures represent different values of entanglement as measured by concurrence as a function of @xmath46 and @xmath140 . all quantities are dimensionless , except the concurrence , which is in ebits . ] the two - qubit hamiltonian describing the @xmath2 model in a transverse field is given by @xcite @xmath141 + h\sum_{i=1}^{2}\sigma_{i}^{z } , \label{xy - ham}\end{aligned}\ ] ] where @xmath142 is proportional to the interaction strength , @xmath50 is the anisotropy parameter , and @xmath143 is the strength of the external transverse magnetic field . note that @xmath144 can be written as a sum of an interacting and a local hamiltonian as @xmath145 where @xmath146 ; h_{xy}^{l}=\sum_{i=1}^{2}\sigma_{i}^{z}. \nonumber \\\end{aligned}\ ] ] the average energy of an arbitrary two - qubit state , @xmath8 , in the present case , has lower and upper bounds determined by the minimum and maximum eigenvalues of the hamiltonian @xmath144 . we find that , for a given value of @xmath46 , @xmath147 , where @xmath148 for @xmath149 , and @xmath150 for @xmath151 . here and in the rest of the paper , the calculations of energy are always performed after a division of the hamiltonian by the coupling constant , @xmath142 , so that the ensuing energy expressions are dimensionless . from wcec and using eq . ( [ loc - zero ] ) , we have @xmath152 @xmath119 @xmath153 , which has lower and upper bounds given by the minimum and maximum eigenvalues of @xmath154 . in the present case , @xmath155 , where @xmath156 for @xmath149 , and @xmath157 for @xmath158 . the local unitary operator @xmath159 , involved in obtaining @xmath160 for the qubit @xmath161 ( @xmath41 ) can be parametrized as @xmath162 since @xmath153 is a continuous function of @xmath163 , the parameters of unitary operators with @xmath164 one can always find at least one unitary operator @xmath165 for every value of @xmath153 in the allowed range @xmath166 $ ] . for example , for @xmath167 , choosing @xmath168 , and @xmath169 results in @xmath170 , which can exhaust the range @xmath171 $ ] when @xmath172 and @xmath173 are varied . besides , a choice of @xmath174 , @xmath175 gives @xmath176 by which the range @xmath177 $ ] can be exhausted in a similar fashion . we therefore find that the inclusion of the interaction terms has a drastic effect on canonical distillability . while almost no states are scd for @xmath178 ( proposition i ) , a significant fraction of them are so for @xmath179 . we have performed extensive numerical simulations to check the assumption of independence , and have found it to hold , within numerical accuracy , for the transverse @xmath2 hamiltonian . therefore , via proposition ii , the probability of an arbitrary two - qubit entangled state to be scd with respect to @xmath144 is given by @xmath180 . , where the system is governed by the two - qubit @xmath2 hamiltonian . the shaded regions depict the projections of the volume of physical states on the different cross - sections . the top panels consider projections onto the @xmath181 plane , while the bottom panels consider that onto the @xmath182 plane . on the other hand , the left panels are for @xmath183 , while the right panels are for @xmath184 . the black regions depict the states that are scd , while the dark gray regions depict the states that are not scd but distillable in the usual , non - canonical , sense . the light gray regions indicate undistillable quantum states . note that the region of scd states diminishes with the increase in the value of the external field . this is consistent with proposition i and the fact that the probability , @xmath49 , of a state being scd decreases with increasing external field ( see fig . [ xymixed ] ) . also , an scd state can be obtained by mixing two quantum state that are distillable , but not scd , as indicated by the dotted line in the bottom right panel , implying the non - convexity of the scd states . ] irrespective of the value of @xmath143 , in the limit @xmath185 , the probability of an arbitrary entangled state of a two - qubit system described by the hamiltonian @xmath144 to be scd is unity . this can be understood by noting that @xmath186 as @xmath185 . also , note that for @xmath149 , @xmath187 when @xmath188 . since @xmath156 for @xmath149 , we conclude that @xmath189 in the limit @xmath190 for @xmath149 . on the other hand , for @xmath151 , @xmath191 when @xmath188 . and @xmath157 for @xmath151 . this leads to the following proposition . + * proposition iv . * in a two - qubit system described by the transverse @xmath2 hamiltonian , all entangled states are scd , provided either @xmath192 or @xmath193 . the probability distributions , @xmath44 , of average energy of the state over the space of all two - qubit pure states for different values of @xmath46 are depicted in fig . [ xypure](a ) . the variation of the probability @xmath49 with the field - strength @xmath143 in the case of pure states of the two - qubit @xmath2 model is depicted in fig . [ xypure](b ) , for different values of @xmath50 . note that the value of @xmath49 decreases as @xmath46 increases , and asymptotically vanishes as @xmath194 . this can be understood as a result of proposition i. for a fixed @xmath46 , @xmath49 is a non - monotonic function of @xmath50 with a change in character at @xmath47 , the ising point . for @xmath195 , @xmath49 decreases with increasing @xmath50 for a fixed value of @xmath143 while for @xmath167 , @xmath49 increases as @xmath50 increases , as clearly seen from fig . [ xypure](b ) . , for pure states , as functions of @xmath46 in the transverse- and longitudinal - field ising models . the rate of decay of @xmath49 changes abruptly at @xmath133 in the case of the longitudinal field . the inset exhibits the difference between the probabilities for the two models . see text ( sec . [ longitudexy ] ) for details . all quantities plotted are dimensionless . ] in the case of two - qubit mixed states , we find that the df , @xmath131 , is a function of the rank , @xmath125 , of the state , with @xmath196 , @xmath197 , and @xmath198 . these estimates are obtained via numerical simulation , for which we have generated @xmath199 states ( for each rank ) haar uniformly over the space of quantum states of the corresponding rank . the probability distributions , @xmath44 , of average energy of the two - qubit entangled mixed states with different ranks are shown in fig . [ xymixed](a ) for @xmath133 , @xmath47 . as in the case of pure states , the probability distributions are bell - shaped , and becomes sharply peaked around zero for states with higher ranks . [ xymixed](b ) depicts the variation of @xmath49 against @xmath143 for mixed states of rank @xmath69 , @xmath200 , and @xmath201 with @xmath47 . similar to the case of pure states , the probability @xmath202 as @xmath194 . let us now discuss the canonical distillability of some special types of mixed states . * _ thermal states. _ * we intend to find out whether canonical distillability of a two - qubit mixed entangled state depends on temperature . to pursue this question , we construct the thermal state of the two - qubit system as @xmath203 , where @xmath204 is the partition function of the system and @xmath205 , @xmath206 and @xmath207 being the boltzmann constant and absolute temperature respectively . fig . [ xythermal ] depicts the canonical distillation phase diagram on the @xmath139 plane for different values of @xmath50 in case of the @xmath2 model in a transverse field . for every value of @xmath143 , there is a critical value of @xmath207 above which the state satisfies the wcec ( eq . ( [ cdeq ] ) ) , whereas below the critical value , it does not . we call this value as the _ scd temperature_. note that in fig . [ xythermal ] , the vertical axis is proportional to @xmath208 , i.e. , proportional to inverse of @xmath207 . for the zero temperature cases , we find that increase of entanglement tends to facilitate canonical distillability . however , after thermal mixing , there is a trade - off between temperature and entanglement , which aids in canonical distillability at higher temperatures where entanglement is typically low . the entanglement of the state , as measured by concurrence @xcite , is mapped onto the @xmath139 plane and is represented by different shades in fig . [ xythermal ] . note that for low values of @xmath46 , the thermal states have low scd temperatures , whereas the trend is opposite for higher values of @xmath46 . the figure clearly shows that there exist thermal states that are not entangled , but satisfies eq . ( [ cdeq ] ) . similarly , thermal entangled states exist which do not satisfy eq . ( [ cdeq ] ) and hence can not be scd . , for pure states , as a function of @xmath46 for different values of @xmath209 in the case of two - qubit @xmath3 model in an external field . all quantities employed are dimensionless . ] * _ bell - diagonal states. _ * next , we explore the canonical distillability of bell - diagonal ( bd ) states given by @xmath210 , with @xmath211 , where @xmath212 are the diagonal elements of the correlation matrix ( @xmath213 ) , and where @xmath214 and @xmath215 are the identity matrices in the hilbert spaces of the qubits @xmath4 and @xmath69 , respectively . the positivity of the bd state dictates that the correlators , @xmath216 , are constrained to vary within a strict subset of the hypercube . the average energy of the state is given by @xmath217 . hence , @xmath218 @xmath219 @xmath220 $ ] for @xmath149 , and @xmath218 @xmath219 @xmath221 $ ] when @xmath158 . since the range of the left - hand side of eq . ( [ cdeq ] ) coincides with that of the right - hand side , the bd states are always scd for the transverse @xmath2 model . * _ mixed states with fixed magnetizations. _ * introduction of magnetization in the @xmath62-direction to the state @xmath222 makes the two - qubit mixed state to be of the form @xmath223 . here , @xmath224 and @xmath225 represent the magnetizations of the qubits @xmath4 and @xmath69 , respectively , with @xmath226 , @xmath227 , @xmath228 being the local density matrix of the qubit @xmath136 . similar to the correlators @xmath229 , @xmath230 . the average energy of this state is given by @xmath231 . the canonical distillability of the state depends explicitly on the values of the correlators and the magnetizations . [ bellvolume ] depicts the projections of the volume in the @xmath232 and @xmath233 planes , in which all the states of the form @xmath234 are scd ( red regions ) for fixed values of the other parameters ( @xmath235 , @xmath236 ) . with increasing @xmath46 , the region shrinks , thereby indicating a decrease in the probability that a mixed state of the form @xmath234 is distillable and scd this is consistent with our earlier findings regarding the dependence of special canonical distillability of mixed entangled states on @xmath46 . in the case of @xmath3 model in an external field . all other considerations are the same as in fig . [ xythermal ] . ] to investigate whether a change in the direction of the external field alters the probability of a state to be scd , we consider the two - qubit @xmath2 model in a longitudinal field , described by the hamiltonian @xmath237 + h\sum_{i=1}^{2}\sigma_{i}^{x},\nonumber \\ \label{xyl - ham}\end{aligned}\ ] ] where @xmath143 is the strength of the longitudinal field . note that we are using the same symbol @xmath143 for the longitudinal field that we had used in the preceding case for the transverse field . the probability @xmath49 , in the case of the longitudinal - field ising model @xmath238 , is plotted against @xmath46 in fig . [ longxy ] . note that the rate of decay of @xmath49 with @xmath143 changes abruptly at @xmath133 . this can be understood by noting that the average energy of an entangled state of a two - qubit system described by the hamiltonian @xmath239 for @xmath47 is bounded below and above by the minimum and maximum eigenvalues of the ising hamiltonian , respectively . they are given by @xmath240 , @xmath241 for @xmath242 , and @xmath243 , @xmath241 for @xmath244 . due to the energy level crossing at @xmath245 , the allowed range of values for the average energy changes whereas the range of @xmath246 remains fixed at @xmath220 $ ] irrespective of the values of @xmath247 , thereby changing variation of @xmath49 abruptly . in fig . [ longxy ] , we compare the result of longitudinal - field ising model with that obtained from the transverse - field ising model where the same field strength is applied in the transverse direction . we observe that over a certain interval of the field values , the probability is greater in the case of the longitudinal field than that in the case of the transverse field , while the opposite is true in the complementary region . this clearly indicates that the probability depends on the direction of the applied field . for comparison between the longitudinal and transverse models , we introduce the quantity @xmath248 , and plot it as a function of @xmath46 and @xmath50 in the inset of fig . [ longxy ] . the existence of both positive and negative values of @xmath249 reveals the interesting fact that @xmath49 can be increased or decreased by changing only the direction of the field . , for pure states , as a function of the interaction parameter @xmath250 and the field - strength @xmath46 for the two - qutrit model described by the hamiltonian @xmath251 . a general two - qutrit state is less possible to be scd with the increase of @xmath46 for a fixed value of @xmath250 . all quantities used are dimensionless . ] the two - qubit @xmath3 model in an external field is described by the hamiltonian @xcite @xmath252 where @xmath142 is proportional to the interaction strength , @xmath209 is the anisotropy in @xmath62 direction , and @xmath143 is the strength of the external field . the average energy of any entangled state of a two - qubit system defined by the hamiltonian @xmath253 must be in the range @xmath254 $ ] , where @xmath255 and @xmath256 are the minimum and maximum eigenvalues , respectively , of @xmath253 , with @xmath255 , @xmath256 @xmath219 @xmath257 . the choice of @xmath258 and @xmath259 depends on the values of @xmath209 and @xmath46 . the right hand side of the wcec ( eq . ( [ cdeq ] ) ) lies in the range @xmath96 $ ] , where @xmath260 and @xmath261 are the minimum and maximum eigenvalues , respectively , of the interacting part of the @xmath3 hamiltonian given by @xmath262 . for all values of @xmath209 , @xmath263 whereas @xmath264 for @xmath265 , and @xmath266 for @xmath267 . the probability of a two - qubit state being scd can be obtained following proposition ii . note that with unitary operators of the form ( [ unitary ] ) , one can obtain all the values of @xmath268 in the allowed range @xmath269 $ ] since @xmath270 is a continuous function of the parameters @xmath271 , @xmath272 . [ xxzpure ] represents the decay of the probability @xmath49 as a function of @xmath46 for different values of @xmath209 in the case of pure states . note that the dependence of @xmath49 on @xmath209 , similar to that of @xmath49 on @xmath50 in the case of the longitudinal - field @xmath2 model , is non - monotonic for a fixed value of @xmath46 . when @xmath273 , @xmath49 decreases with increasing @xmath209 for a fixed @xmath46 whereas for @xmath274 , the opposite trend is observed . also , the decay rate of @xmath49 with @xmath46 changes abruptly for @xmath267 due to the ground state energy level crossing that changes the limits of the distribution @xmath44 , similar to the case of the @xmath2 model in a longitudinal field . in the case of two - qubit mixed states , one can find features similar to that in the case of transverse - field @xmath2 hamiltonian ( as depicted in fig . [ xymixed ] ) using the same methodology . we conclude the discussion on the @xmath3 model by examining the canonical distillability of the thermal states constructed from the eigen - spectrum of the model . we find that similar to the transverse - field @xmath2 model , a critical temperature exists for all values of @xmath46 , for a fixed @xmath209 , above which the thermal state satisfies eq . ( [ cdeq ] ) , but below which it does not . ( see fig . [ xxzthermal ] ) . the zero - entanglement regions above the white line indicates that there exist thermal states that satisfy eq . ( [ cdeq ] ) , but are non - distillable . , of the average energy , @xmath45 , in the case of three - qubit states belonging to the ghz and w classes . the hamiltonian parameters are set to @xmath47 and @xmath133 . ( b ) the variation of the probability @xmath49 as a function of @xmath46 in the ghz and w classes . all quantities employed are dimensionless . ] we conclude the discussion on canonical distillability of bipartite systems with an example of a two - qutrit system defined by a bilinear - biquadratic hamiltonian @xcite in the presence of a field term . the hamiltonian is given by @xmath275 + h\sum_{i=1}^{2}s_{i}^{z},\end{aligned}\ ] ] where @xmath276 , @xmath227 , are the spin operators on the qutrit @xmath136 , with @xmath277 here , @xmath278 and @xmath279 are the relative strengths of the bilinear and biquadratic interactions , respectively . the probability that a randomly chosen two - qutrit state @xmath8 is scd is determined by an equation similar to eq . ( [ cdeq ] ) where @xmath10 is replaced by @xmath33 ( eq . ( [ qudit_state ] ) ) with @xmath280 . the left hand side of the equation represents the average energy of the two - qutrit state which is bounded below and above by the minimum and the maximum eigenenergy of the hamiltonian , given by @xmath281 and @xmath282 , respectively . here , @xmath283 and @xmath284 where @xmath285 is the set of eigenvalues of the hamiltonian , the maximum and the minimum being determined by the values of @xmath46 and @xmath250 . the probability of a randomly chosen two - qutrit pure state to be scd is plotted on the @xmath286 plane in fig . [ qutrit ] . the probability decreases as @xmath46 increases for a fixed value of @xmath250 , as consistent with proposition i. note that at @xmath287 , the probability @xmath49 , unlike in the case of the two - qubit pure states , is not always unity but depends on @xmath250 and therefore on the strengths of the bilinear and biquadratic interactions . canonical distillability of a multipartite system constituted of @xmath51 parties and described by a hamiltonian @xmath18 can be investigated using the same methodology as that used in the case of bipartite systems . for the purpose of demonstration , we restrict ourselves to three - qubit states belonging to the well - known ghz and w classes @xcite , which are mutually disjoint sets that collectively exhaust the entire set of three - qubit pure states . starting with three - qubit states chosen from the ghz class as resource states , we consider the distillation of the multiparty entangled three - qubit ghz state , @xmath288 , using the wcec . similarly , while considering the w class of states as the resource states , the target state is the three - qubit w state , @xmath289 . let the multiqubit system be described by the @xmath2 hamiltonian in an external transverse field , given by @xmath290 + h\sum_{i=1}^{n}\sigma_{i}^{z}.\nonumber \\ \label{xy_ham_n}\end{aligned}\ ] ] a general three - qubit state belonging to ghz class is given by @xcite @xmath291 where @xmath292 is a normalization constant , and @xmath293 , @xmath294 , with @xmath295 . let us consider the case of @xmath47 ( transverse ising model ) . one side of the wcec in this case consists of the average energy of the three - qubit states in the ghz class and has the minimum and maximum allowed values , @xmath296 and @xmath297 , as the minimum and maximum eigenvalues , respectively , of @xmath298 . choosing the standard state on the other side of the wcec as @xmath299 , we find that the part @xmath300 of the hamiltonian @xmath298 does not contribute and the bounds on this side are decided by the minimum and maximum eigenvalues of @xmath301 $ ] : @xmath302 where @xmath303 is the state @xmath299 up to local unitary operators . with the choice of unitary operators as @xmath304 one obtains @xmath305 , a continuous function , using which one can exhaust all possible real values in the allowed range @xmath306 $ ] by varying @xmath250 . therefore , the probability that a three - qubit state of the ghz class is scd to the three - qubit ghz state can be obtained by determining the normalized probability distribution @xmath44 for the average energy @xmath45 of the ghz states and integrating it within proper limits : @xmath307 the df is unity by virtue of ref . the probability distribution @xmath44 in the case of the three - qubit transverse ising model and the ghz class of states is shown in fig . [ three](a ) . [ three](b ) shows the variation of the probability @xmath49 as a decreasing function of @xmath247 . a three - qubit state belonging to the w class is given by @xcite @xmath308 with @xmath309 . we intend to canonically distill three - qubit w states , @xmath310 , where the system is described by the hamiltonian @xmath298 . similar to the ghz class , we consider @xmath47 . both sides of the wcec in this case are bounded by @xmath296 and @xmath297 . however , we find that the effective range is a strict subset of @xmath311 $ ] . denoting the ends of that subset by @xmath312 and @xmath313 , the probability @xmath49 , in the present case , is given by @xmath314 , where the probability distribution @xmath44 is exhibited in fig . [ three](a ) . the df is again unity by virtue of ref . the variation of @xmath49 as a function of @xmath143 is shown in fig . [ three](b ) . note that , in contrast to the case of ghz class , up to @xmath133 , we have @xmath315 . when @xmath316 , @xmath49 decreases with increasing @xmath46 . this is therefore an occasion where it is advantageous to create w - class states than ghz - class states ( cf . distillation of entanglement from shared quantum states is a useful technique for several quantum information protocols . it is important , for both fundamental and practical reasons , to consider this protocol in a scenario of limited resources . we have considered distillation of entanglement in bipartite and multipartite quantum states in the situation where there is a limited amount of energy that is accessible for the process to be accomplished . in particular , we consider constraints on average energy : a canonical energy constraint and a weak canonical energy constraint , that naturally lead to the concepts of canonical distillability and special canonical distillability . we have shown that for a bipartite system described by a non - interacting hamiltonian , almost no states are special canonically distillable . this , in turn , implies that almost no states are canonically distillable , as canonically distillable states form a subset of the set of special canonically distillable states . significant understanding about the set of special canonically distillable states can be obtained by looking at the probability distributions of the average energies of the shared states . the concept has been applied to a number of bipartite and multipartite systems described by well - known spin hamiltonians , namely , the spin-@xmath1 @xmath2 model in transverse and longitudinal fields , the spin-@xmath1 @xmath3 model in an external field , and the bilinear - biquadratic hamiltonian of a spin-@xmath4 system in the presence of an external field . we find that the probability that a randomly chosen state is special canonically distillable can be manipulated by altering the direction of the external field . the canonical distillability of a number of mixed states such as the thermal state , the bell - diagonal states , and mixed bipartite states with fixed magnetizations are also investigated . it has been shown that for a fixed external field value , the thermal state can be special canonical distillable only above a critical temperature , which we have called the special canonically distillable temperature . the concept of canonical distillability of three - qubit ghz and w class states have also been introduced . the results are expected to be of importance in realization of quantum communication channels . 100 r. horodecki , p. horodecki , m. horodecki , and k. horodecki , rev . phys . * 81 * , 865 ( 2009 ) , and the references therein . h. j. briegel , d. e. browne , w. dr , r. raussendorf , and m. van den nest , nature physics * 5 * , 19 - 26 ( 2009 ) . a. sen(de ) and u. sen , physics news * 40 * , 17 - 32 ( 2010 ) ( arxiv:1105.2412 [ quant - ph ] ) . a. acn , n. brunner , n. gisin , s. massar , s. pironio , and v. scarani , phys . 98 , 230501 ( 2007 ) ; s. pironio , a. acn , n. brunner , n. gisin , s. massar , and v. scarani , new j. phys . * 11 * , 045021 ( 2009 ) ; n. gisin , s. pironio , and n. sangouard , phys . 105 , 070501 ( 2010 ) ; l. masanes , s. pironio , and a. acn , nature commun . * 2 * , 238 ( 2011 ) ; h. k. lo , m. curty , and b. qi , phys . 108 , 130503 ( 2012 ) ; c. h. bennett , g. brassard , s. popescu , b. schumacher , j. a. smolin , and w. k. wootters , phys . rev . lett . * 76 * , 722 ( 1996 ) ; m. horodecki , p. horodecki , and r. horodecki , phys . rev . lett . * 78 * , 574 ( 1997 ) ; m. murao , m. b. plenio , s. popescu , v. vedral , and p. l. knight , phys . rev . a * 57 * , r4075 ( 1998 ) ; w. dr , j. i. cirac , and r. tarrach , phys . . lett . * 83 * , 3562 ( 1999 ) ; m. horodecki and p. horodecki , phys . rev . a * 59 * , 4206 ( 1999 ) ; w. dr and j. i. cirac , phys . rev . a * 61 * , 042314 ( 2000 ) ; w. dr , j. i. cirac , m. lewenstein , and d. bru , phys . rev . a * 61 * , 062313 ( 2000 ) ; g. alber , a. delgado , n. gisin , and i. jex , j. phys . a. * 34 * , 8821 ( 2001 ) ; j. dehaene , m. van den nest , b. d. moor , and f. vestraete , phys . rev . a * 67 * , 022310 ( 2003 ) ; w. dr , h. aschaur , and h. j. briegel , phys . rev . lett . * 91 * , 107903 ( 2003 ) ; e. hostens , j. dehaene , and b. d. moor , arxiv : quant - ph/0406017 ( 2004 ) ; k. g. h. vollbrecht and f. vestraete , phys . a * 71 * , 062325 ( 2005 ) ; a. miyake and h. j. briegel , phys . rev . 95 * , 220501 ( 2005 ) ; h. aschaur , w. dr , and h. j. briegel , phys . rev . a * 71 * , 012319 ( 2005 ) ; e. hostens , j. dehaene , and b. d. moor , phys . a * 73 * , 042316 ( 2006 ) ; e. hostens , j. dehaene , and b. d. moor , phys . a * 74 * , 062318 ( 2006 ) ; a. kay , j. k. pachos , w. dr , and h. j. briegel , new . j. phys . * 8 * , 147 ( 2006 ) ; c. kruszynska , a. miyake , h. j. briegel , and w. dr , phys . a * 74 * , 052316 ( 2006 ) ; s. glancy , e. knill , and h. m. vasconcelos , phys . a * 74 * , 032319 ( 2006 ) . p. horodecki , phys . a * 232 * , 333 ( 1997 ) ; m. horodecki , p. horodecki , and r. horodecki , phys . rev . lett . * 80 * , 5239 ( 1998 ) ; w. dr and j. i. cirac , phys . a * 62 * , 022302 ( 2000 ) ; p. horodecki , j. a. smolin , b. m. terhal , and a. v. thapliyal , theor . comput . sc . * 292 * , 589 ( 2003 ) ; j. lavoie , r. kaltenbaek , m. piani , and k. j. resch , nature physics * 6 * , 827 ( 2010 ) ; j. lavoie , r. kaltenbaek , m. piani , and k. j. resch , phys . lett . * 105 * , 130501 ( 2010 ) ; j. diguglielmo , a. samblowski , b. hage , c. pineda , j. eisert , and r. schnabel , phys . rev . 107 , 240503 ( 2011 ) ; e. amselem , m. sadiq , and m. bourennane , scientific reports * 3 * , 1966 ( 2013 ) . d. p. divincenzo , p. w. shor , j. a. smolin , b. m. terhal , and a. v. thapliyal , phys . a * 61 * , 062312 ( 2000 ) ; b. kraus , m. lewenstein , and j. i. cirac , phys . a * 65 * , 042327 ( 2002 ) ; j. watrous , phys . lett . * 93 * , 010502 ( 2004 ) . a. ekert , phys . lett . * 67 * , 661 ( 1991 ) ; t. jennewein , c. simon , g. weihs , h. weinfurter , and a. zeilinger , phys . lett . * 84 * , 4729 ( 2000 ) ; d. s. naik , c. g. peterson , a. g. white , a. j. berglund , and p. g. kwiat , phys . rev . * 84 * , 4733 ( 2000 ) ; w. tittel , t. brendel , h. zbinden , and n. gisin , phys . rev . lett . * 84 * , 4737 ( 2000 ) ; n. gisin , g. ribordy , w. tittel , and h. zbinden , rev . * 74 * , 145 ( 2002 ) . c. h. bennet and s. j. wiesner , phys * 69 * , 2881 ( 1992 ) ; k. mattle , h. weinfurter , p. g. kwiat , and a. zeilinger , phys . * 76 * , 4656 ( 1996 ) . c. h. bennett , g. brassard , c. crpeau , r. jozsa , a. peres , and w. k. wootters , phys . lett . * 70 * , 1895 ( 1993 ) ; d. bouwmeester , j. w. pan , k. mattle , m. eibl , h. weinfurter , and a. zeilinger , nature * 390 * , 575 ( 1997 ) ; j. w. pan , d. bouwmeester , h. weinfurter , and a. zeilinger , phys . lett . * 80 * , 3891 ( 1998 ) ; d. bouwmeester , j. w. pan , h. weinfurter , and a. zeilinger , j. mod . opt . * 47 * , 279 ( 2000 ) . j. briegel , w. dr , j. i. cirac , and p. zoller , phys . * 81 * , 5932 ( 1998 ) . h. bombin and m. a. martin - delgado , phys . rev . lett . * 97 * , 180501 ( 2006 ) . p. g. kwiat , s. barraza - lopez , a. stefanov , and n. gisin , nature * 409 * , 1014 ( 2001 ) ; j. w. pan , c. simon , c. brukner , and a. zeilinger , nature * 410 * , 1067 ( 2001 ) ; j. w. pan , s. gasparoni , r. ursin , g. weihs , and a. zeilinger , nature * 423 * , 417 ( 2003 ) . p. horodecki , r. horodecki , and m. horodecki , acta physica slovaca * 48 * , 141 ( 1998 ) . p. m. hayden , m. horodecki , and b. m. terhal , j. phys . a * 34 * , 6891 ( 2001 ) ; m. b. plenio and s. virmani , quant . * 7 * , 1 ( 2006 ) . s. hill and w. k. wootters , phys . lett . * 78 * , 5022 ( 1997 ) ; w. k. wootters , phys . lett . * 80 * , 2245 ( 1998 ) ; v. coffman , j. kundu , and w. k. wootters , phys . a * 61 * , 052306 ( 2000 ) ; w. k. wootters , quant . * 1 * , 27 ( 2001 ) . j. p. gordon , in proceedings of the international school of physics , `` enrico fermi , course xxxi , '' edited by p. a. miles ( academic press , new york , 1964 ) , p. 156 ; l. b. levitin , in proceedings of the vi national conference on inf . theory , p. 111 ( tashkent 1969 ) . h. p. yuen , in quantum communication , computing , and measurement , edited by o. hirota et al . , ( plenum , new york , 1997 ) . a. s. holevo , probl . peredachi inf . * 9 * , 3 ( 1973)[probl . inf . transm . 9 , 110 ( 1973 ) ] ; b. schumacher , m. westmoreland , and w. k. wootters , phys . * 76 * , 3452 ( 1996 ) ; p. badziag , m. horodecki , a. sen(de ) , and u. sen , phys . rev.lett . * 91 * , 117901 ( 2003 ) ; m. horodecki , j. oppenheim , a. sen(de ) , and u. sen , _ ibid_. * 93 * , 170503 ( 2004 ) . h. p. yuen and m. ozawa , phys . lett . * 70 * , 363 ( 1993 ) ; c. m. caves and p. d. drummond , rev 66 * , 481 ( 1994 ) ; a. s. holevo , m. sohma , and o. hirota , phys . rev . a * 59 * , 1820 ( 1999 ) ; a. s. holevo and r. f. werner , ibid . * 63 * , 032312 ( 2001 ) ; s. lloyd , phys . rev . lett . * 90 * , 167902 ( 2003 ) ; v. giovannetti , s. lloyd , l. maccone , and p. w. shor , ibid . * 91 * , 047901 ( 2003 ) ; v. giovannetti , s. lloyd , and l. maccone , ibid . * 70 * , 012307 ( 2004 ) ; v. giovannetti , s. guha , s. lloyd , l. maccone , and j. h. shapiro , ibid . * 70 * , 032315 ( 2004 ) ; v. giovannetti , s. guha , s. lloyd , l. maccone , j. h. shapiro , and h.p . yuen , ibid . * 92 * , 027902 ( 2004 ) ; a. sen(de ) , u. sen , b. gromek , d. bru , and m. lewenstein , phys . lett . * 95 * , 260503 ( 2005 ) ; a. sen(de ) , u. sen , b. gromek , d. bru , and m. lewenstein , phys . a * 75 * , 022331 ( 2007 ) ; m. m. wolf , d. prez - garca , and g. giedke , phys . 98 , 130501 ( 2007 ) ; c. lupo , v. giovannetti , and s. mancini , phys . rev . lett . 104 , 030501 ( 2010 ) ; v. giovannetti , r. garca - patrn , n. j. cerf , and a. s. holevo , nature photonics * 8 * , 796 ( 2014 ) . e. lieb , t. schultz , and d. mattis , ann . * 16 * , 407 ( 1961 ) ; e. barouch , b.m . mccoy , and m. dresden , phys . rev . * 2 * , 1075 ( 1970 ) ; e. barouch and b.m . mccoy , phys . rev . * 3 * , 786 ( 1971 ) . b. k. chakrabarti , a. dutta , and p. sen , quantum ising phases and transitions in transverse ising models ( springer , heidelberg , 1996 ) ; s. sachdev , quantum phase transitions ( cambridge university press , cambridge , 2011 ) ; s. suzuki , j. -i . inou , b. k. chakrabarti , quantum ising phases and transitions in transverse ising models ( springer , heidelberg , 2013 ) . c. n. yang and c. p. yang , phys . rev . * 150 * , 321 ( 1966 ) ; c. n. yang and c. p. yang , _ ibid . _ * 150 * , 327 ( 1966 ) . e. demler and f. zhou , phys . lett . * 88 * , 163001 ( 2002 ) ; s. k. yip , phys . . lett . * 90 * , 250402 ( 2003 ) ; o. romero - isart , k. eckert , and a. sanpera , phys . rev . a * 75 * , 050303(r ) ( 2007 ) . d. m. greenberger , m. a. horne , and a. zeilinger , in bell s theorem , quantum theory , and conceptions of the universe , ed . m. kafatos ( kluwer academic , dordrecht , the netherlands , 1989 ) . a. zeilinger , m. a. horne , and d. m. greenberger , in proceedings of squeezed states and quantum uncertainty , edited by d. han , y. s. kim , and w. w. zachary , nasa conf . publ . * 3135 * , 73 ( 1992 ) . w. dr , g. vidal , j. i. cirac , phys . a * 62 * , 062314 ( 2000 ) . p. w. shor , j. a. smolin , and b. m. terhal , phys . lett . * 86 * , 2681 ( 2001 ) . a. peres , phys . lett . * 77 * , 1413 ( 1996 ) ; m. horodecki , p. horodecki , and r. horodecki , phys . a * 223 * , 1 ( 1996 ) . a. sen(de ) , u. sen , m. wiesniak , d. kaszlikowski , and m. ukowski , phys . a , * 68 * , 062306 ( 2003 ) .

we calculate explicit speeds for random walks in uniform degenerate random environments . for certain non - uniform random environments , we calculate speeds that are non - monotone . # 1 # 1 # 1 # 1

1304.7520

in @xcite the authors study random walk in an iid random environment , where the environment need not satisfy any ellipticity condition . in other words , where various nearest neighbour transitions may have quenched probability @xmath0 . if such a walk can get stuck on a finite set of vertices with positive probability , then it will get stuck with probability one . there are necessary and sufficient conditions for such a walk not to get stuck in this way , and @xcite studies transience and speed questions for such walks . there are many interesting models in which such properties are non - trivial . there are also examples in which transience is essentially trivial , but in which speeds can be calculated explicitly , because of a renewal structure . @xcite gives a table of such speeds , for random walks in _ uniform _ degenerate random environments . that is , environments in which the walker chooses at random from the ( random ) set of allowed steps . the purpose of this note is to supply details for the latter calculations . we also include details of some calculations , for speeds and other quantities , related to examples of non - monotone behaviour . for fixed @xmath1 let @xmath2 be the set of unit vectors in @xmath3 . let @xmath4 denote the set of probability measures on @xmath5 , and let @xmath6 be a probability measure on @xmath7 . let @xmath8 be equipped with the product measure @xmath9 ( and the corresponding product @xmath10-algebra ) . a random environment @xmath11 is an element of @xmath12 . we write @xmath13 for @xmath14 . note that @xmath15 are i.i.d . with law @xmath6 under @xmath16 . the random walk in environment @xmath17 is a time - homogeneous markov chain with transition probabilities from @xmath18 to @xmath19 defined by @xmath20 given an environment @xmath17 , we let @xmath21 denote the law of this random walk @xmath22 , starting at the origin . let @xmath23 denote the law of the annealed random walk , i.e. @xmath24 . since @xmath25 $ ] and @xmath26 , @xmath27 if and only if @xmath28 for @xmath16-almost every @xmath17 . similarly @xmath29 if and only if @xmath30 for @xmath16-almost every @xmath17 . if we start the rwre at @xmath31 instead , we write @xmath32 for the corresponding probability , so @xmath33 . we associate to each environment @xmath17 a directed graph @xmath34 ( with vertex set @xmath3 ) as follows . for each @xmath35 , the directed edge @xmath36 is in @xmath37 if and only if @xmath38 , and the edge set of @xmath34 is @xmath39 . for convenience we will also write @xmath40 . note that under @xmath16 , @xmath41 are i.i.d . subsets of @xmath5 . the graph @xmath34 is equivalent to the entire graph @xmath3 , precisely when the environment is _ elliptic _ , i.e. @xmath42 for each @xmath43 . much of the current literature assumes either the latter condition , or the stronger property of _ uniform ellipticity _ , ie that @xmath44 such that @xmath45 for each @xmath43 . on the other hand , given a directed graph @xmath40 ( with vertex set @xmath3 , and such that @xmath46 for each @xmath18 ) , we can define a _ uniform _ random environment @xmath47 . let @xmath48 denote the cardinality of @xmath49 , and set @xmath50 the corresponding rwre then moves by choosing uniformly from available steps at its current location . this gives us a way of constructing rather nice and natural examples of random walks in non - elliptic random environments : first generate a random directed graph @xmath40 where @xmath37 are i.i.d . , then run a random walk on the resulting random graph ( choosing uniformly from available steps ) . [ def:2-val ] we say that the environment is _ @xmath51-valued _ when @xmath6 charges exactly two points , i.e. there exist @xmath52 and @xmath53 such that @xmath54 , @xmath55 . we say that the graph is _ @xmath51-valued _ when there exist @xmath56 and @xmath57 such that @xmath58 and @xmath59 . @xcite proves that the following simple criterion is equivalent to the statement that the random walk visits infinitely many sites almost surely . @xmath60 the following is stated in ( * ? ? ? * lemma 5.1 ) : [ lem : trivialtransience ] assume ( [ unstuckcondition ] ) and suppose that @xmath61 but @xmath62 . then the rwre is transient in direction @xmath63 , @xmath23-almost surely . let @xmath64 be the first time the rwre follows direction @xmath63 . if @xmath65<\infty$ ] then @xmath22 has an asymptotic velocity @xmath66 } , \dots , v^{[d]})$ ] , in the sense that @xmath67 . moreover , @xmath68}=e[x^{[i]}_t]/e[t].$ ] the random walk visits infinitely many sites , and at each visit to a new site there is positive ( non - vanishing ) probability of then taking a step in direction @xmath63 . thus the second coordinate of the random walk converges monotonically to @xmath69 . let @xmath70 be the @xmath71th time that @xmath22 moves in direction @xmath63 , and @xmath72 . let @xmath73 . since the environment seen by the random walker is refreshed at every time @xmath70 , the @xmath74 are iid , and the @xmath70 are sums of iid random variables with distribution that of @xmath64 . because @xmath65<\infty$ ] , it follows that @xmath75<\infty$ ] as well . by the law of large numbers , @xmath76 $ ] and @xmath77 $ ] almost surely . moreover @xmath78 . thus @xmath79}e[y_1]=v\quad p\text{-almost surely.}\ ] ] table [ tab : walks ] summarizes what we know about uniform rwdre in 2-dimensional 2-valued random environments . it reproduces and updates table 1 of @xcite . there is a related table in @xcite , giving percolation properties for the directed graphs @xmath80 and @xmath81 . the latter includes 2-valued environments such as @xmath82 ( site percolation ) , in which one of the possible environments has no arrows . these environments do not appear in the present table , because ( as remarked in section 3 of @xcite ) , the walk gets stuck on a finite set of vertices ( in this case 1 vertex ) . the rwre setup we have chosen requires that motion be possible in at least one direction . .table of results for uniform rwdre in 2-dimensional 2-valued degenerate random environments , where the first configuration occurs with probability @xmath83 and the other with probability @xmath84 . [ cols="<,<,<",options="header " , ] * notes to table [ tab : walks ] * @xmath85 it follows from results of berger & deuschel @xcite that @xmath81 is recurrent @xmath86 . @xmath87 bounds on the critical probability are given in @xcite . improved bounds are in preparation.@xmath88 improved ranges of values giving transience and speeds are in preparation . @xmath89 an expansion in terms of @xmath90-hypergeometric functions is described below.@xmath91 we do not have a closed form expression for this . but asymptotic expressions are given below.@xmath92 the expansion is as in the case @xmath93 but messier , so the formula is not included in the table . the non - trivial 2-valued uniform models , in which one must turn to the results of @xcite for existence of a speed , and in which we can at present say very little about the speed ( other than mononicity ) are as follows : * @xmath94 @xmath95 * @xmath96 @xmath97 * @xmath96 @xmath98 * @xmath96 @xmath94 * @xmath96 @xmath99 * @xmath96 @xmath100 * @xmath101 @xmath97 * @xmath101 @xmath94 * @xmath101 @xmath96 there are two further models , which are also non - trivial , but for which , once we know that the velocity exists , it must be @xmath102 by symmetry , namely : * @xmath103 * @xmath101 @xmath104 the simplest models where one can explicitly calculate the speed are : * @xmath105 : because the rwdre sees a new environment every time , the velocity is simply @xmath106 . * @xmath107 : let @xmath70 be the @xmath71th time @xmath108 that @xmath109 , with @xmath110=0 . let @xmath111}_{\tau_k}$ ] . at each time @xmath70 the process starts exploring a new independent environment , so @xmath112 are iid ( for @xmath113 ) , as are @xmath114 . by the strong law , @xmath115/e[t_2]$ ] as @xmath116 . if @xmath117 is the last @xmath71 such that @xmath118 then @xmath119}_n}{n}\le \frac{\eta_{n_{n+1}}}{\tau_{n_n}}\ ] ] so that @xmath120/e[t_2]$ ] as well . + if @xmath121 then @xmath122 , since that is the mean time to reach @xmath123 of a random walk on @xmath124 $ ] with reflection at 0 . thus @xmath125=e[m_2 ^ 2]$ ] . but @xmath126 is geometric , @xmath127=\sum_{m=1}^\infty mp^{m-1}(1-p)=1/(1-p)$ ] and @xmath128=(1+p)/(1-p)^2 $ ] . so @xmath129}=(1-p)/(1+p)$ ] . * @xmath130 : in this model also , each step of @xmath22 explores a new environment . so we essentially have a regular random walk , whose step distribution is @xmath131 with probability @xmath132 and @xmath97 with probability @xmath133 . so @xmath134 . in the remaining examples , we use the setup of lemma [ lem : trivialtransience ] . there is a direction @xmath135 for which the first time @xmath64 that @xmath22 moves in direction @xmath135 is a renewal time what happens starting at time @xmath64 is independent of what came before . if @xmath136 , then @xmath129}=\pm1/e[t]$ ] . then if @xmath137}_t$ ] , then @xmath138}=e[y]/e[t]$ ] . with corresponding formulae if @xmath139 . in the following example we calculate @xmath140 $ ] to get the speed . * @xmath141 : for @xmath142 , let @xmath143}=n\}$ ] . then for @xmath144 , @xmath145 are i.i.d . geometric@xmath146 random variables ( with mean @xmath51 ) , and @xmath147}_{\tau_i -1}-x^{[1]}_{\tau_{i-1}}$ ] are i.i.d . random variables , independent of the @xmath148 . so @xmath149=2 $ ] and @xmath138}=1/2 $ ] . let @xmath150 . here @xmath151 , and the first time @xmath64 that we move upwards is geometric with parameter @xmath152 . + then almost surely , @xmath153}}{n}=\frac{\sum_{i=1}^{n_n}y_i+\sum_{i=\tau_{n_n}+1}^{n}(x^{[1]}_i - x^{[1]}_{i-1})}{n}=\frac{n_n}{n}\frac{\sum_{i=1}^{n_n}y_i}{n_n}+\frac{\sum_{i=\tau_{n_n}+1}^{n}(x^{[1]}_i - x^{[1]}_{i-1})}{n}{\rightarrow}\frac{e[y_1]}{e[t_1]},\ ] ] as @xmath154 , where we have used the fact that @xmath155}_i - x^{[1]}_{i-1})|\le t_{n_{n+1}}$ ] . + now let @xmath156}_t$ ] , so @xmath129}=e[y]/2 $ ] . for @xmath157 , we can have @xmath158 three ways reaching no @xmath159 vertex , reaching a @xmath159 vertex at @xmath160 , or reaching a @xmath159 vertex at @xmath161 . thus @xmath162 likewise , we can have @xmath163 , @xmath157 three ways , depending on where if anywhere @xmath22 reaches a @xmath94 vertex , giving @xmath164 the case @xmath165 would be similar , but is not needed . summing over @xmath166 gives that @xmath167&=\frac{p(4-p^2)}{12}\cdot \frac{1}{(1-p/2)^2}-\frac{(1-p)(4-(1-p)^2)}{12}\cdot\frac{1}{(1-(1-p)/2)^2}\\ & = \frac{p(2+p)}{3(2-p)}-\frac{(1-p)(3-p)}{3(1+p ) } = \frac{(2p-1)(p^2-p+6)}{3(2-p)(1+p)}.\end{aligned}\ ] ] comparing this with the speed @xmath168}=p-{\frac{1}{2}}$ ] of a true random walk that goes up with probability @xmath152 , right with probability @xmath132 , and left with probability @xmath169 , we see that the speeds agree for @xmath170 , and 1 , but the rwre is slower in between . see figure [ fig : introcase ] . + , @xmath171 , and the first coordinate of the velocity as a function of @xmath172.,title="fig : " ] , @xmath171 , and the first coordinate of the velocity as a function of @xmath172.,title="fig:",width=302,height=302 ] in some cases , we can avoid calculating @xmath140 $ ] directly . again , assume @xmath136 . there are two generators , @xmath173 and @xmath174 , depending on the environment . if we apply them to the functions @xmath175}$ ] we get that @xmath176}-\sum_{k < n}({\mathcal{l}}_1f_j1_{\{{\mathcal{g}}_{x_k}=\gamma_1\}}+{\mathcal{l}}_2f_j1_{\{{\mathcal{g}}_{x_k}=\gamma_2\}})$ ] is a martingale . therefore @xmath177}e[t]=e[x_t^{[j]}]={\mathcal{l}}_1f_j\alpha_1 + { \mathcal{l}}_2f_j \alpha_2\ ] ] where @xmath178 $ ] . when @xmath179 the lhs is @xmath180 , which usually lets us solve for @xmath181 in terms of @xmath65 $ ] . we know that @xmath182-\alpha_1 $ ] , so putting @xmath183 will then give us @xmath138}$ ] . thus all that remains is to calculate @xmath65 $ ] . ( we could have done the previous example this was as well . ) * @xmath184 : here @xmath151 . we have @xmath129}=0 $ ] by symmetry , and @xmath138}=1/e[t]$ ] . if the origin is @xmath97 , then @xmath185 . otherwise , suppose there are @xmath97 at @xmath186 and at @xmath160 , with only @xmath104 in between . then @xmath187 , since this is the mean exit time for a simple random walk on @xmath188 $ ] . given that the origin is @xmath104 ( which happens with probability @xmath189 ) , @xmath190 and @xmath166 are independent geometric random variables , with means @xmath191 . thus @xmath65=1+p/(1-p)^2 $ ] . * @xmath192 : here @xmath151 , and the martingale equations are that @xmath193}e[t]=\alpha_2/2 $ ] and @xmath129}e[t]=\alpha_2/2 $ ] . in other words , @xmath129}=v^{[2]}=1/e[t]$ ] . so we must now find @xmath65 $ ] . + first consider a random walk @xmath194 on @xmath195 with the following boundary conditions : at @xmath108 there is absorption , and at @xmath196 we reflect with probability @xmath152 and die otherwise . let @xmath197 be the time of death or absorption , and let @xmath198 $ ] . then @xmath199 for @xmath200 , @xmath201 , and @xmath202 . the solution to the recurrence is @xmath203 , and substituting the boundary conditions gives @xmath204 . likewise let @xmath205 . then @xmath206/2 $ ] for @xmath200 , with boundary conditions @xmath207 and @xmath208 . this has solution @xmath209 . + now think of how @xmath210 evolves . let the first @xmath94 to the left of @xmath211 be at @xmath212 , where @xmath213 . let successive @xmath94 to the right of @xmath211 be at @xmath214 , @xmath215 , etc . , where @xmath216 . on the horizontal interval @xmath217 , @xmath210 performs a simple random walk till it hits @xmath218 or @xmath219 . when it hits @xmath218 it either moves upward ( making this time @xmath64 ) , or it reflects back into the interval . if it reaches @xmath219 it leaves this interval forever , and starts the same process over again on the interval @xmath220 . let the interval being visited at time @xmath64 be @xmath221 , where @xmath222 , and let @xmath223 be the total time spent in @xmath224 . therefore @xmath225 , and @xmath226=\sum_{j=0}^\infty e[s_j 1_{\{n\ge j\}}].\ ] ] let @xmath227 be the event that @xmath22 exits @xmath224 on the right , ie . at @xmath228 . then for @xmath157 , @xmath229 . moreover , there is a renewal every time @xmath22 enters a new interval , because a new environment starts getting explored . the interval @xmath217 is different from the rest , because we start at @xmath211 . but for all other @xmath230 the process starts walking at @xmath231 . therefore the cases @xmath157 are actually independent replications of the same procedure . in other words , @xmath226=e[s_0]+\sum_{j=1}^\infty e[s_j \prod_{k=0}^{j-1}1_{a_k}]=e[s_0]+\sum_{j=1}^\infty e[s_1\mid a_0 ] p(a_0)p(a_1\mid a_0)^{j-1}.\ ] ] we use the expressions for @xmath232 and @xmath233 to work out these factors . by the expression for @xmath232 , we have @xmath234 . take @xmath235 and @xmath236 . then @xmath237=\sum_{i=0}^\infty\sum_{k=1}^\infty p^2(1-p)^{i+k-1}k(1+i ) = \big(\sum_{k=1}^\infty p(1-p)^{k-1}k\big)^2=\frac{1}{p^2}.\ ] ] likewise @xmath238 . write @xmath71 for this quantity , so @xmath239=\sum_{k=1}^\infty ( 1-p)^{k-1}pk=\frac{p}{p^2}=\frac{1}{p}.\ ] ] similarly , @xmath240 , so ( letting @xmath241 ) @xmath242 and @xmath243 , so @xmath244 putting this together , @xmath226=\frac{1}{p^2}+\frac{(1-p)^2}{2p(1-p+p\log p)}.\ ] ] * @xmath245 : here @xmath151 , and the martingale equations are that @xmath193}e[t]=\alpha_1/2 $ ] and @xmath129}e[t]=\alpha_1/2-\alpha_2=-e[t]+3\alpha_1/2 $ ] . in other words , @xmath138}=1/e[t]$ ] and @xmath129}=-1 + 3/e[t]$ ] . so we must now find @xmath65 $ ] . + suppose that there is a @xmath94 at @xmath186 for @xmath144 , and @xmath246 s at @xmath211 and all points in between ( a scenario with probability @xmath247 . then @xmath22 takes @xmath190 steps to the left , and then oscillates between @xmath186 and @xmath248 a random number of times , before @xmath64 occurs . + the other possibility is that there is a @xmath246 at @xmath160 for @xmath157 , and @xmath94 s at @xmath211 and all points in between . this scenario has probability @xmath249 . now @xmath22 steps right , and @xmath64 may occur before it reaches @xmath250 , or it may reach @xmath160 and then oscillate until time @xmath64 . the various scenarios lead to the following expression : @xmath251&=\sum_{i=1}^\infty p(1-p)^i\sum_{k=0}^\infty ( 1/2)^{k+1}[i+2k+1]\\ & \qquad\qquad+\sum_{j=1}^\infty ( 1-p)p^j\big(\sum_{k=0}^{j-2}(1/2)^{k+1}[k+1 ] + \sum_{k=0}^\infty ( 1/2)^{j+k}[j+2k]\big)\\ & = \sum_{i=1}^\infty p(1-p)^i(i+3)+\sum_{j=1}^\infty ( 1-p)p^j\big(\frac12\frac{d}{dt}\big|_{t=1/2}\frac{1-t^j}{1-t}+\frac{j+2}{2^{j-1}}\big)\\ & = 3(1-p)+\frac{1-p}{p}+2\sum_{j=1}^\infty ( 1-p)p^j\big(1+\frac{1}{2^j}\big)\\ & = 2 - 3p+\frac{1}{p}+2p+\frac{p(1-p)}{1-p/2 } = \frac{2 + 3p-2p^2-p^3}{p(2-p)}.\end{aligned}\ ] ] * @xmath252 : the velocity is @xmath253)$ ] , where @xmath64 is the time of the first step in the @xmath254 direction . + to find @xmath65 $ ] , first consider random walk on @xmath255 $ ] , with the probability of death in 1 step starting from @xmath200 being @xmath256 , and the probability of death in 1 step being 1 starting from 0 or @xmath108 . let @xmath257 be the mean time of death , starting from @xmath71 . then @xmath258 , and otherwise @xmath259 the solution is @xmath260 where @xmath261 are solutions of @xmath262 . in other words , @xmath263 and @xmath264 . from the boundary conditions , we get @xmath265 but @xmath266 , so this simplifies to @xmath267 if there is a @xmath254 at @xmath211 then @xmath185 . otherwise , suppose there are @xmath254 s at @xmath186 and @xmath160 , with @xmath96 s in between , where @xmath268 . then @xmath269 . therefore @xmath251 & = ( 1-p)\cdot 1 + p\sum_{i , j=1}^\infty p^{i+j-2}(1-p)^2\big(3 - 2\frac{\gamma^i+\gamma^j}{\gamma^{i+j}+1}\big)\\ & = 1-p+3p(1-p)^2\big(\sum_{i=1}^\infty p^{i-1}\big)^2 -\frac{2(1-p)^2}{p}\sum_{k=2}^\infty \frac{p^k}{\gamma^k+1}\sum_{j=1}^{k-1}(\gamma^j+\gamma^{k - j})\\ & = 1 + 2p-\frac{4(1-p)^2}{p}\sum_{k=2}^\infty \frac{p^k}{\gamma^k+1}\big(\frac{\gamma^k-1}{\gamma-1}-1\big)\\ & = 1 + 2p-\frac{4(1-p)^2}{p}\sum_{k=2}^\infty \big(\frac{1}{\gamma-1}p^k-\frac{\gamma+1}{\gamma-1}\frac{p^k}{\gamma^k+1}\big)\\ & = 1 + 2p-\frac{4p(1-p)}{\gamma-1}+\frac{4(1-p)^2(\gamma+1)}{p(\gamma-1)}\sum_{k=2}^\infty \frac{p^k}{\gamma^k+1}.\end{aligned}\ ] ] note that the expression @xmath270 is known as a unilateral @xmath90-hypergeometric series , and in the theory of special functions would be written @xmath271.\ ] ] see @xcite . we are indebted to martin muldoon for pointing this out . we could therefore also write @xmath251&=\frac{2(1-p)^2(\gamma+1)}{p(\gamma-1)}q(\gamma;p)+1 + 2p\\ & \qquad\qquad\qquad-\frac{4p(1-p)}{\gamma-1}-\frac{4(1-p)^2}{\gamma-1}-\frac{2(1-p)^2(\gamma+1)}{p(\gamma-1)}\\ & = \frac{2(1-p)^2(\gamma+1)}{p(\gamma-1)}q(\gamma;p)+5-\frac{2}{p}-\frac{4(1-p)}{p(\gamma-1)}\end{aligned}\ ] ] * @xmath272 : here @xmath273 , and the martingale equations are that @xmath274}e[t]=-\alpha_1/3 $ ] and @xmath129}e[t]=\alpha_2=e[t]-\alpha_1=e[t]-3 $ ] . in other words , @xmath138}=-1/e[t]$ ] and @xmath129}=1 - 3/e[t]=1 + 3v^{[2]}$ ] . so we must now find @xmath65 $ ] . + first consider a random walk @xmath194 on @xmath255 $ ] with the following boundary conditions : it reflects at @xmath196 , and it is absorbed at @xmath108 . at points in between there is killing with probability @xmath256 , and otherwise @xmath194 performs a simple symmetric random walk . let @xmath197 be the time of death or absorption , and let @xmath275 $ ] . then @xmath276 for @xmath200 , @xmath201 , and @xmath277 . the solution to the recurrence is @xmath278 where as above , @xmath279 are @xmath280 . from the boundary conditions , we get @xmath281 but @xmath266 , so this simplifies to @xmath282.\end{aligned}\ ] ] likewise let @xmath283 . then @xmath206/3 $ ] for @xmath200 , so @xmath284 with boundary conditions @xmath285 and @xmath208 . as above , this has solution @xmath286 + now consider how @xmath210 evolves . let the first @xmath131 to the left of @xmath211 be at @xmath212 , where @xmath213 . let successive @xmath131 to the right of @xmath211 be at @xmath214 , @xmath215 , etc . , where @xmath216 . at interior points of the horizontal interval @xmath217 , @xmath210 leaves the interval with probability @xmath256 at every step , by moving downwards ( ie at time @xmath64 ) . otherwise it evolves as a simple random walk till it hits @xmath218 or @xmath219 . if it hits @xmath218 it reflects back with probability 1 . if it reaches @xmath219 it leaves this interval forever , and starts the same process over again on the interval @xmath220 . let the interval being visited at time @xmath64 be @xmath221 , where @xmath222 , and let @xmath223 be the total time spent in @xmath224 . then @xmath225 , and @xmath226=\sum_{j=0}^\infty e[s_j 1_{\{n\ge j\}}].\ ] ] if @xmath227 is the event that @xmath22 exits @xmath224 at @xmath228 ( ie before @xmath64 ) , then for @xmath157 we have @xmath229 . moreover , there is a renewal every time @xmath22 enters a new interval , because we start exploring a new environment . the interval @xmath217 is different from the rest , because we start at @xmath211 . but for all other @xmath230 the process starts walking at @xmath231 . in other words , the cases @xmath157 are independent replications of the same procedure . therefore @xmath251&=e[s_0]+\sum_{j=1}^\infty e[s_j \prod_{k=0}^{j-1}1_{a_k}]=e[s_0]+\sum_{j=1}^\infty e[s_1\mid a_0 ] p(a_0)p(a_1\mid a_0)^{j-1}\\ & = e[s_0]+e[s_1\mid a_0 ] p(a_0)/(1-p(a_1\mid a_0)).\end{aligned}\ ] ] we can work out all these factors using the expressions for @xmath232 and @xmath233 . first define @xmath287\ ] ] where @xmath288 is the @xmath90-hypergeometric series defined earlier . then using that @xmath289 , @xmath290&=\sum_{i_0=0}^\infty \sum_{i_1=1}^\infty p^{i_0+i_1 - 1}(1-p)^2 f_{i_0+i_1}(i_0)\\ & = 3-(1-p)^2\sum_{n=1}^\infty p^{n-1}\sum_{k=0}^{n-1}\big[\frac{3(\gamma^{n - k}+\gamma^{n+k-1})}{\gamma^{2n-1}+1 } + \frac{\gamma^k-\gamma^{2n - k}}{(\gamma-1)(\gamma^{2n-1}+1)}\big]\\ & = 3-(1-p)^2\sum_{n=1}^\infty\frac{p^{n-1}(\gamma^n-1)}{(\gamma-1)^2(\gamma^{2n-1}+1)}\big[3(\gamma-1)(\gamma+\gamma^{n-1})+1-\gamma^{n+1}\big]\\ & = 3-\frac{(1-p)^2}{(\gamma-1)^2}\sum_{n=1}^\infty\frac{p^{n-1}}{\gamma^{2n-1}+1}\big[(\gamma^{2n-1}+1)(3\gamma-3-\gamma^2)+\gamma^{n-1}(3\gamma^3 - 2\gamma^2 - 2\gamma+3)-2(\gamma^2 - 1)\big]\\ & = 3-\frac{(1-p)(3\gamma-3-\gamma^2)}{(\gamma-1)^2}+\left(\frac{1-p}{\gamma-1}\right)^2\big[2(\gamma^2 - 1)\theta(p)-(3\gamma^3 - 2\gamma^2 - 2\gamma+3)\theta(p\gamma)\big],\\ & = 3+\frac{2(1-p)}{\gamma}+\frac{2(1-p)^2}{\gamma}\big[(3\gamma-2)\theta(p)-(8\gamma-2)\theta(p\gamma)\big],\\ e[s_1\mid a_0]&=\sum_{n=1}^\infty p^{n-1}(1-p)f_n(0 ) = 3-\sum_{n=1}^\infty \frac{p^{n-1}(1-p)}{\gamma^{2n-1}+1}\big[3\gamma^{n-1}(1+\gamma)-\frac{\gamma^{2n}-1}{\gamma-1}\big]\\ & = 3 - 3(1-p)(1+\gamma)\theta(p\gamma)+\frac{\gamma}{\gamma-1}-\frac{(1-p)(1+\gamma)}{\gamma-1}\theta(p ) , p(a_0)&=\sum_{i_0=0}^\infty \sum_{i_1=1}^\infty p^{i_0+i_1 - 1}(1-p)^2 g_{i_0+i_1}(i_0 ) = ( 1-p)^2\sum_{n=1}^\infty p^{n-1}\sum_{k=0}^{n-1}\frac{\gamma^{n - k}+\gamma^{n+k-1}}{\gamma^{2n-1}+1}\\ & = \frac{(1-p)^2}{\gamma-1}\sum_{n=1}^\infty \frac{p^{n-1}(\gamma^{2n-1}+\gamma^{n+1}-\gamma^{n-1}-\gamma)}{\gamma^{2n-1}+1 } = \frac{(1-p)}{\gamma-1}+(1-p)^2(\gamma+1)\big[\theta(p\gamma)-\frac{\theta(p)}{\gamma-1}\big],\\ p(a_1\mid a_0)&=\sum_{n=1}^\infty p^{n-1}(1-p)g_n(0)=(1-p)(\gamma+1)\theta(p\gamma).\end{aligned}\ ] ] therefore @xmath291=3+\frac{2(1-p)}{\gamma}+\frac{2(1-p)^2}{\gamma}\big[(3\gamma-2)\theta(p)-(8\gamma-2)\theta(p\gamma)\big]\\ + \frac{\big[3 - 3(1-p)(1+\gamma)\theta(p\gamma)+\frac{\gamma}{\gamma-1}-\frac{(1-p)(\gamma+1)}{\gamma-1}\theta(p)\big]\big[\frac{(1-p)}{\gamma-1}+(1-p)^2(\gamma+1)\big\{\theta(p\gamma)-\frac{\theta(p)}{\gamma-1}\big\}\big]}{1-(1-p)(\gamma+1)\theta(p\gamma)}.\end{gathered}\ ] ] simplifying this , we have @xmath226=4-p-(1-p)^2\frac{4\gamma-1}{\gamma}\theta(p\gamma ) + \frac{1-p}{\gamma}\frac{\big[\gamma-(1-p)(\gamma+1)\theta(p)\big]^2}{1-(1-p)(\gamma+1)\theta(p\gamma)}.\ ] ] substituting for @xmath292 gives @xmath226=4-p-\frac{5+\sqrt{5}}{2}(1-p)^2\theta(p\gamma ) + \frac{(1-p)\big[3+\sqrt{5}-(1-p)(5+\sqrt{5})\theta(p)\big]^2}{(3+\sqrt{5})\big[2-(1-p)(5+\sqrt{5})\theta(p\gamma)\big]}.\ ] ] * @xmath293 : here @xmath273 , and the martingale equations are that @xmath274}e[t]=-\alpha_1/3-\alpha_2/2=-e[t]/3-\alpha_2/6 $ ] and @xmath129}e[t]=-\alpha_2/2=e[t]-3 $ ] . in other words , @xmath138}=-1/e[t]$ ] and @xmath129}=1 - 3/e[t]=1 + 3v^{[2]}$ ] . so we must now find @xmath65 $ ] . this is just like the previous example , but messier . + it is slightly more convenient to work out @xmath65 $ ] using @xmath294 , since the structure of the previous example can be maintained with minor changes . for @xmath232 and @xmath233 , the difference is that now there is killing at 0 at rate @xmath152 . this changes the boundary condition for @xmath295 to be @xmath296 , from which we get @xmath297 and therefore @xmath298.\ ] ] the changed boundary condition for @xmath233 is that @xmath299 , from which we get @xmath300 and therefore @xmath301 for the walk @xmath210 there is now reflection at @xmath302 with probability @xmath303 and @xmath254 with probability @xmath303 . set @xmath304 @xmath290&=\sum_{i_0=0}^\infty \sum_{i_1=1}^\infty p^{i_0+i_1 - 1}(1-p)^2 f_{i_0+i_1}(i_0)\\ & = 3+(1-p)^2\sum_{n=1}^\infty \frac{p^{n-1}}{\gamma^{2n-1}(1 - 2\gamma)-\gamma+2}\sum_{k=0}^{n-1}\big [ \gamma^{n - k}(\gamma^n+3\gamma-6)+\gamma^k(6\gamma^n-3\gamma^{n-1}-1)\big]\\ & = 3+\frac{(1-p)^2}{\gamma-1}\sum_{n=1}^\infty \frac{p^{n-1}(\gamma^n-1)}{\gamma^{2n-1}(1 - 2\gamma)-\gamma+2}\big [ \gamma(\gamma^n+3\gamma-6)+6\gamma^n-3\gamma^{n-1}-1\big]\\ & = 3+\frac{(1-p)^2}{\gamma-1}\sum_{n=1}^\infty \frac{p^{n-1}}{\gamma^{2n-1}(1 - 2\gamma)-\gamma+2}\big [ \gamma^{2(n-1)}(\gamma^3 + 6\gamma^2 - 3\gamma)+\\ & \hspace{3in}+\gamma^{n-1}(3\gamma^3 - 7\gamma^2 - 7\gamma+3)-(3\gamma^2 - 6\gamma+1)\big]\\ & = 3+\frac{(1-p)^2}{\gamma-1}\big [ ( \gamma^3 + 6\gamma^2 - 3\gamma)\tilde\theta(p\gamma^2)+(3\gamma^3 - 7\gamma^2 - 7\gamma+3)\tilde\theta(p\gamma)-(3\gamma^2 - 6\gamma+1)\tilde\theta(p)\big]\\ e[s_1\mid a_0]&=\sum_{n=1}^\infty p^{n-1}(1-p)f_n(0)\\ & = 3+\sum_{n=1}^\infty \frac{p^{n-1}(1-p)}{\gamma^{2n-1}(1 - 2\gamma)-\gamma+2}\big[\gamma^{n}(\gamma^n+3\gamma-6)+(6\gamma^n-3\gamma^{n-1}-1)\big]\\ & = 3+\sum_{n=1}^\infty \frac{p^{n-1}(1-p)}{\gamma^{2n-1}(1 - 2\gamma)-\gamma+2}\big[\gamma^2\gamma^{2(n-1)}+3\gamma^{n-1}(\gamma^2 - 1)-1\big]\\ & = 3+(1-p)\big[\gamma^2\tilde\theta(p\gamma^2)+3(\gamma^2 - 1)\tilde\theta(p\gamma)-\tilde\theta(p)\big]\\ p(a_0)&=\sum_{i_0=0}^\infty \sum_{i_1=1}^\infty p^{i_0+i_1 - 1}(1-p)^2 g_{i_0+i_1}(i_0 ) = ( 1-p)^2\sum_{n=1}^\infty p^{n-1}\sum_{k=0}^{n-1}\frac{\gamma^{n - k}(2-\gamma)+\gamma^{n+k-1}(1 - 2\gamma)}{\gamma^{2n-1}(1 - 2\gamma)-\gamma+2}\\ & = \frac{(1-p)^2}{\gamma-1}\sum_{n=1}^\infty \frac{p^{n-1}}{\gamma^{2n-1}(1 - 2\gamma)-\gamma+2}\big[\gamma(\gamma^n-1)(2-\gamma)+\gamma^{n-1}(\gamma^n-1)(1 - 2\gamma)\big]\\ & = \frac{(1-p)^2}{\gamma-1}\sum_{n=1}^\infty \frac{p^{n-1}}{\gamma^{2n-1}(1 - 2\gamma)-\gamma+2}\big[\gamma^{2(n-1)}\gamma(1 - 2\gamma)-\gamma^{n-1}(\gamma^3 - 2\gamma^2 - 2\gamma+1)+\gamma(\gamma-2)\big]\\ & = \frac{(1-p)^2}{\gamma-1}\big[(\gamma(1 - 2\gamma)\tilde\theta(p\gamma^2)-(\gamma^3 - 2\gamma^2 - 2\gamma+1)\tilde\theta(p\gamma)+\gamma(\gamma-2)\tilde\theta(p)\big]\\ p(a_1\mid a_0)&=\sum_{n=1}^\infty p^{n-1}(1-p)g_n(0)=(1-p)\sum_{n=1}^\infty \frac{p^{n-1}}{\gamma^{2n-1}(1 - 2\gamma)-\gamma+2}\big[\gamma^{n-1}(1-\gamma^2)\big]\\ & = ( 1-p)(1-\gamma^2)\tilde\theta(p\gamma).\end{aligned}\ ] ] therefore @xmath291=3+\frac{(1-p)^2}{\gamma-1}\big [ ( \gamma^3 + 6\gamma^2 - 3\gamma)\tilde\theta(p\gamma^2)+(3\gamma^3 - 7\gamma^2 - 7\gamma+3)\tilde\theta(p\gamma)-(3\gamma^2 - 6\gamma+1)\tilde\theta(p)\big]+\\ + \big[3+(1-p)\big[\gamma^2\tilde\theta(p\gamma^2)+3(\gamma^2 - 1)\tilde\theta(p\gamma)-\tilde\theta(p)\big]\big]\frac{(1-p)^2}{\gamma-1}\cdot\\ \cdot\big[(\gamma(1 - 2\gamma)\tilde\theta(p\gamma^2)-(\gamma^3 - 2\gamma^2 - 2\gamma+1)\tilde\theta(p\gamma)+\gamma(\gamma-2)\tilde\theta(p)\big]/\big[1-(1-p)(1-\gamma^2)\tilde\theta(p\gamma)\big]\end{gathered}\ ] ] finally , we give an asymptotic argument , in the one case in which there is an elementary renewal structure for which we do nt know how to find the speed analytically . * @xmath305 : here @xmath273 , and @xmath306)$ ] . though we do nt know how to find @xmath307 analytically , here is an approach that should give an asymptotic expansion in powers of @xmath308 . + embed @xmath309 as @xmath310 , and let @xmath311}_n$ ] for @xmath312 . we can fill in new independent increments after time @xmath313 to make @xmath314 into a simple symmetric random walk started from @xmath196 , and then recover an independent copy of @xmath64 by killing @xmath315 at a rate that depends on the environment . write @xmath316 for this extension of @xmath23 . let @xmath317 be 1 ( resp . @xmath318 ) if @xmath319 ( resp . @xmath96 ) . then by feynman - kac , @xmath226=\sum_{k=1}^\infty p(t\ge k)=\sum_{k=1}^\infty \tilde p(\prod_{i=0}^{k-2 } v_{y_i}).\ ] ] this equals @xmath320 , where @xmath321 counts the number of visits of @xmath314 to @xmath166 , for @xmath322 . integrating out the environment , and setting @xmath323 , we get @xmath324 ) & = \sum_{k=0}^\infty \tilde p(\prod_{j\in{{{\mathbb z } } } } \gamma^{n_j(k)}[q\gamma^{-n_j(k)}+1-q])\\ & = \sum_{k=0}^\infty \gamma^{k}\tilde p(\prod_{j\in{{{\mathbb z } } } } [ 1+q(\gamma^{-n_j(k)}-1)])\end{aligned}\ ] ] ( since @xmath325 ) . this expression can in principle be expanded as a series in @xmath90 . what complicates this is that the @xmath326 term involves knowing the joint distributions of the @xmath327 for @xmath190 choices of @xmath166 . still , it is easy to work out the constant and linear terms ( and with some work one could in principal take this further ) . in particular , the following should be true : @xmath251&=\sum_{k=0}^\infty \gamma^{k } + q\sum_{k=0}^\infty \sum_{j\in{{{\mathbb z } } } } \gamma^{k } \tilde p(\gamma^{-n_j(k)}-1)+o(q^2)\\ & = 3+q \sum_{j\in{{{\mathbb z } } } } \big[\sum_{k=0}^\infty \gamma^{k } \tilde p(\gamma^{-n_j(k)})-3\big]+o(q^2).\end{aligned}\ ] ] set @xmath328 . conditioning on the first step @xmath329 , we get the recurrence @xmath330 , & j\neq 0\\ 1+\big[\frac{f_{j+1}+f_{j-1}}{2}\big ] , & j= 0 . \end{cases}\ ] ] the solution is @xmath331 where @xmath332 . since @xmath333 it follows that @xmath226=3+\frac{5+\sqrt{5}}{2}q+o(q^2)\ ] ] as @xmath334 . the authors thank neal madras for his comments about the above calculation . + one can also approximate @xmath65 $ ] for small @xmath189 . for example , to first order , @xmath65 $ ] is the mean time to reach the first @xmath96 site on either side of the origin . so @xmath226\sim\sum_{j=1}^\infty\sum_{k=1}^\infty p^2q^{j+k-1}\big(\frac{j+k}{2}\big)^2\sim\frac{3}{2p^2}\ ] ] as @xmath335 . this can be improved on by considering additional @xmath96 sites . + one can show in general that @xmath138}<0 $ ] , ie . that @xmath65<\infty$ ] for @xmath336 . ( this approach was suggested by remco van der hofstad ) . let @xmath337}_n=-1\}$ ] , and @xmath338 . then @xmath138}=-e[t]^{-1}$ ] , and @xmath339 , where @xmath340 for each @xmath341 , and @xmath342 denotes the local time of a _ simple random walk _ in @xmath343 ( as above , the rwre behaves as a random walk in @xmath343 killed as soon as it chooses to take a @xmath254 step from a @xmath96 site ) . therefore , letting @xmath344 denote the range of simple random walk we have by donsker - varadhan and markov s inequality , @xmath345=e\left[\left(\frac23\right)^{\sum_x\ell_n(x)i_{x}}\right]\\ \le & p(-r_n>-n^{\frac14})+e\left[\left(\frac23\right)^{\sum_x\ell_n(x)i_{x}}\big|r_n\ge n^{1/4}\right]p(r_n\ge n^{1/4})\\ \le & e^{-cn^{\frac13}}e^{n^\frac14}+e\left[\left(\frac23\right)^{\sum_x\ell_n(x)i_{x}}\big|r_n\ge n^{1/4}\right ] . \ ] ] the second term is at most @xmath346=e\left[\prod_{x=1}^{n^{1/4}}\left(\frac23\right)^{i_{x}}\right]=\big(\frac{2}{3}p+(1-p)\big)^{n^{1/4}}.\end{aligned}\ ] ] therefore there exists @xmath347 ( independent of @xmath189 ) such that @xmath348 which is summable in @xmath108 for all @xmath349 , whence @xmath138}<0 $ ] for @xmath349 . in @xcite there are several examples of non - monotonicity , for which details of the calculations are not given . we present those here . * @xmath350 : _ a 2-valued example in which @xmath129}$ ] is not monotone in @xmath351 _ + here @xmath352 is an environment @xmath353 , with @xmath354 . @xmath355 is the environment @xmath246 with @xmath356 . we take @xmath357 . this example appears as example 4.6 of @xcite . let @xmath64 be the first time we move @xmath97 . we first calculate @xmath65 $ ] . + suppose that there is a @xmath352 at @xmath186 for @xmath144 , and @xmath246 s at @xmath211 and all points in between ( a scenario with probability @xmath247 ) . then @xmath22 takes @xmath190 steps to the left , and then oscillates between @xmath186 and @xmath248 a random number of times , before @xmath64 occurs . + the other possibility is that there is a @xmath246 at @xmath160 for @xmath157 , and @xmath352 s at @xmath211 and all points in between . this scenario has probability @xmath249 . now @xmath22 steps right , and @xmath64 may occur before it reaches @xmath160 , or it may reach @xmath160 and then oscillate until time @xmath64 . the various scenarios lead to the following expression : @xmath251&=\sum_{i=1}^\infty p(1-p)^i\sum_{k=0}^\infty \alpha^k(1-\alpha)[i+2k+1]\\ & \qquad\qquad+\sum_{j=1}^\infty ( 1-p)p^j\big(\sum_{k=0}^{j-2}\alpha^k(1-\alpha)[k+1 ] + \sum_{k=0}^\infty \alpha^{j+k-1}(1-\alpha)[j+2k]\big)\\ & = \sum_{i=1}^\infty p(1-p)^i(i+\frac{1+\alpha}{1-\alpha})+\sum_{j=1}^\infty ( 1-p)p^j\big((1-\alpha)\frac{d}{d\alpha}\frac{1-\alpha^j}{1-\alpha}+j\alpha^{j-1}+\frac{2\alpha^j}{1-\alpha}\big)\\ & = \frac{1+\alpha}{1-\alpha}(1-p)+\frac{1-p}{p}+\sum_{j=1}^\infty ( 1-p)p^j\frac{1+\alpha^j}{1-\alpha}\\ & = \frac{1+\alpha}{1-\alpha}(1-p)+\frac{1-p}{p}+\frac{p}{1-\alpha}+\frac{(1-p)p\alpha}{(1-\alpha)(1-p\alpha)}\\ & = \frac{\alpha}{1-\alpha}+p+\frac{1-p}{p}+\frac{(1-p)}{(1-\alpha)(1-p\alpha)}.\end{aligned}\ ] ] we expect this to be decreasing in @xmath189 and increasing in @xmath351 . rewriting the first term as @xmath358 we see that indeed , each term is increasing in @xmath351 . it can also be rewritten as @xmath359 and both @xmath360 and @xmath361 are indeed decreasing in @xmath189 , so our expectations are realized . this gives us that @xmath138}=1/e[t]$ ] is monotone in both parameters . + the martingale argument given above shows that @xmath177}e[t]={\mathcal{l}}_1f_jn_1 + { \mathcal{l}}_2f_jn_2\ ] ] where the @xmath362 are the generators for the two environments , @xmath175}$ ] , and @xmath363 is the expected number of visits to environment @xmath190 before @xmath64 . in particular , @xmath364 , @xmath365 , @xmath366 , @xmath367 , so @xmath368}e[t]=\alpha n_1-n_2 , \qquad 1=v^{[2]}e[t]=(1-\alpha ) n_1 , \qquad n_1+n_2=e[t].\ ] ] therefore @xmath369 which implies that @xmath370}&=\frac{1+\alpha}{1-\alpha}\frac{1}{e[t]}-1\\ & = ( 1+\alpha)\left(\alpha+(1-\alpha)\big(p+\frac1p-1\big)+\frac{(1-p)}{(1-p\alpha)}\right)^{-1}-1.\end{aligned}\ ] ] + when @xmath371 , @xmath129}=0 $ ] . when @xmath372 , @xmath129}=p-1<0 $ ] . graphing @xmath129}$ ] , we see that it is increasing in @xmath351 for @xmath373 . but for @xmath374 , @xmath129}$ ] has a positive interior maximum . + likewise , consider the fraction of sites @xmath375 which are of type @xmath376 , namely @xmath377=1/w$ ] , where @xmath378 = p+\frac1p-1+\alpha(2-(p+\frac1p))+\frac{1-p}{1-\alpha p}.\ ] ] when plotted , this function is not monotone in @xmath351 , at least when @xmath379 . + when @xmath372 we have @xmath380 . when @xmath371 we have @xmath381 , and clearly @xmath382 is where these two values agree . in other words , for small @xmath189 the fraction of @xmath352 is increasing with @xmath351 , whereas for large @xmath189 it is decreasing with @xmath351 . in other words , for small @xmath189 an increase in @xmath351 means we tend to linger more around the @xmath352 we ve managed to find . whereas when @xmath189 is big , increasing @xmath351 means we get stuck for a longer time , which brings the fraction of @xmath376 we see down towards @xmath303 . * @xmath383 _ a 3-valued example in which the frequency @xmath129}$ ] is not monotone in @xmath384 . _ + this example contains 4 parameters @xmath385 ^ 4 $ ] . we let @xmath386 , @xmath387 , and @xmath388 . we set @xmath389 , @xmath390 , and @xmath391 . the claim is that there exist @xmath392 such that @xmath129}=v^{[1]}(\alpha,\beta , p , q)$ ] is not monotone increasing in @xmath90 . this appears as example 4.4 of @xcite + we need to calculate @xmath65 $ ] and @xmath393}_t]$ ] . + let @xmath394 be the number of consecutive @xmath246 starting from the origin ( going to the left ) . then for @xmath395 , @xmath396 . let @xmath397 be the number of consecutive @xmath94 ( of either type ) starting from the origin ( going to the right ) . then for @xmath395 , @xmath398 . let @xmath399 and @xmath400 . let @xmath401 be a random variable that is equal to @xmath402 with probability @xmath90 and @xmath403 otherwise . let @xmath404 be a random variable that ( conditional on @xmath401 ) is geometric with parameter @xmath401 . + let @xmath405}$ ] . then @xmath406={\mathbb{e}}[\delta\mathbbm{1}_{y^->0}]+{\mathbb{e}}[\delta\mathbbm{1}_{y^+>0}],\ ] ] where the first term on the right is @xmath407=-{\mathbb{e}}[y^-]=\frac{-p}{1-p}.\ ] ] similarly @xmath408={\mathbb{e}}[t\mathbbm{1}_{y^->0}]+{\mathbb{e}}[t\mathbbm{1}_{y^+>0}],\ ] ] where the first term on the right is equal to @xmath409=p(1 + 2\xi)+{\mathbb{e}}[y^-\mathbbm{1}_{y^->0}]=p(1 + 2\xi)+\frac{p}{1-p}.\ ] ] + consider now @xmath410 $ ] and @xmath411 $ ] . if @xmath412 , the walker has @xmath413 independent opportunities to go up , each with probability @xmath414 , before reaching the last @xmath94 site before the first @xmath246 site . under this conditioning , the probability that @xmath415 $ ] ( and @xmath416 ) is then @xmath417 . the probability that @xmath418 is @xmath419.\ ] ] the probability that @xmath420 is @xmath421 ( note that this is also the sum over @xmath422 of the above ) . thus , using @xmath423 and letting @xmath424 we have + @xmath425=&\sum_{k=1}^{\infty}(1-p)^{k}p\left((k-1)\eta^{k-1 } + \sum_{j=0}^{k-2}j\eta^{j}(1-\eta)\right)\\ = & \frac{p(1-p)^2\eta}{(1-a)^2}+\frac{\eta ( 1-\eta)(1-p)^3}{(1-a)^2}=\frac{(1-p)^2\eta}{1-a } \ ] ] + moreover , @xmath426=&\sum_{k=1}^{\infty}(1-p)^{k}p\bigg(\sum_{m=0}^{\infty}(k+2m)\eta^{k-1}\left[q\alpha^m(1-\alpha ) + ( 1-q)\beta^m(1-\beta ) \right]\\ & + \sum_{j=0}^{k-2}(j+1)\eta^{j}(1-\eta)\bigg)\\ = & ( 1-p)p\left[q(1-\alpha)h(\alpha)+(1-q)(1-\beta ) h(\beta ) \right ] + \frac{(1-p)^2(1-\eta)}{(1-a)^2},\end{aligned}\ ] ] where @xmath427 + thus , @xmath426=\frac{1-p}{1-a}\left[1 + 2p\xi\right].\end{aligned}\ ] ] + finally we have that @xmath428=&\frac{(1-p)^2\eta}{1-a}-\frac{p}{1-p}\\ { \mathbb{e}}[t]=&p(1 + 2\xi)+\frac{p}{1-p } + \frac{1-p}{1-a}\left[1 + 2p\xi\right],\end{aligned}\ ] ] where @xmath429 and @xmath400 .

we argue the sign problem of the fermion determinant at finite density . it is unavoidable not only in monte - carlo simulations on the lattice but in the mean - field approximation as well . a simple model deriving from quantum chromodynamics ( qcd ) in the double limit of large quark mass and large quark chemical potential exemplifies how the sign problem arises in the polyakov loop dynamics at finite temperature and density . in the color su(2 ) case our mean - field estimate is in excellent agreement with the lattice simulation . we combine the mean - field approximation with a simple phase reweighting technique to circumvent the complex action encountered in the color su(3 ) case . we also investigate the mean - field free energy , from the saddle - point of which we can estimate the expectation value of the polyakov loop .

hep-ph0610323

quantum chromodynamics ( qcd ) exhibits various states of matter depending on the environment @xcite : a strongly coupled quark - gluon plasma ( sqgp ) has been discovered at the relativistic heavy icon collider ( rhic ) @xcite and various color superconductors presumably exist in the cores of the compact stellar object @xcite . these are widely - known extreme states of equilibrated qcd matter at high temperature and at high ( baryon or quark ) density , respectively . the study on sqgp properties at high temperature has been led by the monte - carlo simulation of qcd on the lattice @xcite . the lattice qcd results provide us with fundamental information such as the phase transition temperature @xmath0 @xcite , equation of state @xcite , susceptibility @xcite , mesonic correlation above @xmath0 @xcite , etc . in contrast , at finite density , the lattice technique is not quite successful so far ; it is hindered by the notorious problem that the fermion determinant in the presence of nonzero quark chemical potential @xmath1 is not positive semidefinite ( i.e. not nonnegative ) . this problem is commonly refereed to as the fermion sign problem @xcite . the monte - carlo simulation based on importance sampling requires positive semidefinite probability for each gauge configuration . in the presence of @xmath2 , however , the probability is no longer a well - defined quantity due to a negative fermion determinant arising . there have been several techniques proposed to handle the sign problem , e.g. the reweighting method @xcite , the taylor expansion method @xcite , the analytical continuation from an imaginary chemical potential @xcite . although these proposals have achieved partial successes when @xmath3 is small , no prescription applicable at high density has been established yet . this work is aimed to point out that the sign problem is relevant even in the mean - field treatment of gauge fields . one can intuitively understand it in the following way ; in the mean - field approximation the partition function is estimated from the most dominant contribution of particular configurations that have the largest probability or the smallest free energy . when the probability for each gauge configuration is ill - defined due to the sign problem , hence , the mean - field free energy may well be problematic . in the present work we shall focus on a specific manifestation of the sign problem appearing in the simplest model at finite temperature and density . because qcd is a highly nontrivial theory , applicability of the mean - field approximation is quite limited . we choose a finite temperature model because the mean - field description presumably works well to sketch the hot qcd phase transition . the polyakov loop plays an essential role as an order parameter there @xcite . the dynamics of the polyakov loop was closely examined some years ago both in the lattice simulation @xcite and in the mean - field approximation @xcite . recently , in addition , the polyakov loop dynamics near @xmath0 is specifically paid attention @xcite . there is also an interesting observation that the entanglement between the chiral and polyakov loop dynamics turns out to be indispensable to understand the nature of the qcd phase transitions @xcite . we have already known several indications from the mean - field studies about the polyakov loop behavior at @xmath2 . in the best of our knowledge , the effective potential of the polyakov loop at finite temperature and density was first derived in ref . @xcite in perturbative qcd in the one - loop order . ( see also ref . @xcite for the potential with an imaginary chemical potential . ) it is obvious from eq . ( 5 ) in ref . @xcite that the polyakov loop variable is augmented to a complex valued variable when @xmath2 , and thus the effective potential turns complex . it is quite nontrivial how to derive meaningful information from such a complex effective potential ; the expectation value of the polyakov loop can not be fixed directly by a minimum of the complex effective potential . the same problem occurs also in the chiral effective models with the polyakov loop coupling which was first formulated by one of the present authors in ref . @xcite and has been investigated extensively @xcite , though the sign problem has been almost overlooked . here , we would refer to a closely related work , ref . @xcite , in which the authors mentioned on the sign problem in a matrix model of the polyakov loop . it should be noted that the mean - field free energy as seen in refs . @xcite is not complex unlike the effective potential in ref . @xcite so that one can determine the expectation value of the polyakov loop without ambiguity . we will go into details of this issue later . our findings are all consistent with what has been foreseen in ref . @xcite . the concept of this work lies not in solving the sign problem but in observing the sign problem in a manageable way , so to speak , in a clean environment . we also demonstrate that a technique which we call the _ phase reweighting method _ works fine as a practical prescription in the similar but not the same spirit as in the finite density lattice simulation . we will employ the dense - heavy model obtained from qcd in the double limit of large quark mass and large quark chemical potential @xcite ( see also discussions in ref . the reasons we adopt the dense - heavy model are as follows : first , the fermion determinant is exactly calculable as a function of the gauge field . second , neither the chiral condensate nor the diquark condensate is involved in the dynamics owing to large quark mass . third , lattice data is available from ref . @xcite . from these three reasons the dense - heavy model is considered to be an appropriate implement for our purpose to scrutinize the sign problem within the mean - field approximation . this paper is organized as follows . in sec . [ sec : sign_problem ] we make a brief overview on the sign problem . we formulate the model in sec . [ sec : dense_heavy ] and the mean - field approximation in sec . [ sec : mean_field ] . then , in sec . [ sec : su2 ] , we examine the color su(2 ) case first which is free from the sign problem in order to make sure if the mean - field approximation is reasonable . we next proceed to the su(3 ) calculation . in sec . [ sec : su3 ] we explain the phase reweighting method to circumvent the su(3 ) complex fermion determinant and then discuss the validity of our method by viewing the mean - filed effective potential in sec . [ sec : mffreeenergy ] . section [ sec : summary ] is devoted to the summary . here are general discussions on the sign problem of the fermion determinant at finite density . readers who are already familiar with it can skip to the model study starting from sec . [ sec : model ] . we will later demonstrate how the dense - heavy model concretely embodies the general features mentioned in this section . the fermion determinant in euclidean space - time with a quark mass @xmath4 and a quark chemical potential @xmath1 takes the form of @xmath5 \,,\ ] ] where @xmath6 is the covariant derivative . it is a well - known argument that @xmath7 and thus the fermion determinant is real in the zero density ( @xmath8 ) case . alternatively , we can explicitly look into the eigenvalue spectrum of the dirac operator to confirm that the determinant is positive semidefinite as follows : for @xmath8 , when @xmath9 is an eigenstate of @xmath10 , the eigenvalue @xmath11 is pure imaginary because @xmath10 is anti - hermitian where @xmath12 s are hermitian in our convention . since the mass term is simply proportional to unity , @xmath9 is also an eigenstate of @xmath13 with the eigenvalue @xmath14 . one can show that @xmath15 is an eigenstate of @xmath13 having the eigenvalue @xmath16 from the property @xmath17 . therefore , the determinant consists of a pair of @xmath14 and @xmath18 which makes the whole determinant real and nonnegative as @xmath19 . in the presence of the chemical potential term the fermion determinant is not necessarily positive semidefinite . when @xmath9 is an eigenstate of @xmath20 with the eigenvalue @xmath11 , in the same was as above , one can show that @xmath21 is also an eigenstate of @xmath20 with the eigenvalue @xmath22 which is not @xmath23 because @xmath20 is no longer anti - hermitian . it is apparent from this explanation that , if @xmath1 is pure imaginary @xcite , @xmath20 is anti - hermitian so that the fermion determinant turns nonnegative . if a considered theory has degenerate fermions associated with an internal symmetry under the transformation @xmath24 and the chemical potential @xmath1 is replaced by a matrix @xmath25 in internal space transforming like @xmath26 then @xmath27 , so that the determinant becomes real . a well - known realization of this comes from the isospin degrees of freedom in flavor ( @xmath28,@xmath29 ) space . the isospin chemical potential @xmath30 and @xmath31 ( or @xmath32 ) certainly satisfies eq . ( [ eq : change ] ) where @xmath33 s are the pauli matrices in ( @xmath28,@xmath29 ) space @xcite . another realization is the color su(2 ) case @xcite that is relevant to our model study as we will argue later . the @xmath34-transformation changes the quark chemical potential @xmath1 as in eq . ( [ eq : change ] ) . in general cases , however , the @xmath34-transformation does not give an escape from the sign problem because it also swaps color and anticolor . that is , @xmath35^\ast\ ] ] with @xmath36 . special for @xmath37 is that anticolor is not distinguishable from color since two doublets can make a singlet . actually @xmath38 where @xmath39 s are the pauli matrices in color space . therefore the su(2 ) case is free from the sign problem . we should note that the fermion determinant being complex is not necessarily harmful on its own . rather , it is important whether the real or imaginary part of the fermion determinant is positive semidefinite or not . even though the determinant evaluated for a certain @xmath40 is a complex number , the functional integral over @xmath40 amounts to a real value for physical observables . it is understood in view of the relation , @xmath41 \\ = & \bigl\{\det\bigl[\gamma_\mu d^\mu+\gamma_4\mu_q + m_q\bigr]\bigr\}^\ast \ , . \end{split } \label{eq : cconj}\ ] ] we see clearly that the real ( imaginary ) part of the determinant is @xmath34-even ( @xmath34-odd ) . for a @xmath34-even ( @xmath34-odd ) observable , thus , the imaginary ( real ) part of the determinant vanishes after integration over @xmath40 . accordingly the genuine problem stems from that the real or imaginary part of the determinant may change its sign depending on the configuration @xmath40 . we will analyze a simple model to see the sign problem occurring in the mean - field level . as a practice to study the model , we will make the mean - field approximation in the su(2 ) case for which we do not have to face the sign problem . we will then observe the sign problem in the su(3 ) calculation and attempt the phase reweighting method to deal with the complex phase of the fermion determinant . we will closely analyze a lattice model with dense heavy quarks @xcite . let us consider qcd in the limit of @xmath42 so that we can drop antiquarks . we shall simultaneously take another limit of @xmath43 which renders all quarks static . under such limits we can evaluate the staggered fermion determinant exactly to reach the fermion action , @xmath44 } \equiv & \det\bigl[\gamma_\mu d^\mu+\gamma_4\mu_q + m_q\bigr]\notag\\ \to & \bigl[\det(1+\epsilon l)\bigr]^{{n_{\text{f}}}/4 } \,,\end{aligned}\ ] ] where @xmath45 is the number of flavors . not to go into subtlety of the flavor counting inherent to the staggered formalism , which is not of our interest , we set @xmath46 throughout this paper . we take those two limits in a way characterized by the parameter @xmath47 ranging from zero to infinity ; @xmath48 we will often call this model parameter as the `` density '' parameter because @xmath47 has strong correlation to the quark number density as seen in fig . [ fig : edep_density_su2 ] for the su(2 ) case and fig . [ fig : edep_density_su3 ] for the su(3 ) case . here @xmath49 is the number of the lattice sites in the temporal direction , that is , the inverse temperature @xmath50 is given by @xmath51 with the lattice spacing @xmath52 . in this model quarks are allowed to propagate only in the positive temporal direction . each time a quark with mass @xmath4 travels by one temporal lattice , it picks up the hopping parameter @xmath53 and the gauge invariant chemical potential factor @xmath54 @xcite . after @xmath49 hops , a quark winds around the temporal circle . it results in the weight @xmath47 for the quark excitation represented by the polyakov loop @xmath55 in the fundamental representation , @xmath56 \,.\end{aligned}\ ] ] the first line is the expression of the polyakov loop in terms of link variables on the lattice and second is in the continuum . it does not matter whichever expression we use since the su(@xmath57 ) matrix @xmath55 plays the role of the dynamical variable in our model and we will not return to its definition . the calculation of the determinant in color space is explicitly doable . after all we have the fermion action , @xmath58 } = \prod_{\vec{x}}\bigl[1+\epsilon^2 + 2\epsilon\ell\bigr ] \label{eq : sf_su2}\ ] ] in the color @xmath37 case ( and @xmath46 is implicit as we already noted ) and @xmath58 } = \prod_{\vec{x}}\bigl[1+\epsilon^3 + 3\epsilon\ell+3\epsilon^2\ell^\ast\bigr ] \label{eq : sf_su3}\ ] ] in the color @xmath59 case . in the above expressions we employed the traced polyakov loop defined as @xmath60 where @xmath61 is taken in fundamental color space . we remark that the @xmath34-transformation changes @xmath62 to @xmath63 and vice versa , where @xmath62 is real for @xmath37 and generally complex for @xmath64 . it is obvious that the @xmath37 determinant ( [ eq : sf_su2 ] ) is real and positive semidefinite for any @xmath47 because @xmath65 , while the @xmath59 expression ( [ eq : sf_su3 ] ) suffers the sign problem for nonzero @xmath47 ; the determinant can be complex except when either @xmath66 ( zero density ) , @xmath67 ( half - filling ) , or @xmath68 ( full - filling ) . we can intuitively understand why the su(3 ) determinant becomes real for @xmath67 . one - quark excitation represented by @xmath62 and two - quark excitation that is equivalent to one - antiquark excitation represented by @xmath63 occur with the common weight @xmath69 because of half - filling ( see @xmath70 at @xmath67 in figs . [ fig : edep_density_su2 ] and [ fig : edep_density_su3 ] ) . therefore , in effect , the system has equality in number of quarks and ( effective ) antiquarks just like in the zero density case . it is not an escape from the sign problem , however . in the @xmath59 case @xmath71 can take a value ranging from @xmath72 to @xmath73 , and thus the determinant at @xmath67 , i.e.@xmath74 is real but can be negative for @xmath75 . we note here one more important feature of the model . the fermion determinant is invariant under the duality transformation @xcite , @xmath76 by which it is sufficient for us to investigate the model in the region @xmath77 $ ] and the outer region @xmath78 $ ] can be deduced by means of the duality . regarding the gluodynamics , we assume the nearest neighbor interaction between the polyakov loops , @xmath79=-{n_{\text{c}}}^2 j\sum_{\text{n.n . } } \ell(\vec{x})\ , \ell^\ast(\vec{y } ) . \label{eq : sg}\ ] ] this action looks pretty simple and still reproduces the fundamental nature of the phase transition in the pure gluonic sector ; the action ( [ eq : sg ] ) leads to a second - order phase transition for the su(2 ) case and a first - order phase transition for the su(3 ) case @xcite , which is in agreement with the lattice qcd simulation and the theoretical expectation from center symmetry @xcite . one can interpret @xmath80 as a model parameter specifying the `` temperature '' of the system . in the strong coupling expansion , in fact , @xmath80 is related to @xmath24 through @xmath81 $ ] where @xmath39 is the string tension . in this work we shall leave @xmath80 as a model parameter as it is , for we do not want to introduce any further modeling into our analyses . in other words , our aim is to form a model not to imitate qcd itself but to mimic the qcd sign problem . the effective action that defines our model is eventually given by @xmath82={s_{\text{g}}}[l]+{s_{\text{f}}}[l]\ ] ] with the `` density '' parameter @xmath47 contained in @xmath83 $ ] and the `` temperature '' parameter @xmath80 in @xmath84 $ ] . in the finite - temperature field theory the free energy is evaluated in the functional integral form like the effective action as @xmath85 } \;.\ ] ] we make use of the mean - field technique to approximate the free energy . our ansatz for the mean - field action is @xcite @xmath86 \equiv -\frac{x}{2}\sum_{\vec{x}}\bigl[\ell(\vec{x } ) + \ell^\ast(\vec{x})\bigr ] \;. \label{eq : ansatz}\ ] ] in case of pure gluonic theories , the mean - field @xmath87 is simply proportional to the polyakov loop expectation value ; @xmath88 . in the presence of fermionic contributions , however , there is no simple relation between them . besides , @xmath89 and @xmath90 in the su(3 ) case have different dependence on @xmath1 . one might think that there should be two independent mean - fields to deal with differing @xmath89 and @xmath90 . this idea has much to do with the sign problem actually , and we will come back to this point later . then the mean - field free energy can be estimated as @xmath91-{s_{\text{mf}}}[l ] \rangle_{\text{mf } } -\ln\!\int{\mathcal{d}}l\,e^{-{s_{\text{mf}}}[l ] } , \label{eq : free_energy}\ ] ] where the average @xmath92 is taken by the mean - field action @xmath93 $ ] . roughly speaking , the first part corresponds to the internal energy and the logarithmic part is the entropy . we fix @xmath87 so as to minimize @xmath94 . once @xmath93 $ ] is known with @xmath87 determined , the expectation value of any observable @xmath95 $ ] as a function of the polyakov loop can be estimated by the group integration over @xmath55 with the mean - field action @xmath93 $ ] ; @xmath96\rangle \simeq \langle\mathcal{o}[l ] \rangle_{\text{mf } } \equiv \frac{\displaystyle\int\!dl\ , \mathcal{o}[l]\,e^{-{s_{\text{mf}}}[l]}}{\displaystyle \int\!dl\ , e^{-{s_{\text{mf}}}[l ] } } \,.\ ] ] for instance , the quark number density per color degrees of freedom is available by calculating @xmath97 one can calculate the polyakov loop susceptibility @xmath98 in the same way which reflects information of the deconfinement phase transition . if one directly uses @xmath95=\ell^2 $ ] in the mean - field approximation , however , nothing becomes singular at the critical point unless the fluctuation of @xmath62 is taken into account . equivalently one can estimate the susceptibility from the inverse curvature of the effective potential because the one - loop fluctuation leads to the trace of propagator for @xmath99 , that is , the inverse screening mass . since @xmath89 and @xmath90 are uniquely determined given @xmath87 is fixed by the free energy ( [ eq : free_energy ] ) , one can regard @xmath94 as a function of @xmath89 or @xmath90 . the polyakov loop susceptibility is then @xmath100 where we used @xmath101 . of course , the susceptibility defined in terms of @xmath63 is available with @xmath102 replaced by @xmath103 . the difference thus lies only in the nonsingular coefficient we are not interested in . for our purpose it is rather convenient to focus on the singular part alone discarding the difference of @xmath89 and @xmath90 . hence , we define the susceptibility as @xmath104 for presenting our numerical results . now that we finish explaining our approximations and computational procedures , let us step forward to the model analysis . we consider the model in the color su(2 ) case first , as we mentioned , to see how the mean - field approximation works apart from the sign problem . and the quark number density per lattice site ( i.e. density in the unit of the lattice spacing ) divided by @xmath37 . it is obvious from the figure that @xmath66 is the zero - density ( @xmath105 ) state and @xmath67 is the half - filling ( @xmath70 ) one.,width=283 ] as a function of the coupling @xmath80 at various density parameters ; @xmath1060 ( solid ) , 0.1 ( dotted ) , 0.2 ( short - dashed ) , 0.5 ( dashed ) , and 1.0 ( dotted - dashed ) . the second - order phase transition at @xmath66 occurs at @xmath107 . the transitional behavior is gradually smeared by the center symmetry breaking terms in the fermion determinant as @xmath47 grows larger.,width=283 ] as a function of the density parameter @xmath47 at various temperature parameters ; @xmath80=0 ( solid ) , 0.05 ( short - dashed ) , 0.1 ( dashed ) , and 0.2 ( dotted - dashed ) . the polyakov loop becomes insensitive to @xmath47 as @xmath80 goes larger , which is consistent with fig . [ fig : jdep_su2 ] in which the results at various @xmath47 converge at large @xmath80.,width=283 ] figure [ fig : edep_density_su2 ] shows the results for the quark number density @xmath108 as a function of the density parameter @xmath47 using eq . ( [ eq : density ] ) . for @xmath109 the duality relation @xmath110 enables us to deduce the number density . we can immediately confirm from this plot and the duality relation that the density parameter specifies the quark number density uniquely which monotonously approaches unity as @xmath68 . it should be noted that the half - filling @xmath70 realizes at @xmath67 and there is no @xmath80 dependence at all then . let us look at the phase transition seen in the polyakov loop behavior with increasing @xmath80 . the deconfinement phase transition in the su(2 ) pure gluonic theory is known to be second - order belonging to the same universality class as the ising model @xcite . in our model at @xmath66 we have a continuous transition at @xmath111 as indicated by the solid curve in fig . [ fig : jdep_su2 ] . the presence of dynamical quarks acts on the polyakov loop variable as an external field breaking center symmetry . in fact , the results at nonzero @xmath47 in fig . [ fig : jdep_su2 ] are not of transition but of crossover . we plot the density dependence of the polyakov loop behavior in fig . [ fig : edep_su2 ] . the density effects generally tend to make the polyakov loop larger , and eventually , the polyakov loop becomes insensitive to the density in the large @xmath80 ( i.e. high temperature ) region . it is because both the temperature and the density break center symmetry spontaneously and explicitly , respectively , having the polyakov loop saturated . therefore , the @xmath80 dependence is less for larger @xmath47 and the @xmath47 dependence is less for larger @xmath80 . ( circle ) and @xmath112 ( square ) taken from fig . 1 in ref . @xcite.,width=283 ] ( circle ) and @xmath112 ( square ) taken from fig . 2 in ref . @xcite.,width=283 ] our mean - field outputs are to be compared with the lattice simulation in ref . @xcite ; our figs . [ fig : edep_density_su2 ] and [ fig : edep_su2 ] correspond to figs . 1 and 2 presented in ref . @xcite , respectively . we can not expect a quantitative coincidence because our ansatz for the pure gluonic action @xmath84 $ ] is only a crude approximation of qcd and besides we neglect the renormalization of the polyakov loop in the mean - field treatment . nevertheless , the agreement turns out to be surprisingly good beyond our expectation if we treat the model parameter @xmath80 as a fitting parameter incorporating the undetermined effect of the polyakov loop renormalization , which is implied by the ansatz ( [ eq : sg ] ) . in such a way , we can fix @xmath113 and @xmath114 to reproduce the su(2 ) polyakov loop _ only _ at @xmath67 for @xmath115 and @xmath112 , respectively . we would emphasize that we did _ not _ use the data of the polyakov loop at @xmath116 and not the data of the number density at all . nevertheless , as clearly seen from the comparisons in figs . [ fig : tom_fit_num ] and [ fig : tom_fit_p ] , our numerical results fit _ all _ of the lattice data pretty well . we can conclude from this observation that the main qcd corrections to our ansatz ( [ eq : sg ] ) could be absorbed into the renormalization of the coupling alone . we are now confident that the mean - field treatment is a fairly acceptable approximation for this type of problem . as a function of @xmath80 at various density parameters ; @xmath1060.0 ( solid ) , 0.1 ( short - dashed ) , 0.2 ( dashed ) , and 1.0 ( dotted - dashed ) . the @xmath66 result has a divergence at @xmath117 characteristic to the second - order phase transition.,width=283 ] finally let us check that the susceptibility ( [ eq : sus ] ) diverges at @xmath66 and @xmath117 . figure [ fig : jdep_sus_su2 ] shows the susceptibility as a function of @xmath80 at various @xmath47 . in the plot @xmath98 becomes greater with increasing @xmath80 because we did not include @xmath118 in our definition of @xmath98 in eq . ( [ eq : sus ] ) . it is intriguing to remark that the @xmath119 result is not really critical in view of @xmath98 , while the crossover at @xmath119 in fig . [ fig : jdep_su2 ] looks rather close to a phase transition . actually @xmath98 is a more informative quantity to judge how critical the crossover is in fact . in summary of the su(2 ) polyakov loop dynamics at finite temperature and density , we shall depict a three - dimensional plot of @xmath89 in fig . [ fig:3d_su2 ] as a function of the `` temperature '' @xmath80 and the `` density '' @xmath47 . we immediately see general tendency that @xmath89 grows up with increasing @xmath80 and @xmath47 . on the temperature parameter @xmath80 at various density parameters ; @xmath47=0.0 ( solid ) , 0.1 ( dotted ) , 0.2 ( short - dashed ) , 0.5 ( dashed ) , and 1.0 ( dotted - dashed).,width=283 ] before closing this subsection we will comment a bit on the adjoint polyakov loop whose definition is @xmath120 the adjoint polyakov loop is no longer an order parameter because the adjoint representation is not faithful by the center of the gauge group . this quantity has been closely argued in ref . @xcite and there still remain subtleties . not in the su(2 ) but in the su(3 ) case , it has been found that @xmath121 takes an infinitesimally small value in the confined phase . one possible explanation would be that the group integration over @xmath55 is important in the center symmetric regime where nonperturbative phenomena like color confinement are relevant ( see also ref . @xcite ) . as a matter of fact , the su(@xmath57 ) group integration over @xmath122 turns out to be zero , if we are allowed to disregard the dynamics . even with the dynamics taken into account within the mean - field approximation , as shown in fig . [ fig : jdep_ad_su2 ] , the argument should hold so @xmath123 in the low temperature side . however , the adjoint polyakov loop behavior in our study should be understood only up to a qualitative level . it is pointed out in ref . @xcite that the renormalization for @xmath122 is significant , which is not considered in our present treatment . in case of @xmath59 we have to tackle the sign problem . we are not capable of solving qcd exactly like the lattice simulation , so one might think that within the framework of approximations one is allowed to impose the mean - field ansatz ( [ eq : ansatz ] ) to get some results anyhow . the free energy after the integration over @xmath55 could be real in terms of @xmath87 . it seems to work at least as a rough estimate that is worth trying first . the serious flaw in such a simple strategy is that @xmath124 is inevitably concluded . it is , however , contradict to the lattice results @xcite and the model analyses @xcite where @xmath125 has been observed at finite density . if the mean - field ansatz ( [ eq : ansatz ] ) is extended to having two variables @xmath87 and @xmath126 in order to take account of the difference between @xmath89 and @xmath90 , the price to pay is that the mean - field free energy is not convex . we will revisit this issue latter . in any case , though the appearance might be unalike , the difficulty of the sign problem is conserved even in the mean - field approximation unless the difference @xmath125 is neglected . in our work we shall elucidate that the `` phase reweighting method '' is one way to resolve these difficulties . we should , however , note that the reweighting method in the present context is one _ approximation _ scheme unlike in the lattice simulation . that is , the reweighting method is expected to be precise if the number of configurations is infinitely large , and thus the lattice simulation with infinite number of configurations generated could provide us with the exact answer in principle , while the mean - field approximation picking up only the most dominant configuration can not . the point of the method is that we decompose the fermion determinant into one part that gives the positive semidefinite probability and the other part that is regarded as the observable whose average is taken by configurations . the complex fermion determinant consists of the @xmath34-even magnitude and the @xmath34-odd phase . accordingly , the fermion action can be rewritten as @xmath127= { s_{\text{f}}^{\text{mag}}}[l]+i\theta[l ] \,,\ ] ] where @xmath128 & = -\sum_{\vec{x } } \ln\bigl|1+\epsilon^3 + 3\epsilon\ell+3\epsilon^2\ell^\ast\bigr| \ , , \\ \theta[l ] & = -\sum_{\vec{x}}\arg\bigl(1+\epsilon^3 + 3\epsilon\ell + 3\epsilon^2\ell^\ast\bigr ) \ , . \label{eq : phase}\end{aligned}\ ] ] with these definitions we approximate the expectation value of @xmath95 $ ] by the one obtained as follows ; @xmath96\rangle\simeq \frac{\displaystyle \bigl\langle\mathcal{o}[l]\,e^{-i\theta[l]}\bigr\rangle_{\text{mf } } } { \displaystyle \bigl\langle e^{-i\theta[l]}\bigr\rangle_{\text{mf } } } \;. \label{eq : rew}\ ] ] here @xmath93 $ ] or @xmath87 is fixed from the free energy with the action @xmath84+{s_{\text{f}}^{\text{mag}}}[l]$ ] , so that @xmath87 encompasses the information of @xmath129 $ ] implicitly . this scheme is the same as what has been adopted in the lattice simulation in ref . @xcite . here we would draw attention to a related work ; in ref . @xcite the correlation between @xmath130 and @xmath131 $ ] was investigated in the lattice simulation , which is apparent in our case from the expression ( [ eq : phase ] ) . and the quark number density @xmath108 per lattice site divided by @xmath59 . the gross feature is similar to fig . [ fig : edep_density_su2 ] ; @xmath66 is the zero density state ( @xmath105 ) and @xmath67 is the half - filling state ( @xmath70 ) in this su(3 ) case as well as in the su(2 ) case.,width=283 ] as a function of the temperature parameter @xmath80 at various density parameters ; @xmath1060 ( solid ) , 0.1 ( dotted ) , 0.2 ( short - dashed ) , 0.5 ( dashed ) , and 1.0 ( dotted - dashed ) . the first - order phase transition at @xmath66 and @xmath119 occurs at @xmath132 and @xmath133 , respectively.,width=283 ] let us start with checking the monotonous correlation between @xmath47 and @xmath108 as in the su(2 ) case . this allows us to regard increase ( or decrease ) in the density parameter @xmath47 as increase ( or decrease ) in the quark number density @xmath108 . the positive correlation is obvious from the results we show in fig . [ fig : edep_density_su3 ] . the `` temperature '' or @xmath80 dependence is slightly greater than the su(2 ) results . we can give a possible account for this as follows ; in the confined phase at small @xmath80 , the quark number density is suppressed as compared with high @xmath80 results . this suppression comes from the group integration that forces the thermally excited particles to be not quarks but ( nearly ) color - singlet baryons consisting of @xmath57 quarks @xcite . in general larger @xmath57 leads to stronger suppression by heavier excitation quanta . therefore , the stronger @xmath80 dependence presented in fig . [ fig : edep_density_su3 ] originates from the stronger suppression at small @xmath80 . the suppression is physically interpreted as _ effective _ tendency toward confinement @xcite . figure [ fig : jdep_su3 ] is the polyakov loop as a function of @xmath80 corresponding to the su(2 ) result in fig . [ fig : jdep_su2 ] and to be compared qualitatively with the lattice result of fig . 7 in ref . @xcite . we find a first - order phase transition for @xmath66 at @xmath134 and for @xmath119 at @xmath135 . the effect of nonzero @xmath47 smears the transitional behavior and the phase transition eventually ceases to be of first - order at a certain @xmath47 . the end - point of the first - order phase boundary is a second - order critical point called the critical end - point , which is of much interest in attempts to clarify the qcd phase diagram @xcite . we have crossover at larger @xmath47 . the global picture is well consistent with what has been already clarified in the potts system as a toy model of finite temperature and density qcd @xcite . as a function the density parameter @xmath47 at various temperature parameters ; @xmath80=0 ( solid ) , 0.05 ( dotted ) , 0.10 ( short - dashed ) , 0.13 ( dashed ) , and 0.2 ( dotted - dashed ) . the gross feature is similar to the su(2 ) case in fig . [ fig : edep_su2].,width=283 ] we plot the `` density '' dependence of the su(3 ) polyakov loop in fig . [ fig : edep_su3 ] which is the su(3 ) counterpart of fig . [ fig : edep_su2 ] . the su(2 ) and su(3 ) results are qualitatively similar except for that the polyakov loop is suppressed at small @xmath80 and @xmath47 just as we found in the quark number density . it would be interesting if we could compare our results with the lattice data , but unfortunately , the su(3 ) data as a function of @xmath47 is not available from ref . we can not argue the @xmath80 ( or @xmath136 ) dependence because it involves unknown renormalization effects . and @xmath90 as a function of the temperature parameter @xmath80 at various density parameters ; @xmath1060 ( solid ) which is zero entirely , 0.1 ( dotted ) , 0.2 ( short - dashed ) , and 0.5 ( dashed).,width=283 ] and @xmath90 as a function of the density parameter @xmath47 at various temperature parameters ; @xmath1370 ( solid ) , 0.05 ( short - dashed ) , 0.1 ( dashed ) , and 0.2 ( dotted - dashed ) . the thin curves represent the results from the taylor expansion method in the confine phase at @xmath138 and in the deconfined phase at @xmath139 which is almost overlaid on the result from the phase reweighting.,width=294 ] in the phase reweighting calculation we can see how @xmath89 and @xmath90 become distinct at @xmath2 . the observable @xmath140 is @xmath34-odd and so the imaginary part of the fermion determinant is responsible for a nonvanishing difference . when @xmath47 is small in the fermionic determinant ( [ eq : sf_su3 ] ) the imaginary part comes from @xmath141 . consequently the expectation value of the difference is proportional to @xmath142 where @xmath143 is taken at zero density @xcite . in fig . [ fig : diff ] we present our numerical results for the difference @xmath144 as a function of @xmath80 . the difference is trivially zero at @xmath66 and @xmath67 where the fermion determinant is real . as long as the density parameter stays smaller than @xmath145 , a larger density parameter @xmath47 leads to a bigger difference . for example , we find the difference at @xmath146 as large as @xmath147 which is comparable to @xmath148 . one can intuitively understand why @xmath90 is greater than @xmath89 at nonzero @xmath1 , which agrees with what has been observed in the lattice simulation @xcite . it is because , as discussed also in ref . @xcite , the presence of quarks at finite density enhances the screening effect for antiquarks so that the antiquark excitation costs less energy . the dense - heavy model , or @xmath149 , originally takes a form of power series in @xmath47 as seen in the expression ( [ eq : sf_su3 ] ) . we have thus performed the taylor expansion also to estimate the polyakov loop difference . the expectation value of the coefficient of the @xmath47 series is calculated at zero density @xmath66 . in this static model the fermionic contribution is just vanishing at @xmath66 , and so the taylor expansion method gives rise to the identical output for any @xmath138 in the confined phase where @xmath87 remains zero . in the deconfined phase @xmath150 the @xmath80 dependence is brought in by nonzero @xmath87 . figure [ fig : edep_diff ] shows the difference @xmath144 as a function of @xmath47 at various @xmath80 with the taylor expansion results in the confined and deconfined phase . we can see excellent agreement between two methods from the comparison at @xmath151 and @xmath152 . however , the taylor expansion can not reproduce the results at @xmath153 at all . the lesson from our model study is that the taylor expansion in terms of _ density _ breaks down in the presence of the first - order phase transition with respect to _ temperature_. still , in the deconfined phase at high temperature , the expansion is validated . we do not think that this finding is trivial in our model . in the realistic qcd lattice simulation with @xmath154 flavors , the zero density result is most likely crossover , and thus the taylor expansion method should be reliable at all temperatures . as we showed in the previous subsection , we shall present the three - dimensional plot of the su(3 ) polyakov loop as a function of the temperature parameter @xmath80 and the density parameter @xmath47 in fig . [ fig:3d_su3 ] . the figure is qualitatively consistent with the lattice result shown in fig . 5 in ref . @xcite . as a function of @xmath80 at various density parameters ; @xmath155 ( solid ) , 0.1 ( short - dashed ) , 0.2 ( dashed ) , and 1.0 ( dotted - dashed).,width=283 ] as a function of the temperature parameter @xmath80 at various density parameters ; @xmath47=0.0 ( solid ) , 0.1 ( dotted ) , 0.2 ( short - dashed ) , 0.5 ( dashed ) , and 1.0 ( dotted - dashed).,width=283 ] we shall turn to the susceptibility @xmath98 to examine the phase transition more closely . because the phase transition at small @xmath47 is of first - order , @xmath98 jumps discontinuously at @xmath117 as seen in fig . [ fig : jdep_sus_su3 ] . the susceptibility @xmath98 grows as the model parameters approach the critical end - point where the second - order phase transition takes place . figure [ fig : jdep_sus_su3 ] implies that the zero density ( @xmath66 ) result is significantly affected by the critical end - point which should be located nearby at small @xmath47 . in view of figs . [ fig : jdep_sus_su2 ] and [ fig : jdep_sus_su3 ] the critical region in the su(3 ) case is wider than the su(2 ) case . let us comment on the adjoint polyakov loop here again . as we mentioned in discussions on the su(2 ) results , the adjoint polyakov loop should require proper renormalization beyond the mean - field treatment . we show the adjoint polyakov loop result in fig . [ fig : jdep_ad_su3 ] just because we can do that . the first - order phase transition is located at the same point ; @xmath156 , of course . the group integration over @xmath55 suppresses @xmath121 at low @xmath80 even for large @xmath47 , which makes the crossover in fig . [ fig : jdep_ad_su3 ] look shaper than in case of the fundamental polyakov loop in fig . [ fig : jdep_su3 ] . per lattice site as a function of the density parameter @xmath47 at various temperature parameters ; @xmath1370 ( solid ) , 0.05 ( short - dashed ) , 0.10 ( dashed ) , and 0.20 ( dotted - dashed).,width=283 ] finally we present the results for the expectation value of the phase factor of the fermionic determinant , @xmath157 . we plot @xmath158 as a function of @xmath47 in fig . [ fig : edep_phase ] , where @xmath159 is the phase at each lattice site ; @xmath160 ( i.e. @xmath161 . ) comparing it with fig . 9 in ref . @xcite , we can surely check that our results qualitatively reproduce the lattice data . for more quantitative arguments , let us put the volume @xmath162 of the lattice simulation in ref . @xcite into our results . the expected phase factor @xmath163 can be , as a rough estimate , approximated as @xmath164 . for instance , our @xmath165 result has a minimum at @xmath166 where @xmath167 , and we get @xmath168 . the minimum value in fig . 9 in ref . @xcite is read as of order @xmath169 , which is not quite far from our estimate viewed on the logarithmic plot . here we will pursue another strategy to deal with the difference between @xmath89 and @xmath90 as a double - check . it is possible to extend the mean - field ansatz as @xmath86 = -\frac{x}{2}\sum_{\vec{x}}\bigl[\ell(\vec{x } ) \!+\!\ell^\ast(\vec{x})\bigr ] -\frac{y}{2}\sum_{\vec{x}}\bigl [ \ell(\vec{x})\!-\!\ell^\ast(\vec{x})\bigr ] \ , , \label{eq : ansatz2}\ ] ] and then we can compute the mean - field free energy from eq . ( [ eq : free_energy ] ) as a function of @xmath87 and @xmath126 . after the integration over @xmath55 , the free energy @xmath170 is a real function of real variables @xmath87 and @xmath126 . we can thus fix the mean - fields by @xmath171 nonzero @xmath172 appears in the presence of nonzero @xmath1 . it turns out that around @xmath173 the free energy @xmath170 has a minimum in the @xmath87 direction , while it has a maximum in the @xmath126 direction . that is , the solution to eq . ( [ eq : mean_field_eq ] ) is a saddle - point of @xmath170 , which is consistent with ref . the instability with respect to @xmath126 should be a remnant of the sign problem might be that @xmath140 is pure imaginary though @xmath126 corresponding to @xmath144 is real . ] we here point out that this instability also exists in the zero density limit ; @xmath170 has a saddle - point at @xmath174 even at zero density . it should be instructive to take a closer look at how the saddle - point arises even at zero density . if we expand the free energy in terms of @xmath89 and @xmath90 , the leading term dependent on @xmath89 and @xmath90 is quadratic @xmath175 as explicitly seen in a simple estimate in ref . this form of the free energy implies an instability inducing @xmath125 because it can written as @xmath176 . the situation is somewhat analogous to thermodynamics in the finite density njl - model calculation . if the free energy @xmath177 is calculated as a function of the renormalized chemical potential @xmath1 @xcite , then the value of @xmath1 is fixed by the condition @xmath178 which corresponds to not a minimum but a maximum of @xmath177 as a function of @xmath1 . this is not problematic , however , because the condition @xmath179 means a _ constraint _ equation of number density . likewise , we might as well think that the determination of @xmath126 in the polyakov loop dynamics is not energetic but the second equation of ( [ eq : mean_field_eq ] ) has something to do with a constraint equation of number density of gauge charge , namely , the gauss law constraint . if this conjecture is the case , though the rigorous proof is beyond our current scope , the saddle - point nature of the free energy is no longer an obstacle to fix @xmath87 and @xmath126 . estimated by the saddle - point of the mean - field free energy at various densities ; @xmath1060.1 ( solid and dashed ) and 0.5 ( dotted and dotted - dashed).,width=283 ] let us see what comes out if we calculate the polyakov loop using the ansatz ( [ eq : ansatz2 ] ) with the solution to eq . ( [ eq : mean_field_eq ] ) assuming that the second equation of ( [ eq : mean_field_eq ] ) stems from a constraint associated with gauge dynamics . in fig . [ fig : diff_comp ] we show the difference @xmath144 as a function of @xmath80 . the solid and dotted curves at @xmath119 and @xmath146 respectively are just the same as already presented in fig . [ fig : diff ] . the dashed and dotted - dashed curves are the counterparts derived from the mean - field energy with @xmath87 and @xmath126 . in view of two @xmath146 curves , on the one hand , the mean - field result is entirely consistent with what we found in the phase reweighting method . on the other hand , the coincidence is not very good for the @xmath119 results except for @xmath180 and @xmath150 . this discrepancy comes from a singularity of the dense - heavy model located at @xmath66 ; when @xmath47 is small , the fermion action in the dense - heavy model is approximated as @xmath83\sim-3\epsilon\sum_{\vec{x}}\ell(\vec{x})$ ] . hence , nonzero @xmath47 tends to align the vacuum into the direction of @xmath181 ( see eq . ( [ eq : ansatz2 ] ) ) , and the model does not reduce to a pure gluonic theory with @xmath174 smoothly in the limit of @xmath182 . this situation makes a sharp contrast to finite density qcd . in the strong coupling expansion , for a simple example , the fermion action is @xmath83\sim h\sum_{\vec{x}}\bigl[\ell(\vec{x})e^{\mu_q / t}+\ell^\ast ( \vec{x})e^{-\mu_q / t}\bigr]$ ] . in the limit of @xmath183 , therefore , the fermionic action acts as an external field toward @xmath184 and @xmath174 , so that the vacuum alignment of a pure gluonic theory is smoothly retrieved in the @xmath183 and @xmath185 limit . apart from the discrepancy inherent to the singular behavior of the dense - heavy model near @xmath66 , the mean - field free energy leads to the results in accord with the phase reweighting method . this is an indirect evidence for that the saddle - point is not harmful actually , as we conjectured . we shall comment on a possible clue to resolve the sign problem based on what we have seen here . we changed the sign problem into a form of the saddle - point of the mean - field free energy . the saddle - point appears harmless from our analyses , presumably stabilized by the gauss law , and then the sign problem is resolved . of course , it is highly nontrivial how to map this analytical procedure to the lattice qcd simulation . still , we would point out that the clustering method developed in ref . @xcite is a philosophically similar idea along this line ; partial integration over the cluster domains wipes away the sign problem , just like seen in our mean - field free energy free from the sign problem after the group integration . the field - theoretical approach to reveal the relation between the saddle - point of the free energy and the gauss law is beyond our current scope , but it definitely deserves further investigation . we investigated the dense - heavy model and observed the sign problem at finite density within the framework of the mean - field approximation . we calculated the quark number density , the fundamental and adjoint polyakov loop , the susceptibility , and the phase of the fermion determinant as a function of the model parameters specifying the temperature and the density . all the mean - field results are reasonable and even in quantitative agreement with the lattice data . in the environment free from the sign problem in the su(2 ) case we found that the mean - field approximation goes better than expected . our mean - field results nicely reproduced the quark number density and the polyakov loop measured on the lattice once the unknown polyakov loop renormalization is fixed by fitting . then , we proceeded into the su(3 ) case where the sign problem is relevant . we saw that the approximation scheme which is capable of describing the different @xmath1 dependence of @xmath89 and @xmath90 can not avoid the sign problem even at the mean - field level . we applied the phase reweighting method as a practical resolution in order to handle the complex fermion determinant . we acquired @xmath144 as a function of the density parameter or the temperature parameter . so far , there are not many lattice data available for this quantity , but our results @xmath186 are consistent with other model studies as well as what has been obtained from the taylor expansion method on the lattice . the important message from our work is the following . the sign problem may be a serious obstacle even in the mean - field model studies at finite temperature and density . we would say , at least , that one should be careful enough when one deals with the polyakov loop behavior at finite density . the chiral effective models with the polyakov loop coupled are examples that definitely need more or less caution for application to the finite density problem . the phase reweighting method is one prescription in order to approximate the expectation value in a manageable way . this prescription is , however , only practical and not any solution to the sign problem . we would insist that one can not solve the qcd sign problem until one can at least find a solution to the sign problem appearing even in that simple model setting . the converse is not necessarily true , though . in the future , we would like to apply other ideas than the phase reweighting method for the sign problem at the mean - field level . the bottom line here is , thus , that we have formulated an analytical testing ground to think about the sign problem , and we believe that our clarification would be useful for further developments . the authors thank y. aoki , t. blum , ph . de forcrand , r. pisarski , c. schmidt for useful conversations . the authors specially thank s. ejiri for discussions on the dense - heavy model which urged us to initiate this work . this research was supported in part by riken bnl research center and the u.s . department of energy under cooperative research agreement # de - ac02 - 98ch10886 . for theoretical reviews on qcd in extreme environments , see , h. meyer - ortmanns , rev . phys . * 68 * , 473 ( 1996 ) [ arxiv : hep - lat/9608098 ] ; f. wilczek , arxiv : hep - ph/0003183 ; d. h. rischke , prog . phys . * 52 * , 197 ( 2004 ) [ arxiv : nucl - th/0305030 ] . for reviews on color superconductivity and the phase diagram of cold and dense quark matter , see , k. rajagopal and f. wilczek , arxiv : hep - ph/0011333 ; k. fukushima , arxiv : hep - ph/0510299 ; m. g. alford , arxiv : hep - lat/0610046 ; s. b. ruster , v. werth , m. buballa , i. a. shovkovy and d. h. rischke , arxiv : nucl - th/0602018 . for reviews on lattice qcd simulations , see , f. karsch , lect . notes phys . * 583 * , 209 ( 2002 ) [ arxiv : hep - lat/0106019 ] ; k. kanaya , nucl . a * 715 * , 233 ( 2003 ) [ arxiv : hep - ph/0209116 ] ; e. laermann and o. philipsen , ann . sci . * 53 * , 163 ( 2003 ) [ arxiv : hep - ph/0303042 ] ; c. schmidt and t. umeda , arxiv : hep - lat/0609032 . t. umeda , k. nomura and h. matsufuru , eur . * 39s1 * , 9 ( 2005 ) [ arxiv : hep - lat/0211003 ] ; m. asakawa and t. hatsuda , phys . rev . lett . * 92 * , 012001 ( 2004 ) [ arxiv : hep - lat/0308034 ] ; s. datta , f. karsch , p. petreczky and i. wetzorke , phys . 094507 ( 2004 ) [ arxiv : hep - lat/0312037 ] . for reviews on the sign problem , see , s. muroya , a. nakamura , c. nonaka and t. takaishi , prog . phys . * 110 * , 615 ( 2003 ) [ arxiv : hep - lat/0306031 ] ; m. p. lombardo , prog . suppl . * 153 * , 26 ( 2004 ) [ arxiv : hep - lat/0401021 ] ; s. aoki , int . j. mod . a * 21 * , 682 ( 2006 ) [ arxiv : hep - lat/0509068 ] . c. r. allton _ _ , phys . d * 66 * , 074507 ( 2002 ) [ arxiv : hep - lat/0204010 ] ; c. r. allton , s. ejiri , s. j. hands , o. kaczmarek , f. karsch , e. laermann and c. schmidt , phys . d * 68 * , 014507 ( 2003 ) [ arxiv : hep - lat/0305007 ] . m. g. alford , a. kapustin and f. wilczek , phys . rev . d * 59 * , 054502 ( 1999 ) [ arxiv : hep - lat/9807039 ] ; p. de forcrand and o. philipsen , nucl . b * 642 * , 290 ( 2002 ) [ arxiv : hep - lat/0205016 ] ; nucl . b * 673 * , 170 ( 2003 ) [ arxiv : hep - lat/0307020 ] ; [ arxiv : hep - lat/0607017 ] ; m. delia and m. p. lombardo , phys . rev . d * 67 * , 014505 ( 2003 ) [ arxiv : hep - lat/0209146 ] . n. weiss , phys . d * 24 * , 475 ( 1981 ) ; phys . d * 25 * , 2667 ( 1982 ) ; v. m. belyaev , phys . b * 241 * , 91 ( 1990 ) ; phys . b * 254 * , 153 ( 1991 ) ; c. p. korthals altes , nucl . b * 420 * , 637 ( 1994 ) [ arxiv : hep - th/9310195 ] ; k. fukushima and k. ohta , j. phys . g * 26 * , 1397 ( 2000 ) [ arxiv : hep - ph/0011108 ] ; e. megias , e. ruiz arriola and l. l. salcedo , phys . d * 74 * , 065005 ( 2006 ) [ arxiv : hep - ph/0412308 ] ; s. k. ghosh , t. k. mukherjee , m. g. mustafa and r. ray , phys . d * 73 * , 114007 ( 2006 ) [ arxiv : hep - ph/0603050 ] ; h. hansen , w. m. alberico , a. beraudo , a. molinari , m. nardi and c. ratti , arxiv : hep - ph/0609116 ; s. mukherjee , m. g. mustafa and r. ray , arxiv : hep - ph/0609249 . t. c. blum , j. e. hetrick and d. toussaint , phys . * 76 * , 1019 ( 1996 ) [ arxiv : hep - lat/9509002 ] . ( we note that the published version of this paper does not contain the su(2 ) results which are present in the preprint version . we refer only to the preprint version . ) e. dagotto , f. karsch and a. moreo , phys . b * 169 * , 421 ( 1986 ) ; j. u. klatke and k. h. mutter , nucl . b * 342 * , 764 ( 1990 ) ; s. hands , j. b. kogut , m. p. lombardo and s. e. morrison , nucl . b * 558 * , 327 ( 1999 ) [ arxiv : hep - lat/9902034 ] ; j. b. kogut , d. toublan and d. k. sinclair , phys . lett . b * 514 * , 77 ( 2001 ) [ arxiv : hep - lat/0104010 ] ; j. b. kogut , d. toublan and d. k. sinclair , nucl . b * 642 * , 181 ( 2002 ) [ arxiv : hep - lat/0205019 ] ; j. b. kogut , d. toublan and d. k. sinclair , phys . rev . d * 68 * , 054507 ( 2003 ) [ arxiv : hep - lat/0305003 ] ; s. muroya , a. nakamura and c. nonaka , phys . b * 551 * , 305 ( 2003 ) [ arxiv : hep - lat/0211010 ] ; b. vanderheyden and a. d. jackson , phys . d * 64 * , 074016 ( 2001 ) [ arxiv : hep - ph/0102064 ] ; j. b. kogut , m. a. stephanov and d. toublan , phys . b * 464 * , 183 ( 1999 ) [ arxiv : hep - ph/9906346 ] ; j. b. kogut , m. a. stephanov , d. toublan , j. j. m. verbaarschot and a. zhitnitsky , nucl . b * 582 * , 477 ( 2000 ) [ arxiv : hep - ph/0001171 ] ; y. nishida , k. fukushima and t. hatsuda , phys . rept . * 398 * , 281 ( 2004 ) [ arxiv : hep - ph/0306066 ] ; s. chandrasekharan and f. j. jiang , phys . d * 74 * , 014506 ( 2006 ) [ arxiv : hep - lat/0602031 ] ; b. alles , m. delia and m. p. lombardo , nucl . b * 752 * , 124 ( 2006 ) [ arxiv : hep - lat/0602022 ] ; s. hands , s. kim and j. i. skullerud , arxiv : hep - lat/0604004 . a. barducci , r. casalbuoni , s. de curtis , r. gatto and g. pettini , phys . b * 231 * , 463 ( 1989 ) ; f. wilczek , int . j. mod . phys . a * 7 * , 3911 ( 1992 ) [ erratum - ibid . a * 7 * , 6951 ( 1992 ) ] ; a. barducci , r. casalbuoni , g. pettini and r. gatto , phys . d * 49 * , 426 ( 1994 ) ; j. berges and k. rajagopal , nucl . b * 538 * , 215 ( 1999 ) [ arxiv : hep - ph/9804233 ] ; m. a. halasz , a. d. jackson , r. e. shrock , m. a. stephanov and j. j. m. verbaarschot , phys . d * 58 * , 096007 ( 1998 ) [ arxiv : hep - ph/9804290 ] ; m. a. stephanov , k. rajagopal and e. v. shuryak , phys . lett . * 81 * , 4816 ( 1998 ) [ arxiv : hep - ph/9806219 ] ; m. a. stephanov , k. rajagopal and e. v. shuryak , phys . rev . d * 60 * , 114028 ( 1999 ) [ arxiv : hep - ph/9903292 ] ; k. fukushima , phys . c * 67 * , 025203 ( 2003 ) [ arxiv : hep - ph/0209270 ] ; y. hatta and t. ikeda , phys . d * 67 * , 014028 ( 2003 ) [ arxiv : hep - ph/0210284 ] .

a model for pulsars with polar - cap magnetic flux density antiparallel with spin is described . it recognizes the significance of two elementary processes , proton production in electromagnetic showers and photoelectric transitions in ions accelerated through the blackbody radiation field , which must be present at the polar cap in the case , but not for pulsars of the opposite spin direction . the two populations are likely to be indistinguishable observationally until curvature radiation pair creation ceases to be possible . the model generates , and provides a physically realistic framework for , the polar - cap potential fluctuations and their time - scales that can produce mode - changes and nulls . the rrats are then no more than an extreme case of the more commonly observed nulls . the model is also able to support the basic features of subpulse drift and to some extent the null - memory phenomenon that is associated with it . unfortunately , it appears that the most important neutron - star parameter for quantitative predictive purposes is the whole - surface temperature , a quantity which is not readily observable at the neutron - star ages concerned . [ firstpage ] instabilities - plasma - stars : neutron - pulsars : general

1302.5260

the discovery within the last decade of further complex pulsar phenomena , particularly the rotating radio transients ( rrats ; mclaughlin et al 2006 ) , has extended the problem of understanding these systems by introducing new time - scales . in a number of previous papers ( jones 2010a , 2011 , 2012a , 2012b ; hereafter papers i - iv ) physical processes at the polar - cap surface in pulsars with spin @xmath1 and magnetic magnetic flux density @xmath0 such that @xmath2 have been examined to see if they are relevant . the processes that introduce important time - scales are those associated with the formation of electromagnetic showers by reverse - electrons from either electron - positron pair formation or from photoelectric transitions in accelerated ions . they are of no significance in @xmath4 pulsars with goldreich - julian charge density @xmath5 and outward electron acceleration . the work has assumed that the space - charge - limited flow ( sclf ) boundary condition @xmath6 is satisfied on the polar - cap surface at all times . the aim of these papers has been to determine the composition and energy distribution of the accelerated particle flux and to look for instabilities that might be relevant to the phenomena which are observed in the radio - frequency spectrum . these properties can then be compared with those needed to produce the observed emission spectra . this has a large bandwidth , of the order of @xmath7 hz , and efficient conversion of kinetic energy to radio frequencies . for example , the peak power of the brightest pulses in psr b0656 + 14 ( weltevrede et al 2006a ) , is of the order of @xmath8 erg s@xmath9 , which is equivalent to approximately @xmath10 mev per unit charge accelerated at the polar caps . growth of a collective mode able to transfer energy at this rate to the radiation field constrains the longitudinal effective mass , @xmath11 , of beam particles with mass @xmath12 and lorentz factor @xmath13 . for a secondary electron - positron pair plasma , it is well known that this requires a low energy , @xmath14 , which is also of the same order as that needed if coherent curvature radiation , in a dipole - field , were the source of the observed radiation . papers iii and iv showed that , under the sclf boundary condition , the creation of a reverse flux of electrons by photoelectric transitions in the accelerated ions limits the acceleration potential , analogously with the effect of electron - positron pair creation . thus the outward particle flux has two principal components : protons formed in the electromagnetic showers and ions with lorentz factors @xmath15 which are relativistic , but not ultra - relativistic as they would be in the absence of photoelectric transitions . they can have longitudinal effective masses that allow the rapid growth of the quasi - longitudinal langmuir mode considered by asseo , pelletier & sol ( 1990 ) . as noted by asseo et al , the quasi - longitudinal mode couples directly with the radiation field and so introduces , in principle , a second source of coherent radio - frequency emission which is not present in @xmath16 pulsars . electromagnetic shower theory ( see landau & rumer 1938 , nordheim & hebb 1939 ) makes it possible to calculate the total photon track length per unit interval of photon frequency . it is an almost linear function of primary electron energy in the ultra - relativistic limit . then known partial cross - sections for the formation and decay of the giant dipole resonance enable @xmath17 , the number of protons formed per unit incident electron energy , to be estimated with adequate reliability for the present work . proton formation is concentrated at shower depths of the order of @xmath18 radiation lengths and it was shown in paper i that the time @xmath19 for diffusion to the top of the polar - cap atmosphere introduces the likelihood of instability in the composition of the accelerated particle flux . estimates in paper ii , under more appropriate assumptions about the depth of the atmosphere , were that @xmath19 would be of the order of @xmath20 s. with this result in mind , it was evident that the photoelectric transitions considered in papers iii and iv would introduce large - scale fluctuations in the the acceleration potential above the polar cap . the time - scale @xmath19 is sufficiently long that the electric field is simply given by solving poisson s equation for the charge densities instantaneously present above the polar cap . it follows that the conditions necessary for coherent radio emission can appear or disappear with time - scales related to @xmath19 . before proceeding further , it is appropriate to place this paper in the context of recent studies of the pulsar magnetosphere based on computational techniques in numerical plasma kinetics . it is true that the inductance per unit length of the tube of open magnetic flux lines varies little with altitude above the polar cap so that the magnetosphere as a whole limits current - density time derivatives . the further hypothesis that the actual current densities are determined by the whole magnetosphere rather than by considerations specific to the small region immediately above the polar cap appeared in a paper by mestel et al ( 1985 ) and was later extended by beloborodov ( 2008 ) . solutions for the magnetic flux density external to a sphere , typically of radius @xmath21 , where @xmath22 is the light - cylinder radius , have been obtained by kalapotharakos & contopoulos ( 2009 ) and by bai & spitkovsky ( 2010 ) under the assumption of a force - free magnetosphere . current densities can then be derived from the @xmath0-distribution as functions of basic parameters ; the dipole moment , the rotation period @xmath23 , and the angle @xmath24 subtended by @xmath1 and @xmath0 . those found by bai & spitkovsky deviate considerably from @xmath25 , the goldreich - julian value , and were followed by one - dimensional time - dependent numerical studies of the polar cap with the @xmath26 surface boundary condition ( timokhin 2010 ) and later with the sclf boundary condition ( chen & beloborodov 2013 , timokhin & arons 2013 ) assuming fixed values of the time - averaged current density that are either larger than @xmath27 or of the opposite sign . these produced evidence for the development of high potential differences and microsecond bursts of pair creation . all authors assumed the case @xmath28 which is not the subject of the present paper . some care will be needed in translating the results obtained by chen & beloborodov and by timokhin & arons to the @xmath2 case . for this , the basic unit of length in the description of the one - dimensional current and charge density , derived from the plasma frequency , is almost two orders of magnitude larger than for electrons . the one - dimensional potential must form a continuous function with the three - dimensional potential derived from the lense - thirring effect ( see muslimov & tsygan 1992 ) . in this case , owing to the larger scale - length , there is no possibility of the particle backflow noted by beloborodov ( 2008 ) and chen & beloborodov ( 2013 ) so that only the stable zero - temperature solution they refer to is realistic . the present paper does not accept the assumption of a precisely force - free magnetosphere on which the computational work is based . instead , it retains the assumption that the current density is determined in the polar - cap region and that it flows toward the light cylinder through a magnetosphere that is not precisely force - free . paper iv attempted to describe , qualitatively , the relevance of physical processes at the surface of the polar cap to the complex phenomena that are observed in radio pulsars , but a number of numerical estimates made earlier , in papers i and ii , have been superseded by later work . thus the aim of the present paper is to introduce an elementary mathematical model of the polar cap which embodies those results obtained in papers i - iv that remain relevant . the model itself and the information which has been derived from it are described in section 3 . section 2 summarizes the physical properties of the polar - cap surface that are essential for its construction and also considers the non - static aspects of charge - to - mass ratio fractionation of atmospheric composition which were neglected in the earlier papers . it is the scale of the acceleration potential fluctuations which is the most significant feature displayed by the model . we relate this with the observed properties of the rrats and nulls in section 4 and with the phenomena of subpulse drift and null memory in section 5 . our conclusions are summarized in section 6 along with a brief discussion of observed phenomena on which , at the present time , our model appears to have no impact . it is assumed , unlike papers i and ii , that the actual polar - cap magnetic flux density is of the same order of magnitude as that inferred from the spin - down rate , which is usually less than the critical field @xmath29 g. thus the ion separation energy ( medin & lai 2006 ) is small enough for the mass of the atmosphere at a polar - cap temperature @xmath30 k to be significant . this atmosphere is very compact and in local thermodynamic equilibrium ( lte ) ; its scale height is of the order of @xmath31 cm . the total mass is poorly known owing to its exponential dependence on the ion separation energy , but is possibly in an interval equivalent to @xmath31 to @xmath32 radiation lengths . thus it may contain the whole or some part of the electromagnetic showers formed by inward accelerated electrons . the extent of a shower is itself uncertain because the landau - pomeranchuk - migdal effect is present at the energies , densities and magnetic fields concerned ( see jones 2010b ) . at depths immediately below the atmosphere , the state of matter is uncertain and could be either liquid or solid ( see paper ii ) . the nuclear charge is not known , but we assume the canonical value @xmath33 . electromagnetic showers reduce this to a mean @xmath34 and mass number @xmath35 at the top of the atmosphere with an lte ion charge @xmath36 . there is a fractionation of ion charge - to - mass ratio with the largest values at the top of the atmosphere . the relatively small number of protons created in showers are nowhere in equilibrium within the lte ion atmosphere and , under the influence of the small electric field @xmath37 present , move outward and are either accelerated or , if their flux exceeds the goldreich - julian current density @xmath27 , form an atmosphere above the ions . its scale height is @xmath38 , at local temperature @xmath39 and gravitational acceleration @xmath40 . the chemical potential gradient that causes their motion within the ion sector of the atmosphere is initially mostly an entropy gradient , but at lower densities changes to that derived from the electric field . this fractionation is of central importance to the model described in this paper , but we have not previously considered the adequacy of our elementary static treatment of the problem . also , the presence of partial ionization means that the possibility of convective instability has to be considered ( for a simple explanation see , for example , rast 2001 ) . bearing in mind the functioning of a laboratory high - vacuum diffusion pump , it is also necessary to ask if the upward flux of protons is likely to carry with it sufficient ions to interfere with fractionation . in order to examine this , we require two transport relaxation times . we employ approximate expressions using lowest - order zero - field perturbation theory , satisfactory here because the proton cyclotron energy quantum is small compared with the polar - cap thermal energy @xmath41 . the first , for upward movement of a proton in the ion atmosphere at ion number density @xmath42 is , @xmath43 in which the function @xmath44 is , @xmath45 here , @xmath46 is the debye wavenumber for the ions , @xmath47 and , @xmath48 the equivalent ion relaxation time for movement relative to the proton atmosphere above the ions , if it exists , is also required . at proton number density @xmath49 it is , @xmath50 in which @xmath51 is given by equation ( 2 ) with the substitutions , @xmath52 it is typically several orders of magnitude longer than @xmath53 . the proton drift time from a depth @xmath54 with ion number density @xmath42 to the top of the atmosphere is , @xmath55 under the influence of a chemical potential gradient fixed by the ions , @xmath56 equivalent to an upward directed force . the upward velocity is given by , @xmath57 and the scale height of the atmosphere is , @xmath58 for @xmath59 , @xmath60 , @xmath61 cm s@xmath62 and @xmath63 k. evaluation for @xmath64 @xmath65 gives @xmath66 s@xmath9 and a drift time of @xmath67 s. this should be regarded only as a tentative order of magnitude because equations ( 1 ) , ( 3 ) and ( 4 ) are strictly valid only at number densities much lower than @xmath68 @xmath65 . the position of the change of phase from gas to liquid or solid is also uncertain . the proton sector number densities are also sufficient for it to form an lte atmosphere , but during phases of proton emission into the magnetosphere it has the goldreich - julian outward flux . thus as a first approximation , its kinetic distribution function is isotropic in a frame of reference moving outward with a velocity @xmath69 until the proton number density @xmath49 becomes so small that the lte condition breaks down . if this is not to interfere with fractionation , the force exerted on an ion by the upward flux of protons must be small compared with its chemical potential gradient in the static proton lte atmosphere , which is , @xmath70 a proton atmosphere able to maintain a typical goldreich - julian flux for one second would have a density at base of @xmath71 @xmath65 at which its ion transport relaxation time would be @xmath72 s@xmath9 giving a force several orders of magnitude smaller than equation ( 10 ) and of small effect on fractionation . the existence of convective instability at any density within the atmosphere is unlikely because the lateral motion implicit in a convective cell is strongly suppressed by magnetic fields of the order of @xmath73 g. we refer to miralles , urpin & van riper ( 1997 ) for a full treatment of this problem . but even if convective cells exist , the presumption must be that the electric field @xmath37 , reflecting the internal equilibrium of the adiabatically - moving volume , is still present within it . thus the velocity of the protons , averaged over many circulations , would remain as calculated above and there would be no interference with fractionation . the polar - cap radius , as in papers iii and iv , denotes the division between open and closed magnetic flux lines , and is that given by harding & muslimov ( 2001 ) , @xmath74 where @xmath23 is the rotation period . we assume a neutron - star mass @xmath75 and radius @xmath76 cm , for which @xmath77 . our approximation for the electrostatic potential @xmath78 is based on the lense - thirring effect described by muslimov & tsygan ( 1992 ) , @xmath79 in cylindrical polar coordinates , at altitude @xmath80 and radius @xmath81 , for the specific case of a charge density @xmath82 that is independent of @xmath83 and is only a slowly varying function of @xmath80 . ( it is almost identical with the potential that would be present given a time - independent outward flow of electrons under sclf boundary conditions in @xmath28 pulsars . ) it also assumes approximate forms , valid at altitudes well within the light - cylinder radius , for the more precise charge densities which were given by harding & muslimov ( 2001 ) . these are independent of @xmath84 as required by equation ( 12 ) and are , @xmath85 and @xmath86 in which @xmath24 is the angle between @xmath1 and @xmath0 , @xmath87 is the dimensionless lense - thirring factor , and @xmath88 . equation ( 12 ) is certainly a satisfactory approximation to the true potential at altitudes @xmath89 , but for lower values of @xmath80 , we must assume that the h - m potential changes continuously to the one - dimensional potential that exists at @xmath90 , which was originally described by michel ( 1974 ) and then further investigated by mestel et al ( 1985 ) and by beloborodov ( 2008 ) . we have already noted , in section 1 , that the length scale associated with the form of the one - dimensional potential is large , in the @xmath2 case . thus equation ( 12 ) is a fair representation of the true potential , which would remove any possibility of the backflow described by beloborodov . under the assumption that equation ( 12 ) is valid , the condition @xmath91 ensures that the sclf condition @xmath92 is satisfied at all @xmath83 on the polar - cap surface . the maximum potential available for acceleration above the polar cap occurs if @xmath82 is independent of altitude as would be the case if the particles were ultra - relativistic and there were no charge - separating interactions . expressed in convenient energy units , it is , @xmath93 per unit charge on a flux line with radial coordinate @xmath83 . the blackbody temperatures responsible for photoelectric transitions are those of the polar cap , @xmath94 , and of the whole surface , @xmath3 . the polar cap is not significant at altitudes @xmath95 because the photon lorentz transformation to the ion rest frame becomes unfavourable . there is a little ionization at @xmath96 to a mean charge @xmath97 but the ion lorentz factors there are small , given the sclf boundary conditions , as is the contribution to the total reverse - electron energy per ion . we assume a fixed value for this component of @xmath98 gev . lorentz transformations of whole - surface photons are much more favourable at higher altitudes and produce the major part of the total reverse - electron energy per ion accelerated , @xmath99 . ionization may or may not be complete and we define the mean final charge as @xmath100 . we have shown previously that the flux of shower - produced protons reaching the top of the neutron - star atmosphere at any point @xmath83 is given by , @xmath101 in terms of the ion flux @xmath102 ( see equation ( 20 ) of paper iv ) . the first component @xmath103 can not exceed the goldreich - julian flux ; the remainder @xmath104 accumulates at the top of the atmosphere . paper i assumed the neutron - star atmosphere to be of negligible depth so that @xmath105 was assumed to be the standard diffusion function . but for the depth considered in paper iv and here , an expression based on the drift time given by equation ( 7 ) is a better approximation . if the proton atmosphere is exhausted , ion emission commences extremely rapidly in order to satisfy the sclf condition @xmath106 at the polar - cap surface . in general , the screening of any electric field in the @xmath2 case for which a positive charge density is needed occurs preferentially through ion or proton emission in a single relativistic - particle transit time , whereas the process of electron - positron pair multiplication requires many transit times . protons accelerated to @xmath107 have only a very small probability of creating electron - positron pairs through interaction with blackbody photons and , with ions of the same energy per unit charge , would have only negligible growth rates for the quasi - longitudinal langmuir mode . there are three ways in which particle beams capable of giving strong coherent radio - frequency emission might be produced . the two considered here in the first instance are as follows . \(a ) @xmath108 is so large that self - sustaining curvature - radiation ( cr ) electron - positron pair production occurs . a permanent proton atmosphere exists over much of the polar cap giving a primary current density of protons and positrons which , at least for times long compared with @xmath109 , may be in a steady state . a plasma of low - energy secondary electrons and positrons forms . \(b ) the potential is so reduced from @xmath108 by photo - electron backflow that ion and proton lorentz factors are either in a region that allows rapid growth of the quasi - longitudinal mode or are of magnitude such that downward fluctuations to the necessary values can occur . in this case , the lorentz factors are such that the mode wave - vector component perpendicular to @xmath0 is unlikely to be negligible so that coherent radio - frequency emission need not be exactly parallel with local flux lines . the third set of processes are more obscure in the case of @xmath110 pulsars and concern inverse compton scattering ( ics ) of blackbody photons above the polar cap . pair production by the conversion of outward - moving high - energy ics photons is known to be significant in @xmath28 pulsars ( see hibschman & arons 2001 , harding & muslimov 2002 ) even if dipole - field geometry is assumed . also , deviations from such a field , to the extent that they exist , can greatly enhance pair densities ( harding & muslimov 2011 ) . but in @xmath2 pulsars , high - energy ics photons are directed inward and the sources of outward - moving positrons to scatter blackbody photons are more obscure . clearly , a fraction of the photons may have sufficient momentum transverse to @xmath0 to produce pairs at low altitudes . a further possible source of low - altitude pairs is the conversion of neutron - capture @xmath111-rays at the top of the atmosphere . showers produce approximately the same numbers of neutrons and protons , but reliable calculation of the probability that a neutron would reach the top of the atmosphere is not easy . it is also not obvious how the density of electron - positron pair production via these processes would respond to fluctuations in the acceleration potential . however , although the model described in section 3 has been considered in the context of either low - energy ion - proton beams or cr pair production , it is necessary to bear the more obscure pair - production processes in mind . the previous section summarized those specific physical properties of the polar cap that are important for the model , but we shall also give a brief description of its framework . this is addressed primarily to the typical radio pulsar with an age of the order of @xmath112 myr , or older , and not to the small-@xmath23 case in which continuous pair production is anticipated . photoelectric transitions in accelerated ions are a source of electrons which flow inward to the polar - cap surface and partially screen the potential given by equation ( 12 ) . data presented in table 1 of paper iv , for specific values of @xmath113 and @xmath3 , gave @xmath100 , the final charge of an accelerated ion , and the energy of its reverse electrons @xmath114 , as functions of the acceleration potential @xmath115 experienced by that ion . then with the inclusion of the small energy @xmath116 arising from photoelectric transitions at altitudes @xmath117 , as described in section 2.3 , the total energy @xmath118 gives the number of protons created in the electron showers . this is the approximately linear function @xmath119 ( see papers i and ii ) . it is convenient here to define @xmath120 as the number of protons formed per unit ion charge . ( in papers i and ii , it was defined as the number of protons per unit nuclear charge of the ion . ) the model described here uses an elementary form of equation ( 16 ) which describes the temporal distribution of the protons reaching the top of the lte atmosphere . but the distribution of electron back - flow and the values of @xmath100 over the polar - cap also define through solution of poisson s equation , at any instant , the actual electrostatic potential which , of course , is not identical with equation ( 12 ) because @xmath121 is not independent of @xmath122 . the model then finds by relaxation procedures finite - element approximations to the self - consistent functions @xmath123 and @xmath124 for the polar cap . this is circular , with radius given by equation ( 11 ) , which assumption is not essential but is made for ease of calculation . the polar cap is divided into @xmath125 elements of equal area by first defining @xmath126 annuli of outer radius @xmath127 , where @xmath128 and then further dividing each of these into @xmath129 equal azimuthal elements . within each annulus , the elements are displaced , by a random azimuthal angle , with respect to the axis @xmath130 . it is assumed that the charge density is independent of @xmath83 within any element . equation ( 16 ) is much simplified by the model assumption @xmath131 giving , @xmath132 within any element , so that the state of the element alternates between proton and ion emission . the total number of protons produced at time @xmath133 on unit area of surface at the top of the atmosphere following an ion - emission interval of length @xmath19 is , @xmath134 the ion current density is @xmath135 and the goldreich - julian charge density is @xmath136 . therefore , if we define the length of the proton phase by @xmath137 , @xmath138 an event is defined as a change of state in any element from proton to ion emission or vice - versa . the length of an ion phase is @xmath19 , which is assumed constant . the length of the following proton phase is @xmath139 , and is a function of the state of the whole polar cap within the integration time over the preceding ion phase . table 1 of paper iv gave values of @xmath114 and of @xmath100 as functions of a cut - off potential @xmath140 for different values of the surface magnetic flux density @xmath141 , in units of @xmath73 g , and the whole - surface temperature @xmath3 . for reasons which we outline later , we assume a surface nuclear charge @xmath142 . the cut - off potential is a representation of the effect of photo - electric transitions on the maximum possible acceleration potential difference @xmath108 given by equation ( 15 ) . from table 1 , with some additional values , we have been able to tabulate the functions @xmath143 and @xmath144 for the interval @xmath145 . they are not strongly dependent on @xmath83 and we can assume that they are functions only of the magnitude of @xmath146 , the potential at the upper end of the polar - cap acceleration zone which is represented here by a cut - off at @xmath147 . in the model , it is assumed to be the potential on the central axis of an element . the excess charge density within an element during the ion phase is , @xmath148 and is the source of a potential deviation downward from @xmath149 . in order to calculate this with some economy , we use the approximation on which equation ( 12 ) is based , specifically that at an altitude @xmath150 , a section of the long narrow open magnetosphere can be approximated locally by a cylinder of constant radius @xmath151 . then the green function satisfying the sclf boundary condition for a line source at cylindrical polar coordinates @xmath152 is , @xmath153 the potential on the axis of an element derived from the charge excess within that element is the most important factor and is obtained by numerical integration using the green function . the potential generated on its axis by any other ion - phase element is found directly from the green function by using a line approximation for the excess charge . the potential within a proton - phase element is of no interest and is not calculated . after each event , the @xmath100 in all ion - phase elements , also referred to as ion - zones , are recalculated by a relaxation procedure so that there is consistency between the poisson equation solution for @xmath146 in terms of all the @xmath100 and the tabulated function @xmath144 obtained from table 1 of paper iv which is based purely on the photoelectric transition rates . this also gives the function @xmath114 and hence @xmath120 which , with the values of @xmath100 and their times of duration , then allow calculation of the integral for @xmath154 and the duration of the subsequent proton phase . the initial state of the system has all elements in the proton phase . all @xmath155 elements , selected in random order , are then assigned sequential times for transition to the ion phase . the interval between any two adjacent times is @xmath156 , where @xmath157 is a random number in the interval @xmath158 . our initial choice was @xmath159 . throughout the calculation , a list is maintained , in temporal order , of the time of the next event in each of the @xmath155 elements . this procedure has the advantage that the calculation of every ion phase is entirely self - contained so that cumulative errors are not carried forward . the model has been run for intervals equivalent to real times of the order of ten days . for a given random initial state , the subsequent states of the model polar cap are not , of course , random but are chaotic , and variation of @xmath160 appears not to affect their character . the proton production parameter @xmath17 is a slowly varying function of @xmath113 but we have chosen the conservative value @xmath161 gev@xmath9 and an ion charge @xmath162 as in paper iv . the numbers of elements have been in the interval @xmath163 . the usual procedure has been to let the model run for a time @xmath164 before sampling its state at intervals of @xmath19 . the presence of the instability described in paper i has been confirmed . the system has shown no sign of settling down to other than a chaotic state . its main characteristics are as follows . \(i ) fluctuations in the central potential @xmath165 and in the number of ion - emission elements can be very large and are dependent , principally , on the whole - surface temperature @xmath3 but also , to a lesser extent , on the parameter @xmath166 which scales @xmath108 . \(ii ) the elements most likely to be in an ion - emission phase are those near the periphery @xmath167 . \(iii ) peripheral ion - emission elements have a significant autocorrelation function in the angle @xmath168 , unaffected by running time , indicating the formation of clusters . at this stage , we should note that the model as precisely defined by equations ( 17 ) - ( 19 ) does not produce , at any point on the polar - cap surface , the mixture of protons and ions which is required for growth of the quasi - longitudinal langmuir mode . but the growth rate is only a slowly varying function of the proton - ion ratio ( see papers iii and iv ) and we do not doubt that , given a physically realistic form of the diffusion function @xmath105 in equation ( 16 ) , there will be a sufficient interval of time , of the order of @xmath19 , within which the current density contains both components . .the first two columns give values of the rotation period @xmath23 and whole - surface temperature @xmath3 . the next five columns histogram the acceleration potential @xmath169 sampled after @xmath170 successive intervals of @xmath19 in order to show the scale of the fluctuations that are present in the model for various @xmath23 and @xmath3 . the bin size is @xmath171 and the number of elements here is @xmath172 ; the polar - cap magnetic flux density is @xmath173 . the final column gives @xmath174 , the probability that the whole polar cap is in the proton phase at any instant . [ cols= " > , > , > , > , > , > , > , > " , ] the potential fluctuations given in the table are , of course , dependent on @xmath155 . the autocorrelation function in @xmath168 indicates that a smaller of @xmath155 should be preferred and , within the limited framework of our model , would be a better approximation to reality . but @xmath175 is the smallest value for which our approximation for the interaction between two elements can be reasonably adequate . table 1 therefore underestimates the true scale of fluctuation . the values of @xmath141 and @xmath23 have been chosen to be representative of the typical pulsar exhibiting nulls as listed by wang , manchester & johnston ( 2007 ) . the scale of the downward fluctuations in @xmath176 is , as might be anticipated , very strongly dependent on @xmath3 and hence , presumably on pulsar age . at @xmath177 k , significant downward fluctuations would be present only for very large values of @xmath166 , the parameter that scales @xmath108 . with increasing rotation period , downward fluctuations also increase . the most significant feature of the model is the size of the potential fluctuations that occur for larger values of @xmath3 . but this is superimposed on a further instability , with medium time - scale , described in paper ii . showers reduce the atomic number from @xmath178 , possibly @xmath179 , to @xmath34 at the top of the atmosphere . provided we ignore capture of shower - produced neutrons and subsequent @xmath180-decay of the neutron - rich nuclei formed , this liberates @xmath181 protons which diffuse with time - scale @xmath19 . at any instant , there is naturally a distribution of @xmath34 values , but over a time long compared with the ablation time ( the interval in which one radiation length of matter is removed from the polar cap at the goldreich - julian current density ) @xmath182 the average quantities must satisfy @xmath183 this means that @xmath34 and @xmath154 defined by equation ( 19 ) can not be independent variables when considered over medium time - scales of the order of @xmath184 . the reason is that large values of @xmath120 within some interval of time imply reduction of @xmath178 at the shower maximum to very small @xmath34 so producing nuclei able to move to the top of the atmosphere either by diffusion on a time - scale possibly some orders of magnitude larger than @xmath19 , or by rayleigh - taylor instability . in particular , values smaller than @xmath185 have a high probability of being completely ionized in the lte atmosphere and the consequent absence of a reverse - electron flux stops shower and proton formation . the condition of these surface layers is obviously not well understood , but in paper ii a case was advanced that it would be one of instability rather than a steady - state value of @xmath34 . this is the case for our model assumption of a compromise value @xmath142 and is also the basis for the observed phenomena , mode - changes and long nulls , with time - scales of the order of @xmath184 . our proposal is that the potential fluctuations found in the model provide a basis for understanding why the extent of null lengths and fractions seen in both pulsars and the rrats occur quite naturally . we refer to burke - spolaor & bailes ( 2010 ) , keane et al ( 2011 ) and keane & mclaughlin ( 2011 ) for reviews of the observational data on these sources . values of @xmath186 , the fraction of periods in which a pulse is detected , are broadly in the interval @xmath187 within which their distribution is approximately uniform in @xmath188 . the bursts of emission are short , usually one but occasionally several periods , and are consistent with being of the same order of magnitude as @xmath19 . the rrats listed in table 3 of keane et al are all quite distant with the exception of j1840 - 1419 which also has a period @xmath189 s and a value of the acceleration potential parameter @xmath190 , that is , below the cut - off value @xmath191 which can be found from the atnf catalogue ( manchester et al 2005 ) . thus weltevrede et al ( 2006a ) proposed that the rrat pulses are simply analogues of the giant pulses seen in the nearby pulsar b0656 + 14 and that their emission , apart from the giant pulses , is unobservable owing to distance . however , there is a difference whose significance becomes obvious in view of the model described here . with the exception of j1554 - 5209 , the @xmath192 rrat listed with surface magnetic fields by keane et al are well below the threshold for cr secondary pair production as defined by harding & muslimov ( 2002 ) . from fig.1 of their paper , this condition can be represented approximately in terms of the parameter @xmath193 : the threshold is given by the critical value @xmath194 . psr 0656 + 14 lies well above this threshold , which assumes dipole - field flux - line curvature , as does the exceptional rrat j1554 - 5209 , and it is reasonable to suppose that these pulsars are capable of supporting cr secondary pair creation over at least a fraction of their polar caps . therefore , we suggest that the giant pulses of b0656 + 14 are produced by process ( a ) through a rare upward fluctuation in @xmath176 toward @xmath108 and its more general emission by process ( b ) . some rrats may , of course , have sufficient non - dipole flux - line curvature to enable process ( a ) , but discounting this possibility , the rrats listed by keane et al are so far below the harding - muslimov threshold that we suggest their isolated pulses are a consequence of rare but large downward fluctuations in @xmath176 . it might be argued that , owing to the boundary condition satisfied by equation ( 12 ) , lorentz factors @xmath195 small enough for growth of the quasi - longitudinal langmuir mode always exist , in principle , at @xmath122 close to @xmath167 so that there should be some coherent emission at all times . but in this case it is not obvious that the lateral ( perpendicular to @xmath0 and to @xmath83 ) depth of the particle beam would be suitably large in relation to the wavelength of the mode to allow the necessary growth rate . there is an obvious need for some analogue of the relativistic penrose condition ( see buschauer & benford 1977 ) extended to allow for a beam with lateral velocity gradients and limited lateral depths . the question of the conditions necessary for observable emission will be further considered in section 5.1 in relation to the structure and drift of subpulses . the ages and surface magnetic fields of rrats given by keane et al can be compared with those of the nulling pulsars listed in tables 1 and 2 of wang , manchester & johnston ( 2007 ) . it is immediately seen that rrat fields are typically an order of magnitude larger than those given in the atnf catalogue ( manchester et al 2005 ) for the wang et al list . but the distributions of the acceleration parameter @xmath166 are very similar because the rrat have long periods . the age distributions are also similar though both are very wide . the relation between age and proper - frame surface temperature @xmath3 is , unfortunately , obscure in the region assumed in table 1 , which is well within the photon - cooling epoch . potekhin & yakovlev ( 2001 ; fig.7 ) have shown that increasing surface fields accelerates cooling , but an even more important factor would be the presence of low-@xmath178 elements in the outer crust , possibly from fall - back , which have the same effect ( potekhin et al 2003 ) . with reference to table 1 , it is possible to see qualitatively how nulls change with age during the cooling from high-@xmath3 to low-@xmath3 . we consider pulsars which lie below the pair - creation threshold . initially , values of @xmath196 are usually small enough to give the mode growth rates necessary for emission by process ( b ) ( see section 2.3 ) . upward fluctuations of @xmath196 increase the ion and proton lorentz factors and , owing to the exponential dependence of amplitude growth on them , can produce short nulls . upward fluctuations become more frequent as cooling proceeds , giving higher null fractions @xmath197 . eventually , cooling reaches the stage at which the mean @xmath196 approaches unity and downward fluctuations are required to produce the growth rates necessary for observable emission . null fractions are then near unity and , as the limit is approached , there is no observable emission . we emphasize that these fluctuations are superimposed on medium time - scale fluctuations ( see section 3.2 ) in the surface atomic number @xmath34 which are unlikely to be uniform over the whole polar - cap surface . values smaller than @xmath198 produce a negligible reverse - electron energy flux and hence , if present over a substantial area of the polar cap , much reduced deviations of @xmath196 from unity . it is proposed here that these are responsible for null and burst lengths more nearly of the order of @xmath184 rather than @xmath19 . isolated neutron stars ( ins ) are a small group of radio - quiet thermal x - ray emitting sources positioned close to the rrat in the @xmath199 plane ( see keane et al 2011 ) but at slightly greater periods . the six sources for which timing solutions exist have been listed by zhu et al ( 2011 ) . given their kinematic ages , the observer - frame temperatures are large ( @xmath200 k ) compared with those considered in table 1 but , with the exception of the anomalous j0420 - 5022 , their @xmath166 are typically close to the general cut - off value of 0.22 found from the atnf catalogue ( manchester et al 2005 ) . if they form part of the @xmath16 population , they should support pair formation by the ics mechanism ; for @xmath2 , extrapolation from table 1 shows that small @xmath196 would be expected satisfying the condition necessary for rapid growth of the ion - proton beam instability . however , the fact that no radio emission has been observed from this small number of sources is , perhaps , not surprising in view of their anticipated small beaming fractions compared with the @xmath201-observability of the whole - surface x - rays . subpulse drift , and the null memory which , in some pulsars , is observed with it , is obviously an important diagnostic of polar - cap physics . unfortunately , our model is incomplete in that it does not have the capacity to spontaneously exhibit this phenomenon following the randomly - constructed initial state described in section 3 . its finite - element structure and the very elementary approximation represented by equation ( 17 ) are the reasons for this . however , the phenomenological model developed by deshpande & rankin ( 1999 ) from the classic ruderman & sutherland ( 1975 ) model has proved so useful that any physical polar - cap model must be able to support it . it was noted in section 3 that the model system settles down to a chaotic state , but with ion - emission elements most likely to be near the periphery @xmath167 and an autocorrelation function in the azimuthal angle @xmath168 indicating the formation of clusters . this is unsurprising , because ion - emission elements at smaller @xmath202 have larger reverse - electron energy fluxes and so lead to longer phases of proton emission . thus our model generates quite naturally the conal structure that is required for the deshpande - rankin carousel model and we can therefore proceed to examine the extent to which it is capable of supporting its organized subpulse motion . for convenience , we shall consider a circular path of radius @xmath202 on the polar cap and associate a moving ion zone with the formation of a subpulse . then it is possible to envisage organized circular motion in which an ion - emission zone of angular width @xmath203 is followed by a proton - emission zone of @xmath204 . thus at any instant , @xmath205 with @xmath206 in which @xmath126 is here the number of elements on the carousel and each @xmath154 is determined according to equation ( 19 ) by the preceding ion - zone . comparison with measured values of the longitudinal subpulse separation @xmath207 indicates that the order of magnitude of model values should not be large , @xmath208 , but this order of magnitude is associated with values of the acceleration potential , on the ion - zone flux lines , that are unlikely to enable significant electron - positron pair creation . this is then consistent with mechanism ( b ) , based on protons and ions , for the production of coherent radio emission . it is easy to see that a steady uniform state of equation ( 23 ) with @xmath209 and constant @xmath210 that is independent of @xmath202 does not , in general , exist . however , we have to bear in mind the limitations of our model based on equation ( 16 ) , which is local in @xmath122 , and on its very elementary approximation given by equation ( 17 ) . thus in a model with a more realistic diffusion function @xmath105 and without the finite element structure , we would expect that at constant @xmath168 , @xmath154 would increase as a function of @xmath202 , starting from a negligible value at @xmath167 and reaching a maximum before declining as a proton - emission zone is entered . the presence of this extremum is a possible basis for the existence of short - term quasi - steady - state solutions of equation ( 23 ) within a finite interval of @xmath202 at @xmath211 which also coincides with conditions suitable for the growth of the instability giving observable emission . it may be that the tendency to form a cluster , present in our model , indicates that a more realistic model would counteract the dispersion inherent in equation ( 23 ) and maintain subpulse shape , but it is not possible to assert that it would be so . however , it is possible to envisage this motion on a polar cap of any plausible geometrical shape , elliptical or approximately semi - circular . the motion is , of course , independent of @xmath212 drift and could be in either direction , as is seen in the survey of weltevrede , edwards & stappers ( 2006b ) , or may even be bi - directional at some instant on different areas of the polar cap . there are many ways in which potential fluctuations could disrupt this organized motion . we have seen in the previous section that a whole - surface temperature @xmath3 large enough to give low values of @xmath196 favours the occurrence of nulls through upward potential fluctuations . the most simple fluctuation is one that increases the potential over much of the polar cap and in particular at @xmath213 . the functions @xmath143 , and hence @xmath154 , increase quite rapidly with @xmath176 . there are two consequences . firstly , the exponent in the expression for the mode growth rate ( paper iv , equation 17 ) is dependent on @xmath214 so that the conditions necessary for observable coherent radio emission are not reached and a null occurs . secondly , the ( unobserved ) motion of the ion zones is substantially changed . it is possible to see the nature of these changes by considering how two variables , the proton density @xmath215 at the top of the atmosphere and the ion current density @xmath102 satisfying equations ( 17 ) - ( 19 ) on a section of the path at @xmath216 , evolve over a sequence of times . in doing this , we assume that the ion - zone is able to move continuously and abandon the fixed finite - element calculation of the model described in section 3.1 the ion - zone motion during a null shows how a form of null memory occurs . it is convenient , for this problem , to represent the density @xmath217 in units of @xmath218 and @xmath102 in units of @xmath219 . the ion - zones at time @xmath133 have @xmath220 , width @xmath221 and move with velocity @xmath222 . each generates a proton density @xmath215 at the top of the atmosphere which is depleted by the proton - zone current density @xmath223 . thus at an instant @xmath133 , and as a function of @xmath168 , the proton density following an ion zone rises linearly to a maximum of @xmath224 in a distance @xmath221 and falls linearly to zero in a further distance @xmath225 . the velocity of an ion zone is therefore determined by the gradient @xmath226 of the preceding proton zone . this is the quasi - steady - state motion at time @xmath133 , at which instant , an upward potential fluctuation increases proton production to @xmath227 . the increase in ion lorentz factor reduces the quasi - longitudinal mode growth rate to a sub - critical value and so initiates the null . a new maximum proton density @xmath228 is reached at a later time @xmath229 but the system velocity remains @xmath230 until @xmath231 . the width of an ion zone is then reduced . in detail , and with reference to equations ( 17 ) - ( 19 ) , this happens because the velocity of its leading edge is reduced whilst that of the rear edge remains at @xmath230 until @xmath232 at which time the ion - zone width and velocity are both reduced by a factor of @xmath233 . the ion - zone velocity then tends to its new steady - state value @xmath234 . hence subpulse drift does not stop in the model , but is much reduced after a null time interval of @xmath235 . thus for short nulls , there is a subpulse memory , but it does not correspond precisely with the conservation of subpulse longitude over a short null that has been observed , for example in psr b0809 + 74 , by van leeuwen et al ( 2002 ) . we are not unduly disturbed by this , because there are many possible spatial distributions of potential fluctuation over the polar cap other than the uniform case considered above . it is also not clear that the precise form of memory seen in b0809 + 74 is universally observed . the upward potential fluctuation that caused this could reverse at any time during or after the sequence described in the previous paragraph . in relation to null memory , the question then is , when does observable coherent radio emission recommence ? the problem arises owing to the very elementary nature of our model defined by equations ( 17 ) - ( 19 ) and is that @xmath236 in the model can be very small . the acceleration potential within a very small interval of @xmath168 may be too large to permit adequate growth of the mode and the conditions discussed briefly in section 4.1 may not be satisfied . published observations on mode - changes and subpulse drift demonstrate the extremely heterogeneous nature of these phenomena . some features are quite distinct in a small number of pulsars , but much less distinct or not present at all in others . for this reason , we have used the results of the extensive survey made by weltevrede et al ( 2006b ) based on observations of 187 pulsars selected only by signal - to - noise ratio , and list some of their conclusions below . \(i ) roughly equal numbers of pulsars have subpulse drift to smaller or greater longitudes , and direction reversal is observed . \(ii ) bi - directional drifting is observed in a small number of pulsars . j0815 + 09 and b1839 - 04 have mirrored drift bands . \(iii ) the drift rate can be mode - dependent , having one of a number of discrete values of the band separation @xmath237 , and nulls may also be confined to a particular drift - mode . \(iv ) drift bands can be curved or non - linear with pulse longitude - dependent spacing of subpulses : they are often indistinct and can be found only by two - dimensional spectral analysis . \(v ) the band separation @xmath237 is uncorrelated with age ( @xmath238 ) , or with @xmath23 and @xmath113 . conclusions ( i ) and ( ii ) are quite naturally consistent with the model . we emphasize again that @xmath212 drift is not involved and there is no reason why , given the chaotic state of the polar cap , there should not be reversals or even bi - directionality . the nature of the chaotic state also means that it is unrealistic to suppose that these phenomena can be predictable . conclusion ( iii ) is also a natural consequence of medium time - scale instability in the value of the surface atomic number @xmath34 and hence in the potential as mentioned in section 4.2 . the existence of a polar - cap area emitting nuclei with @xmath34 too small to produce a significant reverse - electron flux is obviously associated with an upward displacement of @xmath196 for times very broadly of the order of @xmath184 . this is quite consistent with the observed time intervals of @xmath239 s for mode - changes . non - linear drift bands ( iv ) are simply a consequence of non - uniformity of @xmath240 values in equation ( 23 ) . band separation @xmath237 was discussed in paper ii . it is given by @xmath241 . in this expression , @xmath19 is dependent only on the properties of the lte atmosphere at the polar - cap surface , but @xmath154 may have some dependence on rotation period and magnetic flux density ; also on whole - surface temperature @xmath3 and hence pulsar age . it was noted in paper ii that the distribution of @xmath237 is quite compact , most of the values listed by weltevrede et al ( 2006b ) being within the interval @xmath242 s. the fact that the carousel path is restricted to being near the polar - cap periphery @xmath167 may well act as a constraint on the values of @xmath154 and so @xmath237 that occur . finally , kloumann & rankin ( 2010 ) have observed randomly distributed pseudo - nulls in b1944 + 17 of length less than @xmath243 with weak emission , but well above noise . they interpreted them as a state of no subpulse on that band of polar - cap flux lines from which photons can enter the line of sight . these occur naturally in the model . the final column of table 1 gives @xmath174 , the probability that the whole polar cap is in the proton phase at any instant . there would then be no possible way of producing particle beams satisfying conditions ( a ) or ( b ) of section 2.3 and capable of generating coherent radio emission . this contributes to the total null fraction that is observed . the probability @xmath244 that there will be no ion - zones merely in the part of the polar cap which is observable along the line of sight is therefore a natural parameter of the model . papers i - iv , followed by the present paper , have attempted to predict some of the consequences of there being two populations of isolated pulsars having opposite spin directions . in many important respects , the physical state of the polar - cap has been found to be dependent on the sign of the goldreich - julian charge density . in the @xmath110 case considered in these papers , sclf boundary conditions have been assumed rather than the @xmath245 condition of the original ruderman & sutherland ( 1975 ) model . then the state of the polar cap is unstable on time - scales of the order of @xmath246 and of the longer polar - cap ablation time @xmath184 . these are both many orders of magnitude longer than the characteristic polar - cap time - scale of @xmath109 and there can be little doubt that the system should be able to allow the movements of charge necessary to maintain the sclf boundary conditions for them . we therefore need to look at published observational data to see if there is any real evidence for the existence of two populations . the weltevrede et al ( 2006b ) survey is an obvious starting point . it lists 187 pulsars selected only by signal - to - noise ratio and examines both subpulse modulation ( the wide variations of intensity at a fixed longitude in a sequence of observed pulses ) and subpulse drift . subpulse modulation is an almost universal characteristic and we assume it to be a direct consequence of the plasma turbulence that itself is now widely believed to be the source of the radio emisssion ( see asseo & porzio 2006 ) . subpulse drift with measurable values of both longitudinal separation of successive subpulses @xmath207 and band separation @xmath237 was detected in 72 pulsars and these can be compared with a set of 113 which do not show detectable subpulse drift . the distributions of both sets as a function of age are wide and are broadly similar except that only 7 of the 72 have an age less than 1 myr as opposed to 31 of the set of 113 . the distributions as functions of the parameter @xmath247 whose critical value @xmath248 defines the cr pair creation threshold are also wide but the set of 113 has an excess at @xmath249 . there are 35 pulsars with @xmath249 as opposed to only 7 of the 72 . this is consistent with , but does not prove , the existence of two populations which separate as @xmath247 moves with age to values below @xmath250 . we propose that the @xmath2 set show the phenomena of mode - changes , nulls and subpulse drift as they age through growth of the ion - proton beam quasi - longitudinal mode ( mechanism ( b ) of section 2.3 but see also the comments there on more obscure processes of pair creation for pulsars of this spin direction ) . pair creation by ics photons in the @xmath28 case has been very fully investigated by hibschman & arons ( 2001 ) and by harding & muslimov ( 2002 , 2011 ) and our assumption is that this is the source of coherent emission in the population that has no nulls or subpulse drift . having divided the weltevrede et al list into two populations , we should compare these with the extensive survey of nulls made by wang et al ( 2007 ) . no pulsar in table 1 of wang et al appears in the earlier weltevrede et al list , presumably because they were not known or did not satisfy the selection criteria . of the 46 pulsars with measured null fractions given in table 2 of wang et al , 18 also do not appear in weltevrede et al , but 20 are listed with a measured value of @xmath237 and 8 have no detected @xmath237 . a smaller but more recent list of nulling pulsars in the paper of gajjar , joshi & kramer ( 2012 ) includes a further 8 that were not considered by wang et al , of which 2 have measured values of @xmath237 but 6 do not appear in the paper of of weltevrede et al . of the 8 pulsars that have nulls but no detected @xmath237 , 4 have null fractions @xmath251 and b0656 + 14 is the special case described in section 4.1 . there are only 3 pulsars having substantial null fractions but no detected @xmath237 . these are the otherwise unremarkable b1112 + 50 , b2315 + 21 and b2327 - 20 . but it is not necessarily correct to regard them as anomalous because , as noted by weltevrede et al , drift bands are often indistinct and detectable only by two - dimensional spectral analysis . we suggest that the results of this comparison are by no means inconsistent with our division of the weltevrede et al pulsars into two populations . a different , but perhaps more anomalous set are the three pulsars known to exhibit nulls of very long duration , of the order of @xmath18 d. the first of these to be found ( b1931 + 24 ; kramer et al 2006 ) enabled spin - down rate measurements to be made separately in both on and off states of emission . this was followed by similar measurements for j1832 + 0029 ( lorimer et al 2012 ) and j1841 - 0500 ( camilo et al 2012 ) . in each case , the off - state spin - down rate was about half that of the on state . but these pulsars have quite large values @xmath252 , @xmath253 and @xmath254 respectively , close to the critical value @xmath248 , and may quite possibly support self - sustaining cr pair creation during the on but not the off state . the consequent difference in the flux and nature of particles passing through the light cylinder appears to be the only physically plausible mechanism for a spin - down torque change of this magnitude . but the @xmath255 s time - scale for both on and off states of emission appears a little too long compared with times of the order of @xmath256 found from equation ( 22 ) . the question of quasi - periodicity with such time - scales should also be addressed , as in the case of the rrats ( palliyaguru et al 2011 ) . we have stressed that our model is not random , but deterministic . quasi - periodicites are therefore not impossible in principle , but remain quite difficult to explain . changes in the polar - cap acceleration potential have little direct effect on the spin - down torque . changes in the proton - ion composition occurring in the process ( b ) nulls described in section 4.2 would lead to a change in the mean charge to mass ratio of particles crossing the light cylinder , but it must be doubted whether mechanism ( b ) could explain the large observed difference in spin - down torque . however , it is not yet clear that this is a universal feature of nulls . the only other pulsar for which measurements have been made ( psr b0823 + 26 : young et al 2012 ) has a fractional upper limit of @xmath257 for the change in spin - down torque . it has a value @xmath258 , rather below the putative critical value @xmath248 denoting the pair creation threshold , so that the relatively small change in torque could be consistent either with relatively weak pair production or with process ( b ) . in the absence of electron and positron densities large enough to adjust and so cancel an electric field , as in mechanism ( a ) , we have to remember that acceleration or deceleration remains at higher altitudes beyond @xmath259 as a consequence of natural flux - line curvature . but at this stage , growth of the quasi - longitudinal mode has already occurred so that further acceleration would have negligible effect on the coherent radio emission . this is a further distinction between the two populations . in the @xmath2 case , the polar cap of open magnetic flux lines may well have an approximately semi - circular shape . in this case , the observed drift of subpulses would be along a diameter instead of an arc of a circle . although all the above problems remain , the model @xmath260 polar cap described here has some positive features . it does not require that neutron - star magnetic fields lie in a particular interval . this is important because radio pulsar inferred polar fields can vary by up to six orders of magnitude . the physical processes in electromagnetic shower development or in the photoelectric transitions exist in the zero - field limit and do not change in any qualitative way with increasing field . the model is deterministic , but chaotic , but is incomplete in that its use of finite elements for the reason stated at the end of section 3.1 and the very elementary nature of the approximation made in equation ( 17 ) appear to preclude the spontaneous appearance of subpulse drift from the random initial state which is used . it is possible to do no more than assert that the model can support subpulse drift in a quasi - stable way . it is unfortunate that quantitative model predictions depend so much on surface atomic number and particularly on whole - surface temperature @xmath3 , parameters which are not well - known . cooling calculations ( see the review of yakovlev & pethick 2004 ) show that @xmath3 falls steeply at ages greater than 1 myr . bearing in mind that , for a neutron - star with the mass and radius assumed here , the observer - frame temperature is @xmath261 , we can see that the temperatures assumed in table 1 fall well below the values currently observable . although cooling in this interval is photon - dominated , the whole - surface temperature must be regarded as very uncertain . but having made these reservations , the model does represent a physically - realistic framework for understanding the reasons why rrats and the varied phenomena of mode - changes , nulls and subpulse drift appear during neutron - star aging . it may be that further observations at frequencies below 100 mhz will provide evidence for the existence of a population emitting by process ( b ) and having spectra biassed towards lower frequencies than those for the process ( a ) population . 99 asseo e. , pelletier g. , sol h. , 1990 , mnras , 247 , 529 asseo e. , porzio a. , 2006 , mnras , 369 , 1469 bai x. -n . , spitkovsky a. , 2010 , apj , 715 , 1282 beloborodov a. m. , 2008 , apj , 683 , l41 , burke - spolaor s. , bailes m. , 2010 , mnras , 402 , 855 buschauer r. , benford b. , 1977 , mnras , 179 , 99 camilo f. , ransom s. m. , chatterjee s. , johnston s. , demorest p. , 2012 , apj , 746 - 63 chen a. y. , belobordov a. m. , 2013 , apj , 762:76 deshpande a. a. , rankin j. m. , 1999 , apj , 524 , 1008 gajjar v. , joshi b. c. , kramer m. , 2012 , mnras , 424 , 1197 harding a. k. , muslimov a. g. , 2001 , apj , 556 , 987 harding a. k. , muslimov a. g. , 2002 , apj , 568 , 862 harding a. k. , muslimov a. g. , 2011 , apj , 743 - 181 hibschman j. a. , arons j. , 2001 , apj , 554 , 624 jones p. b. , 2010a , mnras , 401 , 513 ( paper i ) jones p. b. , 2010b , mnras , 409 , 1719 jones p. b. , 2011 , mnras , 414 , 759 ( paper ii ) jones p. b. , 2112a , mnras , 419 , 1682 ( paper iii ) jones p. b. , 2112b , mnras , 423 , 3502 ( paper iv ) kalapotharakos c. , contopoulos i. , 2009 , a&a , 496 , 495 keane e. f. , mclaughlin m. a. , 2011 , bull . astron . soc . india , 39 , 333 keane e. f. , kramer m. , lyne a. g. , stappers b. w. , mclaughlin m. a. , 2011 , mnras , 415 , 3065 klouman i. m. , rankin j. m. , 2010 , mnras , 408 , 40 kramer m. , lyne a. g. , obrien j. t. , jordan c. a. , lorimer d. r. , 2006 , sci , 312 , 549 landau l. , rumer g. , 1938 , proc . soc . , a166 , 213 van leeuwen a. g. j. , kouwenhoven m. l. a. ramachandran r. , rankin j. m. , stappers b. w. , 2002 , a&a , 387 , 169 lorimer d. r. , lyne a. g. , mclaughlin m. a. , pavlov g. g. , kramer m. , chang c. , 2012 , apj , 758:141 manchester r. n. , hobbs g. b. , teoh a. , hobbs m. , 2005 , astron . j. , 129 , 1993 mclaughlin m. a. , et al , 2006 , nat . , 439 , 817 medin z. , lai d. , 2006 , phys . rev . a , 74 , 062508 mestel l. , robertson j. a. , wang y. -m . , westfold k. c. , 1985 , mnras , 217 , 443 michel f. c. , 1974 , apj , 192 , 713 miralles j. a. , urpin v. , van riper k. , 1997 , apj , 480 , 358 muslimov a. g. , tsygan a. i. , 1992 , mnras , 255 , 61 nordheim l. w. , hebb m. h. , 1939 , phys . , 56 , 494 palliyaguru n. t. , et al , 2011 , mnras , 417 , 1871 potekhin a. y. , yakovlev d. g. , 2001 , a&a , 374 , 213 potekhin a. y. , yakovlev d. g. , chabrier g. , gnedin o. y. , 2003 , apj , 594 , 404 rast m. p. , 2001 , apj , 561 , l191 ruderman m. a. , sutherland p. g. , 1975 , apj , 196 , 51 timokhin a. n. , 2010 , mnras , 408 , 2092 timokhin a. n. , arons j. , 2013 , mnras , 429 , 20 wang n. , manchester r. n. , johnston s. , 2007 , mnras , 377 , 1383 weltevrede p. , stappers b. w. , rankin j. m. , wright g. a. e. , 2006a , apj , 645 , l149 weltevrede p. , edwards r. t. , stappers b. w. , 2006b , a&a , 445 , 243 yakovlev d. g. , pethick c. j. , 2004 , ara&a , 42 , 169 young n. j. , stappers b. w. , weltevrede p. , lyne a. g. , kramer m. , 2012 , mnras , 427 , 114 zhu w. w. , kaspi v. m. , mclaughlin m. a. , pavlov g. g. , ng c. -y . , manchester r. n. , gaensler b. m. , woods p. m. , 2011 , apj , 734 - 44

we have carried out millimeter interferometric observations of the orion molecular cloud-2 ( omc-2 ) fir 3/4 region at an angular resolution of 3 - 7 with the nobeyama millimeter array ( nma ) in the hco (=10 ) ,co (=10 ) , sio (=0 ,=21 ) , and cs (=21 ) lines and in the 3.3 mm continuum emission . submillimeter single - dish observations of the same region have also been performed with atacama submillimeter telescope experiment ( aste ) in theco (=32 ) and choh (=7 ) lines . our nma observations in the hco emission have revealed 0.07 pc - scale dense gas associated with fir 4 . theco (=32,10 ) emission shows high - velocity blue and red shifted components at the both north - east and south - west of fir 3 , suggesting a molecular outflow nearly along the plane of the sky driven by fir 3 . the sio (=0 ,=21 ) and the submillimeter choh (=7 ) emission , known as shock tracers , are detected around the interface between the outflow and the dense gas . furthermore , theco (=10 ) emission shows an l - shaped structure in the p - v diagram . these results imply presence of the shock due to the interaction between the molecular outflow driven by fir 3 and the dense gas associated with fir 4 . moreover , our high angular - resolution ( 3 ) observations of fir 4 in the 3.3 mm continuum emission with the nma have first found that fir 4 consists of eleven dusty cores with a size of 1500 - 4000 au and a mass of 0.2 - 1.4 m . the separation among these cores ( 5 au ) is on the same order of the jeans length ( 13 au ) , suggesting that the fragmentation into these cores has been caused by the gravitational instability . the time scale of the fragmentation ( 3.8 yr ) , estimated from the separation divided by the sound speed , is similar to the time scale of the interaction between the molecular outflow and the dense gas ( 1.4 yr ) . we suggest that the interaction between the molecular outflow from fir 3 and the dense gas at fir 4 triggered the fragmentation into these dusty cores , and hence the next generation of the cluster formation in fir 4 .

0804.0880

the process of so - called `` isolated '' star formation , where a single star or binary stars form in one molecular cloud core , has been studied intensively @xcite . in this mode , spontaneous gravitational collapse of cloud cores initiates the formation of the protostar at the center , and there is observational evidence of such gravitational collapse @xcite . another mode of star formation is cluster formation , where stellar clusters form simultaneously or sequentially in one cloud core . a vast majority ( @xmath14 90 % ) of stars form as members of clusters in our galaxy @xcite , and the understanding of the cluster formation is crucial for more comprehensive understanding of star formation . previous observational studies of cluster formation have proposed that external effects on cloud cores , such as stellar wind , hii regions @xcite , supernova @xcite , and molecular outflows @xcite , are often required to trigger cluster formation . theoretical studies also suggest that such external effects are often required to trigger cluster formation @xcite . for example , @xcite have found 52 cluster members in `` cloud 2 '' , which locates in the vicinity of the snr h i shell , gsh 138 - 01 - 94 . the age of the cluster in cloud 2 ( @xmath0 1.0 @xmath15 10@xmath16 yr ) is younger than that of gsh 138 - 01 - 94 ( @xmath0 4.3 @xmath15 10@xmath16 yr ) , and they suggest that the star formation in cloud 2 was triggered by the interaction with the snr h i shell . in the 850 @xmath17 m dust continuum map of the ngc 1333 region , there are several cavity and/or shell structures observed , which are most likely created by molecular outflows @xcite . one dusty core , sk-1 , is in one of these shells and is gravitationally bound , and @xcite suggest that this core is a specific example of the outflow - triggered star formation . however , these studies are presumptions mainly based on the overall morphology , and detailed studies of cluster formation processes , including spatially - resolved observations of both triggers and cloud cores and the investigation of the physical processes , have been limited . this is mainly due to the relatively far distance to cluster - forming regions and lack of interferometric observations of cluster - forming regions in millimeter and submillimeter molecular lines as well as continuum emissions . in order to investigate the detailed mechanism of cluster formation , we have initiated millimeter and submillimeter survey observations of the omc-2/3 region with the nobeyama millimeter array ( nma ) and atacama submillimeter telescope experiment ( aste ) . the omc-2/3 region is one of the nearest cluster - forming regions at a distance of 450 pc @xcite , and locates at the north of the orion a giant molecular cloud @xcite . mapping observations of the omc-2/3 region in the 1.3 mm continuum emission have identified 26 dusty cores @xcite , and 33 dusty cores in the 850 @xmath18 m emission @xcite . @xmath5co ( @xmath4=10 ) survey observations of molecular outflows in omc-2/3 have found nine outflows at an average separation of @xmath0 0.2 pc @xcite , and our recent submillimeter @xmath5co ( @xmath4=32 ) survey observations of omc-2/3 have revealed presence of 15 outflows @xcite . extensive mapping observations of the omc-2/3 region in the cs ( @xmath4=10 ) @xcite and h@xmath2co@xmath3 ( @xmath4=10 ) lines @xcite have unveiled physical properties of cloud cores in the cluster - forming region , which are quantitatively different from those in isolated star - forming regions . interferometric h@xmath2co@xmath3(@xmath4=10 ) observations of mms 7 in the omc-3 region , one of the dusty cores , have revealed the disk - like envelope around mms 7 , which is being dispersed by the associated molecular outflow @xcite . in this paper , we focus on observations of the omc-2 fir 3/4 region ( including fir 3 , 4 , and 5 ; * ? ? ? * ) fir 4 is the strongest 1.3 mm dust - continuum source in omc-2 @xcite . in this region , there are three 3.6 cm free - free emission sources of vla 11 ( consistent with fir 3 ) , vla 12 and vla 13 @xcite , and nine mir sources @xcite , suggesting that fir 4 is an active cluster - forming region . a molecular outflow driven by fir 3 has been found in the @xmath5co ( @xmath4=10 , 32 ) lines @xcite , which is one of the candidate triggers of the cluster formation in fir 4 . our millimeter interferometric and submillimeter single - dish observations of the fir 4 region should shed light on the detailed mechanism of the cluster formation , which will be shown in the present paper . we carried out millimeter interferometric observations of the fir 4 region in the h@xmath2co@xmath3 ( @xmath4=10 ; 86.754 ghz ) , @xmath5co ( @xmath4=10 ; 115.271 ghz ) , sio ( @xmath6=0 , @xmath4=21 ; 86.847 ghz ) , and cs ( @xmath4=2 - 1 ; 97.980 ghz ) lines with the nobeyama millimeter array ( nma ) , which consists of six 10 m antennas , during a period from 2005 may to 2007 january . the data in the h@xmath2co@xmath3 ( @xmath4=10 ) , @xmath5co ( @xmath4=10 ) , and sio ( @xmath6=0 , @xmath4=21 ) lines were obtained with the fx correlator , which was configured with 1024 channels per baseline and a bandwidth of 32 mhz . for the @xmath5co ( 10 ) and sio ( 21 ) data we made 5-channel binning to increase the signal - to - noise ratio ( @xmath19 s / n ) of the high - velocity line - wing emission . thus , the velocity resolution in the h@xmath2co@xmath3 , @xmath5co , and sio observations is 0.108 , 0.406 , and 0.539 km s@xmath20 , respectively . the cs ( @xmath4=21 ) data were obtained with the digital spectral correlator , ultra wide band correlator ( uwbc ) @xcite , which has 128 channels and a 1024 mhz bandwidth per baseline ( 1 channel = 24.47 km s@xmath20 ) . the hanning window function was applied to reduce side lobes in the sampling . thus , the effective velocity resolution was widened to 48.94 km s@xmath20 ( corresponding to 2 channels ) . the cs ( @xmath4=21 ) emission is detected over the three uwbc channels with the hanning window function , and we integrated over these three channels ( corresponding to 97.88 km s@xmath20 ) to make the total integrated intensity map . we also obtained continuum data at both the lower ( 87.090 @xmath21 0.512 ghz ) and upper ( 98.418 @xmath21 0.512 ghz ) sidebands with uwbc , simultaneously with the h@xmath2co@xmath3 ( @xmath4=10 ) data . to obtain a higher s / n in the continuum map , the data of both sidebands were co - added ( effective observing frequency = 92 ghz = 3.3 mm ) . using the aips package developed at nrao , we adopted both the uniform and natural uv weighting for the continuum imaging . for the imaging of the molecular emissions we adopted the natural weighting . table [ nmaobs ] and [ continuumobs ] summarize the parameters for the line and continuum observations , respectively . since the minimum projected baseline length of the h@xmath2co@xmath3 ( @xmath4=10 ) , @xmath5co ( @xmath4=10 ) , sio ( @xmath6=0 , @xmath4=21 ) , and cs ( @xmath4=21 ) observations was 2.9 , 3.0 , 2.9 , and 3.2 k@xmath22 , our observations were estimated to be insensitive to structures more extended than 57@xmath1 ( 0.13 pc ) , 55@xmath1 ( 0.12 pc ) , 57@xmath1 ( 0.13 pc ) , and 51@xmath1 ( 0.11 pc ) at the 10 % level , respectively @xcite . the overall uncertainty in the flux calibration was estimated to be @xmath0 15 % . after the calibrations , only the data taken under good weather conditions were adopted in the imaging . the submillimeter ch@xmath7oh ( @xmath8=7@xmath96@xmath9 ; k=-1 , 0 , 2 ; 338.408681 , 338.344629 , 338.72165 ghz ) data in the fir 4 region have been taken with the aste 10 m telescope ( atacama submillimeter telescope experiment ; @xcite ) located at the pampa la bola ( altitude = 4800 m ) , chile . the observed frequency range covers the ch@xmath10oh k - ladder ( 7@xmath96@xmath9 , @xmath23=-2 , -1 , 0 , 2 , and 6 ) from 338.32 ghz to 338.74 ghz , where three of those ( 7@xmath246@xmath24 , 7@xmath256@xmath25 , 7@xmath266@xmath26 ) are detected above 3 @xmath27 level (= 0.51 k ) . observations were remotely made from the aste operation room at mitaka , tokyo on 2006 october 7 , 10 , and 11 , using the network observation system n - cosmos3 developed by naoj @xcite . the half - power beam width of the aste telescope is 22@xmath1 at the observing frequency . the typical system noise temperature with the 345 ghz sis heterodyne receiver in the dsb mode was 270 - 440 k , and the typical atmospheric opacity at 220 ghz was @xmath0 0.05 during our observations . the temperature scale was determined by the chopper - wheel method , which provides us with the antenna temperature corrected for the atmosphere attenuation . as a backend , we used four sets of a 1024-channel auto - correlator , which provide us a frequency resolution of 31.12 khz , corresponding to 0.1 km s@xmath20 at the ch@xmath7oh ( @xmath8=7@xmath96@xmath9 ) frequency . the on - the - fly ( otf ) mapping technique was employed to map the @xmath28 region centered on fir 4 , which covers the entire outflow driven by fir 3 seen in the @xmath5co ( 32 ) emission @xcite . the telescope pointing was checked every two hours by 5-points scans of the point - like @xmath5co ( @xmath4=32 ) emission from o - cet ( @xmath29= 02@xmath30 19@xmath31 20@xmath3280 @xmath33= -02@xmath34 08@xmath35 40@xmath367 ) . the pointing errors were measured to be from 0@xmath362 to 1@xmath362 during the observing run . we also observed ngc 1333 iras 2a and compared the observed ch@xmath10oh spectra to those obtained with the caltech submillimeter observatory ( cso ) telescope , which has the same dish size as that of the aste telescope @xcite , and found that the main beam efficiency of the aste telescope was @xmath0 0.4 . after subtracting linear baselines , the otf data were convolved by a gaussian - tapered bessel function @xcite with a fwhm of 14@xmath1 and were resampled onto a 7@xmath1 grid . since the telescope beam is a gaussian with a fwhm of 22@xmath1 , the effective fwhm resolution in the restored images is 26@xmath1 . the scanning effect " was minimized by combining scans along the r.a . and decl . direction , using the plait algorithm developed by @xcite . the typical rms noise level in the final image is 0.17 k in @xmath37 at a velocity resolution of 0.5 km s@xmath20 . figure [ line_map_co ] shows the distribution of the high - velocity blueshifted and redshifted @xmath5co ( @xmath4=32 ) and @xmath5co ( @xmath4=10 ) emission in the fir 4 region , observed with aste and nma , respectively . the @xmath5co ( @xmath4=32 ) data were taken by our group as a part of our survey project in the omc-2/3 region @xcite . the @xmath5co ( @xmath4=32 ) line profile toward fir 4 is shown in figure [ 12cospec ] . from the peak temperature of the @xmath5co ( @xmath4=32 ) emission , we determine that the gas kinetic temperature in the fir 4 region is @xmath0 52.5 k. the detected velocity range of the @xmath5co ( @xmath4=32 ) and @xmath5co ( @xmath4=10 ) emission is -9.5 @xmath0 29.4 km s@xmath20 and 3.7 @xmath0 18.3 km s@xmath20 , respectively , while that of the h@xmath2co@xmath3 ( @xmath4=10 ) emission , which traces the dense gas in fir 4 , is 10.4 @xmath0 12.5 km s@xmath20 @xcite . there are both blueshifted and redshifted high - velocity components at the north - east ( ne ) and south - west ( sw ) of fir 3 . these results imply that the high - velocity components in the @xmath5co ( @xmath4=32 , 10 ) emission trace a bipolar molecular outflow nearly along the plane of the sky driven by fir 3 . furthermore , the high - velocity components of the @xmath5co emission at the north - east of fir 3 are associated with h@xmath38 knots found by @xcite , which also supports our outflow interpretation . in addition , the velocity range of the @xmath5co ( 32 ) emission is larger than that of the @xmath5co ( 10 ) emission . one of the plausible interpretations of the different velocity range is that the submillimeter line traces warmer gas in the vicinity of the jet , where the higher velocity is expected , due to the different excitation condition between the submillimeter and millimeter lines @xcite . the size of the sw blue lobe measured from the @xmath5co ( @xmath4=32 ) image is @xmath0 0.09 pc , which is twice smaller than that of the ne lobe ( @xmath0 0.19 pc ) . at the south - west of fir 3 both the @xmath5co ( @xmath4=32 ) and @xmath5co ( @xmath4=10 ) emission exhibit blueshifted and redshifted peaks near fir 4 , and that these aste and nma peaks appear to be consistent with each other . on the other hand , at the north - east of fir 3 the prominent blueshifted and redshifted peaks observed with aste are not seen in the nma maps , suggesting that these components are resolved out by the interferometer . these results imply that the sw component of the outflow driven by fir 3 toward fir 4 is more compact than the ne component . from our nma observations , the peak position of the sw red lobe is measured to be @xmath29=05@xmath30 35@xmath31 27@xmath322 , @xmath33=-05@xmath34 09@xmath35 59@xmath365 , and at the tip of the red lobe there is 3.6 cm free - free emission ( vla 12 , open blue circle ) . there are two possibilities on the origin of this free - free emission . one is that there is another protostar located at vla 12 , and the other is that this free - free emission is originated from the ionized jet driven by fir 3 . since there is no 3.3-mm dust continuum counterpart associated with vla 12 ( see @xmath39 3.4 ) , the origin of the 3.6 cm free - free emission is probably the ionized jet driven by fir 3 . figure [ cs ] shows a total integrated intensity map in the cs ( @xmath4=21 ) emission . the peak of this emission is ( @xmath40 ) = ( 05@xmath30 35@xmath31 27@xmath320 , -05@xmath34 09@xmath35 56@xmath360 ) , @xmath0 4@xmath1 north - east of fir 4 . this emission extends from fir 3 to fir 4 , similarly to the high - velocity @xmath5co ( @xmath4=10 , 32 ) emission shown in figure [ line_map_co ] . thus , one of the possible interpretations on the origin of the cs emission is moderately dense gas in the molecular outflow . higher velocity - resolution observations in the cs emission are required to verify the origin of the cs emission in the fir4 region . figure [ h13co+]a shows a total integrated intensity map of the h@xmath2co@xmath3 ( @xmath4=10 ) emission in the fir 4 region , observed with the nma . there exists a @xmath0 0.07 pc - scale ( @xmath0 30@xmath1 ) dense gas structure associated with fir 4 . this dense gas structure has also been identified as ac17 with the single - dish 45 m telescope by @xcite . hereafter we call this structure fir 4 clump " . the peak position of fir 4 clump ( @xmath41=05@xmath3035@xmath3127@xmath320 , @xmath42=-05@xmath34 10@xmath35 03@xmath365 ) , estimated by the 2-dimensional gaussian fitting to the image , is consistent with the position of ac17 , and the systemic velocity ( @xmath43v@xmath44@xmath0 11.3 km s@xmath20 ) and the line width ( @xmath0 1.1 km s@xmath20 ) estimated by the gaussian fitting to the h@xmath2co@xmath3 spectrum at fir 4 are also consistent with the single - dish results @xcite . figure [ h13co+]b shows the distribution of the blue- ( 10.5 km s@xmath20 - 11.2 km s@xmath20 ) and redshifted ( 11.3 km s@xmath20 - 12.4 km s@xmath20 ) h@xmath2co@xmath3 ( @xmath4=10 ) emission in fir 4 . the blueshifted component is distributed from south - east to north - west while the redshifted component is distributed from east to west . we derived the mass of fir 4 clump ( @xmath19 m@xmath45 ) on the assumption of the local thermodynamic equilibrium ( lte ) condition and the optically - thin h@xmath2co@xmath3 ( @xmath4=10 ) emission , using the following equation @xcite , @xmath46 } [ exp\left ( \frac{hb(j+1)(j+2)}{kt_{ex } } \right ) ] \int f_{\nu } d\nu , \label{h13co+lte_mass}\ ] ] where @xmath47 , @xmath48 , @xmath49 , @xmath50 , @xmath51 , x[h@xmath2co@xmath3 ] , @xmath17 , @xmath53 , @xmath54 , @xmath4 , b and @xmath55 are the boltzmann constant , the planck constant , the speed of light , the excitation temperature , distance , fractional abundance of h@xmath2co@xmath3 , dipole moment , the mean mass of the molecule ( 2.33 a.m.u ) , frequency , rotational quantum number , rotational constant , and the flux density , respectively . we assume that the excitation temperature of the h@xmath2co@xmath3 ( @xmath4=10 ) line is equal to 52.5 k , which is derived from the peak brightness temperature of the @xmath5co ( @xmath4=32 ) emission ( see @xmath39 3.1 ) . we adopt @xmath17 = 4.07 debye as the h@xmath2co@xmath3 dipole moment @xcite . the h@xmath2co@xmath3 ( @xmath4=10 ) abundance of 4.5 @xmath15 10@xmath56 is adopted from @xcite , and the derived mass of fir 4 clump is @xmath0 10.5 @xmath57 ( if we adopt the abundance of 1.4 @xmath15 10@xmath58 estimated by @xcite , the mass of fir 4 clump is @xmath0 3.6 @xmath57 ) . we also estimated the momentum ( @xmath59 ) and the internal gas energy ( @xmath60 ) of fir 4 clump as follows @xcite ; @xmath61 @xmath62 @xmath63 where @xmath64 , @xmath65 , t@xmath66 , and m@xmath67 is the fwhm velocity width of the h@xmath2co@xmath3 line , the effective sound speed , the gas kinetic temperature , and the mass mean molecular of the observed molecule . table [ physical_qua ] summarizes our estimates of these parameters . due to the effect of the missing flux in the interferometric observations ( @xmath0 83 % of the total flux is missed ) , these values should be considered as a lower limit , and in fact , the estimated mass from the single - dish data is @xmath0 35 @xmath57 @xcite . we have detected the submillimeter ch@xmath7oh ( @xmath8=7@xmath96@xmath9 ; k=-1 , 0 , 2 ) lines toward fir 4 clump with aste . figure [ line_map_shock ] ( a ) and ( b ) show the distribution of the blue- and redshifted ch@xmath7oh ( @xmath8=7@xmath256@xmath25 ) emission , and figure [ spectral ] shows the spectrum of the ch@xmath7oh ( @xmath8=7@xmath256@xmath25 ) line toward fir 4 . the line width of the submillimeter line is up to 10 km s@xmath20 at the zeroth level , and there is slight indication of the asymmetric line profile , with a more prominent blue wing ( 5.7 km s@xmath20 - 9.9 km s@xmath20 ) than a red wing ( 13.1 km s@xmath20 - 14.2 km s@xmath20 ) . the peak position of the blue and red lobe is ( @xmath41 , @xmath42)=(5@xmath30 35@xmath31 26@xmath326 , -5@xmath34 09@xmath35 55@xmath364 ) and ( 5@xmath30 35@xmath31 27@xmath321 , -5@xmath34 09@xmath35 55@xmath364 ) , respectively . the distribution of the blueshifted ch@xmath7oh emission extends along the east - west direction , while the distribution of the redshifted emission extends toward the direction of the outflow driven by fir 3 , i.e. along sw - ne . we have also detected the sio ( @xmath6=0 , @xmath4=21 ) emission toward fir 4 clump with the nma , as shown in figure [ line_map_shock ] ( c ) . the peak position of the sio emission is ( @xmath41 , @xmath42 ) = ( 05@xmath30 35@xmath31 26@xmath326 , -05@xmath34 09@xmath35 53@xmath360 ) . the sio emission is blueshifted , point - like , and locates at @xmath07@xmath1 north of fir 4 . the peak position of the sio emission is approximately consistent with the peak position of the blueshifted ch@xmath7oh emission . figure [ continuum]a and [ continuum]b show naturally- and uniformly - weighted 3.3-mm continuum images in the fir 4 region observed with the nma . in these images there appears two main components , one is associated with fir 3 and the other with fir 4 . the overall distribution is consistent with that of the single - dish 1.3 mm continuum observations @xcite . the overall ridge distribution of the 3.3-mm emission associated with fir 4 also resembles that of the h@xmath2co@xmath3 ( @xmath4=10 ) emission , although there appears north - eastern extension in the 3.3-mm emission . the north - eastern extension in the 3.3-mm emission may trace the wall " of the molecular outflow , as seen in other observations @xcite however , the high - velocity co lobes are unlikely to be the cloud - core components traced by the h@xmath2co@xmath3 emission , as discussed in section 3.1 . from these considerations , we suggest that most of the 3.3-mm emission traces the dense - gas component in fir 4 , with a contamination from the outflow component to a certain extent . fir 3 was detected by our 3.3 mm dust - continuum observations as a single source , although two mir sources mir 21 and 22 , are present in fir 3 @xcite . @xcite have suggested that mir 21 and 22 are binary stars at a separation of @xmath0 3@xmath1 (= 1350 au ) . however , we consider that mir 21 and mir 22 are reflection nebulae evacuated by the outflow from fir 3 , because the 3.3 mm peak at fir 3 locates between mir 21 and mir 22 and the alignment of mir 21 and 22 is consistent with the direction of the molecular outflow . the dusty component associated with fir 4 is most likely the same identity traced by the h@xmath2co@xmath3 ( @xmath4=10 ) emission , that is , fir4 clump . moreover , the dusty component in fir 4 has subpeaks . these peaks are probably substructures ( @xmath43 cores ) in fir 4 clump . we identified those cores with the following criteria ; ( 1 ) the peak intensity should be higher than 4 @xmath68 noise level , and ( 2 ) the `` valley '' among different cores should be deeper than 1 @xmath27 . we identified five cores in the naturally - weighted image , and six more cores in the uniformly - weighted image . then the total number of the identified cores is eleven . hereafter , we call these smaller - scale cores fir 4a , b, ,k , as labeled in figure [ continuum]b . as we mentioned in @xmath39 3.1 . , there is no 3.3-mm continuum source toward vla 12 . the deconvolved size of these dusty cores was estimated by the 2-dimentional gaussian fitting to the image . the mass of the cores ( @xmath19 @xmath69 ) was derived from the total 3.3 mm flux , @xmath70 , on the assumption that all the 3.3-mm continuum emission arises from dust and that the emission is optically thin , using the formula , @xmath71 where we have adopted a value of the mass opacity , @xmath72 @xmath73 g@xmath20 @xcite and @xmath74=2 . @xcite estimated the value of @xmath75 to be 2 by the sed fitting toward fir 1 and fir 2 in the omc-2 region . since fir 4 locates on the same molecular filament of omc-2 as fir 1 and fir 2 , we consider that @xmath75 = 2 is probably acceptable for fir 4 . for the dust temperature we adopted @xmath76=52.5 k , which is derived from the @xmath5co ( @xmath4=32 ) emission ( see @xmath39 3.1 ) . the mean gas density ( @xmath19 @xmath77 ) in these dusty cores was derived by assuming a spherically - symmetric shape as follows ; @xmath78 where @xmath79 and @xmath80 are the deconvolved size along the major and minor axis . the typical size , mass and the density of these cores are estimated to be @xmath0 5 @xmath1 (= 2250 au ) , 0.78 @xmath57 , and 1.6 @xmath15 10@xmath81 @xmath82 , respectively . the total mass of fir 4 clump estimated from the 3.3-mm dust continuum emission is 9.4 m@xmath11 , which is consistent with that estimated from the h@xmath2co@xmath3 emission ( 10.5 m@xmath11 ) . the average density of fir 4 clump is 9.8 @xmath15@xmath83 @xmath82 , assuming the spherically symmetric core with a radius of 15@xmath1 ( 6750 au ) . table [ ident ] summarizes these physical properties of the identified cores . from the naturally- and uniformly - weighted interferometric 3.3 mm flux in fir 4 , the 850 @xmath17 m flux is estimated to be 1.2 and 0.58 jy , respectively , on the assumption of @xmath75 = 2 , while the 850 @xmath17 m flux measured with jcmt is 7.5 jy @xcite . then , @xmath0 84 and 92 % of the total flux are missed in the naturally- and uniformly - weighted interferometric 3.3 mm continuum images . then , the estimated mass of these cores ( @xmath19 m@xmath84 ) should be considered as lower limits . in table [ ident ] , we also list the mass of these cores corrected for the effect of the missing flux ( @xmath19 m@xmath85 ) , which can be considered as upper limits of the mass . several pieces of evidence for the interaction between the molecular outflow driven by fir 3 and fir 4 clump were found in our observations . the extent of the sw outflow lobe from fir 3 observed in the co ( 10 , 32 ) lines ( @xmath0 0.09 pc ) is shorter than that of the ne lobe ( @xmath0 0.19 pc ) . this suggests that the propagation of the sw outflow is hampered by the material at the head of the outflow , and in fact fir 4 clump locates at the tip of the south - western outflow . these pieces of morphological evidence suggest that the south - western outflow component is dammed by fir 4 clump . we also detected the sio and submillimeter ch@xmath10oh emission at the tip of the south - western outflow . these molecular lines are often observed toward shocked regions , caused by the interaction between outflows and dense gas @xcite . hence , the detection of the sio and submillimeter ch@xmath10oh emission in fir 4 is probably the chemical evidence for the interaction between the molecular outflow and the dense gas . in addition , the line width of the h@xmath2co@xmath3 emission at fir 4 ( dv@xmath86 @xmath0 1.1 km s@xmath20 ) is larger than the average value in other omc-2/3 cores ( dv@xmath86 @xmath0 0.8 km s@xmath20 ; * ? ? ? * ) which may suggest the larger turbulence due to the interaction with the outflow . figure [ pv]a - d show position - velocity ( p - v ) diagrams in the @xmath5co , h@xmath2co@xmath3 , sio , and the ch@xmath10oh emission . the p - v diagrams in the @xmath5co , h@xmath2co@xmath3 , sio , and the ch@xmath10oh emission show successive change of the line width along the direction from fir 3 to fir 4 . first , the @xmath5co ( 10 ) emission shows a distinct l - shaped structure in the p - v diagram , and the location of the line broadening is at the most upstream . second , the ch@xmath10oh emission also shows a similar l - shape with a less broadening , which locates slightly at the downstream from the co l - shape . the sio emission shows the most extreme blueshifted components , which locates at the peak of the blueshifted ch@xmath7oh component . finally , the dense gas component traced by the h@xmath2co@xmath3 emission locates at the most downstream of the successive emission distribution and is pinched by the l - shaped co emission distribution in the p - v diagram . these successive emission distributions and the velocity structures are likely to trace the detail of the interaction . these observational results suggest an interconnection among the molecular outflow , dense gas , and the shock traced by the sio and ch@xmath10oh emission . we consider that these pieces of evidence support the presence of the interaction between the molecular outflow and dense gas . in the last section , we suggest that there is an interaction between the outflow driven by fir 3 and dense gas of fir 4 clump . on the other hand , in @xmath39 3.4 we have demonstrated that fir 4 clump consists of eleven dusty cores . from these results , we presume that the interaction causes the fragmentation of fir 4 clump into these cores and eventually the formation of the next generation of the cluster members . hereafter we will examine this possibility from consideration of the jeans instability , time scale , and the virial analyses . first , we will examine whether the fragmentation into the dusty cores can be caused by the jeans instability , by comparing the average separation of the cores to the jeans length . the average 3-dimensional separation ( @xmath19 @xmath87 ) among these cores can be estimated by the following formula @xcite ; @xmath88 @xmath89 where @xmath90 , @xmath91 , @xmath92 , @xmath93 , and @xmath94 are the volume filling factor , the average radius of the cores , the number of cores , and the average volume of fir4 clump and a single core , respectively . v@xmath95 was estimated from the average of the projected size of the dusty cores ( table [ core ] ) , and v@xmath96 was estimated from the radius of 15@xmath1 , on the assumption of the 3-dimensional spherical symmetry ( see eq . the estimated v@xmath95 , v@xmath96 , and @xmath91 are @xmath97 @xmath98 , @xmath99 @xmath98 , and @xmath100 au , respectively . then , the average separation @xmath87 is estimated to be @xmath0 @xmath101 au . in order to estimate the jeans length ( @xmath19 @xmath102 ) , we use the following equation @xcite ; @xmath103 where @xmath65 , g , and @xmath104 are the effective sound speed , gravitational constant , and the average density of fir 4 clump , respectively . @xmath65 is the effective sound speed in fir 4 clump ( @xmath0 0.62 km s@xmath20 ) derived from equation ( 4 ) , and @xmath104 is estimated to be 9.8 @xmath15 10@xmath105 @xmath82 in @xmath39 3.4 . with these values , the jeans length @xmath106 is estimated to be @xmath0 @xmath107 au , which is on the same order of the average separation of the dusty cores . hence it is possible that the fragmentation of fir 4 clump into cores was caused by the jeans instability . the jeans instability in fir 4 clump may be triggered by the interaction with the outflow from fir 3 , and in the next section we will discuss whether the fragmentation was caused after the interaction between the outflow driven by fir 3 and fir 4 clump . we can estimate the time scale of the fragmentation on the assumption that the fragmentation time scale ( @xmath19 @xmath108 ) is the sound crossing time ; @xmath109 then @xmath108 is estimated to be @xmath0 @xmath110 yr . next , we estimate the time scale of the interaction between the outflow and dense gas , on the assumption that the interaction time scale @xmath111 is similar to the dynamical time @xmath112 of the north - eastern outflow driven by fir 3 , i.e. @xmath111 @xmath0 @xmath112 . @xcite estimated @xmath112 to be @xmath113 yr . then the interaction time scale @xmath111 is estimated to be @xmath113 yr . therefore , the fragmentation time scale ( 3.8 @xmath15 10@xmath114 yr ) is similar to the interaction time scale ( 1.4 @xmath15 10@xmath114 yr ) , and we suggest that the fragmentation was triggered by the interaction with the outflow from fir 3 . moreover , the free - fall time in this region ( @xmath0 1.3 @xmath13 yr ) , which is estimated from the average density of each core ( @xmath0 1.6 @xmath15 10@xmath81 @xmath82 ) , is also similar to the interaction time - scale , and hence it is possible that class 0/i protostars are already formed in the course of the interaction @xcite . in fact , there exists an mir source ( mir 24 ) identified as a class 0 protostar , and it is possible that mir 24 is formed in the course of the interaction . we suggest that in this particular case of fir 4 the interaction between the outflow and the dense gas most likely triggers the fragmentation into cores , while we can not exclude other possibilities such as the induced fragmentation due to the interaction with the hii region ( m43 ) @xcite as well as spontaneous fragmentation due to the gravity @xcite and turbulence @xcite . finally , in order to examine whether these dusty cores have a potential to form protostars inside , we compare the h@xmath26 mass of the cores estimated from the 3.3 mm dust - continuum flux and the virial mass . the virial mass ( @xmath19 m@xmath115 ) is estimated by the following equation ; @xmath116 where @xmath91 , @xmath65 and g are the radius of the core , the effective sound speed ( equation 4 ) and the gravitational constant , respectively . here , we assume that the h@xmath2co@xmath3 emission is associated with all the dusty cores . then we adopted the h@xmath2co@xmath3 line width ( @xmath64 ; see table [ ident ] ) at each peak of the dusty core as the line width in the dusty cores . the estimated virial masses are within the range of the expected h@xmath26 mass of these cores , that is , from m@xmath84 to m@xmath85 ( table [ ident ] ) . these results suggest that the observed dusty cores have a potential to form stars . from the above considerations , we suggest that the interaction between fir 4 clump and the outflow driven by fir 3 caused the fragmentation of fir 4 clump into cores and triggered the next generation of cluster formation . our scenario of the star formation in the fir 4 region is summarized in figure [ scenario ] as ; [ step 1 ] : fir 3 was born and drove the outflow ; [ step 2 ] : the outflow driven by fir 3 started interacting with fir 4 clump ; [ step 3 ] : the interaction caused the fragmentation of fir 4 clump into cores ; [ step 4 ] : these cores will form stars . the present stage of the star formation in the fir 4 region may be between step 3 and step 4 . we speculate that dusty cores will form protostars inside , and eventually , next generation of the stellar cluster in the fir 4 region . possible outflow - triggered star formation has also been reported observationally in ngc 1333 @xcite , l1551 ne @xcite , and ngc 2264 irs 1 @xcite . in ngc 1333 , @xcite found that one of the cores in the 850 @xmath17 m emission ( sk-1 ) locates at the tip of the shell evacuated by the outflow , and they suggest that sk-1 is a possible example of the outflow - triggered star formation . in l 1551 ne , @xcite have found an arc - shape structure of dense gas open toward south - west , or the direction of the molecular outflow driven by l1551 irs 5 , in the cs ( 32 ) emission . they interpret that the dense gas in l1551 ne is affected by the outflow from l1551 irs 5 , creating an arc - shape , and the protostellar formation of l1551 ne . in ngc 2264 irs 1 , @xcite have found that a dense - gas shell seen in the h@xmath2co@xmath3 emission locates around the outflow driven by irs 1 , and at the inner edge of the dense shell there are three compact cores . they suggest that these cores are entrained or compressed material formed by the outflow and that these cores may represent future sites of the formation of a group of the low - mass stars . our observational studies in the omc-2 fir 3/4 region have suggested that the interaction between the outflow and dense gas can cause the formation of stellar clusters . recent 3-dimensional mhd simulations by @xcite suggest that protostellar outflows can trigger cluster formation through shock compression , providing a theoretical support to our scenario . we have carried out high angular - resolution ( @xmath0 3@xmath1 - 7@xmath1 ) millimeter interferometric observations of omc-2 fir 3/4 with the nma in the h@xmath2co@xmath3 ( @xmath4=10 ) , @xmath5co ( @xmath4=10 ) , cs ( @xmath4=21 ) , and sio ( @xmath6=0 , @xmath4=21 ) lines as well as in the 3.3 mm continuum emission , and submillimeter single - dish observations with aste in the @xmath5co ( @xmath4=32 ) and ch@xmath7oh ( @xmath8=@xmath117@xmath118 ) lines . the main results of our new millimeter and submillimeter observations are summarized as follows : 1 . we suggest that the outflow driven by fir 3 interacts with the 0.07 pc - scale dense gas associated with fir 4 ( @xmath19 fir 4 clump ) , from morphological , kinematical , chemical and physical evidence . in the morphological evidence , we found that the length of the sw outflow lobe from fir 3 observed in the co ( 10 , 32 ) lines ( @xmath0 0.09 pc ) is shorter than that of the ne lobe ( @xmath0 0.19 pc ) , and fir 4 clump is located at the tip of the sw molecular outflow . these results suggest that the sw lobe has been dammed by the interaction with fir 4 clump . in the chemical and physical evidence , we detected the sio ( @xmath6=0 , @xmath4=2 - 1 ) emission and the submillimeter ch@xmath7oh ( @xmath8=7@xmath9 - 6@xmath9 ; k=-1,0,2 ) emission around fir 4 clump , which are known to trace shocks caused by the interaction between outflows and dense cores . in the kinematical evidence , the p - v diagrams in the @xmath5co , ch@xmath10oh , sio , and h@xmath2co@xmath3 emission show a successive change of the line width along the direction from fir 3 to fir 4 . the @xmath5co ( 10 ) emission show a distinct l - shaped structure in the p - v diagram , and the location of the broadening is at the most upstream . the ch@xmath10oh emission also shows a similar l - shape at the downstream from the co l - shape , and the sio emission shows the most blueshifted component at the peak of the blueshifted ch@xmath7oh component . the dense gas component traced by the h@xmath2co@xmath3 emission locates at the most downstream of the successive emission distribution and is pinched by the l - shaped co emission distribution in the p - v diagram . these results indicate the presence of the interaction between the outflow and dense gas in fir 4 . 2 . in the 3.3-mm continuum emission , we have first resolved fir 4 clump into eleven dusty cores with a size of 1500 - 4000 au and a mass of 0.2 - 1.4 m@xmath11 . the 3-dementinal separation among these cores , 5000 au , is on the same order of the jeans length , 13000 au . moreover , the estimated time scale of the fragmentation into the cores , 3.8 @xmath119 yr , is also similar to the time scale of the interaction between fir 4 clump and the outflow from fir 3 , that is , 1.4 @xmath13 yr . therefore , it is possible that the fragmentation of fir 4 clump into the cores has been caused by the jeans instability triggered by the interaction with the outflow driven by fir 3 . we suggest that in this particular case of fir 4 the outflow interaction most likely triggers the fragmentation into cores . we speculate that the dusty cores in fir 4 will form protostellar sources eventually , and that the fir 4 region is in the course of cluster formation . we suggest that the outflow driven by fir 3 triggered the next generation of the cluster formation in the fir 4 region , and that the fir 4 region is one of the promising samples of the outflow - triggered cluster formation . we are grateful to the staffs at the nobeyama radio observatory ( nro ) for both operating the nma and helping us with the data reduction , and to m. yamada , y. kurono , and t. tsukagoshi for their helpful comments . nro is a branch of the national astronomical observatory , national institutes of natural sciences , japan . we also acknowledge the aste staffs for both operating aste and helping us with the data reduction . we would like to thank the anonymous referee for providing helpful suggestions to improve the paper . observations with aste were ( in part ) carried out remotely from japan by using ntt s gemnet2 and its partner r&e ( research and education ) networks , which are based on accessnova collaboration of university of chile , ntt laboratories , and national astronomical observatory of japan . this work was supported by grant - in - aid for scientific research a 18204017 . aso , y. , tatematsu , k. , sekimoto , y. , nakano , t. , umemoto , t. , koyama , k. , & yamamoto , s. 2000 , , 131 , 465 avery , l. w. , & chiao , m. 1996 , , 463 , 642 bachiller , r. , prez gutirrez , m. , kumar , m. s. n. , & tafalla , m. 2001 , , 372 , 899 bergin , e. a. , neufeld , d. a. , & melnick , g. j. 1998 , , 499 , 777 chini , r. , reipurth , b. , ward - thompson , d. , bally , j. , nyman , l .- a . , sievers , a. , & billawala , y. 1997 , , 474 , l135 emerson , d. t. , & graeve , r. 1988 , , 190 , 353 ezawa , h. , kawabe , r. , kohno , k. , & yamamoto , s. 2004 , , 5489 , 763 gatley , i. , becklin , e. e. , mattews , k. , neugebauer , g. , penston , m. v. , & scoville , n. 1974 , , 191 , l121 genzel , r. , & stutzki , j. 1989 , , 27 , 41 gueth , f. , bachiller , r. , & tafalla , m. 2003 , , 401 , l5 haese , n. n. , & woods , r. c. 1979 , chem . 61 , 396 hester , j. j. , & desch , s. j. 2005 , chondrites and the protoplanetary disk , 341 , 107 hildebrand , r. h. 1983 , , 24 , 267 hirano , n. , mikami , h. , umemoto , t. , yamamoto , s. , & taniguchi , y. 2001 , , 547 , 899 ikeda , n. , sunada , k. , & kitamura , y. 2007 , , 665 , 1194 inutsuka , s .- i . , & miyama , s. m. 1997 , , 480 , 681 johnson , j. j. , gehrz , r. d. , jones , t. j. , hackwell , j. a. , & grasdalen , g. l. 1990 , , 100 , 518 johnstone , d. , & bally , j. 1999 , , 510 , l49 jrgensen , j. k. , hogerheijde , m. r. , blake , g. a. , van dishoeck , e. f. , mundy , l. g. , & schier , f. l. 2004 , , 415 , 1021 jrgensen , j. k. , et al . 2007 , , 659 , 479 kobayashi , n. , yasui , c. , tokunaga , a. t. , & saito , m. 2005 , protostars and planets v , 8639 kamazaki , t. , et al . 2005 , astronomical data analysis software and systems xiv , 347 , 533 kohno , k. , et al . 2004 , the dense interstellar medium in galaxies , 349 lada , c. j. , & lada , e. a. 2003 , , 41 , 57 li , z .- y . , & nakamura , f. 2004 , , 609 , l83 lis , d. c. , serabyn , e. , keene , j. , dowell , c. d. , benford , d. j. , phillips , t. g. , hunter , t. r. , & wang , n. 1998 , , 509 , 299 myers , p. c. , bachiller , r. , caselli , p. , fuller , g. a. , mardones , d. , tafalla , m. , & wilner , d. j. 1995 , , 449 , l65 mangum , j. , emerson , d. , & greisen , e. 2000 , imaging at radio through submillimeter wavelengths , 217 , 179 mardones , d. , myers , p. c. , tafalla , m. , wilner , d. j. , bachiller , r. , & garay , g. 1997 , , 489 , 719 momose , m. , ohashi , n. , kawabe , r. , nakano , t. , & hayashi , m. 1998 , , 504 , 314 moriarty - schieven , g. h. , johnstone , d. , bally , j. , & jenness , t. 2006 , , 645 , 357 motoyama , k. , umemoto , t. , & shang , h. 2007 , , 467 , 65 mikami , h. , umemoto , t. , yamamoto , s. , & saito , nakamura , f. , & li , z .- y . 2007 , , 662 , 395 nakano , m. , sugitani , k. , & morita , k. 2003 , , 55 , 1 nielbock , m. , chini , r. , mller , s. a. h. 2003 , , 408 , 245 norman , c. , & silk , j. 1980 , , 238 , 158 okumura , s. k. , et al . 2000 , , 52 , 393 ohashi , n. , hayashi , m. , ho , p. t. p. , momose , m. , tamura , m. , hirano , n. , & sargent , a. i. 1997 , , 488 , 317 peng , r. , langer , w. d. , velusamy , t. , kuiper , t. b. h. , & levin , s. 1998 , , 497 , 842 raga , a. , & cabrit , s. 1993 , , 278 , 267 reipurth , b. , rodrguez , l. f. , & chini , r. 1999 , , 118 , 983 saito , m. , kawabe , r. , kitamura , y. , & sunada , k. 1996 , , 473 , 464 saito , m. , sunada , k. , kawabe , r. , kitamura , y. , & hirano , n. 1999 , , 518 , 334 saito , m. , kawabe , r. , kitamura , y. , & sunada , k. 2001 , , 547 , 840 sakamoto , s. , hayashi , m. , hasegawa , t. , handa , t. , & oka , t. 1994 , , 425 , 641 sandell , g. , & knee , l. b. g. 2001 , , 546 , l49 takahashi , s. , saito , m. , takakuwa , s. , & kawabe , r. 2006 , , 651 , 933 takahashi , s. , saito , m. , takakuwa , s. , kusakabe , n. , shimajiri , y. , tamura , m. , & kawabe , r. 2008 , in prep . takakuwa , s. , kamazaki , t. , saito , m. , & hirano , n. 2003a , , 584 , 818 takakuwa , s. , ohashi , n. , & hirano , n. 2003b , , 590 , 932 takakuwa , s. , et al . 2004 , , 616 , l15 takakuwa , s. , kamazaki , t. , saito , m. , yamaguchi , n. , & kohno , k. 2007a , , 59 , 1 takakuwa , s. , et al . 2007b , , 662 , 431 tatematsu , k. , et al . 1993 , , 404 , 643 williams , j. p. , de geus , e. j. , & blitz , l. 1994 , , 428 , 693 williams , j. p. , blitz , l. , & stark , a. a. 1995 , , 451 , 252 williams , j. p. , plambeck , r. l. , & heyer , m. h. 2003 , , 591 , 1025 wilner , d. j. , & welch , w. j. 1994 , , 427 , 898 yasui , c. , kobayashi , n. , tokunaga , a. t. , terada , h. , & saito , m. 2006 , , 649 , 753 yokogawa , s. , kitamura , y. , momose , m. , & kawabe , r. 2003 , , 595 , 266 yu , k. c. , bally , j. , & devine , d. 1997 , , 485 , l45 lcccc configuration@xmath120 & d , c and ab & d & d & d , c and ab + baseline [ k@xmath22 ] & 2.9 - 115 & 3.0 - 30.8 & 2.9 - 23.4 & 3.2 - 113.6 + phase reference center ( j2000 ) & + primary beam hpbw [ arcsec ] & 77@xmath1 & 62@xmath1 & 77@xmath1 & 70@xmath1 + synthesized beam hpbw [ arcsec ] & 9@xmath3654@xmath1225@xmath3615 & 6@xmath3659@xmath15 5@xmath3679 & 10@xmath3622 @xmath15 6@xmath3637 & 6@xmath3668 @xmath15 4@xmath3600 + velocity resolution [ km s@xmath20 ] & 0.108 km s@xmath20 & 0.406 km s@xmath20 & 0.539 km s@xmath20 & 48.94 km s@xmath20 + gain calibrator@xmath123 & + bandpass calibrator@xmath124 & 3c273 , 0420 - 014 , 3c84 , 0528 + 134 & 3c273 , 3c84 & 3c273 , 3c84 & 3c273 , 0420 - 014 , 3c84 , 0528 + 134 + system temperature in dsb [ k]@xmath125 & 100 - 400 k & 150 - 300 k & 100 - 350 k & 100 - 400 k + rms noise level [ jy beam @xmath20 ] & 1.2@xmath12210@xmath20 jy beam @xmath20 & 5.7@xmath12210@xmath20 jy beam @xmath20 & 2.0@xmath12210@xmath20 jy beam @xmath20 & 5.1@xmath12210@xmath127 jy beam @xmath20 + [ nmaobs ] lcc baseline [ k@xmath22 ] & 2.9 - 115 & 2.9 - 115 + weighting & natural & uniform + beamsize ( hpbw ) [ arcsec ] & 6@xmath3696@xmath122 4@xmath3603 ( p.a.=-34.59@xmath34 ) & 6@xmath3655@xmath1223@xmath3634 ( p.a.=-40.31@xmath34 ) + p.a . of the beam [ @xmath34 ] & -34.59 & -40.31 + rms noise level [ jy beam@xmath20 ] & 1.2@xmath12210@xmath127 & 1.4@xmath12210@xmath127 + [ continuumobs ] lcccccccccc fir 3 & 05@xmath129 35@xmath130 27@xmath326 & -05@xmath34 09@xmath35 34@xmath360 & [email protected] & 140.46 & 3.0 & 18.7 & 3.1 & & & + fir 4a@xmath131 & 05 35 27.4 & -05 09 49.0 & [email protected] & 34.05 & 0.51 & 3.2 & 1.8 & 1.1 & 0.53 & 0.47 + fir 4b@xmath131 & 05 35 26.6 & -05 09 51.0 & [email protected] & 150.20 & 1.1 & 6.9 & 2.6 & 2.6 & 1.28 & 0.68 + fir 4c & 05 35 27.0 & -05 09 54.0 & [email protected] & 168.88 & 0.45 & 5.9 & 0.59 & 1.6 & 0.57 & 0.48 + fir 4d & 05 35 26.2 & -05 09 55.9 & [email protected] & 139.87 & 0.76 & 9.6 & 2.2 & 1.2 & 0.55 & 0.47 + fir 4e & 05 35 26.9 & -05 09 57.5 & [email protected] & 4.302 & 1.0 & 14.3 & 0.93 & 3.8 & 1.31 & 0.69 + fir 4f & 05 35 26.6 & -05 09 58.9 & [email protected] & 168.26 & 0.6 & 7.8 & 2.0 & 1.7 & 0.98 & 0.59 + fir 4g@xmath131 & 05 35 26.6 & -05 09 59.4 & [email protected] & 179.53 & 1.2 & 7.5 & 1.3 & 2.1 & 0.81 & 0.54 + fir 4h & 05 35 27.1 & -05 10 04.5 & [email protected] & 156.55 & 0.64 & 8.3 & 0.70 & 3.3 & 1.25 & 0.67 + fir 4i & 05 35 26.5 & -05 10 06.0 & [email protected] & 162.23 & 0.21 & 2.7 & 0.34 & 1.7 & 0.70 & 0.50 + fir 4j@xmath131 & 05 35 27.2 & -05 10 08.0 & [email protected] & 111.31 & 1.38 & 8.8 & 3.7 & 1.7 & 0.90 & 0.56 + fir 4k@xmath131 & 05 35 26.3 & -05 10 16.4 & [email protected] & 165.76 & 0.67 & 4.2 & 1.5 & 2.1 & 1.07 & 0.61 + fir 4 clump & & & & & 9.4 & 58.6 & 0.098@xmath132 & & & + [ core ] [ ident ] co ( @xmath4=32 ) ( left ) and @xmath5co ( @xmath4=10 ) ( right ) emission in the omc-2/fir 4 region . panel ( a ) and ( c ) show the distribution of the blue ( -9.5 km s@xmath20 - 4.6 km s@xmath20 ) and red ( 16.4 km s@xmath20 - 29.4 km s@xmath20 ) lobe in the @xmath5co ( @xmath4=32 ) line observed with aste @xcite , while panel ( b ) and ( d ) show the distribution of the blue ( 3.7 km s@xmath20 to 6.1 km s@xmath20 ) and red ( 15.1 km s@xmath20 to 18.3 km s@xmath20 ) lobe in the @xmath5co ( @xmath4=10 ) line observed with the nma , respectively . open black circles show the field of view of the nma observations , green crosses the position of mir sources @xcite , blue circles the position of 3.6 cm free - free sources @xcite , and red squares the peak position of the 1.3 mm dust - continuum emission @xcite . filled ellipses at the bottom left corner of each panel indicate the beam size . contour levels of these maps start at @xmath133 5 @xmath27 levels with an interval of 5 @xmath27 . the rms noise levels ( 1 @xmath27 ) of the panel ( a ) , ( b ) , ( c ) and ( d ) are 1.2 k , 0.45 jy , 1.2 k , and 0.70 jy beam@xmath20 , respectively.,width=415 ] co ( @xmath4=32 ; 345.705 ghz ) spectrum at fir 4 taken with aste @xcite the rms noise level ( 1@xmath68 ) is 0.5 k in t@xmath134 at a velocity resolution of 1.1 km s@xmath20 . a dashed line shows the systemic velocity ( @xmath0 11.3 km s@xmath20 ) , obtained from the gaussian fit to the spectrum.,width=566 ] = 21 ) emission in the omc-2/fir 4 region , obtained with the nma . symbols in the figure are the same as in figure [ line_map_co ] . contour levels of this map start at @xmath133 3 @xmath27 levels with an interval of 1 @xmath27 . the rms noise level ( 1 @xmath27 ) of this map is 5.1@xmath1510@xmath127 jy beam@xmath20 . , width=566 ] co@xmath3 ( @xmath4=10 ) emission ( left ) , and maps of the blueshifted ( 10.5 - 11.2 km s@xmath20 ) the redshifted ( 11.3 - 12.4 km s@xmath20 ) h@xmath2co@xmath3 components ( right ) in the fir 4 region , obtained with nma . symbols in the figure are the same as in figure [ line_map_co ] . contour levels of these maps start at @xmath133 3 @xmath68 levels with an interval of 1 @xmath27 . the rms noise levels ( 1 @xmath27 ) in panel ( a ) , the blue and red contour in panel ( b ) are 0.81 , 0.49 and 0.49 jy beam@xmath20 , respectively . , width=566 ] oh ( @xmath8=7@xmath256@xmath25 ) and nma sio ( @xmath6=0 , @xmath4=21 ) maps in the fir 4 region . panel ( a ) and ( b ) show the distribution of the blue ( 5.7 km s@xmath20 - 9.9 km s@xmath20 ) and red ( 13.1 km s@xmath20 - 14.2 km s@xmath20 ) lobe in the ch@xmath7oh ( @xmath8=7@xmath256@xmath25 ) line , respectively . panel ( c ) shows the distribution of the blue lobe in the sio ( @xmath6=0 , @xmath4=21 ) line at a velocity range from 3.4 km s@xmath20 to 6.1 km s@xmath20 . symbols in the figure are the same as in figure [ line_map_co ] . contour levels in panel ( a ) and ( b ) start at @xmath133 5 @xmath27 levels with an interval of 5 @xmath27 . contour levels in panel ( c ) start at @xmath133 3 @xmath68 levels with an interval of 1 @xmath68 . the rms noise levels ( 1 @xmath27 ) in panel ( a ) , ( b ) and ( c ) are 0.024 , 0.015 k , and 0.11 jy beam@xmath20 , respectively.,width=566 ] oh ( @xmath8=7@xmath256@xmath25 ; 338.408 ghz ) spectrum at fir 4 taken with aste . the rms noise level ( 1@xmath27 ) is 0.17 k in t@xmath135 at a velocity resolution of 0.5 km s@xmath20 . a dashed line shows the systemic velocity ( @xmath0 11.3 km s@xmath20 ) in the fir 4 region estimated from the nma h@xmath2co@xmath3 spectrum.,width=566 ] . contour levels start at @xmath133 3 @xmath68 noise levels with an interval of 1 @xmath68 . the rms noise level ( 1 @xmath68 ) in panel ( a ) and ( b ) are 1.2@xmath1510@xmath127 and 1.4@xmath1510@xmath127 jy beam@xmath20 , respectively . , width=566 ] co ( @xmath4=10 ) ( upper left ) , h@xmath2co@xmath3 ( @xmath4=10 ) ( upper right ) , sio ( @xmath6=0 , @xmath4=21 ) ( lower left ) and ch@xmath7oh ( @xmath8=7@xmath256@xmath25 ) ( lower right ) emission in the fir 4 region . two horizontal lines in each panel show the position of fir 3 and fir 4 , while a vertical line the systemic velocity of 11.3 km s@xmath20 in fir 4 estimated from the h@xmath2co@xmath3 data . the cut line of panel ( a ) , ( b ) , and ( d ) is adopted to be along p.a . = 30@xmath34 , passing through the position of fir 3 and fir 4 . the cut line of panel ( c ) is taken to be along p.a . = 35@xmath34 , passing through the compact sio emission and fir 3 . contour levels in panel ( a ) and ( d ) start at @xmath133 5 @xmath27 levels with an interval of 5 @xmath27 . contour levels in panel ( b ) start at @xmath133 3 @xmath27 levels with an interval of 3 @xmath27 . contour levels in panel ( c ) start at @xmath133 2 @xmath27 levels with an interval of 1 @xmath27 . the rms noise level ( 1 @xmath27 ) in panel ( a)-(d ) are 0.57 jy beam@xmath20 , 0.12 jy beam@xmath20 , 0.2 jy beam@xmath20 , and 0.17k , respectively . ,

a multi - frequency , multi - configuration study has been made of the compact radio sources in the galactic center quintuplet and arches stellar clusters using the very large array . ten radio sources have been detected in the quintuplet cluster . the majority of these radio sources have rising spectral indices and are positionally coincident with young massive stars that are known to have powerful stellar winds . we conclude that the three most compact of these sources are produced by stellar wind emission ; thus , mass - loss rates can be derived and have an average value of 3 m yr . the remainder of the sources are likely to be a combination of stellar wind emission and free - free emission from surrounding ionized gas . in three cases , the radio sources have no stellar counterpart and the radio emission is thought to arise from compact or ultra - compact h2 regions . if so , these sources would be the first detections of embedded massive stars to be discovered in the galactic center clusters . the radio nebula associated with the pistol star resembles the nebula surrounding the lbv star carina and may be related to the stellar wind of the pistol star . ten compact radio sources are detected in the arches cluster and are interpreted to be stellar wind sources , consistent with previous findings . several of the sources show moderate variability ( 10 - 30% ) in their flux density , possibly related to a nonthermal component in the wind emission . a number of radio sources in both clusters have x - ray counterparts , which have been interpreted to be the shocked , colliding winds of massive binary systems .

astro-ph0508178

the radio arc near the galactic center ( gc ) ( @xmath530 pc in projection from the dynamical center of the galaxy , sgr a@xmath6 ) is considered to be one of the most unusual collections of sources in the galaxy @xcite . both thermal and non - thermal sources comprise the radio arc . the most prominent sources are two sets of curved thermal filaments @xmath7 the arched filaments ( g0.10 + 0.08 ) and the sickle ( [email protected] ) @xcite . these sources are all assumed to be at or very near the gc , which has a distance of 8.0 kpc ; thus 1 @xmath50.04 pc ( reid 1993 ) . one of the most exciting observational results of the last decade has been the near - infrared ( near - ir ) discovery of two extraordinary stellar clusters in the radio arc region , known as the quintuplet and arches clusters @xcite . these clusters have drastically changed our understanding of the stellar content in a region where the visual obscuration is 20@xmath730 magnitudes . the quintuplet cluster contains at least 10 wolf - rayet ( wr ) stars , and more than a dozen luminous `` ob supergiants '' @xcite . the unusual pistol star ( luminosity of @xmath510@xmath8 l@xmath2 ) is thought to be a member of the quintuplet cluster @xcite . this star has been classified ( along with one other quintuplet member , # 362 ) as a luminous blue variable ( lbv ) star @xcite . this type of evolved o - star ( age @xmath5 4 myr ) is thought to be undergoing short lived mass - loss during which up to several m@xmath2 of material are ejected from its surface . the arches cluster appears to be an even more extraordinary cluster with @xmath535 wr stars as well as close to 100 luminous , evolved ob stars , making it one of the most densely - packed clusters in the galaxy @xcite . in total , nearly 50 stars have been classified as wr stellar types in the radio arc region @xcite . such massive clusters are known to have a major impact on the interstellar medium . figure 1 shows the locations of these two clusters relative to the main features in the radio arc . the quintuplet and arches clusters are responsible for the ionization of the sickle and the arched filaments h2 regions @xcite . these luminous clusters with n@xmath9 @xmath510@xmath10 photons s@xmath3 ( quintuplet ) and @xmath54 @xmath0 10@xmath11 photons s@xmath3 ( arches ) can easily account for the average photon fluxes required from 8.5 ghz radio continuum observations ( 3 - 30 @xmath010@xmath12 photons s@xmath3 ) @xcite . radio emission from the individual stellar winds of such massive stars should be detectable at centimeter wavelengths based on the predictions of @xcite and @xcite . indeed , high - resolution radio images made with the very large array ( vla ) of the national radio astronomy observatory ( nrao ) have revealed a number of compact radio sources in these two clusters @xcite . based on their spectral characteristics and positional associations with near - ir stellar sources , the radio sources are interpreted to be stellar wind emission in both clusters . the radio flux densities indicate that many of the sources have high mass - loss rates ( m @xmath13 2 @xmath010@xmath1 m@xmath2 yr@xmath3 ) . the initial results of the quintuplet cluster radio sources presented by @xcite were based on low - resolution ( 2 - 10 ) archival vla data centered on the sickle h2 region and therefore were not ideal for detections of compact sources with sizes @xmath141 . here we present new , high resolution ( 0.3 - 3 ) vla images at 4.9 , 8.5 , 22.5 and 43.4 ghz of the quintuplet and arches stellar clusters that reveal new and variable radio sources and provide more accurate determination of the spectral indices and source sizes . although lang et al . ( 2001b ) presented the first sub - arcsecond imaging of the arches cluster , here we provide a second epoch image at 8.5 ghz to search for new and variable radio sources in this cluster and an image of the cluster at 22.5 ghz to better characterize the spectral indices and sizes of the sources . the observations and data reduction are presented in @xmath152 , the results in @xmath153 , and the interpretation and discussion follow in @xmath154 . vla multifrequency observations were made during 1999@xmath72004 of the quintuplet cluster , pistol star , and the arches cluster using the a , bna , b , d and dnc array configurations . table 1 summarizes the observational details . at 4.9 and 8.5 ghz , j1331 + 305 and j1751 - 253 were used as flux and phase calibrators . standard procedures for calibration , editing and imaging were carried out using the astronomy image processing software ( aips ) of the nrao . at the higher frequencies ( 22.5 and 43.4 ghz ) , the observations were carried out differently . in the bna and b array at both 22.5 and 43.4 ghz , the atmospheric phase fluctuations can be very pronounced , so that a fast cycle time between source and phase calibrator ( sgr a@xmath6 ) was used , with a total cycle time of 2 minutes ( 90 seconds on source , 30 seconds on calibrator ) . the pointing was checked every hour using 8.5 ghz observations of 1751 - 253 . in the dnc and d arrays at 22.5 and 43.4 ghz , a total cycle time of 5 minutes was employed ( 260 seconds on source , 40 seconds on calibrator ) . pointing was also checked every hour by using 8.5 ghz observations of j1751 - 253 . the absolute flux density scale was determined from the observations of j1331 + 305 , with an assumed flux density of 2.52 jy at 22.5 ghz and 1.45 jy at 43.4 ghz and is used to measure a flux density of 1.3 jy at 22.5 ghz and 1.4 jy at 43.4 ghz for sgr a@xmath6 . as the primary beam at 43.4 ghz is only 1 , the 43.4 ghz observations were centered on the extended sources qr1 - 3 ( see figure 2 ) rather than the center of the quintuplet cluster since one of the goals of this study was to determine the flux densities of these unusual sources . all images were created in aips using the standard imaging program imagr . table 2 summarizes the final image resolutions and rms noise levels . the images shown in figures 2 - 6 have not been corrected for primary beam attenuation , but flux densities were measured in images that had been corrected for this attenuation . because the primary beams at 4.9 and 8.5 ghz are very large ( 9 and 5.4 , respectively ) , and because the quintuplet and arches clusters are located near two very extended h2 regions ( see figure 1 ) , it is necessary to filter out part of the extended emission in the images . it is possible to do this with a radio interferometer by removing out the shortest _ ( u , v ) _ baselines in the _ ( u , v ) _ data , which correspond to the largest sized structures in the resulting image . table 2 lists the cutoffs made to the data in the _ ( u , v ) _ plane ( in kilo - wavelengths ) and the corresponding size scales ( in arcseconds ) above which the flux is subsequently filtered out ( i.e. , the largest angular size scale present ) . in addition , at 22.5 ghz , two images of the quintuplet cluster were produced : ( a ) a `` high '' resolution ( @xmath140.5 ) image using _ only _ the b and bna array data to investigate the most compact sources , and ( b ) a `` low '' resolution ( @xmath51.5 ) image using all the available data to investigate the more extended sources . figure 2 shows the 8.5 ghz image of the radio sources within approximately 25 ( @xmath51 pc ) of the quintuplet cluster center . the pistol nebula is a prominent extended radio source in this image and the near - ir position of the pistol is represented by a cross . the pistol nebula is believed to be the ejecta from a previous mass - loss event of the pistol star , but also partly ionized by the quintuplet cluster stars @xcite . nine compact radio sources are detected and we refer to them as qr1 - 9 . qr1 - 5 were previously identified by @xcite . qr6 - 9 are newly identified in this image . the radio sources were identified based on their signal - to - noise ratio ( s / n @xmath135 in most cases ) . the source qr9 has s / n @xmath53 and is located outside of figure 2 , but was identified based on its positional association with a near - ir source ( see @xmath153.1.2 below for more details on near - ir counterparts ) . flux densities at all four frequencies ( 4.9 , 8.5 , 22.5 and 43.4 ghz ) for qr1 - 9 and the pistol star were measured in primary beam corrected images and are listed in table 3 . the aips fitting routine jmfit was used to measure the flux densities and deconvolved sizes for qr4 - 9 . the background levels in the area where qr1 - 3 are located ( to the north of the quintuplet cluster center ; see figure 2 ) vary strongly because of the proximity to the radio arc nonthermal linear filaments . in addition , the sources qr1 - 3 have lower surface brightnesses ( @xmath50.2 mjy beam@xmath3 at 8.5 ghz ) than the more compact sources qr5 - 8 ( average value of @xmath50.6 mjy beam@xmath3 at 8.5 ghz ) , which makes the flux densities more difficult to calculate . in order to estimate accurate flux densities for these sources , the flux densities of several representative background regions were measured using the viewer tool in the _ aips++ _ software package ( http://aips2.nrao.edu ) . this tool was also used to integrate over the region of each source at all four frequencies simultaneously . the source flux densities were then corrected for the background levels at each frequency to determine the final flux densities listed in table 3 . several sources were not detected at the higher frequencies . at 22.5 ghz , the source qr7 is not detected , most likely because of confusion in the vicinity of the pistol nebula . secondly , the source qr9 lies at the edge of the primary beam at 22.5 ghz , and attenuation likely prevents a detection . at 43.4 ghz , detections of sources qr4 - 8 were not made , probably due to the low resolution of the beam at this frequency ( @xmath52 ) , the fact that these sources are fairly compact ( @xmath161 ) , and the larger rms level in this image ( @xmath5100 @xmath17jy beam@xmath3 ) . qr9 is not detected at 43.4 ghz as it lies outside of the 1 primary beam . spectral indices , @xmath18 , are calculated for all sources and are compiled in table 4 . upper limits for flux densities at 4.9 ghz can be estimated for several of the sources not detected at this frequency , with resulting lower limits for the 4.9 to 8.5 ghz and 4.9 to 22.5 ghz spectral indices . qr1 - 3 and the pistol star show spectral indices consistent with being flat ( @xmath18@[email protected] to @xmath70.1 with average errors of @xmath190.4 ) . qr4 - 9 show rising spectral indices in the range of @xmath18 = + 0.4 to + 2.0 , also with large errors ( @xmath190.2 to 0.8 ) . several sources exhibit complex structure , which makes determination of the spectral index difficult . figure 2 ( 8.5 ghz image ) and figure 3 ( the `` high '' resolution 22.5 ghz image ) show that qr5 is a compact source . however , in the `` low '' resolution image at 22.5 ghz and in the 43.4 ghz image , this source has a compact center , but appears to have associated extended emission . therefore , spectral indices are calculated for qr5 by comparing the flux densities in images where the source structure is similar ( i.e. , comparing the 8.5 ghz with the 22.5 ghz `` high '' resolution image and the 22.5 ghz `` low '' resolution image with the 43.4 ghz image ) . qr7 is compact at 4.9 ghz , but fairly extended at 8.5 ghz and thus the spectral index is difficult to measure , but appears to be positive . qr9 is not detected 4.9 ghz and is detected only with a signal to noise ratio of 3.5 in the 8.5 ghz observations . we therefore estimate a lower limit for its spectral index of @xmath18 @xmath13 + 0.2 , implying that this source is likely to have a positive spectral index . the majority of the quintuplet radio sources are resolved and the deconvolved sizes and corresponding linear sizes are presented in table 3 . the deconvolved sizes were derived from the 8.5 ghz image ( figure 2 ) , although the sizes of qr5 - 9 were also measured in the `` high '' resolution image at 22.5 ghz where detected . the major axis sizes range from 0.3 ( 0.015 pc ) to 4.0 ( 0.17 pc ) . three sources ( qr6 , qr8 , and qr9 ) are unresolved at both 8.5 and 22.5 ghz . for comparison , figure 3 shows that qr5 is resolved and qr6 is unresolved in the 22.5 ghz `` high '' resolution image . sources qr1 - 3 are more extended than the other sources in the quintuplet cluster ; their sizes range from 1.5 ( 0.06 pc ) to 4 ( 0.16 pc ) . figure 4 shows a detailed image of the sources qr1 - 3 at 8.5 , 22.5 and 43.4 ghz . the crosses in figure 2 represent the positions of near - ir stars in the quintuplet cluster from @xcite with spectral classifications ( representing only a fraction of the total number of luminous stars in the cluster ) . the positions of 28 stars are shown in figure 2 ( crosses ) and represents the central region of the cluster . there is a systematic shift of positions between the near - ir and radio images ( @xmath20ra=0.9 ) , which is consistent with previous comparisons between _ hst_/nicmos images of the gc clusters and radio data @xcite . all radio sources in the quintuplet cluster appear to have near - ir counterparts within the positional uncertainty of 1 , except for qr1 - 3 ( discussed later in @xmath154.1.3 ) . table 4 gives the stellar counterparts from figer et al . ( 1999b ) . the near - ir counterparts to qr4 , qr5 , and the pistol star were known from the results of lang et al . ( 1999 ) , but the near - ir counterparts of qr6 - 9 are newly associated . the exact spectral classifications of these stars are not well known , although @xcite provide some rough classifications in their table 3 . all the near - ir counterparts to the radio sources in our study appear to be evolving massive stars ( @xmath5 few myr ) , with many classified as wr stellar types ( figer et al . 1999a , b ; cotera et al . in particular , the near - ir counterpart of qr7 is one of five stars in the quintuplet whose nature is unclear because the near - ir spectra are essentially featureless . @xcite suggest that these five stars should be categorized as dusty , wc late - type stars ( dwcl ) , which are the coolest carbon - rich wr stars that tend to form circumstellar dust shells . however , moneti et al . ( 2001 ) present mid - infrared spectroscopic observations of these stars , but find no evidence ( i.e. , circumstellar absorption lines ) to support the dwcl hypothesis . the near - ir counterpart of qr9 is also noteworthy as it has been classified as the second lbv star in the quintuplet ( # 362 ) besides the pistol star @xcite . the radio source associated with the pistol star is a prominent extended source located @xmath510 ( 0.4 pc ) to the south of the pistol nebula . a separate observation pointed in the direction of the pistol star was made in 1999 in order to study the radio emission associated with this extraordinary star . figure 5 shows that the radio source has a bipolar morphology , and that the pistol star lies at the center of this source . the extended emission has a size of approximately 4 ( 0.16 pc ) in the n - s direction and approximately 2 ( 0.08 pc ) in the e - w direction . the flux densities for the pistol radio source are summarized in table 3 . the spectral index of the radio emission is essentially flat ( @[email protected] ) . figure 6 shows the 8.5 ghz image of the arches cluster . nine radio sources with signal - to - noise greater than three are present in the images and are labelled as ar1 - 9 . ar10 is detected with a signal to noise ratio of only two , but is coincident with a near - ir counterpart . sources ar1 - 8 were previously identified by @xcite , and ar9 and 10 are newly detected . five of these sources are also detected at 22.5 ghz . the flux densities at 8.5 and 22.5 ghz were measured for ar1 - 10 using the aips task jmfit and are listed in table 5 . the 8.5 ghz observation presented here is a second epoch observation of this cluster and several sources showed moderate variability compared with the observations of @xcite : ar1 , ar3 , ar4 and ar8 varied by 12% , 29% , 30% , and 25% , respectively . the 22.5 to 8.5 ghz spectral index is calculated and also listed in table 5 . all of the sources except for ar6 show rising spectral indices . ar1 is the only source that is resolved at 8.5 ghz , and has a deconvolved size of [email protected] ( [email protected] pc ) . at 22.5 ghz , ar1 is partially resolved with a deconvolved size of 0.12 ( 0.005 pc ) along its major axis . although the 4.9 to 8.5 ghz spectral index for ar1 is rising ( @[email protected] ; lang et al . ( 2001b ) , the flux density of this source at 22.5 ghz is less than at 8.5 ghz . it is also possible that some of this extended emission may be missing from the image of ar1 at 22.5 ghz as the largest angular size detected in the bna array configuration of the vla at 22.5 ghz could be as small as 2 . for this reason , the 22.5 to 8.5 ghz spectral index is not calculated for ar1 . the crosses in figure 6 represent the positions of near - ir stars in the arches cluster from @xcite . eight of these stars had been previously detected by @xcite and @xcite and classified with emission - line features . the recent observations of @xcite confirm that at least thirty of the stars in the arches cluster have the spectral features characteristic of young , massive stars , including all ten of the near - ir counterparts to the ar1 - 10 radio sources . similar to the quintuplet , there is a systematic shift ( @xmath20ra=0.8 , @xmath20dec=0.5 ) between the near - ir and radio sources in the arches cluster . however , this shift is within 1 , which is typical given the astrometric uncertainty of _ hst_. the position of ar6 is found to be coincident with that of an emission line star ( f19 ) . there is also a weak radio source ( ar10 ; 2@xmath22 detection ) associated with the near - ir source to the w of ar6 and its near - ir counterpart . sources qr6 , qr8 , and qr9 show the major properties of stellar wind emission : ( 1 ) they have rising spectral indices in the range of + 0.5 to + 1.0 , ( 2 ) they are unresolved sources at the resolutions observed , which means their physical sizes are @xmath14 0.2 ( 0.009 pc ) at 22.5 ghz and @xmath140.5 ( 0.02 pc ) at 8.5 ghz , and ( 3 ) they are coincident positionally with young , massive stars . therefore , these three radio sources are interpreted to be stellar winds from quintuplet cluster stars . the flux densities of radio detections of stellar winds can be used to calculate mass loss rates ( @xmath23 ) . based on panagia & felli ( 1975 ) and wright & barlow ( 1975 ) , the mass loss rate can be estimated for an isotropic , homogeneous wind with constant velocity , electron density and chemical composition . however , 20 - 25% of early - type stars are known to have a nonthermal ( and variable ) component in their winds ( abbott et al . 1984 ; bieging et al . 1989 ) . the typical spectral index of such sources is in the range of @xmath18@[email protected] to + 0.3 for frequencies between 5 and 43 ghz ( beiging et al . 1989 ; contreras et al . 1996 ) . contamination of the flux density by the nonthermal component can occur at the lower frequencies . therefore , higher frequency observations ( i.e. , 23 ghz compared to 5 ghz ) may be a better tracer of the thermal component and give more reliable mass - loss rates ( contreras et al . 1996 ) . the mass - loss rate can be estimated from our 22.5 ghz observations using : @xmath24 where s@xmath25 is the flux density of the source at 22.5 ghz , v@xmath26 is the terminal velocity of the wind ( assumed to be 1000 km s@xmath3 ; cotera et al . 1996 ; nagata et al . 1995 ) , and d is the distance to the source ( 8 kpc ) . we have assumed an electron temperature of 10@xmath27 k , z=1 , and a mean molecular weight , @xmath17=2 , due to the enrichment in heavy elements of the late - type wn stars ( leitherer et al . 1997 ) , a classification which figer et al . ( 1999 ) and figer et al . ( 2002 ) has assigned to most of the quintuplet and arches cluster stars . leitherer et al . ( 1997 ) suggest that there is very little variation in the mass - loss rates for wr stars , with an average value of @xmath54 @xmath0 10@xmath1 m@xmath2 yr@xmath3 . cappa et al . ( 2004 ) recently carried out a survey of galactic wr stars and came to similar conclusions concerning the measured radio mass - loss rates ; the average values from their study were 2 - 4 @xmath0 10@xmath1 m@xmath2 yr@xmath3 . values for the mass - loss rates of qr6 , qr8 and qr9 range from 3.2@xmath010@xmath1 m@xmath2 yr@xmath3 to 4.3@xmath010@xmath1 m@xmath2 yr@xmath3 ( see table 4 ) . these values agree well with those found for wr stars described above as well as for other young , massive stellar types . the nature of radio emission from sources qr4 , qr5 and qr7 is less clear than that from qr6 , qr8 and qr9 . although qr4 , qr5 and qr7 show rising spectra and are coincident with near - ir sources , their sizes are more extended ( @xmath131 at 8.5 ghz ) than expected for stellar wind sources . according to rodriguez et al . ( 1980 ) , the angular size of a stellar wind can be related directly to its flux density , independent of mass - loss rate and terminal wind velocity . for the 8.5 ghz flux densities of qr4 , qr5 and qr7 , the expected angular size of the corresponding stellar wind should be @xmath140.05 . the actual sizes of the sources ( 0.3 to 4 ) are clearly larger than the predicted value . however , as mentioned previously ( in @xmath152.2 ) , the quintuplet radio sources are located amidst a very complicated set of extended features , which includes the sickle and pistol h2 regions as well as the linear , nonthermal filaments of the radio arc . although filtering in the _ ( u , v ) _ plane can help by removing the extended structures , the quintuplet sources may still be contaminated by residual background emission , which can make determination of flux densities and sizes difficult . although the errors in size determination are substantial ( [email protected] ; 20 - 50% ) , the sizes are still much more extended than the predicted values for stellar winds ( e.g. , @xmath140.05 ) . mass - loss rates are calculated for these sources as described above , and the values for qr4 , qr5 and qr7 ( average m of 1.2 @xmath0 10@xmath28 m@xmath2 yr@xmath3 ) are significantly larger than for the stellar wind sources qr6 , qr8 and qr9 ( average m of 3.7@xmath010@xmath1 m@xmath2 yr@xmath3 ) . while values for m of @xmath5 10@xmath28 m@xmath2 yr@xmath3 are not unexpected for some mass - losing stars , they typically correspond to epochs when the star may be experiencing an enhanced rate of mass - loss from its surface ( i.e. , the lbv phase ; white et al . 1994 ) . therefore , the simple interpretation of stellar wind emission may not apply for qr4 , qr5 and qr7 . however , the rising spectra and extended sizes of these sources are consistent with partially - thick free - free emission associated with an inhomogeneous , but spatially unresolved , h2 region . therefore , qr4 , qr5 and qr7 could be inhomogeneous h2 regions or even combinations of stellar wind emission and optically - thin , free - free emission from associated ionized material ( perhaps leftover from the formation of the stars ) . based on mezger & henderson ( 1967 ) , we can estimate the physical properties of these regions of ionized gas , assuming they have become optically - thin at the highest frequency at which they are detected , using the deconvolved sizes of these sources , and using an electron temperature of 10@xmath27 k. table 4 lists the electron density , n@xmath29 , mass of ionized hydrogen , m@xmath30 , emission measure , em , and the production rate of ionizing photons , n@xmath9 for these sources . the values of these h2 properties for qr4 , qr5 and qr7 ( in addition to their linear sizes of 0.03 - 0.05 pc ) are consistent with those for compact and ultra - compact h2 regions ( wood & churchwell 1990 ) . the properties of qr1 - 3 differ from the other radio sources in the quintuplet . as described in @xmath153.1.1 , qr1 - 3 are more spatially extended than qr6 , qr8 and qr9 by at least a factor of 2 . more significantly , qr1 - 3 are the only radio sources in the two clusters without closely associated near - ir counterparts . these sources also define the northern edge of the quintuplet cluster ( 5 - 10 north of qr4 - 7 and the main concentration of stellar members ) . table 4 shows that the spectral indices are close to flat with large errors ( i.e. @[email protected]@xmath190.2 for @xmath21 for qr1 ) . such spectra can be attributed to optically thin , free - free emission associated with h2 regions . since there are no stellar counterparts to qr1 - 3 , a likely explanation for these sources is that they may represent embedded massive stellar sources ( churchwell 1990 ) . the linear sizes of qr1 - 3 of @xmath50.1 pc are consistent with the upper end of the size distribution of ultra - compact h2 regions in the sample of kurtz , churchwell , & wood ( 1994 ) . as only @xmath31% of ultra - compact h2 regions are detectable at near - ir wavelengths ( hanson , luhman , & rieke 2002 ) , the lack of near - ir emission associated with qr1 - 3 is also consistent with their being ultra - compact h2 regions . as described above , the h2 properties for qr1 - 3 are derived from the radio continuum and listed in table 4 . the production rate of ionizing photons , n@xmath9 , can provide an estimate of the embedded stellar type that is responsible for the associated radio emission ( assuming a single ionizing star ) . values for n@xmath9 range from @xmath32 s@xmath3 . these ionizing fluxes are consistent with embedded b - type massive stars ( @xmath5 b1v - b2v ; @xcite ) . however , a number of galactic ultra - compact h2 regions are observed to be associated with diffuse extended emission @xcite . these studies conclude that typically @xmath33% of the ionizing flux from the embedded stars in uc h2s escapes to the outer envelope . thus , the inferred stellar content from the @xmath34 values determined for sources qr1 - 3 may be an underestimate , in which case they could have spectral types earlier than b. therefore , we tentatively identify qr1 - 3 as ultracompact or compact h2 regions . the proximity of the quintuplet cluster radio sources ( and in particular , sources qr1 - 3 ) to the edge of a molecular cloud identified by serabyn & gsten ( 1991 ) may indicate that these sources are physically associated with this molecular cloud . in the cs ( j=3 - 2 ) image of serabyn & gsten , a finger of emission extends toward the projected location of qr1 - 3 , indicating the presence of moderately dense gas . the apparent physical association between qr1 - 3 and the edge of the molecular cloud suggests that these young stellar objects have recently emerged from the molecular cloud , probably as a result of erosion due to the more evolved massive stars in the quintuplet cluster . indeed , the quintuplet cluster is thought to be the source of ionization for the larger sickle h2 region ( lang , goss & wood 1997 ; figer et al . 1999a ) , which traces the inner edge of the molecular cloud . alternatively , the radio sources which are not interpreted to be stellar winds ( qr1 - 3 , qr4 , qr5 , and qr7 ) may represent clumps of molecular material leftover from the formation of the stars in the cluster that are externally ionized by the quintuplet cluster members . assuming these clumps are located at approximately the projected distances from the center of the cluster , the quintuplet could easily account for their ionization as it is responsible for the ionization of the much larger sickle h2 region ( located 1 or 2.5 pc to the ne ) and the north side of the pistol nebula ( lang et al . 1997 ; figer et al . if any of the quintuplet radio sources are confirmed to be compact or uc h2s , this detection will lead to the study of the earliest stages of stellar evolution in one of the most extreme environments in the milky way . in particular , these objects are located in a region of strong magnetic fields , high turbulent velocities , and extremely high metallicities ( morris & serabyn 1996 ) . figure 5 shows that the radio emission associated with the pistol star has a bipolar morphology centered around the position of the star and has a flat spectral index ( @xmath21 = + [email protected] ) . one explanation for this source is that it represents the current mass - loss of the star . ( the outer `` pistol - shaped '' nebula is thought to contain the mass associated with an episode of previous mass - loss ( @xmath5 11 m@xmath2 , @xmath55000 years ago ; figer et al . 1998 ) ) . in this case , the integrated flux density would imply a mass - loss rate of 2.7@xmath010@xmath28 m@xmath2 yr@xmath3 . this value is much higher than the average value for wr stellar types ( see @xmath154.1.1 ) , but similarly high mass - loss rates have been detected in other lbv stars during epochs of extreme mass - loss ( e.g. , @xmath4 car ; white et al . 1994 ) and in several wr colliding - wind binary systems ( contreras et al . 1996 ) . however , as for sources qr4 , qr5 , and qr7 , the extended size of the pistol star radio emission is much larger than the expected angular size for a stellar wind source ( @xmath50.1 ) using the methods of rodrguez et al . another problem with the interpretation of the pistol as a stellar wind source is that its double - lobed radio emission is not well modelled by a spherically - symmetric stellar wind surrouding the star . it is possible to calculate a mass - loss rate for the source using the total mass and assuming an expansion velocity for the radio emission . assuming that the radio emission is optically - thin , free - free emission , with t@xmath29@xmath510@xmath27 k , and taking the deconvolved angular sizes of the lobes to be 1@xmath35 , we can estimate the total mass in the lobes to be @xmath5 0.02 @xmath36 . the separation between the pistol star and each of the lobes is @xmath37 cm . assuming that the lobes are expanding at a velocity of 1000 km s@xmath3 , we estimate a mass - loss rate of @xmath38 @xmath36 yr@xmath3 . this value is roughly consistent with the value obtained by using equation ( 1 ) . in addition , it is possible to calculate the ionizing flux required to maintain the bipolar radio source . at a distance of 8 kpc and with the integrated flux density measured , n@xmath9 has a value of @xmath39 s@xmath3 , which can be provided by an early b - type star . however , the two lobes that constitute the radio source are intersecting only a fraction of the 4@xmath40 radians around the star . furthermore , the ionized nebula may be density - bounded , with most of the ionizing flux of the pistol star escaping to the interstellar medium , so the star is likely to be of an earlier spectral type . the physical properties of the radio emission associated with the pistol star are similar to the nebula surrounding the well - known lbv binary system @xmath4 carina ( davidson & humphreys 1997 ) . the bipolar - shaped radio emission around @xmath4 carina has a flat spectral index and a linear size of 0.12 pc ( 10 at 2.5 kpc ; white et al . 1994 ) , similar to the pistol star nebula . in fact , a variety of types of nebular features are detected around other lbv stars and are thought to be the result of stellar wind emission interacting with the surrounding interstellar medium ( nota et al . 1995 ) . in the cases of galactic sources hg car , ag car , wra-751 and he3 - 519 , both extended nebular shells and compact stellar wind emission are detected in the radio continuum ( duncan & white 2002 ) . these authors also calculate substantial mass - loss rates from the central lbv stars ( @xmath135 - 10 @xmath0 10@xmath1 m@xmath2 yr@xmath3 ) . the bipolar morphology of such radio nebulae suggests that the mass - loss is not isotropic ( e.g. , @xmath4 car ) . this type of non - spherical wind emission has been explained by several different mechanisms : ( 1 ) _ internal _ processes , such as the stellar rotation of such massive stars and the resulting mass - loss geometry ( owocki & gayley 1997 ; maeder & maynet 2000 ; smith et al . 2004 ) , and ( 2 ) _ external _ processes , such as the influence of a binary companion on the mass - loss geometry ( iben & livio 1993 ) or expansion of the nebula into a region of high density - contrast interstellar medium ( nota et al . 1995 ) . a comparison of the 8.5 ghz observation of the arches cluster presented here and the observation of lang et al . ( 2001b ) shows that the majority of the sources do not exhibit any strong variability in their flux density . however , moderate variability ( at the level of 12 - 30% ) is detected in several of the arches radio sources previously detected ( ar1 , ar3 , ar4 , ar8 ) and possibly explains the new detection of ar9 . the flux densities of two of the four sources decreased , while the flux density of the other two sources increased . therefore , we can rule out a systematic calibration problem . this moderate variability may indicate ( i ) the presence of a time - variable nonthermal component ( due to shocks in the wind , or colliding winds for a binary system ) and/or ( ii ) variability in the mass - loss rate or wind velocity ( contreras et al . 1996 ; leitherer et al . 1997 ) . nine of the ten arches sources are unresolved in both the 8.5 and 22.5 ghz observations . the major axis of ar1 is resolved at both 8.5 and 22.5 ghz and its deconvolved size is @xmath50.1 . this size is consistent with the expected angular size of 0.06 for an ionized stellar wind source of 1.9 mjy flux density at 8.5 ghz ( rodrguez et al . therefore , all radio sources in the arches cluster exhibit the properties of stellar wind sources ( see @xmath154.1.1 ) . mass - loss rates are also calculated for the arches sources and are listed in table 5 . ar1 has the largest value of m = 2.2 @xmath0 10@xmath28 m@xmath2 yr@xmath3 . such a high value of m is usually attributed to a particular mass - loss epoch during the evolution of a massive star or a colliding wind binary system ( white et al . 1994 ; contreras et al . 1996 ) . the remainder of the arches sources have an average value of @xmath53 @xmath0 10@xmath1 m@xmath2 yr@xmath3 , typical for the types of stars present in the arches and quintuplet clusters ( @xmath154.1.1 ; leitherer et al . 1997 ; cappa et al . 2004 ) . as reported recently by law & yusef - zadeh ( 2004 ) and previously by yusef - zadeh et al . ( 2002 ) , a number of the radio sources in the arches and quintuplet clusters have x - ray counterparts detected with the _ chandra x - ray observatory_. the x - ray emission associated with the arches cluster itself is one of the most luminous x - ray sources in the galactic center region with an x - ray luminosity in the 2 - 10 kev range of @xmath510@xmath41 ergs s@xmath3 ( wang , gotthelf & lang 2002 ) . three x - ray point like sources are detected in the arches cluster : a1s , a1n and a2 ( law & yusef - zadeh 2004 ) . the x - ray sources a1s and a1n can be associated with the radio sources ar4 and ar1 as well as near - ir sources from figer et al . the x - ray emission in these sources is thought to arise from shocks formed in the winds of massive stars ( o- and wr - types ) as are found in the arches and quintuplet clusters ( law & yusef - zadeh 2004 ) . the characteristics of several arches radio sources that have x - ray counterparts suggest that these sources are likely to have shocks in their winds and/or be colliding wind binaries . radio emission from the colliding winds of binary massive stars is well known to have a nonthermal component that flattens the spectral index ( dougherty & williams 2000 ) as well as showing time - variable emission ( abbott et al . 1984 ) . in the case of the radio source ar1 , the spectral index ( @[email protected] ; lang et al . 2001b ) is flatter than the canonical @xmath18=+0.6 value for stellar wind emission and the flux density of the source varies by 12% between epochs . ar4 has a steeper spectral index ( although the errors are larger ) , @[email protected] , but its flux density varied by 30% in the two observed epochs . in the quintuplet cluster , law & yusef - zadeh ( 2004 ) report that two of the radio sources ( qr6 and qr7 ) have probable x - ray counterparts . it is not possible to determine whether the spectral index of qr6 is flattened ( @[email protected] ) or consistent with the value of @xmath18=+0.6 . however , this source was not detected in the results of lang et al . ( 1999 ) , either because of its time - variable nature or because the resolution was too coarse . the possible detection ( 2.6@xmath22 as reported by law & yusef - zadeh 2004 ) of x - ray emission associated with quintuplet source qr7 is interesting as not much is known about the nature of the near - ir counterpart ( @xmath153.1.2 ; figer et al . 1999 ) . the arches and quintuplet clusters are two of the most massive and luminous clusters in our galaxy ( figer et al . 1999a ) . in particular , the stellar density of the arches cluster ( @xmath42 @xmath5 3 @xmath010@xmath43 m@xmath2 pc@xmath44 ) is comparable only to that found in the centers of clusters such as hd in ngc3603 ( moffat et al . 1994 ) and r136 in the large magellanic cloud , associated with the 30 doradus nebula ( massey & hunter 1998 ) . therefore , these clusters should have a profound effect on the surrounding ism . indeed , these clusters are thought to be responsible for the ionization of the adjacent h2 regions , the sickle and the arched filaments ( lang et al . 1997 ; lang et al . 2001a ) . the presence of x - ray emission associated with individual stellar wind sources ( see above ) also indicates that the winds of these massive stars are powerful . in addition to the point - like x - ray sources , both clusters also show diffuse x - ray emission ( yusef - zadeh et al . 2002 ; law & yusef - zadeh 2004 ) . in the arches cluster , the diffuse x - ray emission is apparent for @xmath51 pc surrounding the cluster , and in the quintuplet , the diffuse x - ray emission has a slightly larger extent ( @xmath51.2 pc ) but is not detected with high signal to noise ( law & yusef - zadeh 2004 ) . several authors have suggested that the diffuse emission is related to a `` cluster wind '' triggered by the collective effect of a large number of massive star winds in the cluster core ( cant et al . 2000 ; raga et al . 2001 ; yusef - zadeh et al . such processes can lead to shock acceleration of the wind particles and heating of the gas to high temperatures ( 10@xmath8 to 10@xmath45 k ) . nonthermal radio emission has been detected within a 9 radius ( 0.3 pc ) of the cluster core and is believed to provide additional evidence that particle acceleration occurs here ( yusef - zadeh et al . 2003 ) . the physical parameters of the `` cluster wind '' and the resulting region of x - ray emitting gas depend primarily on the individual mass - loss rates of the stellar wind sources ( stevens & hartwell 2003 ; cant et al . 2000 ; raga et al . the total stellar wind mass injection rate is dependent on the sum of mass - loss rates of individual stars multiplied by the square of the terminal wind velocity . using an estimate for the velocity , predicted by the models in stevens & hartwell ( 2003 ) , or by using the average value of @xmath51000 km s@xmath3 , measured by cotera et al . ( 1996 ) , it is possible to derive the total stellar wind kinetic energy injection rate . the observations presented here provide the most reliable mass - loss rate estimates for the stellar wind sources in the arches and quintuplet clusters . based on the mass - loss rates presented here , the total stellar wind mass injection rate for the arches cluster is @xmath54 @xmath0 10@xmath28 m@xmath2 yr@xmath3 , and for the quintuplet ( including the pistol star as well as sources qr4 , qr5 , and qr7 ) is @xmath57 @xmath0 10@xmath28 m@xmath2 yr@xmath3 . these values for total m are in close agreement with those used by stevens & hartwell ( 2003 ) and raga et al . ( 2001 ) in their calculation of the total stellar wind kinetic energy injection rate . two types of numerical calculations have also been made recently to characterize the x - ray `` cluster wind '' that arises from the arches ( and presumably , the quintuplet ) clusters . cant et al . ( 2000 ) and raga et al . ( 2001 ) present a solution for an adiabatic flow and predict an envelope of x - ray emission of t@xmath55 @xmath010@xmath8 k , extending well beyond ( 10 s of pc ) the cluster core ( 0.2 pc ) . in contrast , silich et al . ( 2004 ) employ radiative cooling in their work , and predict , for the arches cluster , a quasi - adiabatic solution that suggests a much smaller boundary around the cluster of x - ray emission . depending on the values chosen for m and the terminal velocity , silich et al . ( 2004 ) estimate 3 - 7 pc for the extent of the x - ray emitting region . these authors argue that this smaller size for the x - ray emitting region is in part due to the slowing of wind emission by interactions of the wind with cooler surrounding ism material . the arches cluster , in particular , is located in the midst of a dense molecular and ionized environment ( lang et al . 2002 ) and there is a large reservoir of cooler material which may decelerate the wind . morphologically , the arched filaments h2 regions and molecular cloud extend for as much as 20 pc in projection around the cluster . x - ray observations near the arched filaments show that the diffuse emission appears to be confined to the area near the easternmost of the arched filaments ( @xmath14 10 pc ) , as predicted by silich et al . however , the short - integration nature of these observations ( @xmath14 50 ks ) with _ chandra _ and the subsequent low sensitivity to diffuse objects have so far prevented detailed spectral fitting and spatial imaging of the diffuse x - ray emission associated with these clusters ( yusef - zadeh et al . 2002 ; wang , gotthelf & lang 2002 ) . several examples of radio clusters associated with regions of recent star formation have been reported in the literature over the last ten years ( e.g. , garay et al . 1987 ; churchwell et al . 1987 ; becker & white 1988 ; stine & oneal 1998 ; rodrguez et al . 1999 ; gmez et al . 2000 ; lang et al . the compact radio sources in the stellar clusters have a variety of origins : ionized stellar winds , ultracompact h2 regions , externally ionized proplyds , and magnetically active stars with gyrosynchrotron emission are the dominant types of sources . in table 6 we summarize the parameters of the radio sources in these clusters . although more of these clusters should be studied to have a reliable statistical base , some interesting trends are evident . the diameters of the radio clusters are in the 0.2 to 1.0 pc range ( with the exception of the quintuplet cluster with the pistol star included , that reaches up to 2 pc ) . the radio luminosity of the most luminous member is correlated with the bolometric luminosity of the cluster . clearly , the arches and quintuplet cluster have their most luminous radio member with luminosities one to three orders of magnitude larger than typical young clusters in the galactic disk . these clusters of radio sources could be a valuable observational tool to study the stellar population of heavily obscured regions of star formation when combined with high resolution spectroscopic observations . we have made multi - frequency radio observations of both the quintuplet and arches clusters using the vla . our conclusions can be summarized as follows : \(1 ) ten radio sources have been detected in the quintuplet cluster . the majority of these sources have rising spectral indices and are positionally coincident with near - ir sources which have been classified as young , massive stars . sources qr6 , qr8 and qr9 are unresolved and thought to be very compact in size ( @xmath140.01 pc ) ; therefore , these three sources are interpreted to be stellar winds . sources qr4 , qr5 and qr7 are more extended ( [email protected] ; 0.015 to 0.06 pc ) and likely represent a combination of stellar wind and free - free emission associated with h2 regions . sources qr1 - 3 do not have near - ir counterparts and these are interepreted to be the first detections of embedded massive stars in the quintuplet cluster . \(2 ) an observation of the pistol star shows that its structure is extended and distinctly bipolar , centered on the pistol star near - ir source . the radio emission from this nebula has a flat spectral index . assuming that this emission is optically thin , free - free emission , we can roughly calculate a mass - loss rate of 5 @xmath0 10@xmath28 m@xmath2 yr@xmath3 . the pistol star radio source resembles the nebula that surrounds the luminous blue variable star @xmath4 carina . \(3 ) ten compact radio sources have also been detected in the arches cluster . nine of the sources have spectral indices consistent with a rising spectrum and only one of the sources may be resolved . as found in lang et al . ( 2001b ) , these sources are interpreted to be stellar winds from the young massive stars in the cluster . the average mass - loss rate for these stellar winds is 3 @xmath0 10@xmath1 m@xmath2 yr@xmath3 . \(4 ) both point - like and diffuse x - ray emission has been identified with the arches and quintuplet clusters and thought to be related to the presence of colliding wind binaries in these clusters . the mass - loss estimates from our radio observations here provide the best estimates of one of the parameters used to model the x - ray `` cluster winds '' which are observed to be associated with a number of massive stellar clusters . c.c.l . acknowledges that this material is based on work supported by the national science foundation ( nsf ) under grant no . 0307052 . k.e.j gratefully acknowledges support for this work provided by the nsf through an astronomy & astrophysics postdoctoral fellowship , and nasa through hubble fellowship grant # hf-01173.02-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 . lcccccc quintuplet&17 46 15.00 & @xmath728 49 35.0&4.9 & bna&june 2002 & 4 + quintuplet&17 46 15.00 & @xmath728 49 35.0&8.5 & bna&june 2002 & 4 + quintuplet&17 46 15.00 & @xmath728 49 35.0&22.5 & bna&sept 2003&2 + quintuplet&17 46 15.00 & @xmath728 49 35.0&22.5 & b&jan . 2004&3 + quintuplet&17 46 15.00&@xmath728 49 35.0&22.5 & dnc , d&june 2004&3.5 + qr1 - 3 sources&17 46 14.60&@xmath728 49 20.0&43.4 & dnc&june 2004&2.5 + qr1 - 3 sources&17 46 14.60&@xmath728 49 20.0&43.4 & d & july 2004&3 + pistol & 17 46 15.24 & @xmath728 50 04.0 & 8.5 & b , bna & nov . 1999 & 4 + arches&17 45 50.41 & @xmath728 49 21.8 & 8.5 & a&jan . 2002&4 + arches&17 45 50.41 & @xmath728 49 21.8 & 22.5 & bna&sept . 2003 & 2.5 + lclccc quintuplet & 4.9 & 1.15 @xmath0 0.85 , pa=62.8&75&10&20.6 + quintuplet & 8.5 & 0.64 @xmath0 0.50 , pa=62.0&55&10&20.6 + quintuplet&22.5 & 0.39 @xmath0 0.22 , pa=52.7&65&15&13.8 + quintuplet & 22.5 & 1.45 @xmath0 0.87 , pa=06.8&75&@xmath7&@xmath7 + quintuplet ( qr1 - 3 ) & 43.4 & 2.54 @xmath0 1.35 , pa=03.3&100&@xmath7&@xmath7 + pistol & 8.5 & 1.17 @xmath0 0.59 , pa=15.2&75&25&8.2 + arches&8.5 & 0.42 @xmath0 0.20 , pa=01.6&25&20&10.3 + arches&22.5 & 0.33 @xmath0 0.20 , pa=47.8&85&10&20.6 + lcccccccc qr1 & 17 46 15.16 & @xmath728 49 22.5 & [email protected] & [email protected]&[email protected]&[email protected]&[email protected] & [email protected] + qr2 & 17 46 14.95 & @xmath728 49 20.2 & [email protected] & [email protected] & [email protected] & [email protected]&[email protected] & [email protected] + qr3 & 17 46 14.42 & @xmath728 49 19.4 & [email protected] & [email protected] & [email protected] & [email protected]&[email protected] & [email protected] + qr4 & 17 46 15.06 & @xmath728 49 29.4 & [email protected] & [email protected] & ... & ... & [email protected]&[email protected] + qr5 & 17 46 15.11 & @xmath728 49 37.2 & [email protected] & [email protected] & [email protected] & ... & [email protected] & [email protected] + qr5ext ... & ... & ... & ... & ... & [email protected] & [email protected] & [email protected] & [email protected] + qr6 & 17 46 15.14 & @xmath728 49 32.7 & [email protected] & [email protected] & [email protected] & ... & ps & ... + qr7 & 17 46 14.71 & @xmath728 49 40.8 & @xmath140.24 & [email protected] & ... & ... & [email protected]&[email protected] + qr8 & 17 46 14.04 & @xmath728 49 16.7 & @xmath140.24 & [email protected] & [email protected] & ... & ps & ... + qr9 & 17 46 17.98 & @xmath728 49 03.4 & @xmath140.24 & [email protected] & ... & ... & ps & ... + pistol & 17 46 15.32 & @xmath728 50 04.2 & [email protected] & [email protected]&[email protected] & [email protected] & [email protected] & [email protected] + lccccccccccc qr1 & @[email protected]&@[email protected] & @[email protected]&@[email protected] & @xmath7 & 1.5@xmath010@xmath46 & 0.023&2.4@xmath010@xmath43&1.2@xmath010@xmath47 & @xmath7 + qr2 & @[email protected] & @[email protected]&@[email protected]&@[email protected] & @xmath7 & 0.9@xmath010@xmath46 & 0.065&1.6@xmath010@xmath43 & 2.1@xmath010@xmath47 & @xmath7 + qr3 & @[email protected] & @[email protected]&@[email protected]&@[email protected] & @xmath7 & 0.9@xmath010@xmath46 & 0.030&1.1@xmath010@xmath43&1.0@xmath010@xmath47 & @xmath7 + qr4 & + [email protected] & @xmath7 & @xmath7 & @xmath7 & 1.5@xmath010@xmath28 & 2.7@xmath010@xmath46 & 0.008&4.8@xmath010@xmath43&7.9@xmath010@xmath48 & f270s + qr5 & + [email protected] & + [email protected] & + [email protected] & [email protected] & 1.1@xmath010@xmath28 & 3.0@xmath010@xmath46 & 0.017&7.1@xmath010@xmath43&1.8@xmath010@xmath47 & f241 + qr6 & + [email protected]&[email protected]&[email protected]&@xmath7 & 3.7@xmath010@xmath1 & @xmath7&@xmath7&@xmath7&@xmath7 & f257 + qr7 & + [email protected]&@xmath7&@xmath7&@xmath7 & 9.5@xmath010@xmath1 & 2.7@xmath010@xmath46 & 0.005&4.1@xmath010@xmath43&4.8@xmath010@xmath48 & f231 + qr8 & @xmath13 + 0.3&[email protected]&@xmath13 + 0.6&@xmath7 & 4.3@xmath010@xmath1 & @xmath7&@xmath7&@xmath7&@xmath7 & f320 + qr9 & @xmath13 + 0.2 & @xmath7&@xmath7&@xmath7 & 3.2@xmath010@xmath1 & @xmath7&@xmath7&@xmath7&@xmath7 & f362 + pistol & + [email protected]&[email protected]&[email protected]&[email protected] & 2.7@xmath010@xmath28 & @xmath7 & @xmath7 & @xmath7 & @xmath7 & f134 + lcccccccc ar1 & 17 45 50.42&@xmath728 49 22.3 & [email protected]&[email protected]&@xmath7&2.2@xmath010@xmath28&f6 , n8 , c8 & 0.14 @xmath00.10 + ar2 & 17 45 50.39&@xmath728 49 21.3 & [email protected]&[email protected]&[email protected] & 4.6@xmath010@xmath1 & f8 , n7 , c6&ps + ar3 & 17 45 50.20&@xmath728 49 22.3 & [email protected]&@xmath140.25&@xmath130.8&1.9@xmath010@xmath1&f1 , n4 , c9&ps + ar4 & 17 45 50.47&@xmath728 49 19.5 & [email protected]&[email protected]&[email protected]&4.8@xmath010@xmath1&f37 , n10 , c5&ps + ar5 & 17 45 50.57&@xmath728 49 17.5 & [email protected]&@xmath140.25&@xmath130.5&1.9@xmath010@xmath1 & f4 , n11,c2&ps + ar6 & 17 45 49.76&@xmath728 49 26.0 & [email protected]&@xmath140.25&@xmath130.0&3.0@xmath010@xmath1&f19&ps + ar7 & 17 45 50.83&@xmath728 49 26.4 & [email protected]&[email protected]&[email protected]&4.3@xmath010@xmath1&f3 , n14 , c11&ps + ar8 & 17 45 50.45&@xmath728 49 31.9 & [email protected]&[email protected]&[email protected]&3.7@xmath010@xmath1&f5 , n9&ps + ar9 & 17 45 50.47&@xmath728 49 17.9 & [email protected]&@xmath140.25&@xmath130.5&1.9@xmath010@xmath1&f18&ps + ar10 & 17 45 49.69&@xmath728 49 25.9 & [email protected]&@xmath140.25&@xmath131.5&1.9@xmath010@xmath1&f2 , n1 , c13&ps + lccccc orion & @xmath52@xmath010@xmath43 & 77 & 8.8 & 0.3 & zapata et al . ( 2004 ) + ngc 1579 & @xmath52@xmath010@xmath46 & 16 & 0.6 & 0.3 & stine & oneal ( 1998 ) + ggd 14 & @xmath510@xmath27 & 6 & 0.2 & 0.2 & gmez et al . ( 2002 ) + ngc 1333 & @xmath5120 & 44 & 0.3 & 0.7 & rodrguez et al . ( 1999 ) + ngc 2024 & @xmath55@xmath010@xmath27 & 25 & 3.0 & 0.5 & rodrguez et al . ( 2003 ) + w40 & @xmath55@xmath010@xmath27 & 14 & 2.8 & 0.2 & rodrguez & reipurth ( 2005 ) + quintuplet & @xmath5 10@xmath45 & 10 & 218 & 2.0 & this paper + arches & @xmath510@xmath45 & 10 & 122 & 0.7 & this paper + ngc3603 & @xmath55@xmath0 10@xmath8 & 2 & 147 & 0.7 & moffat et al . ( 2002 ) + ngc3603 & @xmath5 5@xmath0 10@xmath8 & 4 & 687 & 5.0 & mcke et al . ( 2002 ) + 0.22 , pa=52.7 . the rms noise level in this image is 65 @xmath17jy beam@xmath3 and the contours correspond to -3 , 3 , 5 , 7 , 9 and 12 times this level . this image has not been corrected for primary beam attenuation . [ fig3 ] ] ra=0.9 was introduced to align the radio and _ hst _ near - ir images ) . the 8.5 ghz image ( top ) is an inset from figure 2 , with a resolution of [email protected] , pa=62 ; the 22.5 ghz image ( middle ) is the `` low '' resolution image and has a resolution of [email protected] , pa=6.8 ; the 43.4 ghz ( bottom ) image has a resolution of [email protected] , pa=3.3 . contours in each case correspond to -3 , 3 , 5 , 7 , 9 , and 12 times the rms levels of 55 @xmath17jy beam@xmath3 ( 8.5 ghz ) , 75 @xmath17jy beam@xmath3 ( 22.5 ghz ) and 100 @xmath17jy beam@xmath3 ( 43.4 ghz ) . none of these images have been corrected for primary beam attenuation .

we study distributed learning with the least squares regularization scheme in a reproducing kernel hilbert space ( rkhs ) . by a divide - and - conquer approach , the algorithm partitions a data set into disjoint data subsets , applies the least squares regularization scheme to each data subset to produce an output function , and then takes an average of the individual output functions as a final global estimator or predictor . we show with error bounds in expectation in both the-metric and rkhs - metric that the global output function of this distributed learning is a good approximation to the algorithm processing the whole data in one single machine . our error bounds are sharp and stated in a general setting without any eigenfunction assumption . the analysis is achieved by a novel second order decomposition of operator differences in our integral operator approach . even for the classical least squares regularization scheme in the rkhs associated with a general kernel , we give the best learning rate in the literature . distributed learning , divide - and - conquer , error analysis , integral operator , second order decomposition .

1608.03339

in the era of big data , the rapid expansion of computing capacities in automatic data generation and acquisition brings data of unprecedented size and complexity , and raises a series of scientific challenges such as storage bottleneck and algorithmic scalability @xcite . to overcome the difficulty , some approaches for generating scalable approximate algorithms have been introduced in the literature such as low - rank approximations of kernel matrices for kernel principal component analysis @xcite , incomplete cholesky decomposition @xcite , early - stopping of iterative optimization algorithms for gradient descent methods @xcite , and greedy - type algorithms . another method proposed recently is distributed learning based on a divide - and - conquer approach and a particular learning algorithm implemented in individual machines @xcite . this method produces distributed learning algorithms consisting of three steps : partitioning the data into disjoint subsets , applying a particular learning algorithm implemented in an individual machine to each data subset to produce an individual output ( function ) , and synthesizing a global output by utilizing some average of the individual outputs . this method can successfully reduce the time and memory costs , and its learning performance has been observed in many practical applications to be as good as that of a big machine which could process the whole data . theoretical attempts have been recently made in @xcite to derive learning rates for distributed learning with least squares regularization under certain assumptions . this paper aims at error analysis of the distributed learning with regularized least squares and its approximation to the algorithm processing the whole data in one single machine . recall @xcite that in a reproduced kernel hilbert space ( rkhs ) @xmath1 induced by a mercer kernel @xmath2 on an input metric space @xmath3 , with a sample @xmath4 where @xmath5 is the output space , the least squares regularization scheme can be stated as @xmath6 here @xmath7 is a regularization parameter and @xmath8 is the cardinality of @xmath9 . this learning algorithm is also called kernel ridge regression in statistics and has been well studied in learning theory . see e.g. @xcite . the regularization scheme ( [ all estimate ] ) can be explicitly solved by using a standard matrix inversion technique , which requires costs of @xmath10 in time and @xmath11 in memory . however , this matrix inversion technique may not conquer challenges on storages or computations arising from big data . the distributed learning algorithm studied in this paper starts with partitioning the data set @xmath9 into @xmath12 disjoint subsets @xmath13 . then it assigns each data subset @xmath14 to one machine or processor to produce a local estimator @xmath15 by the least squares regularization scheme ( [ all estimate ] ) . finally , these local estimators are communicated to a central processor , and a global estimator @xmath16 is synthesized by taking a weighted average @xmath17 of the local estimators @xmath18 . this algorithm has been studied with a matrix analysis approach in @xcite where some error analysis has been conducted under some eigenfunction assumptions for the integral operator associated with the kernel , presenting error bounds in expectation . in this paper we shall use a novel integral operator approach to prove that @xmath16 is a good approximation of @xmath19 . we present a representation of the difference @xmath20 in terms of empirical integral operators , and analyze the error @xmath20 in expectation without any eigenfunction assumptions . as a by - product , we present the best learning rates for the least squares regularization scheme ( [ all estimate ] ) in a general setting , which surprisingly has not been done for a general kernel in the literature ( see detailed comparisons below ) . our analysis is carried out in the standard least squares regression framework with a borel probability measure @xmath21 on @xmath22 . the sample @xmath9 is independently drawn according to @xmath21 . a primary assumption in our integral operator approach is the compactness of the input metric space @xmath3 . under this assumption , the mercer kernel @xmath23 defines an integral operator @xmath24 on @xmath25 as @xmath26 where @xmath27 is the function @xmath28 in @xmath25 and @xmath29 is the marginal distribution of @xmath21 on @xmath3 . our error bounds in expectation for the distributed learning algorithm ( [ distrilearn ] ) require the uniform boundedness condition for the output @xmath30 , that is , for some constant @xmath31 , there holds @xmath32 almost surely . our bounds are stated in terms of the approximation error @xmath33 where @xmath34 is the data - free limit of ( [ all estimate ] ) defined by @xmath35 @xmath36 denotes the norm of @xmath37 , the hilbert space of square integrable functions with respect to @xmath29 , and @xmath38 is the regression function ( conditional mean ) of @xmath21 defined by @xmath39 with @xmath40 being the conditional distribution of @xmath21 induced at @xmath41 . since @xmath2 is continuous , symmetric and positive semidefinite , @xmath24 is a compact positive operator of trace class and @xmath42 is invertible . define a quantity measuring the complexity of @xmath25 with respect to @xmath29 , the _ effective dimension _ , to be the trace of the operator @xmath43 as @xmath44 in section [ proofs ] we shall prove the following first main result of this paper concerning error bounds in expectation of @xmath20 in @xmath25 and in @xmath37 . denote @xmath45 . [ mainexpected ] assume @xmath32 almost surely . if @xmath46 for @xmath47 , then we have @xmath48 and @xmath49 where @xmath50 is a constant depending only on @xmath51 . for minimax rates of sobolev spaces with @xmath52 , the parameters satisfy @xmath53 . a more general case with an arbitrary @xmath12 is covered as follows . [ mainequalsize ] assume @xmath32 almost surely . if @xmath46 for @xmath47 , and @xmath54 satisfies @xmath55 for some constant @xmath56 , then we have @xmath57 and @xmath58 where @xmath59 is a constant depending only on @xmath51 , @xmath60 , and the largest eigenvalue of @xmath24 . in the special case that @xmath61 , the approximation error can be bounded as @xmath62 . a more general condition can be imposed for the regression function as @xmath63 where the integral operator @xmath64 is regarded as a compact positive operator on @xmath65 and its @xmath66th power is well defined for any @xmath67 . the condition ( [ approxr ] ) means @xmath38 lies in the range of @xmath68 , and the special case @xmath61 corresponds to the choice @xmath69 . we shall see in the proof given in section [ proofsconf ] that ( [ approxr ] ) with @xmath70 implies @xmath71 . the effective dimension of a sobolev space on a closed domain of a euclidean space with the uniform distribution @xmath29 satisfies @xmath72 for some @xmath73 . these facts together with corollary [ mainequalsize ] yield the following nice convergence rates for the distributed learning algorithm . [ specialequalsize ] assume ( [ approxr ] ) for some @xmath74 , @xmath32 almost surely and @xmath72 for some @xmath75 . if @xmath46 for @xmath47 with @xmath76 and @xmath77 , then we have @xmath78 and @xmath79 in particular , when @xmath61 and @xmath80 , the choice @xmath81 for the parameter yields @xmath82 and @xmath83 the second main result of this paper is a sharp error bound for the least squares regularization scheme ( [ all estimate ] ) . we can even relax the uniform boundedness to a moment condition that for some constant @xmath84 , @xmath85 where @xmath86 is the conditional variance defined by @xmath87 . the following learning rates for the least squares regularization scheme ( [ all estimate ] ) will be proved in section [ proofsconf ] . the existence of @xmath88 is ensured by @xmath89<\infty$ ] . [ mainconfid ] assume @xmath89<\infty$ ] and ( [ uniformbound ] ) for some @xmath90 . then we have @xmath91 \leq \left(1 + 58\kappa^4 + 58\kappa^2\right ) ( 1 + \kappa ) \left(1 + \frac{1}{(n\lambda)^2 } + \frac{{\mathcal n}(\lambda)}{n\lambda}\right ) \nonumber \\ & & \quad \biggl\{\biggl(1 + \frac{1}{\sqrt{n\lambda}}\biggr ) \|f_\lambda - f_\rho\|_\rho + \sqrt{\left\|\sigma^2_\rho\right\|_p } \left(\frac{{\mathcal n}(\lambda)}{n}\right)^{\frac{1}{2 } ( 1-\frac{1}{p } ) } \left(\frac{1}{n\lambda}\right)^{\frac{1}{2p}}\biggr\}. \label{confboundgen}\end{aligned}\ ] ] if the parameters satisfy @xmath53 , we have the following explicit bound . [ specialpara ] assume @xmath89<\infty$ ] and ( [ uniformbound ] ) for some @xmath90 . if @xmath54 satisfies ( [ lambdacond ] ) with @xmath52 , then we have @xmath92 = o\left(\|f_\lambda - f_\rho\|_\rho + \left(\frac{{\mathcal n}(\lambda)}{n}\right)^{\frac{1}{2 } ( 1-\frac{1}{p } ) } \left(\frac{1}{n\lambda}\right)^{\frac{1}{2p}}\right).\ ] ] in particular , if @xmath93 , that is , the conditional variances are uniformly bounded , then @xmath92 = o\left(\|f_\lambda - f_\rho\|_\rho + \sqrt{\frac{{\mathcal n}(\lambda)}{n}}\right).\ ] ] in particular , when ( [ approxr ] ) is satisfied , we have the following learning rates . [ mainminmax ] assume @xmath89<\infty$ ] , ( [ uniformbound ] ) for some @xmath90 , and ( [ approxr ] ) for some @xmath70 . if @xmath72 for some @xmath75 , then by taking @xmath94 we have @xmath92 = o\left(n^{- \frac{2r \alpha}{2 \alpha \max\{2r , 1\ } + 1 } + \frac{1}{2p } \frac{2 \alpha -1}{2 \alpha \max\{2r , 1\ } + 1}}\right).\ ] ] in particular , when @xmath93 ( the conditional variances are uniformly bounded ) , we have @xmath92 = o\left(n^{-\frac{2 r \alpha}{2 \alpha \max\{2r , 1\ } + 1}}\right).\ ] ] combing bounds for @xmath95 and @xmath96 , we can derive learning rates for the distributed learning algorithm ( [ distrilearn ] ) for regression . [ finalrate ] assume @xmath32 almost surely and ( [ approxr ] ) for some @xmath97 . if @xmath72 for some @xmath75 , @xmath46 for @xmath47 , and @xmath12 satisfies the restriction @xmath98 then by taking @xmath99 , we have @xmath100 = o\left(n^{-\frac{2 \alpha r}{4 \alpha r + 1}}\right).\ ] ] the least squares regularization scheme ( [ all estimate ] ) is a classical algorithm for regression and has been extensively investigated in statistics and learning theory . there is a vast literature on this topic . here for a general kernel beyond the sobolev kernels , we compare our results with the best learning rates in the existing literature . denote the set of positive eigenvalues of @xmath24 as @xmath101 arranged in a decreasing order , and a set of normalized ( in @xmath25 ) eigenfunctions @xmath102 of @xmath24 corresponding to the eigenvalues @xmath101 . under the assumption that the orthogonal projection @xmath103 of @xmath38 in @xmath65 onto the closure of @xmath25 satisfies ( [ approxr ] ) for some @xmath104 , and that the eigenvalues @xmath105 satisfy @xmath106 with some @xmath107 , it was proved in @xcite that @xmath108 = 0,\ ] ] where @xmath109 and @xmath110 denotes a set of probability measures @xmath21 satisfying some moment decay condition ( which is satisfied when @xmath32 ) . this learning rate is suboptimal due to the limitation taken for @xmath111 and the logarithmic factor in the case @xmath112 . in particular , to have @xmath113 with confidence @xmath114 , one needs to restrict @xmath115 to be large enough and has the constant @xmath116 depending on @xmath117 to be large enough . our learning rates in corollary [ mainminmax ] have the advantages of holding in expectation and having no logarithmic factor in the case @xmath112 . under the assumption that @xmath32 almost surely , the eigenvalues @xmath105 satisfy @xmath118 with some @xmath107 and @xmath119 , and for some constant @xmath120 , the pair @xmath121 satisfies @xmath122 for every @xmath123 , it was proved in @xcite that for some constant @xmath124 depending only on @xmath125 and @xmath126 , with confidence @xmath114 , for any @xmath127 , @xmath128 here @xmath129 is the projection onto the interval @xmath130 $ ] defined @xcite by @xmath131 and @xmath132 is the approximation error defined by @xmath133 when @xmath61 , @xmath134 and the choice @xmath135 gives a learning rate of order @xmath136 . but one needs to impose the condition ( [ interpolationc ] ) for the functions spaces @xmath65 and @xmath25 , and to take the projection onto @xmath130 $ ] . our learning rates in corollary [ mainminmax ] do not require such a condition for the function spaces , nor do we take the projection . learning rates for the least squares regularization scheme ( [ all estimate ] ) in the @xmath25-metric have been investigated in the literature @xcite . for the distributed learning algorithm ( [ distrilearn ] ) with subsets @xmath13 of equal size , under the assumption that for some constants @xmath137 and @xmath138 , the eigenfunctions @xmath102 satisfy @xmath139 \leq a^{2 k } , \qquad i=1 , 2 , \ldots,\ ] ] that @xmath61 and @xmath118 for some @xmath107 and @xmath119 , it was proved in @xcite that for @xmath140 and @xmath12 satisfies the restriction @xmath141 with a constant @xmath142 depending only on @xmath125 , there holds @xmath143 = o\left(n^{-\frac{2\alpha}{2 \alpha + 1}}\right)$ ] . this interesting result was achieved by a matrix analysis approach for which the eigenfunction assumption ( [ eigenfcnc ] ) played an essential role . this eigenfunction assumption generalizes the classical case that the eigenfunctions are uniformly bounded : @xmath144 . an example of a @xmath145 mercer kernel was presented in @xcite to show that smoothness of the mercer kernel does not guarantee the uniform boundedness of the eigenfunctions . our learning rates stated in corollary [ specialequalsize ] do not require such an eigenfunction assumption . also , our restriction ( [ mrestrict ] ) for the number @xmath12 of local processors is more general when @xmath125 is close to @xmath146 . notice that the learning rates stated in corollary [ specialequalsize ] are for the difference @xmath20 between the output function of the distributed learning algorithm ( [ distrilearn ] ) and that of the algorithm ( [ all estimate ] ) using the whole data . in the special case of @xmath112 , we can see that @xmath147 , achieved by choosing @xmath81 , is smaller as @xmath12 becomes larger . this is natural because the error @xmath148 reflects more the sample error and should become smaller when we use more local processors . on the other hand , as one expects , increasing the number @xmath12 of local processors would increase the approximation error for the regression problem , which can also be seen from the bound with @xmath81 stated in theorem [ mainconfid ] . the result in corollary [ finalrate ] with @xmath149 compensates and gives the best learning rate @xmath150 = o\left(n^{-\frac{2 r \alpha}{4 \alpha r + 1}}\right)$ ] by restricting @xmath12 as in ( [ restrictmfinal ] ) . in this paper we only consider distributed learning with the regularized least squares ( [ all estimate ] ) . it would be nice to extend our analysis to distributed learning with other learning algorithms @xcite . our error estimates are achieved by a novel second order decomposition of operator differences in our integral operator approach . we approximate the integral operator @xmath24 by the empirical integral operator @xmath151 on @xmath25 defined with the input data set @xmath152 as @xmath153 where the reproducing property @xmath154 for @xmath123 is used . since @xmath2 is a mercer kernel , @xmath155 is a finite - rank positive operator and @xmath156 is invertible . the operator difference in our study is @xmath157 with @xmath158 and @xmath159 . our second order decomposition for the difference @xmath157 is stated as follows . [ operatordifference ] let @xmath160 and @xmath161 be invertible operators on a banach space . then we have @xmath162 in particular , we have @xmath163 we can decompose the operator @xmath157 as @xmath164 this is the first order decomposition . write the last term @xmath165 as @xmath166 and apply another first order decomposition similar to ( [ operatordiffii ] ) as @xmath167 it follows from ( [ operatordiffii ] ) that @xmath168 then the desired identity ( [ operatordiffform ] ) is verified . the lemma is proved . to estimate norms of various operators involving the approximation of @xmath24 by @xmath151 , we need the following probability inequality for vector - valued random variables in @xcite . [ pinelislemma ] for a random variable @xmath169 on @xmath170 with values in a hilbert space @xmath171 satisfying @xmath172 almost surely , and a random sample @xmath173 independent drawn according to @xmath21 , there holds with confidence @xmath174 , @xmath175\biggr\| \leq { 2 \widetilde{m } \log \bigl(2/\widetilde{\delta}\bigr ) \over s } + \sqrt{{2 e ( \|\xi\|^2 ) \log \bigl(2/\widetilde{\delta}\bigr ) \over s}}.\ ] ] note that @xmath176 and the @xmath66th power of the compact positive operator @xmath42 or @xmath156 is well defined for any @xmath177 . the following lemma provides estimates for the operator @xmath178 in the second order decomposition ( [ secondorder ] ) . as in @xcite , we use effective dimensions defined by ( [ effectdim ] ) to estimate operator norms . [ keynormsoperator ] let @xmath9 be a sample drawn independently according to @xmath21 . then the following estimates for the operator norm @xmath179 hold . \(a ) @xmath180 \leq \frac{\kappa^2 { \mathcal n}(\lambda)}{|d|}.$ ] \(b ) for any @xmath181 , with confidence at least @xmath182 , there holds @xmath183 where we denote the constant @xmath184 \(c ) for any @xmath185 , there holds @xmath186\right\}^{\frac{1}{d } } \leq ( 2 d \gamma ( d))^{\frac{1}{d } } { \mathcal b}_{|d| , \lambda},\ ] ] where @xmath187 is the gamma function defined for @xmath188 by @xmath189 . we apply lemma ( [ pinelislemma ] ) to the random variable @xmath190 defined by @xmath191 it takes values in @xmath192 , the hilbert space of hilbert - schmidt operators on @xmath25 , with inner product @xmath193 here tr denotes the trace of a ( trace - class ) linear operator . the norm is given by @xmath194 where @xmath195 is an orthonormal basis of @xmath25 . the space @xmath196 is a subspace of the space of bounded linear operators on @xmath197 , denoted as @xmath198 , with the norm relations @xmath199 now we use effective dimensions to estimate norms involving @xmath190 . the random variable @xmath190 defined by ( [ eta1 ] ) has mean @xmath200 and sample mean @xmath201 . recall the set of normalized ( in @xmath25 ) eigenfunctions @xmath102 of @xmath24 . it is an orthonormal basis of @xmath25 . if we regard @xmath24 as an operator on @xmath65 , the normalized eigenfunctions in @xmath65 are @xmath202 and they form an orthonormal basis of the orthogonal complement of the eigenspace associated with eigenvalue @xmath203 . by the mercer theorem , we have the following uniform convergent mercer expansion @xmath204 take the orthonormal basis @xmath102 of @xmath25 . by the definition of the hs norm , we have @xmath205 for a fixed @xmath206 , @xmath207 and @xmath208 can be expended by the orthonormal basis @xmath209 as @xmath210 hence @xmath211 combining this with ( [ mercer ] ) , we see that @xmath212 and @xmath213 \leq \kappa^2 e\left[\sum_\ell \frac{\left(\varphi_\ell ( x)\right)^2}{\lambda_\ell + \lambda}\right ] = \kappa^2 \sum_\ell \frac{\int_{\mathcal x } \left(\varphi_\ell(x)\right)^2 d \rho_x}{\lambda_\ell + \lambda}.\end{aligned}\ ] ] but @xmath214 so we have @xmath215 \leq \kappa^2 \sum_\ell \frac{\lambda_\ell}{\lambda_\ell + \lambda } = \kappa^2 \hbox{tr}\left(\left(l_k+\lambda i\right)^{-1 } l_k\right ) = \kappa^2 { \mathcal n}(\lambda)\ ] ] and @xmath216\right\|^2_{hs } = e\left[\left\|\left(l_{k } + \lambda i\right)^{-1/2 } \left\{l_k- l_{k , d(x)}\right\}\right\|_{hs}^2\right ] \leq \frac{\kappa^2 { \mathcal n}(\lambda)}{|d|}.\ ] ] then our desired inequality in part ( a ) follows from the first inequality of ( [ normrelation ] ) . from ( [ expandkx ] ) and ( [ eta1bound ] ) , we find a bound for @xmath190 as @xmath217 applying lemma [ pinelislemma ] to the random variable @xmath190 with @xmath218 , we know by ( [ normrelation ] ) that with confidence at least @xmath182 , @xmath219 - \frac{1}{|d| } \sum_{x\in d(x ) } \eta_1 ( x)\right\| & \leq & \left\|e[\eta_1 ] - \frac{1}{|d| } \sum_{x\in d(x ) } \eta_1 ( x)\right\|_{hs } \\ & \leq & { 2 \kappa^2 \log \bigl(2/\delta\bigr ) \over |d|\sqrt{\lambda } } + \sqrt{\frac{2 \kappa^2 { \mathcal n}(\lambda)\log \bigl(2/\delta\bigr)}{|d|}}.\end{aligned}\ ] ] writing the above bound by taking a factor @xmath220 , we get the desired bound ( [ normprepareeta2 ] ) . recall @xmath221 defined by ( [ bdef ] ) . apply the formula @xmath222 = \int_0^\infty \hbox{prob}\left[\xi > t\right ] d t\ ] ] for nonnegative random variables to @xmath223 and use the bound @xmath224 = \hbox{prob}\left[\xi^{\frac{1}{d } } > t^{\frac{1}{d}}\right ] \leq 2 \exp\left\{-\frac{t^{\frac{1}{d}}}{{\mathcal b}_{|d| , \lambda}}\right\}\ ] ] derived from ( [ normprepareeta2 ] ) . we find @xmath225 \leq \int_0^\infty 2 \exp\left\{-\frac{t^{\frac{1}{d}}}{{\mathcal b}_{|d| , \lambda}}\right\ } d t\ ] ] which equals @xmath226 . then the desired bound in part ( c ) follows from @xmath227 and the lemma is proved . to analyze the error @xmath20 , we need the following representation in terms of the difference of inverse operators denoted by @xmath228 and @xmath229 for the data subset @xmath14 . the empirical integral operator @xmath155 is defined with @xmath9 replaced by the data subset @xmath14 . define two random variables @xmath230 and @xmath231 with values in the hilbert space @xmath25 by @xmath232 [ difference ] assume @xmath89 < \infty$ ] . for @xmath7 , we have @xmath233 \delta_j \nonumber \\ & = & \sum_{j=1}^m \frac{|d_j|}{|d| } \left[q_{d_j(x)}\right ] \delta'_j + \sum_{j=1}^m \frac{|d_j|}{|d| } \left[q_{d_j(x)}\right ] \delta''_j - \left[q_{d(x)}\right ] \delta_d , \label{expressdifflk}\end{aligned}\ ] ] where @xmath234 , \quad \delta_d : = \frac{1}{|d| } \sum_{z \in d } \xi_\lambda ( z ) - e[\xi_\lambda],\ ] ] and @xmath235.\ ] ] a well known formula ( see e.g. @xcite ) asserts that @xmath236 so we know that @xmath237 also , with the whole data @xmath9 , we have @xmath238 but @xmath239 = \sum_{j=1}^m \frac{|d_j|}{|d| } \left\{\frac{1}{|d_j| } \sum_{z \in d_j } \xi_\lambda ( z ) - e[\xi_\lambda]\right\ } = \sum_{j=1}^m \frac{|d_j|}{|d| } \delta_j.\ ] ] hence @xmath240 then the first desired expression for @xmath20 follows . by adding and subtracting the operator @xmath241 , writing @xmath242 , and noting @xmath243=0 $ ] , we know that the first expression implies ( [ expressdifflk ] ) . this proves lemma [ difference ] . to apply ( [ expressdifflk ] ) for our error analysis , we also need to bound norms involving @xmath244 and @xmath245 . we are able to give the following estimates even after multiplying with @xmath246 taken from the operator @xmath247 or @xmath248 . [ keynorms ] let @xmath9 be a sample drawn independently according to @xmath21 and @xmath249 be a measurable bounded function on @xmath250 and @xmath251 be the random variable with values on @xmath25 given by @xmath252 for @xmath253 . then the following statements hold . \(a ) @xmath254 = { \mathcal n}(\lambda).$ ] \(b ) for any @xmath181 , with confidence at least @xmath182 , there holds @xmath255\right)\right\|_k \leq \frac{2 \|g\|_\infty \log \bigl(2/\delta\bigr ) } { \sqrt{|d| } } \left\{\frac{\kappa } { \sqrt{|d|\lambda } } + \sqrt{{\mathcal n}(\lambda)}\right\}.\end{aligned}\ ] ] consider the random variable @xmath256 defined by @xmath257 it takes values in @xmath25 . by ( [ expandkx ] ) , it satisfies @xmath258 so @xmath259 = e\left[\sum_{\ell } \frac{(\varphi_\ell ( x))^2}{\lambda_\ell + \lambda}\right ] = { \mathcal n}(\lambda).\ ] ] this is the statement of part ( a ) . for part ( b ) , we consider another random variable @xmath260 defined by @xmath261 it takes values in @xmath25 and satisfies @xmath262 so @xmath263 and @xmath264 \leq \|g\|_\infty^2 e\left[\sum_\ell \frac{\left(\varphi_\ell ( x)\right)^2}{\lambda_\ell + \lambda}\right ] = \|g\|_\infty^2 { \mathcal n}(\lambda).\ ] ] applying lemma [ pinelislemma ] proves the statement in part ( b ) . to illustrate how to apply the second order decomposition ( [ secondorder ] ) for operator differences in our integral operator approach , we prove in this section our main result on error bounds for the least squares regularization scheme ( [ all estimate ] ) . [ mainpropls ] assume @xmath89 < \infty$ ] and ( [ uniformbound ] ) for some @xmath90 . then we have @xmath265 & \leq & \left(1 + 58\kappa^4 + 58\kappa^2\right ) \left(1 + \frac{1}{(|d|\lambda)^2 } + \frac{{\mathcal n}(\lambda)}{|d|\lambda}\right ) \\ & & \left\{\kappa^{\frac{1}{p}}\sqrt{\left\|\sigma^2_\rho\right\|_p } \left(\frac{{\mathcal n}(\lambda)}{|d|}\right)^{\frac{1}{2 } ( 1-\frac{1}{p } ) } \left(\frac{1}{|d|\lambda}\right)^{\frac{1}{2p } } + \kappa \frac{\|f_\lambda - f_\rho\|_\rho}{\sqrt{|d|\lambda}}\right\}.\end{aligned}\ ] ] we recall the expression ( [ fdlambda ] ) for @xmath266 and the notation @xmath267 defined by ( [ qdnotation ] ) for the operator difference @xmath268 . then we see @xmath269 \delta_d + \left(l_{k}+\lambda i\right)^{-1 } \delta_d.\ ] ] to estimate the @xmath65 norm , we use the identity @xmath270 and get @xmath271 \delta_d\right\|_k + \left\|l_k^{1/2}\left(l_{k}+\lambda i\right)^{-1 } \delta_d\right\|_k.\ ] ] we apply the second order decomposition ( [ secondorder ] ) , use the bounds @xmath272 , @xmath273 , @xmath274 , and know that @xmath275 \delta_d\right\|_k \leq \left\|\left(l_{k } + \lambda i\right)^{-1/2 } \left\{l_k- l_{k , d(x)}\right\ } \left(l_{k } + \lambda i\right)^{-1 } \delta_d\right\|_k + \\ & & \left\|\left(l_{k } + \lambda i\right)^{-1/2 } \left\{l_k- l_{k , d(x)}\right\ } \left(l_{k , d(x)}+\lambda i\right)^{-1 } \left\{l_k- l_{k , d(x)}\right\ } \left(l_{k } + \lambda i\right)^{-1 } \delta_d\right\|_k \\ & & \leq \left\|\left(l_{k } + \lambda i\right)^{-1/2 } \left\{l_k- l_{k , d(x)}\right\}\right\| \frac{1}{\sqrt{\lambda } } \left\|\left(l_{k } + \lambda i\right)^{-1/2 } \delta_d\right\|_k + \\ & & \left\|\left(l_{k } + \lambda i\right)^{-1/2 } \left\{l_k- l_{k , d(x)}\right\}\right\| \frac{1}{\lambda } \left\|\left\{l_k- l_{k , d(x)}\right\}\left(l_{k } + \lambda i\right)^{-1/2}\right\| \biggl\|\left(l_{k } + \lambda i\right)^{-1/2 } \delta_d\biggr\|_k.\end{aligned}\ ] ] for convenience , we introduce the notation @xmath276 then the above bound can be restated as @xmath277 \delta_d\right\|_k \leq \left(\frac{\xi_d}{\sqrt{\lambda } } + \frac{\xi_d^2}{\lambda}\right)\left\|\left(l_{k } + \lambda i\right)^{-1/2 } \delta_d\right\|_k.\ ] ] hence @xmath278 and by the schwarz inequality we have @xmath279 \leq \left\{e\left[\left(1 + \frac{\xi_d}{\sqrt{\lambda } } + \frac{\xi_d^2}{\lambda}\right)^2\right]\right\}^{1/2 } \left\{e\left[\left\|\left(l_{k } + \lambda i\right)^{-1/2 } \delta_d\right\|_k^2\right]\right\}^{1/2}.\ ] ] to deal with the expected value in ( [ firstdecom ] ) , as in lemma [ difference ] , we separate @xmath245 as @xmath280 where @xmath281.\ ] ] then @xmath282\right\}^{1/2 } \nonumber \\ & & \leq \left\{e\left[\left\|\left(l_{k } + \lambda i\right)^{-1/2 } \delta'_d\right\|_k^2\right]\right\}^{1/2 } + \left\{e\left[\left\|\left(l_{k } + \lambda i\right)^{-1/2 } \delta''_d\right\|_k^2\right]\right\}^{1/2}. \label{separate}\end{aligned}\ ] ] observe that @xmath283 each term in this expression is unbiased because @xmath284 . this unbiasedness and the independence tell us that @xmath285\right\}^{1/2 } & = & \left\{\frac{1}{|d| } e\left[\left\|(y - f_\rho ( x))\left[\left(l_{k } + \lambda i\right)^{-1/2}\right ] ( k_x)\right\|_k^2\right]\right\}^{1/2 } \nonumber \\ & = & \left\{\frac{1}{|d| } e\left[\sigma^2_\rho ( x ) \left\|\left[\left(l_{k } + \lambda i\right)^{-1/2}\right ] ( k_x)\right\|_k^2\right]\right\}^{1/2}. \label{deltad'}\end{aligned}\ ] ] if @xmath286 , then @xmath287 and by lemma [ keynorms ] we have @xmath288\right\}^{1/2 } \leq \sqrt{\left\|\sigma^2_\rho\right\|_\infty } \sqrt{{\mathcal n}(\lambda)/|d|}.\ ] ] if @xmath289 with @xmath290 , we take @xmath291 ( @xmath292 for @xmath293 ) satisfying @xmath294 and apply the hlder inequality @xmath295 \leq \left(e[|\xi|^p]\right)^{1/p } \left(e[|\eta|^{q}]\right)^{1/q}$ ] to @xmath296 , @xmath297 to find @xmath298 ( k_x)\right\|_k^2\right ] \leq \left\|\sigma^2_\rho\right\|_{p } \left\{e\left[\left\|\left[\left(l_{k } + \lambda i\right)^{-1/2}\right ] ( k_x)\right\|_k^{2q}\right]\right\}^{1/q}.\ ] ] but @xmath299 ( k_x)\right\|_k^{2q-2 } \leq \left(\kappa/\sqrt{\lambda}\right)^{2 q -2}\ ] ] and @xmath300= { \mathcal n}(\lambda)$ ] by lemma [ keynorms ] . so we have @xmath285\right\}^{1/2 } & \leq & \left\{\frac{1}{|d| } \left\|\sigma^2_\rho\right\|_{p } \left\{\frac{\kappa^{2q -2}}{\lambda^{q-1 } } { \mathcal n}(\lambda)\right\}^{1/q}\right\}^{1/2 } \\ & = & \sqrt{\left\|\sigma^2_\rho\right\|_p } \kappa^{\frac{1}{p } } \left(\frac{{\mathcal n}(\lambda)}{|d|}\right)^{\frac{1}{2 } ( 1-\frac{1}{p } ) } \left(\frac{1}{|d|\lambda}\right)^{\frac{1}{2p}}.\end{aligned}\ ] ] combining the above two cases , we know that for either @xmath93 or @xmath301 , @xmath288\right\}^{1/2 } \leq \sqrt{\left\|\sigma^2_\rho\right\|_p } \kappa^{\frac{1}{p } } \left(\frac{{\mathcal n}(\lambda)}{|d|}\right)^{\frac{1}{2 } ( 1-\frac{1}{p } ) } \left(\frac{1}{|d|\lambda}\right)^{\frac{1}{2p}}.\ ] ] the second term of ( [ separate ] ) can be bounded easily as @xmath302\right\}^{1/2 } & \leq & \frac{1}{\sqrt{|d| } } \left\{e\left[\left(f_\rho ( x ) - f_\lambda ( x)\right)^2 \left\|\left(l_{k } + \lambda i\right)^{-1/2 } ( k_x ) \right\|_k^2\right]\right\}^{1/2 } \\ & \leq & \frac{1}{\sqrt{|d| } } \left\{e\left[\left(f_\rho ( x ) - f_\lambda ( x)\right)^2 \frac{\kappa^2}{\lambda}\right]\right\}^{1/2 } = \frac{\kappa \|f_\rho - f_\lambda\|_\rho}{\sqrt{|d|\lambda}}.\end{aligned}\ ] ] putting the above estimates for the two terms of ( [ separate ] ) into ( [ firstdecom ] ) and applying lemma [ keynormsoperator ] to get @xmath303\right\}^{1/2 } & \leq & 1 + \left\{e\left[\frac{\xi_d^2}{\lambda}\right]\right\}^{1/2 } + \left\{e\left[\frac{\xi_d^4}{\lambda^2}\right]\right\}^{1/2 } \\ & \leq & 1 + \left\{\frac{\kappa^2 { \mathcal n}(\lambda)}{|d|\lambda}\right\}^{1/2 } + \left\{\frac{48 { \mathcal b}_{|d| , \lambda}^4}{\lambda^2}\right\}^{1/2 } \\ & \leq & 1 + \frac{58\kappa^4}{(|d|\lambda)^2 } + \frac{58\kappa^2 { \mathcal n}(\lambda)}{|d|\lambda},\end{aligned}\ ] ] we know that @xmath304 $ ] is bounded by @xmath305 then our desired error bound follows . the proof of the proposition is complete . * proof of theorem [ mainconfid ] * combining proposition [ mainpropls ] with the triangle inequality @xmath306 , we know that @xmath307 & \leq & \|f_\lambda - f_\rho\|_\rho + \left(1 + 58\kappa^4 + 58\kappa^2\right ) \left(1 + \frac{1}{(|d|\lambda)^2 } + \frac{{\mathcal n}(\lambda)}{|d|\lambda}\right ) \\ & & \left\{\kappa^{\frac{1}{p}}\sqrt{\left\|\sigma^2_\rho\right\|_p } \left(\frac{{\mathcal n}(\lambda)}{|d|}\right)^{\frac{1}{2 } ( 1-\frac{1}{p } ) } \left(\frac{1}{|d|\lambda}\right)^{\frac{1}{2p } } + \frac{\kappa}{\sqrt{|d|\lambda } } \|f_\lambda - f_\rho\|_\rho\right\}.\end{aligned}\ ] ] then the desired error bound holds true , and the proof of theorem [ mainconfid ] is complete . * proof of corollary [ specialpara ] * by the definition of effective dimension , @xmath308 combining this with the restriction ( [ lambdacond ] ) with @xmath52 , we find @xmath309 and @xmath310 . putting these and the restriction ( [ lambdacond ] ) with @xmath52 into the error bound ( [ confboundgen ] ) , we know that @xmath91 \leq \left(1 + 58\kappa^4 + 58\kappa^2\right ) ( 1 + \kappa ) \left(1 + \frac{(\lambda_1 + c_0)^2c_0 ^ 2}{\lambda_1 ^ 2 } + c_0\right ) \nonumber \\ & & \biggl\{\biggl(1 + \sqrt{(\lambda_1 + c_0 ) c_0/\lambda_1}\biggr ) \|f_\lambda - f_\rho\|_\rho + \sqrt{\left\|\sigma^2_\rho\right\|_p } \left(\frac{{\mathcal n}(\lambda)}{n}\right)^{\frac{1}{2 } ( 1-\frac{1}{p } ) } \left(\frac{1}{n\lambda}\right)^{\frac{1}{2p}}\biggr\}.\end{aligned}\ ] ] then the desired bound follows . the proof of corollary [ specialpara ] is complete . * proof of corollary [ mainminmax ] * the representation @xmath311 for @xmath88 , with @xmath312 , tells us the well known fact that the condition ( [ approxr ] ) with @xmath70 yields a bound for @xmath313 as @xmath314 if @xmath315 for some constant @xmath316 , then the choice @xmath94 yields @xmath317 so ( [ lambdacond ] ) with @xmath52 is satisfied . with this choice we also have @xmath318 putting these estimates into corollary [ specialpara ] , we know that @xmath307 & = & o\left(n^{-\frac{2\alpha r}{2 \alpha \max\{2r , 1\}+1 } } + n^{- \frac{\alpha \max\{2r , 1\}}{2 \alpha \max\{2r , 1\ } + 1 } + \frac{1}{2p } \frac{2 \alpha -1}{2 \alpha \max\{2r , 1\ } + 1}}\right ) \\ & = & o\left(n^{- \frac{\alpha \min\left\{2 r , \max\{2r , 1\}\right\}}{2 \alpha \max\{2r , 1\ } + 1 } + \frac{1}{2p } \frac{2 \alpha -1}{2 \alpha \max\{2r , 1\ } + 1}}\right).\end{aligned}\ ] ] but we find @xmath319 by discussing the two different cases @xmath320 and @xmath321 . then our conclusion follows immediately . the proof of corollary [ mainminmax ] is complete . in this section , we prove our first main result on the error @xmath20 in the @xmath25 metric and @xmath322 metric . the following result is more general , allowing different sizes for data subsets @xmath323 . [ mainexpectedgeneral ] assume that for some constant @xmath324 , @xmath325 almost surely . then we have @xmath326 \leq c'_\kappa \left(\frac{1}{(n \lambda)^2 } + \frac{{\mathcal n}(\lambda)}{n\lambda}\right ) \biggl\{\frac{\|f_\lambda - f_\rho\|_\rho}{\sqrt{n\lambda } } \sum_{j=1}^m \left(\frac{|d|}{|d_j|}\right)^{\frac{3}{2 } } + m \sqrt{\lambda } \sqrt{\frac{{\mathcal n}(\lambda)}{n \lambda}}\biggr\ } \\ & & + c'_\kappa m \sqrt{\lambda } \left\{\sum_{j=1}^m \left(\frac{|d_j|}{|d|}\right)^2 \left(\frac{1}{|d_j|^2 \lambda^2 } + \frac{{\mathcal n}(\lambda)}{|d_j|\lambda}\right)^2 \left\{1 + \left(\frac{1}{|d_j|^2 \lambda^2 } + \frac{{\mathcal n}(\lambda)}{|d_j|\lambda}\right)^2 \right\}\right\}^{1/2}\end{aligned}\ ] ] and @xmath327 \leq c'_\kappa \left(\frac{1}{(n \lambda)^2 } + \frac{{\mathcal n}(\lambda)}{n\lambda}\right ) \biggl\{\frac{\|f_\lambda - f_\rho\|_\rho}{\sqrt{n}\lambda } \sum_{j=1}^m \left(\frac{|d|}{|d_j|}\right)^{\frac{3}{2 } } + m \sqrt{\frac{{\mathcal n}(\lambda)}{n \lambda}}\biggr\ } \\ & & + c'_\kappa m \left\{\sum_{j=1}^m \left(\frac{|d_j|}{|d|}\right)^2 \left(\frac{1}{|d_j|^2 \lambda^2 } + \frac{{\mathcal n}(\lambda)}{|d_j|\lambda}\right)^2 \left\{1 + \left(\frac{1}{|d_j|^2 \lambda^2 } + \frac{{\mathcal n}(\lambda)}{|d_j|\lambda}\right)^2 \right\}\right\}^{1/2},\end{aligned}\ ] ] where @xmath328 is a constant depending only on @xmath51 . recall ( [ expressdifflk ] ) in lemma [ difference ] . it enables us to express @xmath329 where the terms @xmath330 are given by @xmath331 \delta'_j , \ j_2 = \sum_{j=1}^m \frac{|d_j|}{|d| } \left[l_k^{1/2}q_{d_j(x)}\right ] \delta''_j , \ j_3 = -\left[l_k^{1/2}q_{d(x)}\right ] \delta_d.\ ] ] these three terms will be dealt with separately in the following . for the first term @xmath332 of ( [ fdrhodecom ] ) , each summand with @xmath333 can be expressed as @xmath334 ( k_x)$ ] , and is unbiased because @xmath284 . the unbiasedness and the independence tell us that @xmath335 \leq \left\{e\left[\left\|j_1\right\|_k^2\right]\right\}^{1/2 } \leq \left\{\sum_{j=1}^m \left(\frac{|d_j|}{|d|}\right)^2 e\left[\left\|\left[l_k^{1/2}q_{d_j(x)}\right ] \delta'_j\right\|_k^2\right]\right\}^{1/2}.\end{aligned}\ ] ] let @xmath333 . the relation ( [ fdlambdafirst ] ) derived from the second order decomposition ( [ secondorder ] ) in the proof of proposition [ mainpropls ] yields @xmath336 \delta'_j\right\|_k^2 \leq \left(\frac{\xi_{d_j}}{\sqrt{\lambda } } + \frac{\xi_{d_j}^2}{\lambda}\right)^2 \left\|\left(l_{k } + \lambda i\right)^{-1/2 } \delta'_j\right\|_k^2.\ ] ] now we apply the formula ( [ expectedform ] ) to estimate the expected value of ( [ primeone ] ) . by part ( b ) of lemma [ keynormsoperator ] , for @xmath181 , there exists a subset @xmath337 of @xmath338 of measure at least @xmath182 such that @xmath339 applying part ( b ) of lemma [ keynorms ] to @xmath340 with @xmath341 and the data subset @xmath14 , we know that there exists another subset @xmath342 of @xmath338 of measure at least @xmath182 such that @xmath343 combining this with ( [ xidjest ] ) and ( [ primeone ] ) , we know that for @xmath344 , @xmath345 \delta'_j\right\|_k^2 \leq \left(\frac{{\mathcal b}_{|d_j| , \lambda}^2}{\lambda } + \frac{{\mathcal b}_{|d_j| , \lambda}^4}{\lambda^2}\right ) \left(\frac{m}{\kappa}\right)^2 { \mathcal b}_{|d_j| , \lambda}^2 \left(2 \log \bigl(2/\delta\bigr)\right)^6.\ ] ] since the measure of the set @xmath346 is at least @xmath347 , by denoting @xmath348 we see that @xmath349 \delta'_j\right\|_k^2 > { \mathcal c}_{|d_j| , \lambda } \left(\log \bigl(2/\delta\bigr)\right)^6\right ] \leq 2 \delta.\ ] ] for @xmath350 , the equation @xmath351 has the solution @xmath352 when @xmath353 , we have @xmath349 \delta'_j\right\|_k^2 > t\right ] \leq 2 \delta_t = 4 \exp\left\{-\left(t/{\mathcal c}_{|d_j| , \lambda}\right)^{1/6}\right\}.\ ] ] this inequality holds trivially when @xmath354 since the probability is at most @xmath355 . thus we can apply the formula ( [ expectedform ] ) to the nonnegative random variable @xmath356 \delta'_j\right\|_k^2 $ ] and obtain @xmath357 \delta'_j\right\|_k^2\right ] = \int_0^\infty \hbox{prob}\left[\xi > t\right ] d t \leq \int_0^\infty 4 \exp\left\{-\left(t/{\mathcal c}_{|d_j| , \lambda}\right)^{1/6}\right\ } d t\ ] ] which equals @xmath358 . therefore , @xmath335 \leq \left\{\sum_{j=1}^m \left(\frac{|d_j|}{|d|}\right)^2 24 \gamma ( 6 ) { \mathcal c}_{|d_j| , \lambda } \right\}^{1/2 } \\ & & \leq 1536\sqrt{5 } \kappa m \left\{\sum_{j=1}^m \left(\frac{|d_j|}{|d|}\right)^2 \lambda \left(\frac{\kappa^2}{|d_j|^2 \lambda^2 } + \frac{{\mathcal n}(\lambda)}{|d_j|\lambda}\right)^2 \left\{1 + 8 \kappa^2 \left(\frac{\kappa^2}{|d_j|^2 \lambda^2 } + \frac{{\mathcal n}(\lambda)}{|d_j|\lambda}\right)^2 \right\}\right\}^{1/2}.\end{aligned}\ ] ] for the second term @xmath359 of ( [ fdrhodecom ] ) , we use the second order decomposition ( [ secondorder ] ) again and obtain @xmath345 \delta''_j\right\|_k \leq \left(\frac{\xi_{d_j}}{\sqrt{\lambda } } + \frac{\xi_{d_j}^2}{\lambda}\right ) \left\|\left(l_{k } + \lambda i\right)^{-1/2 } \delta''_j\right\|_k.\ ] ] applying the schwarz inequality and lemmas [ keynormsoperator ] and [ keynorms ] , we get @xmath360 \delta''_j\right\|_k\right ] & \leq & \left\{e\left[\left(\frac{\xi_{d_j}}{\sqrt{\lambda } } + \frac{\xi_{d_j}^2}{\lambda}\right)^2\right]\right\}^{1/2 } \\ & & \frac{1}{\sqrt{|d_j| } } \left\{e\left[\left(f_\rho ( x ) - f_\lambda ( x)\right)^2 \left\|\left(l_{k } + \lambda i\right)^{-1/2 } ( k_x ) \right\|_k^2\right]\right\}^{1/2 } \\ & \leq & \left(\left\{\frac{\kappa^2 { \mathcal n}(\lambda)}{|d_j|\lambda}\right\}^{1/2 } + \left\{\frac{48 { \mathcal b}_{|d_j| , \lambda}^4}{\lambda^2}\right\}^{1/2}\right ) \frac{\kappa \|f_\rho - f_\lambda\|_\rho}{\sqrt{|d_j|\lambda } } \\ & \leq & \left(\frac{58\kappa^4}{(|d_j|\lambda)^2 } + \frac{58\kappa^2 { \mathcal n}(\lambda)}{|d_j|\lambda}\right)\frac{\kappa \|f_\rho - f_\lambda\|_\rho}{\sqrt{|d_j|\lambda}}.\end{aligned}\ ] ] it follows that @xmath361 \leq \left(\frac{58\kappa^4}{(|d| \lambda)^2 } + \frac{58\kappa^2 { \mathcal n}(\lambda)}{|d|\lambda}\right ) \sum_{j=1}^m 2\frac{|d|}{|d_j| } \frac{\kappa \|f_\rho - f_\lambda\|_\rho}{\sqrt{|d_j|\lambda}}.\ ] ] the last term @xmath362 of ( [ fdrhodecom ] ) has been handled in the proof of proposition [ mainpropls ] by ignoring the summand @xmath363 in the expression for @xmath266 , and we find from the trivial bound @xmath364 with @xmath93 that @xmath365 \leq \left(\frac{58\kappa^4}{(|d|\lambda)^2 } + \frac{58\kappa^2 { \mathcal n}(\lambda)}{|d|\lambda}\right ) \left(2 m \left(\frac{{\mathcal n}(\lambda)}{|d|}\right)^{\frac{1}{2 } } + \frac{\kappa \|f_\rho - f_\lambda\|_\rho}{\sqrt{|d|\lambda}}\right).\ ] ] combining the above estimates for the three terms of ( [ fdrhodecom ] ) , we see that the desired error bound in the @xmath65 metric holds true . the estimate in the @xmath25 metric follows from the steps in deriving the error bound in the @xmath65 metric except that in the representation ( [ fdrhodecom ] ) the operator @xmath176 in the front disappears . this change gives an additional factor @xmath366 , the bound for the operator @xmath246 , and proves the desired error bound in the @xmath25 metric . since @xmath46 for @xmath47 , the bound in theorem [ mainexpectedgeneral ] in the @xmath65 metric can be simplified as @xmath367 & \leq & c'_\kappa \left(\frac{1}{(n \lambda)^2 } + \frac{{\mathcal n}(\lambda)}{n\lambda}\right ) \biggl\{\frac{\|f_\lambda - f_\rho\|_\rho}{\sqrt{n\lambda } } m^{\frac{5}{2 } } + m \sqrt{\lambda } \sqrt{\frac{{\mathcal n}(\lambda)}{n \lambda}}\biggr\ } \\ & & + c'_\kappa m \frac{\sqrt{\lambda}}{\sqrt{m } } \left(\frac{m^2}{n^2 \lambda^2 } + \frac{m{\mathcal n}(\lambda)}{n\lambda}\right ) \left\{1 + \left(\frac{m^2}{n^2 \lambda^2 } + \frac{m{\mathcal n}(\lambda)}{n\lambda}\right)\right\ } \\ & \leq & c'_\kappa \left(\frac{m}{(n \lambda)^2 } + \frac{{\mathcal n}(\lambda)}{n\lambda}\right ) \biggl\{\frac{\|f_\lambda - f_\rho\|_\rho}{\sqrt{n\lambda } } m^{\frac{5}{2 } } + m \sqrt{\lambda } \sqrt{\frac{{\mathcal n}(\lambda)}{n \lambda } } \\ & & + m \sqrt{\lambda } \sqrt{m } \left\{1 + \left(\frac{m^2}{n^2 \lambda^2 } + \frac{m{\mathcal n}(\lambda)}{n\lambda}\right)\right\}\biggr\}.\end{aligned}\ ] ] notice that the term @xmath368 can be bounded by @xmath369 . then the desired error bound in the @xmath65 metric with @xmath370 follows . the proof for the error bound in the @xmath25 metric is similar . the proof of theorem [ mainexpected ] is complete . as in the proof of corollary [ specialpara ] , the restriction ( [ lambdacond ] ) implies @xmath309 and @xmath371 . it follows that @xmath372 putting these bounds into theorem [ mainexpected ] , we know that the expected value @xmath148 is bounded by @xmath373 and @xmath58 where @xmath374 this proves corollary [ mainequalsize ] . if @xmath315 for some constant @xmath316 , then the choice @xmath77 satisfies ( [ lambdacond ] ) . with this choice we also have @xmath375 since the condition ( [ approxr ] ) yields @xmath376 , we have by corollary [ mainequalsize ] , @xmath377 the inequality @xmath378 is equivalent to @xmath379 and it can be expressed as ( [ mrestrict ] ) . since ( [ mrestrict ] ) is valid , we have @xmath380 this proves the first desired convergence rate . the second rate follows easily . this proves corollary [ specialequalsize ] . the assumption @xmath72 tells us that for some constant @xmath316 , @xmath382 so the choice @xmath383 yields @xmath384 if @xmath12 satisfies @xmath385 then ( [ lambdacond ] ) is valid , and by corollary [ mainequalsize ] , @xmath386 thus , when @xmath12 satisfies @xmath387 we have @xmath388 and thereby @xmath100 = o\left(n^{-\frac{2 \alpha r}{4 \alpha r + 1}}\right).\ ] ] finally , we notice that ( [ restrictmfinal ] ) is equivalent to the combination of ( [ restrictmcor ] ) and ( [ restrictmiicor ] ) . so our conclusion follows . this proves corollary [ finalrate ] . i. steinwart , d. hush , and c. scovel . optimal rates for regularized least squares regression . in proceedings of the 22nd annual conference on learning theory ( s. dasgupta and a. klivans , eds . ) , pp . 79 - 93 , 2009 .

we report on the results of spectroscopic mapping observations carried out in the nuclear region of centaurus a ( ngc5128 ) over the 5.2 - 15 and 20 - 36 m spectral regions using the infrared spectrograph on the _ spitzer space telescope_. we have detected and mapped s(0 ) , s(2 ) , s(3 ) , and s(5 ) pure rotational transition lines of molecular hydrogen and emissions in the fine - structure transitions of [ siii ] , [ siii ] , [ feii ] , [ feiii ] , [ arii ] , [ siv ] , [ neii ] , [ nev ] and [ oiv ] . the 500 pc bipolar dust shell discovered by quillen et al.(2006 ) is even more clearly seen in the 11.3 m dust emission feature than previous broad band imaging . the pure rotational lines of molecular hydrogen other than the s(0 ) line are detected above the dusty disk and associated with the oval dust shell . the molecular hydrogen transitions indicate the presence of warm gas at temperatures 250720k . the column density of the warm molecular hydrogen in the shell is and similar to that estimated from the continuum dust shell surface brightness . the ratio of the dust emission features at 7.7 m and 11.3 m and the ratio of the [ neii](12.8 m ) and 11.3 m dust emission feature are lower in the 500 pc dust shell than in the star forming disk . the clearer shell morphology at 11.3 m , warm molecular hydrogen emission in the shell , and variation in line ratios in the shell compared to those in the disk , confirm spectroscopically that this shell is a separate coherent entity and is unlikely to be a chance superposition of dust filaments . the physical conditions in the shell are most similar to galactic supernova remnants where blast waves encounter molecular clouds . the lines requiring the highest level of ionization , [ nev]( m ) and [ oiv]( m ) , are detected 2025north - east and south - west of the nucleus and at position angles near the radio jet axis . fine structure line ratios and limits from this region suggest that the medium is low density and illuminated by a hard radiation field at low ionization parameter . these higher s molecular hydrogen pure rotational transitions are also particularly bright in the same region as the [ oiv ] and [ nev ] emission . this suggests that the gas associated with the dust shell has been excited near the jet axis and is part of an ionization cone . [ firstpage ]

0710.5087

the nearest of all the giant radio galaxies , centaurus a ( ngc 5128 ) provides a unique opportunity to observe the dynamics and morphology of an active galaxy in detail across the electromagnetic spectrum . for a recent review on this remarkable object see @xcite . in its central regions , ngc 5128 exhibits a well recognized and optically - dark band of absorption across its nucleus . images from the _ spitzer space telescope _ ( sst ) with the infrared array camera ( irac ) and multiple imaging photometer for spitzer ( mips ) cameras in the mid - infrared reveal a 3long parallelogram shape @xcite that has been modeled as a series of folds in a dusty warped thin disk ( e.g. , @xcite ) . the dusty disk is also the site of ongoing star formation at a rate of about @xmath5 yr@xmath6 , ( based on the infrared luminosity estimated by @xcite ) . centaurus a hosts an active nucleus ( e.g. , @xcite ) that has been recently studied using infrared spectra from the sst . @xcite . its nucleus exhibits a strong silicate absorption feature and emission from [ nev ] . here we do not study the nucleus but focus on structure exterior to it . the irac and mips images of centaurus a have revealed another surprise , a bipolar shell - like structure 500pc north and south of the nucleus @xcite . this shell , seen for the first time with spitzer imaging , could be the first extragalactic nuclear shell discovered in the infrared . in this paper we present mapping of the central 500600pc of centaurus a done with the infrared spectrograph ( irs ) on board the sst . the infrared spectral maps were obtained to test the possibility that the apparent dust shell is a separate coherent structure and not a chance superposition of dust filaments . they were also obtained to search for possible interactions between the agn and the interstellar medium either by illumination from the agn or caused by jets or outflows . based on the discussion by @xcite , we adopt a distance to cen a of 3.4 mpc . at this distance , 1 kpc corresponds to 1on the sky . all positions reported in this manuscript are given with respect to epoch 2000 . observations were obtained in spectral mapping mode in the short - low ( sl;514@xmath0 m ) resolution and long - high ( lh ; 19 - 38@xmath0 m ) resolution modules of the infrared spectrograph on board the sst . the spectral resolution of sl is @xmath7 whereas that of lh is @xmath8 . in spectral mapping mode the spitzer space craft points on a raster of discrete steps , settling at each position before the integrations begin . each map was accompanied by a 4 position offset observation taken with the same slit pattern and exposure sequence . the offset ( rogue ) positions were used for sky subtraction . the rogue sky position is at ra=13:25:15.7 ( epoch 2000 ) dec=-42:55:13 , approximately 6north of the nucleus and sufficiently far away that little emission from the galaxy was seen in the previously observed broad band irac and mips images . we mapped in the short low ( sl ) module of the infrared spectrograph using slit overlaps of [email protected] a 60s ramp . one data set covers @xmath10 map positions ( through the guaranteed time observer program ; gto ) , and the other two cover @xmath11 map positions ( through the guest observer program ; go ) . both maps were observed by the sst on mar 13 , 2007 . spectral cube assembly , including background subtraction , bad pixel removal , spatial regridding and calibration was done using the software package cube builder for irs spectra maps , ( cubism ) @xcite producing a single spectral cube . we mapped in the long high spectral resolution ( lh ) module of the infrared spectrograph using slit overlaps of [email protected] , so that central positions in the map were covered by four slit positions . we used a 60s ramp , and each observation contained 2 cycles . for central map positions , this corresponds to a 480s on source integration time . two data sets ( guest observer ; go ) were taken covering 6@xmath920 map positions . the third ( guaranteed time observer ; gto ) data set covers @xmath12 map positions . all three maps were observed by the sst on aug 1 , 2006 . data reduction of the three lh spectral data sets involved bad pixel removal , spectral extraction , spatial regridding , and application of the slit loss correction . for descriptions of the software procedures used see appendix a by @xcite . the resulting three data cubes were then resampled onto the same spatial grid and averaged into a single cube , resulting in 1440s on source exposure times for central positions in the map . we have extracted continuum images from both spectral cubes to check that they are consistent with previous broad band imaging and to bring attention to features noted in previous studies @xcite and that we will discuss below . in figure [ fig : cont25 ] we show the morphology of the continuum emission in @xmath13 m wide bands centered at 25@xmath0 m and 32@xmath0 m from the long - high spectral cube . we chose these regions so as to be free from artifacts caused by order separation and emission lines . the morphology at 25@xmath0 m is consistent with the morphology seen in the broad band mips @xmath14 m image previously presented by @xcite . the broad band of emission approximately 30wide , and stretching approximately along the east - west direction across the galaxy nucleus we refer to as ` the parallelogram ' ( see figure 3 by @xcite ) . the parallelogram shape is due to folds in a thin dusty warped disk . the warped disk is also rich in molecular gas @xcite and so is the site of most active star formation present in the galaxy . in the sl spectral cube , broad and prominent dust emission features are present . these dust emission features are generally attributed to polycyclic aromatic hydrocarbons ( pahs ) . these dominate the broad band emission in the irac broad band images previously studied @xcite except possibly at the nucleus . in figure [ fig : pah11 ] we show the morphology of the peak of the @xmath15 m dust emission feature extracted from the sl spectral cube . this map also shows the parallelogram shape or star forming disk seen in the irac band images . above and below the parallelogram we can see the shell - like bipolar structure described by @xcite that was most prominent in the irac @xmath16 m image . this shell has a major axis of @xmath17 , a minor axis of @xmath18 and position angle of approximately @xmath19 @xcite . the irac 8.0@xmath0 m band image is dominated by the pah dust emission feature at 7.7@xmath0 m and so it is not surprising that the 11.3@xmath0 m image shows similar morphology . we show an image at 11.3@xmath0 m here because it most clearly reveals the dust shell . the morphology at 7.7@xmath0 m and other pah emission features and their ratios will be discussed in more detail below . in the next section we will compare the morphology in line emission maps to that seen in continuum ( figure [ fig : cont25 ] ) and in the dust emission features . in the @xmath20 m continuum image ( figure [ fig : cont25]a ) three peaks in the emission are seen , 2 to the west of the nucleus , one to south - east . a weaker but fourth is located directly east of the nucleus . these are labeled with arrows on the 24@xmath0 m image in figure 2 by @xcite . we bring these to the attention of the reader as we extract spectra from these higher surface brightness regions to represent conditions typical of the star forming disk . we will show that spectra extracted from regions above or below the parallelogram region differ from those extracted from the star forming disk or parallelogram shaped region . line emission images were created by summing emission near the line peak and subtracting nearby continuum . these images are shown in figures [ fig : ssi ] [ fig : h2s0 ] for lines present in the long wavelength spectral data cube and in figures [ fig : h2s3 ] [ fig : neii ] for lines or dust emission features present in the sl spectral data cube . line identifications and rest frame wavelengths are summarized in table [ tab : lines ] . the lines [ siii](@xmath21 m ) and [ siii](@xmath22 m ) ( as displayed in figure [ fig : ssi ] ) are strong in the parallelogram or star forming disk , as would be expected from the presence of hii regions in this disk . peaks in these maps are approximately coincident with peaks seen in the nearby continuum at @xmath23 m ( figure [ fig : cont25]b ) . the [ feii](@xmath24 m ) and [ feiii](22.93@xmath0 m ) lines are also detected in the same region with [ feiii ] weaker than [ feii ] ( figure [ fig : fe ] ) . molecular hydrogen in the h@xmath25s(0)@xmath260(@xmath27 m ) line is also detected associated with star formation in the warped disk or parallelogram ( figure [ fig : h2s0 ] ) though the morphology in this line differs from that seen in the [ siii ] , [ siii ] , and [ feii ] lines . the northwestern and southeastern ridges of the parallelogram are stronger than the southwestern and northern ridges . the northwestern and southeastern ridges also seem to be brighter in the [ feii ] map . the emission lines present in the long wavelength spectral cube are clearly associated with the disk , however there are differences between these maps and the nearby continuum . there is more structure seen in the [ siii ] map than present in nearby continuum at 32@xmath0 m ; peaks in the disk are brighter than the surrounding emission in the parallelogram ( compare figure [ fig : ssi ] to [ fig : cont25]b ) . @xcite suggested that the 4 peaks in the broad band @xmath14 m images were due to a superposition of the shell and the star forming disk . however the emission lines at long wavelengths are not detected in the vicinity of the shell - like structure but do display the four emission peaks ( particularly the [ siii ] line image ) . therefore that explanation for the 4 peaks seen at 24@xmath0 m is probably incorrect . those peaks were not predicted by the warped disk model presented by @xcite . the peaks are symmetrical about the nucleus suggesting that they are due to projection of a warped disk and that improvements in the warp model might succeed in accounting for them . if this were the case then the size of the peaks should be similar in the continuum and line emission . unfortunately the peaks are more prominent in some of the lines ( e.g. , [ siii ] ) than the nearby continuum . spiral structure in the warped disk could cause particular regions in the disk to exhibit higher levels of star formation and could maintain the reflective symmetry . if this were the case then variations in the observed morphology between lines and continuum would require a model in which the strength of spiral structure depends on wavelength . the [ siii ] , [ siii ] and [ feii ] emission are primarily confined to the warped disk ( or parallelogram shape ) . however the [ neii](12.8@xmath0 m ) image ( see figure [ fig : neii ] ) does show faint levels of emission ( at the level of @xmath28erg @xmath2s@xmath6sr@xmath6 ) associated with the dust shell . this line is blended with the nearby 12.7@xmath0 m dust emission feature , nevertheless the shape of the spectrum at different locations suggests that [ neii ] dominates the dust emission and is present above and below the galaxy disk . if so then the [ siii ] , [ siii ] and [ feii ] emission associated with the dust shell is just too faint to see in the long wavelength spectral cube . deeper observations at longer wavelengths or higher spectral resolution short wavelength infrared observations would be needed to confirm the presence of ionized gas in the shell . a spectroscopic study at visible band wavelengths would be quite sensitive as extinction is probably not high in the shell region above and below the parallelogram . lines requiring higher ionization , [ oiv](@xmath29 m ) and [ nev](@xmath30 m ) , are also detected in the lh spectral cube and they exhibit different morphology than the lower ionization lines . emission from [ oiv ] and [ nev ] is strong along position angles of @xmath31 and @xmath32 from the nucleus and is resolved at locations outside the nucleus ; see figure [ fig : oivnev ] . these position angles are within 15@xmath33 of the jet axis , though the jet is seen prominently only on the north - east side in radio emission . the radio and x - ray jets in this region have a position angle of @xmath34 @xcite . in figure [ radioo4 ] we show contours of radio continuum at 5 ghz by @xcite overlayed on the [ oiv](@xmath29 m ) emission map ( shown in grayscale ) . we find that the position angle of the [ oiv ] differs from the jet axis by @xmath35 . the different between the axis of the [ oiv ] emission and radio jet axis is perhaps not surprising . observations of seyfert galaxies have found that the jet axis can differ from that of an ionization cone ( ngc 1068 , for example ) . high angular resolution observations of the central few arcseconds of centaurus a have revealed elongated or cone - shaped emission in pa@xmath36 , pa@xmath37 and [ feii](1.26@xmath0 m ) north of the nucleus that has been interpreted in terms of a rotating disk , rather than in terms of an ionization cone @xcite . the jet axis lies on the western side of the possible ionization cone seen in [ feii](1.26 ) . the [ oiv ] and [ nev ] extension to the north of the galaxy nucleus are approximately oriented in the same direction as the cone seen in [ feii](1.26@xmath0 m ) shown by @xcite . as there is no cone feature seen in the near - infrared [ feii ] observations south of the nucleus ( likely because the band of absorption caused by the warped disk ) it is more difficult to compare the cone to the [ oiv ] emission south - west of the nucleus . the [ oiv ] emission north and south of the nucleus are not @xmath38 apart and the emission south - west of the nucleus would not overlay the [ feii](1.26 ) possibly ionization cone if rotated by @xmath38 . it is interesting to reinspect the morphology of the lower ionization line maps at the position of the peak in [ oiv ] south west of the nucleus . an extension in the contours of the [ siii ] , [ siii ] , [ feii ] [ feiii ] and h@xmath25s(0 ) emission line brightness is seen at the same location as the peak south - west of the nucleus in the [ oiv ] map . the ratio of the surface brightness in the extension compared to that in the disk is higher in the [ feiii ] line compared to the [ feii ] line . this is not surprising as we know these extensions contain higher ionization species such as [ nev ] . the spectral resolution in the lh spectral cube ( a spectral resolution of @xmath39 corresponds to a velocity resolution of @xmath40km / s ) is sufficiently high that we can just barely see the galactic disk s rotation in the brighter lines ( e.g. , [ siii ] , [ siii ] and h@xmath25s(0 ) ) , with the western and eastern sides of the disk redshifted or blueshifted , respectively , compared to the system velocity . we inspected the channels near the line centers and searched for evidence of structure in the radial velocity distribution . we found no clear evidence for a significant ( greater than @xmath41km / s ) red or blue shift in the [ oiv ] emission extensions compared to the system velocity . we also saw no clear evidence for an increase in line width . the line width ( fwhm ) must be smaller than the galactic rotation full width or @xmath42km / s . near the [ oiv ] peak south - west of the nucleus the extensions in the contours of the [ siii ] , [ siii ] , and h@xmath25s(0 ) also appear to have velocities within 200km / s of the system velocity . unfortunately the spectral resolution is just not high enough to be certain about these line widths or allow a search for structure in the radial velocity distribution . in the sl spectral cube we detect four pure rotational molecular hydrogen lines , the s(7)@xmath435 , the s(5)@xmath443 , the s(3)@xmath451 and the s(2)@xmath260 lines . we show the continuum subtracted line emission maps for the h@xmath25s(5)(6.909@xmath0 m ) and h@xmath25s(3)(9.665@xmath0 m ) lines in figure [ fig : h2s3 ] . the s(7 ) is weak and the s(2 ) strongly blended with dust emission features so we have only presented line emission maps from the other two lines . the s(4 ) line at 8.0251@xmath0 m was not detected , probably due to the bright dust emission feature at 7.7@xmath0 m at its expected wavelength and because the even number transitions are singlets rather than triplets and so have one third the number of states . the s(1 ) line at 17.03@xmath0 m was not detected because it lies outside the spectral range covered by the sl and lh modules . to observe this line we would have needed observations with the sh module of the infrared spectrograph . the morphology of the higher s(3 ) and s(5 ) rotational molecular hydrogen lines differs from that we see at longer wavelengths in the s(0 ) line at 28@xmath0 m ( compare figure [ fig : h2s0 ] to figure [ fig : h2s3_ov ] ) . the s(0 ) line is primarily associated with the star forming disk , however the s(3 ) and s(5 ) lines are seen above the disk . an overlay of the h@xmath25s(3)(9.665@xmath0 m ) line and [ oiv](25.89@xmath0 m ) in figure [ fig : h2s3_ov]a shows that the peak in the [ oiv ] south - west of the nucleus coincides with the peak h@xmath25s(3 ) surface brightness in the same region . an overlay shown in figure [ fig : h2s3_ov]b of the s(3 ) line with the dust emission in the 11.3@xmath0 pah emission feature ( previously shown in figure [ fig : pah11 ] ) shows that the warmer molecular hydrogen gas ( traced by the higher s pure rotational transitions ) is near the dust shell . the simplest explanation for the morphology is that warmer molecular hydrogen is associated with the dust shell but that there is additional excitation of the shell material near the jet axis . constraints on the column depth and temperature of the molecular hydrogen are discussed below in section 5 . in figure [ fig : pah ] and figure [ fig : neii ] we show maps of continuum emission at the peak of the 7.7 , and 8.6@xmath0 m dust emission features and the [ neii]12.81@xmath0 m line . we did not subtract continuum from these maps as the [ neii ] line and dust emission feature are 3 - 8 times brighter than the continuum . as expected , the [ neii ] line and dust emission features primarily trace the star forming regions in the warped disk . fainter levels of emission arise from the dust shell above and below the disk and from the region that is bright in [ oiv ] emission ( @xmath4625south west of the nucleus ) . in figure [ fig : ratio ] we show ratios of the 7.7@xmath0 m and 11.3@xmath0 m surface brightness and the ratio of the [ neii]12.8 and 11.3@xmath0 m surface brightness . both of these ratios exhibit variations across the galaxy , with the lowest values in the dust shell and higher values present in the star forming disk . subtraction of the continuum made little difference in the morphology of the ratio maps because the lines and dust emission features are over twice as strong as the continuum . @xcite found a correlation between the 7.7/11.3@xmath0 m ratio of dust emission complexes and the [ neiii](15.6)/[neii](12.8@xmath0 m ) ratio in star forming galaxies . they also noted that the galaxies with low 7.7/11.3 complex ratios were more likely to be classified as liners or seyferts . they suggested that the 7.7@xmath0m/11.3@xmath0 m ratio is an indicator of the hardness of the radiation field . the [ neiii](15.6@xmath0 m ) line lies is outside the wavelength coverage of the sl module so we can not use that ratio to probe the radiation field above the disk . at the nucleus in cen a , this line was detected but was not strong @xcite . we do have higher ionization tracers in the the [ oiv](25.9@xmath0 m ) and [ nev](24.3@xmath0 m ) lines . however , the morphology in those lines differs from that we see in the the 7.7 to 11.3@xmath0 m ratio map . the [ oiv ] line is bright near the jet axis whereas whereas the 7.7/11.3 ratio is low in the dustshell ( see figure [ fig : ratio ] ) . the lack of change in the 7.7/11.3 ratio in the region containing [ oiv]25.9 and [ nev]24.3 emission near the jet axis makes it difficult to interpret variations in the 7.7/11.3 ratio purely in terms of hardness of radiation field . since the 7.7/11.3 ratio drops in the vicinity of the dust shell and this is where we detect warm molecular gas we might instead consider models in which pah emission is affected by the presence of warm gas or the physical process that is responsible for heating the warm gas . we could also consider models where this ratio is affected by the ambient radiation field strength . we have extracted spectra at various positions in the star forming disk from the lh spectral cube and these are shown in figure [ fig : lhspec]a . spectra were extracted from 7square regions centered at 5 regions . 4 of these correspond to peaks evident in figure [ fig : ssi ] excluding the nucleus . from bottom to top the spectra shown in figure [ fig : lhspec]a were extracted from regions centered at ra=13:25:25.8 , dec=-43 00 52 ( peak to the west and north of the nucleus ) , ra=13:25:25.9 , dec=-43:01:09 ( peak west of the nucleus ) , ra=13:25:29.5 , dec=-43:01:27 ( south - east of the nucleus ) , ra=13 25 29.5 , dec=-43:01:10 ( peak east of the nucleus ) , and ra=13:25:28.6 , dec=-43:01:23 ( peak south - east of the nucleus ) . we also extract an lh spectrum along the jet axis at the peak shown in fig.[fig : lhspec]b . this spectrum was extracted over a 11square region centered at ra=13:25:25.7 , dec=-43:01:24 , at the peak in the [ oiv ] emission south - west of the nucleus ( see figure [ fig : oivnev ] ) . the lines we have detected from these spectra are listed in table [ tab : lines ] . line fluxes measured from this region are listed in tables [ tab : emjet ] and [ tab : hjet ] . we see only minor differences in different regions of the parallelogram ( see figure [ fig : lhspec ] ) . however , along the jet axis there are deviations in the spectra , in the relative strengths of the lines requiring higher levels of ionization , [ oiv](@xmath29 m ) and [ nev](@xmath30 m ) , compared to those requiring lower levels ( e.g. , [ siii](@xmath21 m ) , [ siii](@xmath22 m ) ) . to produce the emitting ion from the preceding ionization state , [ oiv ] requires an ionization energy of 54.0ev where as [ nev ] requires 97.1ev . a harder uv radiation field most naturally accounts for these species . in the sl spectral cube we extracted a spectrum from the same bright region along the jet axis south - west of the nucleus as that shown in figure [ fig : lhspec]b and from a representative region in the parallelogram representing the star forming disk . the spectrum representing the star forming disk was extracted from a position centered at ra=13:25:25.9 , dec=-43:01:09 . the location is the same as that for the second from bottom spectrum shown in figure [ fig : lhspec ] . the two spectra extracted from the sl spectral cube are shown in figure [ fig : slspec ] . as can be seen from figure [ fig : slspec]a the star forming region exhibits little molecular hydrogen emission in the higher s rotational lines . however , the s(2 ) , s(3 ) , and s(5 ) lines are detected near the jet axis and above the disk near the dust shell ( see figure [ fig : slspec]b ) . the s(7 ) line is barely detected at 5.511@xmath0 m . the s(4 ) line at 8.0241 is not detected likely because it has been overpowered by the dust emission feature at 7.7@xmath0 m . likewise the s(6 ) line at 6.1086@xmath0 m is not detected as it would have been overpowered by the dust emission feature at 6.2@xmath0 m . the s(5 ) line is detected above the disk but not in the star forming region at 6.909@xmath0 m ( figure [ fig : slspec]b ) . at almost the same wavelength the [ arii ] line at 6.985@xmath0 m is seen in the star forming disk . the spectra are sufficiently high spectral resolution that we can differ between the two lines . a comparison between the two spectra also shows that the relative heights of the 7.7 and 11.3@xmath0 m dust emission features differs . these two dust emission features are approximately the same height in the star forming disk . however the 11.3@xmath0 m emission feature is twice as bright as the 7.7@xmath0 m dust emission feature in the dust shell . the [ neii](12.81@xmath0 m ) and dust emission feature at 11.3@xmath0 m are also similar in height in the star forming region but the 11.3@xmath0 m is almost twice as bright as [ neii ] above the disk . these changes are reflected in the ratio maps shown in figure [ fig : ratio ] and discussed in section 3 . there is a weak feature near 14.2@xmath0 m in both spectra from the star forming disk and jet region ( figure [ fig : slspec]a , b ) . the shape of the spectra are similar in shape at 14.2@xmath0 m suggesting that this feature is a dust emission feature rather than due to emission from [ nev](14.3@xmath0 m ) . a higher spectral resolution spectrum would be required to detect this line . weak [ nev](14.3@xmath0 m ) emission was detected in the sh spectrum of the nucleus @xcite . we have measured line fluxes from the spectra shown in figures [ fig : lhspec]b and [ fig : slspec]b which were extracted at the location of the [ oiv ] peak south - west of the nucleus . table [ tab : emjet ] lists the fluxes of the emission lines from ions and table [ tab : hjet ] lists the fluxes of the pure rotational molecular hydrogen transitions . line fluxes were measured from the 3 spectral cubes in the same aperture using continuum fitting and subtraction routines available in cubism . we have used cubism to measure line fluxes for the lh spectral cube even though the images ( displayed in our figures ) show the spectral cube reduced with the software described by @xcite . this spectral cube had fewer artifacts than the one produced by cubism , however the artifacts in the cubism reduced lh spectral cube should not have affected the measured line fluxes as we were able to choose where to measure the continuum and line emission . also the same software package was used to measure all line fluxes to check and ensure consistency in the calibration between the spectral cubes . pure rotational transitions of molecular hydrogen in the mid - infrared ( between 3 and 20@xmath0 m ) are a major coolant of of warm gas in the temperature range 100 - 1000k @xcite . higher s pure rotation molecular hydrogen lines have previously been detected in extragalactic objects . @xcite detected pure rotational lines from ulirgs including the s(7 ) line in a few objects . @xcite detected these lines in the elliptical galaxy ngc 4435 that is experiencing an off - center encounter with a nearby galaxy . this galaxy hosts a dusty disk as seen from visible band hst images and ongoing star formation at modest rate of @xmath47yr . @xcite observed warm gas in the centers of nearby galaxies with liners and seyferts having warmer h@xmath25 emitting in the excited states or higher s transitions . the above studies did not resolve the h@xmath25 emission or did not study its morphology . higher angular resolution near infrared studies have found emission from molecular hydrogen in vibrational transitions in active galaxies . in some seyferts this emission is resolved on 100pc or larger scales ( e.g. , @xcite ) in others rotation is seen near the massive black hole ( e.g. , @xcite ) . the flux of a pure rotational transition can be written @xmath48 where @xmath49 is planck s constant , @xmath50 is the speed of light , @xmath51 is the wavelength of the transition , @xmath52 is the einstein a - coefficient , @xmath53 is the solid angle and @xmath54 is the column density of the initial @xmath55 quantum state . since we are consider pure rotational @xmath56 transitions the rotational quantum number change is @xmath57 , the vibrational quantum number change is @xmath58 , and @xmath59 . assuming local thermal equilibrium , the fraction of molecules in the @xmath60th state is @xmath61 where @xmath62 is the partition function . here @xmath63 is the statistical weight of the state or @xmath64 times 1 or 3 for the even and odd @xmath55 states , respectively . to estimate the temperature of the warm molecular hydrogen we make an excitation diagram plotting the energy of the initial state @xmath65 versus @xmath66 a constant . the excitation diagram is shown in figure [ fig : h2plot ] . on such a plot a single temperature gas with temperature @xmath67 would have points with slope equal to @xmath68 . we have chosen to plot using @xmath69 rather than the natural log so that the @xmath70-axis gives an estimate for the column density of the states . figure [ fig : h2plot ] shows that a single temperature model does not fit all data points . this is not unexpected as other extragalactic objects also require multi - temperature models ( e.g. , @xcite ) . the points lie along a line for initial states higher than @xmath26 implying that the higher @xmath55 rotational level populations are close to lte . the s(2 ) and higher lines are fit with a temperature of @xmath71 . the s(0 ) and s(2 ) line are connected with a slope corresponding to a temperature of @xmath72 . had we subtracted the warmer component , a somewhat lower temperature would have been measured . unfortunately our spectra do not cover the s(1 ) line . this would have helped in making a 2 component model . it is unlikely that there is a strong deviation from thermalization of ortho levels with para levels as the s(2 ) point lies along the same line as the s(3 ) , s(5 ) and s(7 ) points in the excitation diagram . the column density corresponding to the above excitation model is @xmath73@xmath2 for the cooler molecular component ( @xmath74 ) and @xmath75@xmath2 for the warm component ( @xmath71 ) . the column densities are similar to those measured in the nuclei of nearby galaxies by @xcite and in supernova remnants where blast waves encounter molecular clouds @xcite . the temperatures of the two components we measure are higher than most of those measured by @xcite for most nearby galaxy centers excepting the liners and seyfert galaxies , but lower than the supernova remnants @xcite and ulirgs with two temperature models @xcite . the fraction , @xmath76 , in the warmer component ( that at 720k compared to that at 260k ) is higher than that measured for the ulirgs with 2 component temperature models and most but not all the galaxies studied by @xcite but similar to that measured from the supernova remnants @xcite . the physical conditions in cen a s dustshell are most closely matched by the supernovas studied by @xcite . as we have discussed from the images , the molecular hydrogen emission in the higher s lines arises from the dust shell and along the jet axis , though that in s(0 ) is associated with the disk and so is probably from with photodissasociation regions ( pdrs ) associated with star formation . we did not clearly detect the higher s lines in the disk suggesting that the molecular hydrogen at the hotter @xmath77 temperature is primarily excited in the vicinity of the dust shell or along the jet axis . the column density of cooler molecular hydrogen is similar to that estimated in the dustshell by @xcite from the shell surface brightness . this suggests that gas associated with the dust shell has been heated , in the shell and more so near the jet axis . there is probably more than one excitation process as [ oiv ] and [ nev ] emission are detected primarily along the jet axis whereas the higher s pure rotational molecular hydrogen lines are detected there and in the vicinity of the dust shell . nebular emission lines can constrain the the spectrum of the energy powering the emission . our spectra allow us to make use of the following line ratios : [ oiv]@xmath78/[siii]@xmath22 m , and [ oiv]25.9/[neii]12.8@xmath0 m . we do not detect [ nev ] at @xmath79 m so we will discuss limits on the ratios [ nev]14.3/[neii]12.8@xmath0 m and [ nev]14.3/[nev]24.3@xmath0 m . we measure [ oiv]@xmath78/[siii]@xmath22 m @xmath80 at the [ oiv ] peak . this ratio is most similar to those of the agns considered by @xcite ( see their figure 3 ) . however we have failed to detect a significant [ nev ] line at 14.3@xmath0 m so we estimate a limit on [ nev](14.3)/[neii](12.8 ) @xmath81 . ratios this low are more similar to starbursts than agns . we note that the ratio we measure in this region [ oiv]25.9/[neii]12.8 @xmath82 is lower than expected from the [ oiv]25.9/[siii]33.5 ratio using genzel et al.s corrective factor of 1.7 . our estimated [ oiv]25.9/[neii]12.8 @xmath83 ratio places the emission along the jet axis more similar to ulirgs than agns . [ oiv]25.9/[neii]12.8 is expected to be approximately 1 for agns ( @xcite , figure 3a and figure 7 by @xcite ) . our low estimated ratios of [ nev]14.3/[neii]12.8@xmath0 m and [ oiv]25.9/[neii]12.8@xmath0 m compared to those expected in agns may in part be due to the blending of the [ neii]12.8 emission with dust emission features which may have caused us to overestimate the [ neii](12.8@xmath0 m ) line strength . we do nt detect [ nev](14.3@xmath0 m ) or [ nevi ] at 7.6@xmath0 m and we do detect [ nev](24.3@xmath0 m ) . by comparing upper limits of ratios of these lines with [ oiv](25.9 ) and [ nev](24.3@xmath0 m ) and figure 4 by @xcite ( based on models by @xcite ) we infer the emitting region is likely to have low ionization parameter @xmath84 . our estimated upper limit in the ratio of [ nev]14.3/24.3 of @xmath85 in the jet region and a comparison of figure 3 by @xcite suggests that the electron density is also low @xmath86@xmath87 . the detection of [ nev ] suggests that a hard radiation field is illuminating the dust shell near the jet axis . however we have failed to detect [ nev](14.3@xmath0 m ) or [ nevi](7.65@xmath0 m ) suggesting that the illuminated region is low density and has a low ionization parameter . the difference in the physical conditions compared to those expected in narrow line regions might account for the discrepancy between the [ oiv]25.9/[neii]12.8 and [ oiv]25.9/[siii]33.5 ratios compared to those seen in agns . we consider the possibility that the low electron density and ionization parameter are consistent with illumination of a region distant from the nucleus by hard radiation from the central agn . the position of the [ oiv](25.9@xmath0 m ) peak corresponds to a distance @xmath88 pc from the nucleus . the active nucleus is estimated to have a bolometric luminosity of @xmath89 erg / s @xcite . the ionization parameter , @xmath90 , for isotropic emission is @xmath91 where @xmath50 is the speed of light , @xmath92 is the ionizing photon emission rate ( photons per second ) and @xmath93 is the gas density ( in @xmath87 ) . assuming a mean energy for ionizing photons of 100ev ( the estimated energy of the uv bump and hard enough to produce nev from the preceding ionization state ) at the estimated bolometric luminosity emitted isotropically into @xmath94 radians , @xmath95 this estimate illustrates that a low ionization parameter , low density medium ( such as we infer from the line ratios ) is the only situation that we might expect at a distance of 400 pc from the agn . we now ask : can the agn nucleus provide enough uv photons to account for the flux of the [ oiv ] emission 25from the nucleus ? we summed the [ oiv](25.9@xmath0 m ) flux in a 15wide region centered on the [ oiv ] emission peak , estimating a flux of @xmath96 erg @xmath2 s@xmath6 in the line . this flux corresponds to a luminosity in [ oiv ] of @xmath97 erg / s . we can compare the [ oiv ] luminosity to the mid - infrared luminosity as have @xcite for other objects . since the [ oiv ] flux is illuminated only in a small solid angle from the nucleus we must first estimate the fraction of uv luminosity that could be causing the [ oiv ] emission . the [ oiv ] emission comes from a region that is about 10wide . as viewed from the position of the nucleus , this corresponds to a solid angle of @xmath98 or @xmath99% of @xmath94 steradians . thus we compare the [ oiv ] luminosity to 1% of the mid - infrared luminosity . the mid - infrared luminosity of the agn is @xmath100erg s@xmath6 @xcite . consequently we estimate that the ratio of the [ oiv ] to 1% of the mid - infrared luminosity is @xmath101 . this ratio is similar to that seen for the brighter seyferts studied by @xcite ( see their figure 12 ) . we conclude that as long as they are not absorbed before they run into the dust shell , uv photons from the agn can account for the [ oiv ] and [ nev ] emission near the jet axis . an additional power source ( like shocks associated with the jet ) is not necessarily required to account for the high ionization species seen near the jet axis but distant from the nucleus . if uv photons from the agn are responsible for the high ionization species then the column depth along the sight line between the agn and dust shell must be low . if a blast wave is responsible for sweeping up the dust shell , as proposed by @xcite , then it could also be responsible for evacuating material between the agn and the dust shell . we have carried out a spectroscopic study of the central 2 arcminutes of centaurus a using short low and long high spectral modules of the infrared spectrograph on board the _ spitzer space telescope_. most of the emission lines detected in the spectral cubes ( e.g. , [ siii](33.5@xmath0 m ) , [ siii](34.8 ) , [ feii](26.0 ) , [ feiii](23.9 ) , [ arii](6.98 ) and h@xmath25s(0)(28.2 ) ) and dust emission features primarily trace regions of star formation in the warped disk . our previous study based on irac and mips imaging suggested that centaurus a hosts an oval or bipolar dustshell at a distance approximately 500 pc from the nucleus seen above and below the warped disk . this dust shell , if confirmed , would be the first extragalactic shell to be discovered in the infrared . here we see the dust shell even more clearly and prominently in the 11.3@xmath0 m dust emission feature than we saw previously in the broad band irac images . we have found variations in the dust emission feature 7.7@xmath0m/11.3@xmath0 m ratio and dust 11.3@xmath0m/[neii](12.8@xmath0 ) ratio , with the oval dust shell having the lowest ratios compared to the star forming disk . the clearer shell morphology at 11.3@xmath0 m than previously seen in broad band images , the association of the molecular hydrogen emission in the shell , and the variation in line ratios in the shell compared to those in the disk , confirm spectroscopically that the shell discovered previously @xcite is a separate coherent entity and is unlikely to be a chance superposition of dust filaments . we find evidence for higher ionization species line emission in [ nev](@xmath30 m ) and [ oiv](@xmath29 m ) near the jet axis . emission in these two lines is seen both north - east and south - west of the nucleus along position angles @xmath31 and @xmath102 . these angles are similar to but not exactly the same as the jet axis at 55@xmath33 as seen at 5ghz . outside the nucleus , the peak surface brightness in these lines is 25or 400 pc south - west of the nucleus . emission line ratios and limits at the location of the [ oiv ] peak suggest that the emitting region is at low ionization parameter , @xmath84 , and has low electron density , @xmath86@xmath87 . we crudely estimate that the agn can provide sufficient uv photons to account for the [ oiv ] luminosity 400 pc from the nucleus , as long as uv photons are not absorbed by intervening material as they travel from the agn to the dust shell . a much more detailed photo - ionization study is required to understand the excitation of the [ oiv ] and [ nev ] emission . previous reports of an ionization cone in cen a were based on near - infrared imaging of the central few arcseconds @xcite . subsequent studies interpreted line emission in terms of a disk rather than ionization cone @xcite . unfortunately the morphology of the near - infrared images is strongly affected by extinction and the warp disk models have not been good enough to accurately predict the extinction in the central few arcsecond of the nucleus . cen a might be the only active galaxy in which mid - infrared spectroscopy has found evidence for high ionization species such as nev at hundred pc distances from the nucleus . as far as we know cen a hosts the only ionization cone that has been resolved with observations from the spitzer space telescope . we see evidence for warm molecular hydrogen coincident with the peak in [ oiv ] in the odd pure rotational odd transitions s(3 ) and s(5 ) . the s(7 ) and s(2 ) transitions are also detected but at weaker levels . the s(3 ) and s(5 ) emission also lies in the vicinity of the dust shell that is most prominent in the 11.3@xmath0 m dust ( pah ) emission feature . a two temperature component model can fit the rotational line ratios and implies that there is warm molecular hydrogen with temperatures in the range 250720k . the temperatures are warmer than seen in nucleus of non - active nearby galaxies , similar to those of liners and seyferts but lower than exhibited by supernova remnants ulirgs ( as compared to studies by @xcite , @xcite and @xcite ) . near the jet axis , the column depth of warm molecular hydrogen is @xmath1@xmath2 similar to that estimated from the infrared continuum emission of the dust shell by @xcite . this suggests that gas associated with the dust shell has been heated near the jet axis . there is probably more than one excitation process as [ oiv ] and [ nev ] emission are detected primarily along the jet axis whereas the higher s pure rotational molecular hydrogen lines are detected there and in the vicinity of the dust shell . previous studies of the pure - rotational molecular hydrogen lines in extra - galactic objects ( e.g. , @xcite ) have not well resolved the emission . the association of the warm molecular hydrogen gas with a shell is most similar to phenomena exhibited by galactic supernova remnants where the blast wave encounters molecular clouds @xcite . the physical conditions estimated from the molecular hydrogen observations are similar in properties ( column depth , temperatures and fraction of gas in the two temperature components ) to the parameters estimated by @xcite in 4 galactic supernova remnants . this suggests that theory of interstellar shock waves be applied to interpreting observations of the shell . @xcite associates the warmer molecular hydrogen component responsible for the higher s pure rotational transitions with dissociative shocks . these require shock velocities @xmath103km / s ( e.g. , @xcite ) . a physical scenario and model accounting for the shell s structure and energetics is currently lacking . deep optical spectroscopic and radio studies are particularly needed to better constrain gas motions and physical conditions in this shell . 1.0truein we thank dan watson , paul van der werf , ralph kraft , jacqueline van gorkom , martin hardcastle , and christine jones - forman for helpful discussions and correspondence . we thank martin hardcastle for providing us with images of centaurus a in the radio . support for this work was in part provided by by nasa through an award issued by jpl / caltech , national science foundation grants ast-0406823 @xmath104 phy-0552695 , the national aeronautics and space administration under grant no.@xmath46nng04gm12 g issued through the origins of solar systems program , and hst - ar-10972 to the space telescope science institute . jbh is funded by a federation fellowship from the australian research council . continuum emission from the lh spectral cube from @xmath13 m wide bands . contours are evenly spaced with the lowest contour and spacing at 0.01 mjy / sr . a ) continuum centered at @xmath20 m . b ) continuum centered at @xmath105 m . , title="fig:",width=336 ] continuum emission from the lh spectral cube from @xmath13 m wide bands . contours are evenly spaced with the lowest contour and spacing at 0.01 mjy / sr . a ) continuum centered at @xmath20 m . b ) continuum centered at @xmath105 m . , title="fig:",width=336 ] dust emission feature at 11.3@xmath0 m . no continuum has been subtracted as the emission feature dominates the spectrum by a factor of 38 . the minimum contour and spacing is approximately 10 mjy / sr in the peak of the line . the black contours show the oval dust shell previously described by @xcite . , width=336 ] continuum subtracted line emission images in the lh spectral cube , showing emission in the folded star forming disk . contours are evenly spaced . the lowest contours and spacing for the [ siii](@xmath106 m ) and [ siii](@xmath107 m ) images are 1.0 and 1.5 @xmath108erg @xmath2 s@xmath6 sr@xmath6 , respectively . a ) for [ siii](@xmath106 m ) . b ) for [ siii](@xmath109 m ) . , title="fig:",width=336 ] continuum subtracted line emission images in the lh spectral cube , showing emission in the folded star forming disk . contours are evenly spaced . the lowest contours and spacing for the [ siii](@xmath106 m ) and [ siii](@xmath107 m ) images are 1.0 and 1.5 @xmath108erg @xmath2 s@xmath6 sr@xmath6 , respectively . a ) for [ siii](@xmath106 m ) . b ) for [ siii](@xmath109 m ) . , title="fig:",width=336 ] continuum subtracted line emission images in the lh spectral cube , showing emission in the folded star forming disk . contours are evenly spaced . the lowest contours and spacings for [ feii](25.988@xmath0 m ) and [ feiii](22.925@xmath0 m ) images with lowest contour are 0.05 and 0.025 @xmath108erg @xmath2 s@xmath6 sr@xmath6 , respectively . a ) for [ feii](@xmath110 m ) . b ) for [ feiii](22.925@xmath0 m ) . , title="fig:",width=336 ] continuum subtracted line emission images in the lh spectral cube , showing emission in the folded star forming disk . contours are evenly spaced . the lowest contours and spacings for [ feii](25.988@xmath0 m ) and [ feiii](22.925@xmath0 m ) images with lowest contour are 0.05 and 0.025 @xmath108erg @xmath2 s@xmath6 sr@xmath6 , respectively . a ) for [ feii](@xmath110 m ) . b ) for [ feiii](22.925@xmath0 m ) . , title="fig:",width=336 ] continuum subtracted line emission images in the lh spectral cube , showing emission in the folded star forming disk . for the h@xmath25s(0)(@xmath111 m ) line with lowest contour at and spacing at 0.01 @xmath108erg @xmath2 s@xmath6 sr@xmath6 . , width=336 ] continuum subtracted line emission images in the lh spectral cube , showing emission near the jet axis . contours are evenly spaced . the lowest contours and spacing for the [ oiv](@xmath4 m ) and [ nev](@xmath3 m ) images are 0.5 and 0.1 @xmath108erg @xmath2 s@xmath6 sr@xmath6 , respectively . a ) for [ oiv](@xmath4 m ) . b ) for [ nev](@xmath3 m ) . , title="fig:",width=336 ] continuum subtracted line emission images in the lh spectral cube , showing emission near the jet axis . contours are evenly spaced . the lowest contours and spacing for the [ oiv](@xmath4 m ) and [ nev](@xmath3 m ) images are 0.5 and 0.1 @xmath108erg @xmath2 s@xmath6 sr@xmath6 , respectively . a ) for [ oiv](@xmath4 m ) . b ) for [ nev](@xmath3 m ) . , title="fig:",width=336 ] ( @xmath29 m ) contours . the radio map is a 6resolution map of the inner radio lobes by @xcite . the [ oiv]@xmath29 and [ nev]@xmath30 m line emission are oriented approximately but not exactly along the jet axis . [ radioo4 ] , width=336 ] continuum subtracted line emission images from the sl spectral cube , showing the h@xmath25s(3)(9.665@xmath0 m ) and h@xmath25s(5)(6.909@xmath0 m ) lines with lowest contours at 1 @xmath112erg @xmath2 s@xmath6 sr@xmath6 . the higher s rotational lines from h@xmath25 exhibit different morphology than the h@xmath25s(0)28@xmath0 m line that was seen primarily in the folded star forming disk . emission in the higher s lines is seen above the disk . contours are evenly spaced . lowest contour and spacing is @xmath112erg @xmath2 s@xmath6 sr@xmath6 for both lines . for a ) h@xmath25s(3)(9.665@xmath0 m ) . for b ) h@xmath25s(5)(6.909@xmath0 m ) . , title="fig:",width=336 ] continuum subtracted line emission images from the sl spectral cube , showing the h@xmath25s(3)(9.665@xmath0 m ) and h@xmath25s(5)(6.909@xmath0 m ) lines with lowest contours at 1 @xmath112erg @xmath2 s@xmath6 sr@xmath6 . the higher s rotational lines from h@xmath25 exhibit different morphology than the h@xmath25s(0)28@xmath0 m line that was seen primarily in the folded star forming disk . emission in the higher s lines is seen above the disk . contours are evenly spaced . lowest contour and spacing is @xmath112erg @xmath2 s@xmath6 sr@xmath6 for both lines . for a ) h@xmath25s(3)(9.665@xmath0 m ) . for b ) h@xmath25s(5)(6.909@xmath0 m ) . , title="fig:",width=336 ] a ) emission in h@xmath25s(3)(9.665@xmath0 m ) shown as gray scale overlayed with [ oiv](@xmath4 ) contours . contours are evenly spaced with lowest contours at are 0.5 @xmath108erg @xmath2 s@xmath6 sr@xmath6 and a spacing 4 times this . the gray scale range for the h@xmath25s(3 ) image is 0(white ) to 7(black ) @xmath112erg @xmath2 s@xmath6 sr@xmath6 . b ) same as a ) except the h@xmath25s(3 ) emission is overlayed with contours of the 11.3@xmath0 m pah dust emission feature ( as shown in figure [ fig : pah11 ] ) . contour spacing and lowest level is 20 mjy / sr . , title="fig:",width=336 ] a ) emission in h@xmath25s(3)(9.665@xmath0 m ) shown as gray scale overlayed with [ oiv](@xmath4 ) contours . contours are evenly spaced with lowest contours at are 0.5 @xmath108erg @xmath2 s@xmath6 sr@xmath6 and a spacing 4 times this . the gray scale range for the h@xmath25s(3 ) image is 0(white ) to 7(black ) @xmath112erg @xmath2 s@xmath6 sr@xmath6 . b ) same as a ) except the h@xmath25s(3 ) emission is overlayed with contours of the 11.3@xmath0 m pah dust emission feature ( as shown in figure [ fig : pah11 ] ) . contour spacing and lowest level is 20 mjy / sr . , title="fig:",width=336 ] flux at 7.7 and 8.6@xmath0 m showing dust emission features . contours are evenly spaced . no continuum subtraction has been done . the star forming disk is evident as the parallelogram shaped feature corresponding to the folded disk . the dust shell is seen above and below the parallelogram feature . the minimum contour and spacing is approximately 10 mjy / sr in the peak of the line . a ) the 7.7@xmath0 m dust emission feature . b ) the 8.6@xmath0 m dust emission feature . , title="fig:",width=336 ] flux at 7.7 and 8.6@xmath0 m showing dust emission features . contours are evenly spaced . no continuum subtraction has been done . the star forming disk is evident as the parallelogram shaped feature corresponding to the folded disk . the dust shell is seen above and below the parallelogram feature . the minimum contour and spacing is approximately 10 mjy / sr in the peak of the line . a ) the 7.7@xmath0 m dust emission feature . b ) the 8.6@xmath0 m dust emission feature . , title="fig:",width=336 ] line emission in [ neii](@xmath113 m ) . no continuum has been subtracted as the [ neii ] line dominates the continuum by a factor greater than 2 everywhere . contours are evenly spaced with lowest contour at 4 @xmath112erg @xmath2 s@xmath6 sr@xmath6 . , width=336 ] a ) the dust emission feature at 7.7@xmath0 m divided by that at 11.3@xmath0 m . the lowest and highest contours are shown at a ratios of 0.5 ( in shell ) and 1.0 ( in parallelogram ) , with spacing of 0.1 . black is 1.0 , white is 0.5 . the 7.7 to 11.0@xmath0 m dust feature ratio varies with the dust shell having the lowest ratio . b ) the [ neii](12.8@xmath0 m ) line divided by the @xmath114 m dust emission feature . the lowest and highest contours are shown at ratios of 1.9 ( in shell ) and 3.0 ( in parallelogram ) , with spacing of 0.1 . black is 3.0 , white is 1.9 . the strength of the neii](12.8@xmath0 m ) compared to the 11.3@xmath0 m dust feature also varies with the dust shell having the lowest ratios . , title="fig:",width=336 ] a ) the dust emission feature at 7.7@xmath0 m divided by that at 11.3@xmath0 m . the lowest and highest contours are shown at a ratios of 0.5 ( in shell ) and 1.0 ( in parallelogram ) , with spacing of 0.1 . black is 1.0 , white is 0.5 . the 7.7 to 11.0@xmath0 m dust feature ratio varies with the dust shell having the lowest ratio . b ) the [ neii](12.8@xmath0 m ) line divided by the @xmath114 m dust emission feature . the lowest and highest contours are shown at ratios of 1.9 ( in shell ) and 3.0 ( in parallelogram ) , with spacing of 0.1 . black is 3.0 , white is 1.9 . the strength of the neii](12.8@xmath0 m ) compared to the 11.3@xmath0 m dust feature also varies with the dust shell having the lowest ratios . , title="fig:",width=336 ] a ) spectrum from the star forming parallelogram . d refers to a dust emission feature . b ) spectrum from the jet region . the pure rotational molecular hydrogen lines s(2)s(7 ) are labeled as s2-s7 . molecular hydrogen lines are more prominent in the jet region than in the star forming parallelogram . in the star forming disk the [ arii](6.985@xmath0 m ) line is brighter than the nearby pure rotational molecular hydrogen s(2 ) line at @xmath115 m whereas the opposite is true in the jet region . there is a change in the ratio of the dust emission features in these two spectra and in the ratio of the [ neii]/11.3 dust emission feature . , title="fig:",width=336 ] a ) spectrum from the star forming parallelogram . d refers to a dust emission feature . b ) spectrum from the jet region . the pure rotational molecular hydrogen lines s(2)s(7 ) are labeled as s2-s7 . molecular hydrogen lines are more prominent in the jet region than in the star forming parallelogram . in the star forming disk the [ arii](6.985@xmath0 m ) line is brighter than the nearby pure rotational molecular hydrogen s(2 ) line at @xmath115 m whereas the opposite is true in the jet region . there is a change in the ratio of the dust emission features in these two spectra and in the ratio of the [ neii]/11.3 dust emission feature . , title="fig:",width=336 ] a ) spectra from the star forming warped disk ( seen as a parallelogram in continuum ) are shown at the 4 peaks in the parallelogram and at one additional point also in the parallelogram . each spectrum is offset by @xmath1160.1 mjy / sr from the other . from bottom to top : the northern peak west of the nucleus , the southern peak west of the nucleus , the southern peak east of the nucleus , the northern peak east of the nucleus , a point between the southern and eastern peak and the nucleus . positions are given in the text . in the star forming disk or parallelogram [ nev](@xmath3 m ) is not detected and [ oiv](@xmath4 ) is weaker than [ feii](@xmath110 m ) . b ) spectrum from a region near the jet axis south - west of nucleus . near the jet axis [ nev](@xmath3 m ) is detected and [ oiv](@xmath4 ) is 3 - 5 times brighter than [ feii](@xmath110 m ) . the presence of [ nev ] implies suggests that the radiation field is hard near the jet axis . , title="fig:",width=336 ] a ) spectra from the star forming warped disk ( seen as a parallelogram in continuum ) are shown at the 4 peaks in the parallelogram and at one additional point also in the parallelogram . each spectrum is offset by @xmath1160.1 mjy / sr from the other . from bottom to top : the northern peak west of the nucleus , the southern peak west of the nucleus , the southern peak east of the nucleus , the northern peak east of the nucleus , a point between the southern and eastern peak and the nucleus . positions are given in the text . in the star forming disk or parallelogram [ nev](@xmath3 m ) is not detected and [ oiv](@xmath4 ) is weaker than [ feii](@xmath110 m ) . b ) spectrum from a region near the jet axis south - west of nucleus . near the jet axis [ nev](@xmath3 m ) is detected and [ oiv](@xmath4 ) is 3 - 5 times brighter than [ feii](@xmath110 m ) . the presence of [ nev ] implies suggests that the radiation field is hard near the jet axis . , title="fig:",width=336 ] excitation diagram of the h@xmath25 pure rotational lines . points are based on those measured in the jet region with fluxes listed in table [ tab : hjet ] . solid squares refer to measurements and open triangles to upper limits . the solid line is for a temperature of @xmath71 whereas the dotted one is @xmath117 . , width=336 ] + fluxes were measured in the same region as listed in table [ tab : emjet ] and in the spectra shown in figures [ fig : lhspec]b and [ fig : slspec]b . fluxes are given in units of @xmath118 . for the s(4 ) and s(6 ) lines we estimate upper limits . the upper limits are high because of dust emission features .

big model analytics tackles the training of massive models that go beyond the available memory of a single computing device , e.g. , cpu or gpu . it generalizes big data analytics which is targeted at how to train memory - resident models over out - of - memory training data . in this paper , we propose an in - database solution for big model analytics . we identify dot - product as the primary operation for training generalized linear models and introduce the first array - relation dot - product join database operator between a set of sparse arrays and a dense relation . this is a constrained formulation of the extensively studied sparse matrix vector multiplication ( spmv ) kernel . the paramount challenge in designing the dot - product join operator is how to optimally schedule access to the dense relation based on the non - contiguous entries in the sparse arrays . we prove that this problem is np - hard and propose a practical solution characterized by two technical contributions dynamic batch processing and array reordering . we devise three heuristics lsh , radix , and k - center for array reordering and analyze them thoroughly . we execute extensive experiments over synthetic and real data that confirm the minimal overhead the operator incurs when sufficient memory is available and the graceful degradation it suffers as memory becomes scarce . moreover , dot - product join achieves an order of magnitude reduction in execution time over alternative in - database solutions .

1602.08845

big data analytics is a major topic in contemporary data management and machine learning research and practice . many platforms , e.g. , optiml @xcite , graphlab @xcite , systemml @xcite , vowpal wabbit @xcite , simsql @xcite , glade @xcite and libraries , e.g. , madlib @xcite , bismarck @xcite , mllib @xcite , mahout , have been proposed to provide support for distributed / parallel statistical analytics . we can categorize these solutions into general frameworks with machine learning support mahout , spark s mllib , graphlab dedicated machine learning systems systemml , simsql , optiml , vowpal wabbit and frameworks within databases madlib , bismarck , glade . in this paper , we focus on the last category frameworks for in - databse analytics . a common assumption across all these systems is that the number of model parameters or features is small enough to fit in memory . this is made explicit by the representation of the model as an in - memory array data structure . however , due to the explosive growth in data acquisition and the wide adoption of analytics methods , the current trend is to devise models with an ever - increasing number of features_big models_. a report on industrial machine learning cites models with 100 billion features ( 800 gb in size ) as early as 2012 . scientific applications such as ab initio nuclear structure calculations also generate extremely large models with billions of features @xcite . while these are extreme cases , there are many realistic applications that require big models and are forced to limit the number of features they consider because of insufficient memory resources . we provide two such examples in the following . * example 1 : recommender systems . * a class of analytics models with highly - dimensional feature space are the ones in which the dimensionality grows with the number of observations . low - rank matrix factorization ( lmf ) is a typical example . in lmf , the observations are a set of cells in a sparse @xmath0 matrix @xmath1 . the non - empty cells represent the users ratings of items in a certain category such as songs or movies with each row corresponding to a user and each column to an item . in general , every row is sparse since a typical user rates only a very limited set of items . each column is also sparse since only a restricted number of users rate an item . lmf seeks to decompose matrix @xmath1 into the product of two dense low - rank matrices @xmath2 and @xmath3 with dimensions @xmath4 and @xmath5 , respectively , where @xmath6 is the rank . the prediction accuracy increases with @xmath6 . the lmf features are matrices @xmath2 and @xmath3 which grow with the number of users @xmath7 and the number of items @xmath8 , respectively , and the rank @xmath6 . lmf is heavily used in recommender systems , e.g. , netflix , pandora , spotify . for example , spotify applies lmf for 24 million users and 20 million songs , which leads to 4.4 billion features at a relatively small rank of 100 . * example 2 : topic modeling . * n - grams are a common feature vector in topic modeling . they are extracted by considering sequences of 1-word tokens ( unigrams ) , 2-word tokens ( bigrams ) , up to n - word tokens ( n - grams ) from a fixed dictionary . the feature vector consists of the union of unigrams , bigrams , @xmath9 , n - grams . several analytics models can be applied over this feature space , including latent dirichlet allocation , logistic regression , and support vector machines . for the english wikipedia corpus , a feature vector with 25 billion unigrams and 218 billion bigrams can be constructed @xcite . a similar scale can be observed in genomics where topic modeling is applied to genome sequence analysis . * existing solutions . * the standard model representation across all the big data analytics systems we are aware of in - database or not is a memory - resident container data structure , e.g. , ` vector ` or ` map ` . depending on the parallelization strategy , there can be one or more model instances in the system at the same time . hogwild ! @xcite uses a single non - synchronized instance , while averaging techniques @xcite replicate the model for each execution thread . at the scale of models introduced above , a memory - resident solution incurs prohibitive cost if it is feasible at all . in reality , in - database analytics frameworks can not handle much smaller models . for example , madlib and bismarck are built using the udf - uda functionality available in postgresql . the model is stored as an array attribute in a single - column table . postgresql imposes a hard constraint of 1 gb for the size of an attribute , effectively limiting the model size . high performance computing ( hpc ) libraries for efficient sparse linear algebra such as intel mkl , trilinos , cusparse , and cusp are optimized exclusively for in - memory processing , effectively requiring that both the training dataset and the model fit in memory simultaneously . two approaches are possible to cope with insufficient memory secondary storage processing and distributed memory processing . in secondary storage processing , the model is split into partitions large enough to fit in memory and the goal is to optimize the access pattern in order to minimize the number of references to secondary storage . this principle applies between any two layers of the storage hierarchy cache and memory , memory and disk ( or ssd ) , and texture memory and global memory of a gpu . while we are not aware of any secondary storage solution for data analytics , there have been several attempts to optimize the memory access pattern of the spmv kernel . however , they are targeted at minimizing the number of cache misses @xcite or the number of accesses to the gpu global memory @xcite not the number of disk accesses . in distributed memory processing , the big model is partitioned across several machines , with each machine storing a sufficiently small model partition that fits in its local memory . since remote model access requires data transfer , the objective in distributed processing is to minimize the communication between machines . this can not be easily achieved for the spmv kernel due to the non - clustered access pattern . distributed big data analytics systems built around the map - reduce computing paradigm and its generalizations , e.g. , hadoop and spark , require several rounds of repartitioning and aggregation due to their restrictive communication pattern . to the best of our knowledge , parameter server @xcite is the only distributed memory analytics system capable of handling big models directly . in parameter server , the big model can be transfered and replicated across servers . whenever a model entry is accessed , a copy is transferred over the network and replicated locally . modifications to the model are pushed back to the servers asynchronously . the communication has to be implemented explicitly and optimized accordingly . while parameter server supports big models , it does so at the cost of a significant investment in hardware and with considerable network traffic . our focus is on cost - effective single node solutions . * approach & contributions . * in this paper , we propose an in - database solution for big model analytics . the main idea is to offload the model to secondary storage and leverage database techniques for efficient training . the model is represented as a table rather than as an array attribute . this distinction in model representation changes fundamentally how in - database analytics tasks are carried out . we identify _ dot - product _ as the most critical operation affected by the change in model representation . our central contribution is the first _ dot - product join physical database operator _ optimized to execute secondary storage array - relation dot - products effectively . dot - product join is a constrained instance of the spmv kernel @xcite which is widely - studied across many computing areas , including hpc , architecture , and compilers . the paramount challenge we have to address is how to optimally schedule access to the dense relation buffer management based on the non - contiguous feature entries in the sparse arrays . the goal is to minimize the overall number of secondary storage accesses across all the sparse arrays . we prove that this problem is np - hard and propose a practical solution characterized by two technical contributions . the first contribution is to handle the sparse arrays in _ batches with variable size_determined dynamically at runtime . the second contribution is a _ reordering strategy _ for the arrays such that accesses to co - located entries in the dense relation can be shared . our specific contributions can be summarized as follows : we investigate the big model analytics problem and identify dot - product as the critical operation in training generalized linear models ( section [ sec : problem ] ) . we also establish a direct correspondence with the well - studied sparse matrix vector ( spmv ) multiplication problem . we present several alternatives for implementing dot - product in a relational database and discuss their relative advantages and drawbacks ( section [ sec : baseline ] ) . we design the first array - relation dot - product join database operator targeted at secondary storage ( section [ sec : dot - product ] ) . we prove that identifying the optimal access schedule for the dot - product join operator is np - hard ( section [ ssec : reordering : np - hard ] ) and introduce two optimizations dynamic batch processing and reordering to make the operator practical . we devise three batch reordering heuristics lsh , radix , and k - center ( section [ sec : reordering ] ) inspired from optimizations to the spmv kernel and evaluate them thoroughly . we show how the dot - product join operator is integrated in the gradient descent optimization pipeline for training generalized linear models ( section [ sec : dp - gd ] ) . we execute an extensive set of experiments that evaluate each sub - component of the operator and compare our overall solution with alternative dot - product implementations over synthetic and real data ( section [ sec : experiments ] ) . the results show that dot - product join achieves an order of magnitude reduction in execution time over alternative in - database solutions . in this section , we provide a brief introduction to gradient descent as a general method to train a wide class of analytics models . then , we give two concrete examples logistic regression and low - rank matrix factorization that illustrate how gradient descent optimization works . from these examples , we identify _ vector dot - product _ as the primary operation in gradient descent optimization . we argue that existing in - database solutions can not handle big model analytics because they do not support secondary storage dot - product . we provide a formal statement of the research problem studied in this paper and conclude with a discussion on the relationship with spmv . consider the following optimization problem with a linearly separable objective function : @xmath10 in which a @xmath11-dimensional vector @xmath12 , @xmath13 , known as the model , has to be found such that the objective function is minimized . the constants @xmath14 and @xmath15 , @xmath16 , correspond to the feature vector of the @xmath17 data example and its scalar label , respectively . gradient descent represents by far the most popular method to solve the class of optimization problems given in eq.([eq : optim - form ] ) . gradient descent is an iterative optimization algorithm that starts from an arbitrary model @xmath18 and computes new models @xmath19 , such that the objective function , a.k.a . , the loss , decreases at every step , i.e. , @xmath20 . the new models @xmath19 are determined by moving along the opposite @xmath21 gradient direction . formally , the @xmath21 gradient is a vector consisting of entries given by the partial derivative with respect to each dimension , i.e. , @xmath22 $ ] . the length of the move at a given iteration is known as the step size denoted by @xmath23 . with these , we can write the recursive equation characterizing the gradient descent method : @xmath24 in batch gradient descent ( bgd ) , the update equation is applied as is . this requires the exact computation of the gradient over the entire training dataset . to increase the number of steps taken in one iteration , stochastic gradient descent ( sgd ) estimates the @xmath21 gradient from a subset of the training dataset . this allows for multiple steps to be taken in one sequential scan over the training dataset as opposed to a single step in bgd . in the following , we provide two illustrative examples that show how gradient descent works for two popular analytics tasks logistic regression ( lr ) and low - rank matrix factorization ( lmf ) . * logistic regression . * the lr objective function is : @xmath25 the model @xmath12 that minimizes the cumulative log - likelihood across all the training examples is the solution . the key operation in this formula is the dot - product between vectors @xmath12 and @xmath14 , i.e. , @xmath26 . this dot - product also appears in each component of the gradient : @xmath27 * low - rank matrix factorization . * the general form of the lmf objective function is : @xmath28 two low - rank matrices @xmath29 and @xmath30 , i.e. , the model , have to be found such that the sum of the cell - wise least - squares difference between the data matrix @xmath31 and the product @xmath32 is minimized . @xmath33 is the rank of matrices @xmath2 and @xmath3 . the lmf gradient as a function of row @xmath34 in matrix @xmath35 is shown in the following formula : @xmath36 there is such a dot - product between row @xmath37 and each column @xmath38 , i.e. , @xmath39 . however , only those dot - products for which there are non - zero cells @xmath40 in matrix @xmath1 have to be computed . a similar gradient formula can be written for column @xmath41 in @xmath42 . from these examples , we identify dot - product as the essential operation in gradient descent optimization . with almost no exception , dot - product appears across all analytics tasks . in lr and lmf , a dot - product has to be computed between the model @xmath12 and each example @xmath43 in the training dataset at each iteration . formally , the dot - product of two @xmath11-dimensional vectors @xmath44 and @xmath45 is a scalar value computed as : @xmath46 this translates into a simple algorithm that iterates over the two vectors synchronously , computes the component - wise product , and adds it to a running sum ( algorithm [ alg : dp - form ] ) . in most cases , the @xmath47 function is a simple multiplication . however , in the case of lmf , where @xmath48 and @xmath49 are vectors themselves , @xmath47 is itself a dot - product . @xmath50 @xmath51 @xmath52 @xmath53 in this paper , we focus on _ big model dot - product_. several aspects make this particular form of dot - product an interesting and challenging problem at the same time . first , the vector corresponding to the model say @xmath45 can not fit in the available memory . second , the vector corresponding to a training example say @xmath44 is sparse and the number of non - zero entries is a very small fraction from the dimensionality of the model . in these conditions , algorithm _ dot - product _ becomes highly inefficient because it has to iterate over the complete model even though only a small number of entries contribute to the result . the third challenge is imposed by the requirement to support big model dot - product inside a relational database where there is no notion of order the relational model is unordered . the only solution to implement algorithm _ dot - product _ inside a database is through the udf - uda extensions which process a single tuple at a time . the operands @xmath44 and @xmath45 are declared as table attributes having ` array ` type , while the dot - product is implemented as a udf . for reference , the existing in - database analytics alternatives treat the model as an in - memory uda state variable @xcite . this is impossible in big model analytics . the research problem we study in this paper is _ how to design a database operator for big model dot - product_. given a dataset consisting of sparse training examples ( set of vectors ` u ` = @xmath54 in figure [ fig : dp - operator ] ) and a dense model ( vector ` v ` in figure [ fig : dp - operator ] ) , this operator computes the set of dot - products between the examples and the model ( vector ` dp ` in figure [ fig : dp - operator ] ) optimally . an entry @xmath55 corresponds to the dot - product @xmath56 . following the notation in section [ ssec : problem : grad - descent ] , @xmath57 corresponds to @xmath43 and vector ` v ` corresponds to @xmath12 . since the model ` v ` can not fit in - memory , optimality is measured by the total number of secondary storage accesses . this is a good indicator for the execution time , given the simple computation required in dot - product . in addition to the main objective , we include a functional constraint in the design_results have to be generated in a non - blocking fashion_. as soon as the dot - product @xmath58 corresponding to a training example @xmath57 is computed , it is made available to the calling operator / application . this requirement is essential to support sgd the gradient descent solution used the most in practice . * relationship between big model dot - product and spmv . * if we ignore the non - blocking functional constraint , the result vector ` dp ` is the product of sparse matrix ` u ` and vector ` v ` , i.e. , @xmath59 . this is the standard spmv kernel which is notorious for sustaining low fractions of peak performance and for which there is a plethora of algorithms proposed in the literature . however , none of these algorithms considers the case when vector ` v ` does not fit in memory . they focus instead on the higher levels of the storage hierarchy , i.e. , memory cache @xcite and global memory texture memory in gpu @xcite , respectively . while some of the solutions can be extended to disk - based spmv , they are applicable only to bgd optimization which is less often used in practice . in sgd , ` v ` is updated after computing the dot - product with every vector @xmath57 , i.e. , row of matrix ` u ` . this renders all the spmv algorithms inapplicable to big model dot - product . we propose novel algorithms for this constrained version of spmv . since this is a special instance of spmv , the proposed solution is applicable to the general spmv problem . in this section , we present two database solutions to the big model dot - product problem . first , we introduce a pure relational solution that treats the operands ` u ` and ` v ` as standard relations . second , we give an ` array`-relation solution that represents each of the vectors @xmath57 in ` u ` as an ` array ` data type similar to existing in - database frameworks such as madlib and bismarck . the relational solution represents ` u ` and ` v ` as relations and uses standard relational algebra operators , e.g. , join , group - by , and aggregation and their sql equivalent to compute the dot - product . in order to represent vectors @xmath60 in relational format , we create a tuple for each non - zero dimension . the attributes of the tuple include the index and the actual value in the vector . moreover , since the components of a vector are represented as different tuples , we have to add another attribute identifying the vector , i.e. , _ tid_. the schema for the dense vector ` v ` is obtained following the same procedure . there is no need for a vector / tuple identifier _ tid _ , though , since ` v ` contains a single tuple , i.e. , the model : u(index integer , value numeric , tid integer ) v(index integer , value numeric ) based on this representation , relation ` u ` corresponding to figure [ fig : dp - operator ] contains 19 tuples . the tuples for @xmath61 are @xmath62 @xmath63 . relation ` v ` contains a tuple for each index . an alternative representation is a wide relation with an attribute corresponding to each dimension / index . however , tables with such a large number of attributes are not supported by any of the modern relational databases . the above representation is the standard procedure to map ordered vectors into non - ordered relations at the expense of data redundancy the tuple identifier for sparse vectors and the index for dense vectors . the dot - product vector ` dp ` is computed in the relational data representation with the following standard _ join group - by aggregate _ sql statement : select u.tid , sum(u.value*v.value ) from u , v where u.index=v.index group by u.tid this relational solution supports vectors of any size . however , it breaks the original vector representation since vectors are treated as tuples . moreover , the relational solution can not produce the result dot - product @xmath64 in a non - blocking manner due to the blocking nature of each operator in the query . this is not acceptable for sgd . in the ` array`-relation join solution , the sparse structure of ` u ` is preserved by representing each vector @xmath57 with two ` array ` attributes , one for the non - zero index and another for the value : u(index integer[],value numeric[],tid integer ) the vector identifier _ tid _ is still required . this representation is the database equivalent of the coordinate representation @xcite for sparse matrices and is possible only in database servers with type extension support , e.g. , postgresql . vector ` v ` is decomposed as in the relational solution . the sql query to compute dot - product ` dp ` in postgresql is : select u.tid , sum(u.value[idx(u.index , v.index)]*v.value ) from u , v where v.index = any(u.index ) group by u.tid there are several differences with respect to the relational solution . the equi - join predicate in ` where ` is replaced with an inclusion predicate ` v.index in u.index [ ] ` testing the occurrence of a scalar value in an array . this predicate can be evaluated efficiently with an index on ` v.index ` . the expression in the ` sum ` aggregate contains an indirection based on ` v.index ` which requires specialized support . while the size of the join is the same , the width of the tuples in ` array`-relation join is considerably larger the size of the intermediate join result table is the size of u multiplied by the average number of non - zero entries across the vectors @xmath57 since the entire ` array ` attributes are replicated for every result tuple . this is necessary because the arrays are used in the ` sum ` . finally , the ` array`-relation join remains blocking , thus , it is still not applicable to sgd . in this section , we present the dot - product join operator for big model dot - product computation . essentially , dot - product join combines the advantages of the existing database solutions . it is an in - database operator over an ` array`-relation data representation that computes the result vector without blocking and without generating massive intermediate results . in this section , we present the requirements , challenges , and design decisions behind the dot - product join operator , while a thorough evaluation is given in section [ sec : experiments ] . the dot - product join operator takes as input the operands ` u ` and ` v ` and generates the dot - product vector ` dp ` ( figure [ fig : dp - operator ] ) . ` u ` follows the ` array ` representation in which the index and value are stored as sparse ` array`-type attributes . only the non - zero entries are materialized . ` v ` is stored in range - based partitioned relational format the tuples ` ( index , value ) ` are partitioned into _ pages _ with fixed size and a page contains tuples with consecutive ` index ` values . the overall number of pages in ` v ` is @xmath65 . for example , with a page size of @xmath66 , vector ` v ` in figure [ fig : dp - operator ] is partitioned into @xmath67 pages . page @xmath68 contains indexes @xmath69 , page @xmath66 indexes @xmath70 , and page @xmath67 indexes @xmath71 . the memory budget available to the dot - product join operator for storing ` v ` is @xmath1 pages , which is smaller than @xmath65this is a fundamental constraint . without loss of generality , we assume that each vector @xmath72 accesses at most @xmath1 pages from ` v ` . this allows for atomic computation of entries in vector ` dp ` , i.e. , all the data required for the computation of the entry @xmath55 are memory - resident at a given time instant . moreover , each vector @xmath57 has to be accessed only once in order to compute its contribution to ` dp``u ` can be processed as a stream . as a result , the cost of big model dot - product computation is entirely determined by the number of secondary storage page accesses for ` v`the execution of algorithm _ dot - product _ is considerably faster than accessing a page from secondary storage . thus , the main challenge faced by the dot - product join operator is _ minimizing the number of secondary storage page accesses _ by judiciously using the memory budget @xmath1 for caching pages in ` v ` . a vector @xmath72 contains several non - zero entries . each of these entries requires a request to retrieve the value at the corresponding index in ` v ` . thus , the number of requests can become a significant bottleneck even when the requested index is memory - resident . _ reducing the number of requests _ for entries in ` v ` is a secondary challenge that has to be addressed by the dot - product join operator . in order to illustrate these challenges , we extend upon the example in figure [ fig : dp - operator ] . the total number of requests to entries in ` v ` is @xmath73@xmath67 for @xmath61 , @xmath66 for @xmath74 , and so on . the corresponding number of page accesses to ` v ` when we iterate over ` u ` is @xmath75@xmath66 pages are accessed by each vector @xmath57 . this is also the number of requests when we group together requests to the same page . with a memory budget @xmath76 and lru cache replacement policy , the number of pages accessed from secondary storage reduces to @xmath77 . however , the number of requests remains as before , i.e. , @xmath75 . algorithm _ dot - product join operator _ contains a high - level presentation of the proposed operator for big model dot - product . intuitively , we push the aggregation inside the ` array`-relation join and apply a series of optimizations that address the challenges identified above minimize the number of secondary storage accesses and the number of requests to entries in ` v ` . given ` u ` , ` v ` , and memory budget @xmath1 , _ dot - product join operator _ iterates over the pages of ` u ` and executes a three - stage procedure at _ page - level_. the three stages are : optimization , batch execution , and dot - product computation . optimization minimizes the number of secondary storage accesses . batch execution reduces the number of requests to ` v ` entries . dot - product computation is a direct call to _ dot - product _ that computes an entry of the _ dp _ result . notice that generating intermediate join results and grouping are not required anymore since the entries in ` v ` corresponding to a vector @xmath57 are produced in a contiguous batch . in the following , we give an overview of the optimization and batch execution stages . + u ( index integer[],value numeric[],tid integer ) + v ( index integer , value numeric ) + memory budget @xmath1 dp ( tid integer , product numeric ) reorder vectors @xmath57 to cluster similar vectors together[alg : dot - product : reorder ] group vectors @xmath57 into batches @xmath78 that access at most @xmath1 pages from v[alg : dot - product : batching ] collect pages @xmath79 accessed by vectors @xmath80 into a set @xmath81[alg : dot - product : batch-1 ] request access to pages in @xmath82[alg : dot - product : batch-2 ] [ alg : dot - product : batch-3 ] dp @xmath83 _ dot - product_(@xmath57,v ) append ( @xmath84 , dp ) to dp [ alg : dot - product : batch-4 ] the naive adoption of the ` array`-relation solution inside the dot - product join operator does not address the number of secondary storage accesses to pages in ` v ` . the database buffer manager is entirely responsible for handling secondary storage access . the standard policy to accomplish this is lru the least recently accessed page is evicted from memory . this can result in a significantly suboptimal behavior , as shown in figure [ fig : dp - flow ] . ` u ` is the same as in figure [ fig : dp - operator ] and @xmath76 . as discussed before , the number of secondary storage page accesses is @xmath77 . essentially , every vector @xmath72 , except @xmath85 , requires a page replacement . this is because consecutive vectors access a different set of two pages . the net result is significant _ buffer or cache thrashing_a page is read into memory only to be evicted at the subsequent request . one can argue that the access pattern in this example is the worst - case for lru . the big model dot - product problem , however , exhibits this pattern pervasively because each dot - product computation is atomic . allowing partial dot - product computation incurs a different set of obstacles and does not satisfy the non - blocking requirement . thus , it is not a viable alternative for our problem . @xmath86 in figure [ fig : dp - flow ] contains exactly the same vectors as ` u ` , reordered such that similar vectors vectors with non - zero entries co - located in the same partition / page of ` v ` are clustered together . the number of secondary storage accesses corresponding to @xmath86 is only @xmath87 . the reason for this considerable reduction is that a ` v ` page once read into memory is used by several vectors @xmath57 that have non - zero indexes resident to the page . essentially , vectors share access to the common pages . * reordering . * the main task of the optimization stage is to _ identify the optimal reordering of vectors @xmath72 that minimizes the number of secondary storage accesses to ` v`_line ( [ alg : dot - product : reorder ] ) in algorithm [ alg : dp ] . we prove that this task is np - hard by a reduction from the minimum hamiltonian path problem . we propose three heuristic algorithms that find good reorderings in a fraction of the time depending on the granularity at which the reordering is executed , e.g. , a page , several pages , or a portion of a page . the first algorithm is an lsh - based extension to nearest neighbor the well - known approximation to minimum hamiltonian path . the second algorithm is partition - level radix sort . the third algorithm is standard k - center clustering applied at partition - level . we discuss these algorithms in section [ sec : reordering ] . vector reordering is based on permuting a set of rows and columns of a sparse matrix in order to improve the performance of the spmv kernel . permuting is a common optimization technique used in sparse linear algebra . the most common implementation is the reverse cuthill - mckee algorithm ( rcm ) @xcite which is widely used for minimizing the maximum distance between the non - zeros and the diagonal of the matrix . since the rcm algorithm permutes both rows and columns , it incurs an unacceptable computational cost . to cope with this , vector reordering limits permuting only to the rows of the matrix . we show that even this simplified problem is np - hard and requires efficient heuristics . * batching . * the second task of the optimization stage is to _ reduce the number of page requests to the buffer manager_. even when the requested page is in the buffer , there is a non - negligible overhead in retrieving and passing the page to the dot - product join operator . a straightforward optimization is to group together requests made to indexes co - located in the same page . by applying this strategy to the example in figure [ fig : dp - flow ] , the number of requests in @xmath86 is reduced to @xmath75compared to @xmath73 in ` u ` . we take advantage of the reordering to group consecutive vectors into batches and make requests at batch - level line ( [ alg : dot - product : batching ] ) in algorithm [ alg : dp ] . this strategy is beneficial because similar vectors are grouped together by the reordering . for example , the number of page requests corresponding to @xmath88 in figure [ fig : dp - flow ] is only @xmath89@xmath66 for each batch . formally , we have to identify the batches that minimize the number of page requests to ` v ` given a fixed ordering of the vectors in ` u ` and a memory budget @xmath1 . without the ordering constraint , this is the standard bin packing problem known to be np - hard . due to ordering , a simple greedy heuristic that iterates over the vectors and adds them to the current batch until the capacity constraint is not satisfied is guaranteed to achieve the optimal solution . the output is a set of batches with a variable number of vectors . given the batches @xmath78 extracted in the optimization stage , we compute the dot - products @xmath56 by making a single request to the buffer manager for all the pages accessed in the batch . it is guaranteed that these pages fit in memory at the same time . notice , though , that vectors @xmath57 are still processed one - by - one , thus , dot - product results are generated individually ( lines ( [ alg : dot - product : batch-3])-([alg : dot - product : batch-4 ] ) in algorithm [ alg : dp ] ) . while batch execution reduces the number of requests to the buffer manager , the requests consist of a set of pages not a single page . it is important to emphasize that decomposing the set request into a series of independent single - page requests is sub - optimal . for example , consider ` u ` in figure [ fig : dp - flow ] to consist of @xmath77 batches of a single vector each . vector @xmath90 has a request for page 1 and 2 , respectively . page 2 and 3 are in memory . if we treat the 2-page request as two single - page requests , page 2 is evicted to release space for page 1 , only to be immediately read back into memory . if the request is considered as a unit , page 3 is evicted . in order to support set requests , the functionality of the buffer manager has to be enhanced . instead of directly applying the replacement policy , e.g. , lru , there are two stages in handling a page set request . first , the pages requested that are memory - resident have to be pinned down such that the replacement policy does not consider them . second , the remaining requested pages and the non - pinned buffered pages are passed to the standard replacement policy . _ dot - product join operator _ includes this functionality in lines ( [ alg : dot - product : batch-1])-([alg : dot - product : batch-2 ] ) , before the dot - product computation . in this section , we discuss strategies for reordering vectors @xmath72 that minimize the number of secondary storage accesses . first , we prove that finding the optimal reordering is np - hard . then , we introduce three scalable heuristics that cluster similar vectors . we formalize the reordering problem as follows . assume there are @xmath91 @xmath11-dimensional vectors @xmath54 . each vector @xmath57 contains @xmath92 page requests grouped into a set @xmath93 . given two consecutive vectors @xmath57 , @xmath94 and their corresponding page request sets @xmath95 , @xmath96 , the pages in the set difference @xmath97 potentially require secondary storage access . since we assume that all the pages in any set @xmath95 fit in memory , it is guaranteed that the pages in the set intersection @xmath98 are memory resident . pages in @xmath96 that are not in @xmath95 can be in memory if they have been accessed before and not evicted . let us denote the set difference between any two vectors @xmath57 and @xmath99 as @xmath100 . the cardinality of this set @xmath101 gives the number of potential secondary storage accesses . the goal of reordering is to identify the vector sequence @xmath102 that minimizes the cumulative cardinality @xmath103 of the set difference between all the pairs of consecutive vectors : @xmath104 [ thm : reordering - np - hard ] finding the vector sequence @xmath102 that minimizes the cumulative cardinality @xmath105 is np - hard . we provide a reduction of the minimum hamiltonian path in a weighted directed graph a known np - complete problem @xcite to the reordering problem . given a weighted directed graph the edges are directional and have a weight attached the minimum hamiltonian path traverses all the vertexes once and the resulting path has minimum weight . we define a complete directed graph @xmath106 with @xmath91 vertexes and @xmath107 edges . a vertex corresponds to each vector @xmath57 . for any pair of vertexes , i.e. , vectors @xmath57 , @xmath99 , we introduce two edges @xmath108 , @xmath109 and @xmath110 , @xmath111 , respectively . the weight of an edge @xmath108 , @xmath109 is given by the cardinality @xmath112 . notice that set difference is not commutative . this is the reason for having directed edges . a hamiltonian path in graph @xmath113 corresponds to a reordering because all the vectors are considered only once and all the orders are possible there is an edge between any two vertexes . since the weight of an edge is defined as the cardinality @xmath101 , the weight of the minimum hamiltonian path corresponds to the minimum cumulative cardinality in eq . ( [ eq : optim - order ] ) . several polynomial - time heuristic algorithms for computing an approximation to the minimum hamiltonian path over a complete directed graph exist in the literature @xcite . nearest neighbor is a standard greedy algorithm that chooses the closest non - visited vertex as the next vertex . it has computational complexity of @xmath114 and non - constant approximation factor @xmath115 . the straightforward application of the nearest neighbor algorithm to our problem is too expensive because we have to materialize a complete directed graph . this takes significant time even for values of @xmath91 in the order of thousands not to mention the required space . if we generate the graph on - the - fly , the weight to all the non - visited vertexes has to be computed . while this may seem more efficient , this still has an overall computational complexity of @xmath116 . due to these limitations of the nearest neighbor algorithm for complete graphs , we explore more efficient heuristics that avoid considering all the possible pairs of vertexes . locality - sensitive hashing ( lsh ) @xcite is an efficient method to identify similar objects represented as high - dimensional sparse vectors . similarity between vectors is defined as their jaccard coefficient , i.e. , @xmath117 . the main idea is to build a hash index that groups similar vectors in the same bucket . given an input vector , the most similar vector is found by identifying the vector with the maximum jaccard coefficient between the vectors co - located in the same hash bucket . essentially , the search space is pruned from all the vectors in the set to the vectors in the hash bucket typically considerably fewer . the complexity of lsh consists in finding a hash function that preserves the jaccard coefficient in mapping vectors to buckets the probability of having two vectors in the same bucket approximates their jaccard coefficient . minwise hash functions @xcite satisfy this property on expectation . given a set , a minwise hash function generates any permutation of its elements with equal probability . while such functions are hard to define , they can be approximated with a universal hash function @xcite that has a very large domain , e.g. , @xmath118 . in order to increase the accuracy of correctly estimating the jaccard coefficient , @xmath7 such minwise hash functions are applied to a vector . their output corresponds to the signature of the vector which gives the bucket where the vector is hashed to . the value of @xmath7 is an important parameter controlling the number of vectors which are exhaustively compared to the input query vector . the larger @xmath7 is , the fewer vectors end - up in the same bucket . on the opposite , if @xmath119 , all the vectors that share a common value can be hashed to the same bucket . given a value for @xmath7 , banding is a method that controls the degree of tolerated similarity . the @xmath7-dimensional signature is divided into @xmath120 @xmath121-dimensional bands and a hash table is built independently for each of them . the input vector is compared with all the co - located vectors of at least one hash table . banding decreases the jaccard coefficient threshold acceptable for similarity , while increasing the probability that all the vectors that have a higher coefficient than the threshold are found . + set of vectors @xmath54 with page requests + @xmath7 minwise hash functions grouped into @xmath120 bands reordered set of input vectors @xmath102 compute @xmath7-dimensional signature @xmath122 based on minwise hash functions and group into @xmath120 bands @xmath123 insert vector @xmath57 into hash table @xmath124 , @xmath125 , using minwise hash function of @xmath126 initialize @xmath127 with a random vector @xmath57 let @xmath128 be the set of vectors co - located in the same bucket with @xmath129 in hash table @xmath124 and not selected @xmath130 let @xmath131 be the vector in @xmath132 with the minimum set difference cardinality @xmath103 to the current vector @xmath129 _ compute lsh tables _ section in algorithm [ alg : lsh ] summarizes the construction of the lsh index for the vector reordering problem . although we do not measure similarity using the jaccard coefficient , there is a strong correlation between set difference cardinality and the jaccard coefficient . intuitively , the higher the jaccard coefficient , the larger the intersection between two sets relative to their union . this translates into small set difference cardinality . since the goal of reordering is to cluster vectors with small differences , lsh places them into the same bucket with high probability . we compute the output vector reordering by executing the nearest neighbor heuristic over the lsh index ( section _ lsh - based nearest neighbor search _ in algorithm [ alg : lsh ] ) . we believe this is a novel application of the lsh technique , typically used for point queries . the algorithm starts with a random vector . the next vector is selected from the vectors co - located in the same bucket across at least one other band of the lsh index . the process is repeated until all the vectors are selected . bands play a very important role in reordering because they allow for smooth bucket transition . this is not possible with a single band since there is no strict ordering between buckets . in this situation , choosing the next bucket involves inspecting all the other buckets of the hash table . in general , lsh reduces significantly the number of vector pairs for which the exact set difference has to be computed only @xmath133 vector pairs are considered , i.e. , a constant number for each vector . thus , the overall complexity of _ lsh reordering _ @xmath134 is dominated by the construction of the lsh index . we illustrate how lsh reordering works for the set of vectors @xmath135 depicted in figure [ fig : dp - flow ] . to facilitate understanding , we set @xmath136 and @xmath137 , i.e. , two lsh indexes with 1-d signatures . since @xmath138 , there are many conflicts in each bucket of the two bands . even though this does not reduce dramatically the number of vector pairs that require full comparison , it shows how bucket transition works . let the two minwise hash functions generate the following permutations : @xmath139 and @xmath140 , respectively . remember that the reordering is done at page level . the lsh index for the first band has two buckets , for key @xmath141 and key @xmath142 . the lsh index for the second band also has two buckets , for key @xmath143 and key @xmath144 . let @xmath61 be the random vector we start the nearest neighbor search from . the vectors considered at the first step of the algorithm are @xmath145 . all of them are contained in bucket with key @xmath66 of the first band . since @xmath90 and @xmath146 have set difference 0 to @xmath61 , one of them is selected as @xmath147 and the other as @xmath148 . at this moment , the bucket with key @xmath68 from the second band is exhausted and one of @xmath74 , @xmath149 , and @xmath150 is selected . independent of which one is selected , bucket transition occurs since the new vectors @xmath151 are co - located in bucket with key @xmath67 of the second band . by following the algorithm to termination , @xmath152 in figure [ fig : dp - flow ] is a possible solution . sorting is a natural approach to generate a reordering of a set as long as a strict order relationship can be defined between any two elements of the set . the jaccard coefficient gives an order only with respect to a fixed reference , i.e. , given a reference vector we can quantify which of any other two vectors is smaller based on their jaccard coefficient with the reference . however , this is not sufficient to generate a complete reordering . it is possible to imagine a multitude of ordering strategies for a set of sparse @xmath11-dimensional vectors . the simplest solution is to consider the dimensions in some arbitrary order , e.g. , from left to right . vector @xmath61 is smaller than @xmath74 , i.e. , @xmath153 , for @xmath135 in figure [ fig : dp - flow ] since it has a non - zero entry at index @xmath68 , while the first non - zero entry in @xmath74 is at index @xmath67 . in order to cluster similar vectors together , a better ordering is required . our strategy is to sort the dimensions according to the frequency at which they appear in the set of sparse vectors and compare vectors dimension - wise based on this order . this is exactly how radix sort @xcite works , albeit without reordering the dimensions in descending order of their frequency . a similar idea is proposed for spmv on gpu in @xcite . radix sort reordering _ depicts the entire process . the two stages frequency computation and radix sort are clearly delimited . since their complexity is @xmath154 , the overall is @xmath154 . set of vectors @xmath54 with page requests reordered set of input vectors @xmath102 compute page request frequency across vectors @xmath54 represent @xmath57 by a bitset of 0 s and 1 s where a 1 at index @xmath33 corresponds to the vector requesting page @xmath33 reorder the bitset in decreasing order of the page request frequency , i.e. , index 1 corresponds to the most frequent page apply radix sort to the set of bitsets let @xmath129 be the vector corresponding to the bitset at position @xmath155 in the sorted order we illustrate how radix sort reordering works for the set of vectors @xmath135 in figure [ fig : dp - flow ] . the algorithm operates at page level . since page @xmath66 is the most frequent it appears in @xmath89 vectors it is the first considered . two partitions are generated : @xmath156 accesses page @xmath66 ; and @xmath157 does not . this is exactly @xmath152 in figure [ fig : dp - flow ] . the frequency of page @xmath68 and page @xmath67 is @xmath158 , thus , any of them can be selected in the second iteration . assume that we brake the ties using the original index and we select page @xmath68 . each of the previous two partitions is further split into two . for example , @xmath159 and @xmath160 . the important thing to remark is that vectors accessing page @xmath68 split in the first iteration can not be grouped together anymore . if we follow the algorithm to completion , @xmath152 in figure [ fig : dp - flow ] is generated . the intuition behind radix sort reordering is that by considering pages in decreasing order of their frequency the most requested page is accessed from secondary storage exactly once ; the second most accessed page at most twice ; and so on . this holds true because all the pages accessed by a vector fit together in memory . it is important to notice that although the number of accesses increases the request frequency decreases for pages considered at later iterations . this guarantees that the maximum number of accesses to a page is bounded by @xmath161 , where _ rank _ is the rank of the page and _ freq _ is the access frequency . essentially , radix sort reordering is a dimension - wise greedy heuristic . since the goal of reordering is to cluster similar vectors together , we include a standard k - center clustering @xcite algorithm as a reference . the main challenge is that similar to the jaccard coefficient clustering does not impose a strict ordering between two vectors only with respect to the common center . this is also true for centers . the stopping criterion for our hierarchical k - center clustering is when all the clusters have page requests that fit entirely in memory . as long as this does not hold , we invoke the algorithm recursively for the clusters that do not satisfy this requirement . the resulting clusters are ordered as follows . we start with a random cluster and select as the next cluster the one with the center having the minimum set difference cardinality . notice that reordering the vectors inside a cluster is not necessary since they all fit in memory . this procedure is depicted in algorithm [ alg : k - center ] . it has complexity @xmath162 , where @xmath33 is the resulting number of clusters , i.e. , centers . set of vectors @xmath54 with page requests reordered set of input vectors @xmath102 initialize first set of @xmath33 centers with random vectors @xmath57 assign each vector to the center having the minimum set difference cardinality let @xmath163 be the set of vectors assigned to center @xmath164 , @xmath165 call _ k - center reordering _ recursively for the sets @xmath163 with requests that do not fit in memory reorder centers and their corresponding vectors we propose three heuristic algorithms for the vector reordering problem . lsh and k - center cluster similar vectors together . lsh uses the jaccard coefficient to hash similar vectors to the same bucket of a hash table and then executes nearest neighbor search starting from a random vector . k - center clustering partitions the vectors recursively based on an increasing set of centers . both methods are limited by partial ordering between two vectors and incur overhead to define a strict ordering . moreover , they are randomized algorithms sensitive to the initialization and a handful of parameters . radix sort imposes a strict ordering at the expense of not considering the entire vector in clustering . we alleviate this problem by sorting the dimensions based on their access frequency . this bounds the total number of secondary storage accesses . in the experimental evaluation ( section [ sec : experiments ] ) , we compare these algorithms thoroughly in terms of execution time and reordering quality . in this section , we show how to integrate the dot - product join operator in gradient descent optimization for big model analytics . we discuss the benefits of the operator approach compared to the relational and ` array`-relation solutions presented in section [ sec : baseline ] . figure [ fig : sgd - integration ] depicts the gradient computation required in gradient descent optimization . vector dot - product is only a subroutine in this process . as we move from the relational solution to the proposed dot - product join operator , the query plan becomes considerably simpler . the relational solution consists of two parts . in the first part , the dot - product corresponding to a vector @xmath57 is computed . since vector components are represented as independent tuples with a common identifier , this requires a group - by on _ tid_. however , this results in the loss of the vector components , required for gradient computation ( see eq . ( [ eq : lr - gradient ] ) ) . thus , a second join group - by is necessary in order to compute each component of the gradient . + + + as we show in the experimental evaluation , this turns out to be very inefficient . the ` array`-relation solution is able to discard the second join group - by because it groups vector components as an array attribute of a tuple . the proposed dot - product join operator goes one step further and pushes the group - by aggregation inside the join . while this idea has been introduced in @xcite for bgd over normalized example data , the dot - product join operator considers joins between examples and the model and works for bgd and sgd alike sgd requires only an additional selection . in @xcite , the model @xmath166 is small enough to fit in the state of the uda . the main benefit of pushing the group - by aggregation inside the join is that the temporary join result is not materialized in memory or on secondary storage . the savings in storage can be several orders of magnitude , e.g. , with an average of @xmath167 non - zero indexes per vector @xmath57 , the temporary storage if materialized is 3 orders of magnitude the size of @xmath135 . by discarding the blocking group - by on _ tid _ , the overall gradient computation becomes non - blocking since dot - product join is non - blocking . * sgd considerations . * in order to achieve faster convergence , sgd requires random example traversals . since dot - product join reorders the examples in order to cluster similar examples together , we expect this to have a negative effect on convergence . however , the reordering in dot - product join is only local at page - level . thus , a simple strategy to eliminate the effect of reordering completely is to estimate the gradient at page - level the number of steps in an iteration is equal to the number of pages . any intermediate scheme that trades - off convergence speed with secondary storage accesses can be imagined . to maximize convergence , the data traversal orders across iterations have to be also random . this is easily achieved with lsh and k - center reordering two randomized algorithms . a simple solution for radix is to randomize the order of pages across iterations . we provide experimental evidence that quantifies the impact of radix reordering the strictest solution on convergence . figure [ fig : sgd - convergence ] depicts the behavior of the loss function when 10 sgd iterations are executed for lr and lmf models over two real datasets ( see section [ sec : experiments ] for details ) . radix reordering does not degrade the convergence speed compared to a random data traversal the model is updated after each example . moreover , for the lr model , radix reordering actually improves convergence . these results are in line with those presented in @xcite , where only complete data sorting on the label has a negative impact on convergence . in the section , we first evaluate the effectiveness and efficiency of the three reordering heuristics . then , we apply the reordering heuristics to dot - product join and measure the effect of reordering and batching under different resource constraints . finally , we measure the end - to - end dot - product and gradient descent execution time as a function of the amount of memory available in the system . we also compare dot - product join with the baseline alternatives introduced in section [ sec : baseline ] across several synthetic and real datasets . specifically , the experiments we design are targeted to answer the following questions : how effective are the reordering heuristics and which one should be used under what circumstance ? how do reordering and batching affect the runtime and what is their overhead ? what is the sensitivity of the dot - product join operator with respect to the available memory ? how does dot - product join compare with other solutions ? what is the contribution of dot - product join within the overall big model gradient descent optimization ? * implementation . * we implement dot - product join as a new array - relation operator in glade @xcite a state - of - the - art parallel data processing system that executes analytics tasks expressed with the uda interface . glade has native support for the ` array ` data type . dot - product join is implemented as an optimized index join operator . it iterates over the pages in ` u ` . for each page , it applies the reordering and batching optimizations and then probes the entries in ` v ` at batch granularity . the dot - product corresponding to a vector is generated in a single pass over the vector and pipelined into the gradient uda . * we execute the experiments on a standard server running ubuntu 14.04 smp @xmath168-bit with linux kernel 3.13.0 - 43 . the server has 2 amd opteron 6128 series 8-core processors 16 cores 28 gb of memory , and 1 tb 7200 rpm sas hard - drive . each processor has 12 mb l3 cache , while each core has 128 kb l1 and 512 kb l2 local caches . the average disk bandwidth is 120 mb / s . * methodology . * we perform all experiments at least 3 times and report the average value as the result . in the case of page - level results , we execute the experiments over the entire dataset all the pages and report the average value computed across the pages . we always enforce data to be read from disk in the first iteration by cleaning the file system buffers before execution . memory constraints are enforced by limiting the batch size , i.e. , the number of vectors @xmath57 that can be grouped together after reordering . .datasets used in the experiments . [ cols="<,>,>,>,>",options="header " , ] our experiments identify radix sort as the fastest reordering heuristic and the only one scalable to large batches . the improvement over tuple - at - a - time processing with basic lru replacement is larger in this case . while lsh achieves the largest reduction in page misses , radix is not far behind . the reduction in dot - product computation execution time is as much as an order of magnitude when reordering and batching are combined . independently and when integrated in sgd , the dot - product join operator s performance degrades gracefully when the memory budget reduces below a threshold at which the model can not be buffered in memory . out of all the alternatives , the dot - product join operator is the only solution that can handle big models in a scalable fashion . the relational implementation in glade is in general an order of magnitude or more slower than dot - product join , while the relational and ` array`-relation postgresql versions and scidb do not finish execution even after @xmath169 hours except for small models . monetdb achieves efficient execution time for small and intermediate model sizes with a massive memory budget . however , it fails miserably on truly large models the problem addressed by dot - product join . * in - database analytics . * there has been a sustained effort to add analytics functionality to traditional database servers over the past years . madlib @xcite is a library of analytics algorithms built on top of postgresql . it includes gradient descent optimization functions for training generalized linear models such as lr and lmf @xcite . this is realized through the udf - uda extensions existent in postgresql . the model is represented as the state of the uda . it is memory - resident during an iteration and materialized as an array attribute across iterations . this is not possible for big model analytics because the model can not fit in memory and postgresql has strict limitations on the maximum size of an attribute . glade @xcite follows a similar approach . distributed learning frameworks such as mllib @xcite and vowpal wabbit @xcite represent the model as a program variable and allow the user to fully manage its materialization . thus , they can not handle big models directly . since the vectors are memory - resident , the dot - product is computed by a simple invocation of _ algorithm dot - product_. in the case of madlib , this is done inside the uda code . computing dot - product and other linear algebra operations as udas is shown to be more efficient than the relational and stored procedure solutions for low - dimensional vectors in @xcite . slacid @xcite is a solution to implement sparse matrices and linear algebra operations inside a column - oriented in - memory database in which the emphasis is on the storage format . adding support for efficient matrix operations in distributed frameworks such as hadoop @xcite and spark @xcite has been also investigated recently we consider a more general problem dot - product is a sub - part of matrix multiplication in a centralized environment . moreover , gradient descent optimization does not involve matrix operations only dot - product . * big model parallelism . * parameter server @xcite is the first system that addresses the big model analytics problem . their approach is to partition the model across the distributed memory of multiple _ parameter servers_in charge of managing the model . the training vectors @xmath135 are themselves partitioned over multiple workers . a worker iterates over its subset of vectors and for each non - zero entry makes a request to the corresponding parameter server to retrieve the model entry . once all the necessary model entries are received , the gradient is computed , the model is updated and then pushed back to the parameter servers . instead of partitioning the model across machines , we use secondary storage . minimizing the number of secondary storage accesses is the equivalent of minimizing network traffic in parameter server . thus , the optimizations we propose reordering and batching are applicable in a distributed memory environment . they are not part of parameter server . in strads @xcite , parameter servers are driving the computation , not the workers . the servers select the subset of the model to be updated in an iteration by each worker . in our case , this corresponds to ignoring some of the non - zero entries in a vector @xmath57 to further reduce i / o . we plan to explore strategies to select the optimal parameter subset in the future . * learning over normalized data . * the integration of relational join with gradient , i.e. , dot - product , computation has been studied in @xcite . however , the assumption made in all these papers is that the vectors @xmath57 are vertically partitioned along their dimensions . a join is required to put them together before computing the dot - product . in @xcite , the dot - product computation is pushed inside the join and only applicable to bgd . the dot - product join operator adopts the same idea . however , this operator has to compute the dot - product which is still evaluated inside a uda in @xcite . the join is dropped altogether in @xcite when similar convergence is obtained without considering the dimensions in an entire vertical partition . a solution particular to linear regression is shown to be efficient to compute when joining factorized tables in @xcite . in all these solutions , the model is small enough to fit entirely in memory . moreover , they work exclusively for bgd . * joins . * dot - product join is a novel type of join operator between an array attribute and a relation . we are not aware of any other database operator with the same functionality . from a relational perspective , dot - product join is most similar to index join @xcite . however , for every vector @xmath57 , we have to probe the index on model @xmath166 many times . thus , the number of probes can be several orders of magnitude the size of @xmath135 . the proposed techniques are specifically targeted at this scenario . the batched key access join in mysql is identical to our batching optimization applied at vector level . however , it handles a single probe per tuple and its reordering is aimed at generating sequential storage access for @xmath166 . array joins @xcite are a new class of join operators for array databases . while it is possible to view dot - product join as an array join operator , the main difference is that we consider a relational system and push the aggregation inside the join . this avoids the materialization of the intermediate join result . * spmv kernel . * as we have already discussed in the paper , the dot - product join operator is a constrained formulation of the standard sparse matrix vector ( spmv ) multiplication problem . specifically , the constraint imposes the update of the vector after each multiplication with a row in the sparse matrix . this makes impossible the direct application of spmv kernels to big model analytics beyond bgd . moreover , we consider the case when the vector size goes beyond the available memory . spmv is an exhaustively studied problem with applications to high performance computing , graph algorithms , and analytics . an extended discussion of the recent work on spmv is presented in @xcite on which we draw in our discussion . @xcite and @xcite propose optimizations for spmv in multicore architectures , while @xcite and @xcite optimize distributed spmv for large scale - free graphs with 2d partitioning to reduce communication between machines . array databases such as scidb @xcite and rasdaman @xcite support spmv as calls to optimized linear algebra libraries such as intel mkl and trilinos . there has also been preliminary research on accelerating matrix multiplication with gpus @xcite and ssds @xcite , showing that speedups are limited by i / o and setup costs . none of these works store the vector in secondary storage . in this paper , we propose a database - centric solution to handle big models , where the model is offloaded to secondary storage . we design and implement the first dot - product join physical database operator that is optimized to execute secondary storage array - relation dot - products effectively . we prove that identifying the optimal access schedule for this operator is np - hard . we propose dynamic batching and reordering techniques to minimize the overall number of secondary storage accesses . we design three reordering heuristics lsh , radix , and k - center and evaluate them in terms of execution time and reordering quality . we experimentally identify radix as the most scalable heuristic with significant improvement over basic lru . the dot - product join operator s performance degrades gracefully when the memory budget reduces . out of all the alternatives , the dot - product join operator is the only solution that can handle big models in a scalable fashion . alternative solutions are in general an order of magnitude or more slower than dot - product join . while the focus of the paper is on vector dot - product in the context of gradient descent optimization , the proposed solution is general enough to be applied to any spmv kernel operating over highly - dimensional vectors . essentially , the dot - product join operator is the first spmv kernel that accesses the vector from secondary storage . previous solutions have at most considered accessing the sparse matrix out - of - memory . from a database perspective , dot - product join is the first join operator between an array and a relation . existing joins between an array attribute and a relation are tremendously ineffective . the experimental results in the paper confirm this . thus , we find dot - product join extremely relevant in the context of polystores where multiple storage representations are combined inside a single system . dot - product join eliminates the conversion between representations necessary to execute hybrid operations . in future work , we plan to include data parallelism in the dot - product join operator by partitioning the array over multiple threads that update the model concurrently . while lack of parallelism may be seen as a limitation of the dot - product join operator especially in a big data setting we argue that addressing the sequential problem first is a mandatory requirement in order to design a truly scalable parallel solution . we also plan to investigate partition strategies from spmv kernels to improve the disk access pattern to the vector further . since ssd storage has improved dramatically over the last years , storing the matrix and the vector on ssds is another topic for future research . finally , we plan to expand the range of models covered by dot - product join beyond lr and lmf . as long as a dot - product operation is required between highly - dimensional vectors , the dot - product join operator is a solution to consider .

we propose a new scheme for selecting pool states for the embedded hidden markov model ( hmm ) markov chain monte carlo ( mcmc ) method . this new scheme allows the embedded hmm method to be used for efficient sampling in state space models where the state can be high - dimensional . previously , embedded hmm methods were only applied to models with a one - dimensional state space . we demonstrate that using our proposed pool state selection scheme , an embedded hmm sampler can have similar performance to a well - tuned sampler that uses a combination of particle gibbs with backward sampling ( pgbs ) and metropolis updates . the scaling to higher dimensions is made possible by selecting pool states locally near the current value of the state sequence . the proposed pool state selection scheme also allows each iteration of the embedded hmm sampler to take time linear in the number of the pool states , as opposed to quadratic as in the original embedded hmm sampler . we also consider a model with a multimodal posterior , and show how a technique we term `` mirroring '' can be used to efficiently move between the modes .

1602.06030

consider a non - linear , non - gaussian state space model for an observed sequence @xmath0 . this model , with parameters @xmath1 , assumes that the @xmath2 are drawn from an observation density @xmath3 , where @xmath4 is an unobserved markov process with initial density @xmath5 and transition density @xmath6 . here , the @xmath7 might be either continuous or discrete . we may be interested in inferring both the realized values of the markov process @xmath8 and the model parameters @xmath1 . in a bayesian approach to this problem , this can be done by drawing a sample of values for @xmath9 and @xmath1 using a markov chain that alternately samples from the conditional posterior distributions @xmath10 and @xmath11 . in this paper , we will only consider inference for @xmath9 by sampling from @xmath10 , taking the parameters @xmath1 to be known . as a result , we will omit @xmath1 in model densities for the rest of the paper . except for linear gaussian models and models with a finite state space , this sampling problem has no exact solution and hence approximate methods such as mcmc must be used . one method for sampling state sequences in non - linear , non - gaussian state space models is the embedded hmm method ( neal , 2003 ; neal , beal and roweis , 2004 ) . an embedded hmm update proceeds as follows . first , at each time @xmath12 , a set of @xmath13 `` pool states '' in the latent space is constructed . in this set , @xmath14 of the pool states are drawn from a chosen pool state density and one is the current value of @xmath7 . this step can be thought of as temporarily reducing the state space model to an hmm with a finite set of @xmath13 states , hence the name of the method . then , using efficient forward - backward computations , which take time proportional to @xmath15 , a new sequence @xmath16 is selected from the `` ensemble '' of @xmath17 sequences passing through the set of pool states , with the probability of choosing each sequence proportional to its posterior density divided by the probability of the sequence under the pool state density . at the next iteration of the sampler , a new set of pool states is constructed , so that the chain can sample all possible @xmath7 , even when the set of possible values is infinite . another method is the particle gibbs with backward sampling ( pgbs ) method . the particle gibbs ( pg ) method was first introduced in andrieu , doucet and holenstein ( 2010 ) ; whiteley suggested the backward sampling modification in the discussion following this paper . lindsten and schon ( 2012 ) implemented backward sampling and showed that it improves the efficiency of pg . starting with a current sequence @xmath9 , pgbs first uses conditional sequential monte carlo ( smc ) to construct a set of candidate sequences and then uses backward sampling to select a new sequence from the set of candidate ones . here , conditional smc works in the same way as ordinary smc when generating a set of particles , except that one of the particles at time @xmath12 is always set to the current @xmath7 , similar to what is done in the embedded hmm method , which allows the sampler to remain at @xmath7 if @xmath7 lies in a high - density region . while this method works well for problems with low - dimensional state spaces , the reliance of the smc procedure on choosing an appropriate importance density can make it challenging to make the method work in high dimensions . an important advantage of particle gibbs , however , is that each iteration takes time that is only linear in the number of particles . both the pgbs and embedded hmm methods can facilitate sampling of a latent state sequence , @xmath9 , when there are strong temporal dependencies amongst the @xmath7 . in this case , using a method that samples @xmath7 conditional on fixed values of @xmath18 and @xmath19 can be an inefficient way of producing a sample from @xmath20 , because the conditional density of @xmath7 given @xmath18 and @xmath19 can be highly concentrated relative to the marginal density of @xmath7 . in contrast , with the embedded hmm and pgbs methods it is possible to make changes to blocks of @xmath7 s at once . this allows larger changes to the state in each iteration of the sampler , making updates more efficient . however , good performance of the embedded hmm and pgbs methods relies on appropriately choosing the set of pool states or particles at each time @xmath12 . in this paper , our focus will be on techniques for choosing pool states for the embedded hmm method . when the latent state space is one - dimensional , embedded hmms work well when choosing pool states in a variety of ways . for example , in shestopaloff and neal ( 2013 ) , we choose pool states at each time @xmath12 by constructing a `` pseudo - posterior '' for each latent variable by taking the product of a `` pseudo - prior '' and the observation density , the latter treated as a `` pseudo - likelihood '' for the latent variable . in shestopaloff and neal ( 2014 ) , we choose pool states at each time @xmath12 by sampling from the marginal prior density of the latent process . ways of choosing pool states that work well in one dimension begin to exhibit problems when applied to models with higher - dimensional state spaces . this is true even for dimensions as small as three . since these schemes are global , designed to produce sets of pool states without reference to the current point , as the dimension of the latent space grows , a higher proportion of the sequences in the ensemble ends up having low posterior density . ensuring that performance does nt degrade in higher dimensions thus requires a significant increase in the number of pool states . as a result , computation time may grow so large that any advantage that comes from using embedded hmms is eliminated . one advantage of the embedded hmm method over pgbs is that the embedded hmm construction allows placing pool states locally near the current value of @xmath7 , potentially allowing the method to scale better with the dimensionality of the state space . switching to such a local scheme fixes the problem to some extent . however , local pool state schemes come with their own problems , such as making it difficult to handle models with multiple posterior modes that are well - separated the pool states might end up being placed near only some of the modes . in this paper , we propose an embedded hmm sampler suitable for models where the state space is high dimensional . this sampler uses a sequential approximation to the density @xmath21 or to the density @xmath22 as the pool state density . we show that by using this pool state density , together with an efficient mcmc scheme for sampling from it , we can reduce the cost per iteration of the embedded hmm sampler to be proportional to @xmath23 , as with pgbs . at the same time , we retain the ability to generate pool states locally , allowing better scaling for high - dimensional state spaces . our proposed scheme can thus be thought of as combining the best features of the pgbs and the embedded hmm methods , while overcoming the deficiencies of both . we use two sample state space models as examples . both have gaussian latent processes and poisson observations , with one model having a unimodal posterior and the second a multimodal one . for the multimodal example , we introduce a `` mirroring '' technique that allows efficient movement between the different posterior modes . for these models , we show how our proposed embedded hmm method compares to a simple metropolis sampler , a pgbs sampler , as well as a sampler that combines pgbs and simple metropolis updates . further details on ensemble methods are available in the phd thesis of shestopaloff ( 2016 ) . we review the embedded hmm method ( neal , 2003 ; neal , beal and roweis , 2004 ) here . we take the model parameters , @xmath1 , to be fixed , so we do not write them explicitly . let @xmath24 be the density from which the state at time @xmath25 is drawn , let @xmath26 be the transition density between states at times @xmath12 and @xmath27 , and let @xmath28 be the density of the observation @xmath29 given @xmath7 . suppose our current sequence is @xmath8 . the embedded hmm sampler updates @xmath9 to @xmath16 as follows . first , at each time @xmath30 , we generate a set of @xmath13 pool states , denoted by @xmath31 } , \ldots , x_{i}^{[l]}\}$ ] . the pool states are sampled independently across the different times @xmath12 . we choose @xmath32 uniformly at random and set @xmath33}$ ] to @xmath7 . we sample the remaining @xmath14 pool states @xmath34 } , \ldots , x_{i}^{[l_{i}-1 ] } , x_{i}^{[l_{i}+1 ] } , \ldots , x_{n}^{[l]}$ ] using a markov chain that leaves a pool density @xmath35 invariant , as follows . let @xmath36 be the transitions of this markov chain with @xmath37 the transitions for this markov chain reversed ( i.e. @xmath38 ) , so that @xmath39 for all @xmath9 and @xmath16 . then , starting at @xmath40 , use reverse transitions @xmath41}|x_{i}^{[j+1]})$ ] to generate @xmath42 } , \ldots , x_{i}^{[1]}$ ] and starting at @xmath43 use forward transitions @xmath44}|x_{i}^{[j-1]})$ ] to generate @xmath45 } , \ldots , x_{n}^{[l]}$ ] . at each @xmath30 , we then compute the forward probabilities @xmath46 , with @xmath9 taking values in @xmath47 . at time @xmath48 , we have @xmath49 and at times @xmath50 , we have @xmath51})\alpha_{i-1}(x_{i-1}^{[l ] } ) \label{eq : alphaall}\end{aligned}\ ] ] finally , we sample a new state sequence @xmath16 using a stochastic backwards pass . this is done by selecting @xmath52 amongst the set , @xmath53 , of pool states at time @xmath54 , with probabilities proportional to @xmath55 , and then going backwards , sampling @xmath56 from the set @xmath57 , with probabilities proportional to @xmath58 . note that only the relative values of the @xmath46 will be required , so the @xmath59 may be computed up to some constant factor . alternatively , given a set of pool states , embedded hmm updates can be done by first computing the backward probabilities . we will see later on that the backward probability formulation of the embedded hmm method allows us to introduce a variation of our proposed pool state selection scheme . setting @xmath60 for all @xmath61 , we compute for @xmath62 @xmath63})p(x_{i+1}^{[l]}|x)\beta_{i+1}(x_{i+1}^{[l ] } ) \label{eq : beta}\end{aligned}\ ] ] a new state sequence is then sampled using a stochastic forward pass , setting @xmath64 to one of the @xmath9 in the pool @xmath65 with probabilities proportional to @xmath66 and then choosing subsequent states @xmath67 from the pools @xmath47 with probabilities proportional to @xmath68 . computing the @xmath59 or @xmath69 at each time @xmath70 takes time proportional to @xmath71 , since for each of the @xmath13 pool states it takes time proportional to @xmath13 to compute the sums in ( [ eq : alphaall ] ) or ( [ eq : beta ] ) . hence each iteration of the embedded hmm sampler takes time proportional to @xmath15 . we review the particle gibbs with backward sampling ( pgbs ) sampler here . for full details , see the articles by andrieu , doucet and holenstein ( 2010 ) and lindsten and schon ( 2012 ) . let @xmath72 be the importance density from which we sample particles at time @xmath25 , and let @xmath73 be the importance density for sampling particles at times @xmath70 . these may depend on the current value of the parameters , @xmath1 , which we suppressed in this notation . suppose we start with a current sequence @xmath9 . we set the first particle @xmath74}$ ] to the current state @xmath75 . we then sample @xmath14 particles @xmath76 } , \ldots , x_{1}^{[l]}$ ] from @xmath77 and compute and normalize the weights of the particles : @xmath78 } & = & \frac{p(x_{1}^{[l]})p(y_{1}|x_{1}^{[l]})}{q_{1}(x_{1}^{[l]}|y_{1 } ) } \\ w_{1}^{[l ] } & = & \frac{w_{1}^{[l]}}{\sum_{m=1}^{l}w_{1}^{[m]}}\end{aligned}\ ] ] for @xmath79 . for @xmath70 , we proceed sequentially . we first set @xmath34 } = x_{i}$ ] . we then sample a set of @xmath14 ancestor indices for particles at time @xmath12 , defined by @xmath80 } \in \{1 , \ldots , l\}$ ] , for @xmath81 , with probabilities proportional to @xmath82}$ ] . the ancestor index for the first state , @xmath83}$ ] , is @xmath25 . we then sample each of the @xmath14 particles , @xmath84}$ ] , at time @xmath12 , for @xmath81 , from @xmath85}]})$ ] and compute and normalize the weights at time @xmath12 @xmath86 } & = & \frac{p(x_{i}^{[l]}|x_{i-1}^{[a_{i-1}^{[l]}]})p(y_{i}|x_{i}^{[l]})}{q_{i}(x_{i}^{[l]}|y_{i},x_{i-1}^{[a_{i-1}^{[l ] } ] } ) } \\ w_{i}^{[l ] } & = & \frac{w_{i}^{[l]}}{\sum_{m=1}^{l}w_{i}^{[m]}}\end{aligned}\ ] ] a new sequence taking values in the set of particles at each time is then selected using a backwards sampling pass . this is done by first selecting @xmath52 from the set of particles at time @xmath54 with probabilities @xmath87}$ ] and then selecting the rest of the sequence going backward in time to time @xmath25 , setting @xmath67 to @xmath84}$ ] with probability @xmath88}p(x_{i+1}'|x_{i}^{[l]})}{\sum_{m=1}^{l}w_{i}^{[m]}p(x_{i+1}'|x_{i}^{[m]})}\end{aligned}\ ] ] a common choice for @xmath89 is the model s transition density , which is what is compared to in this paper . note that each iteration of the pgbs sampler takes time proportional to @xmath90 , since it takes time proportional to @xmath13 to create the set of particles at each time @xmath12 , and to do one step of backward sampling . we propose two new ways , denoted @xmath91 and @xmath92 , of generating pool states for the embedded hmm sampler . unlike previously - used pool state selection schemes , where pool states are selected independently at each time , our new schemes select pool states sequentially , with pool states at time @xmath12 selected conditional on pool states at time @xmath27 , or alternatively at time @xmath93 . the first way to generate pool states is to use a forward pool state selection scheme , with a sequential approximation to @xmath94 as the pool state density . in particular , at time @xmath25 , we set the pool state distribution of our proposed embedded hmm sampler to @xmath95 as a result of equation ( [ eq : alphainit ] ) , @xmath96 is constant . at time @xmath70 , we set the pool state distribution to @xmath97})\end{aligned}\ ] ] which makes @xmath46 constant for @xmath70 as well ( see equation ( [ eq : alphaall ] ) ) . we then draw a sequence composed of these pool states with the forward probability implementation of the embedded hmm method , with the @xmath46 s all set to @xmath25 . the second way is to instead use a backward pool state selection scheme , with a sequential approximation of @xmath22 as the pool state density . we begin by creating the pool @xmath53 , consisting of the current state @xmath98 and the remaining @xmath14 pool states sampled from @xmath99 , the marginal density at time @xmath54 , which is the same as @xmath24 if the latent process is stationary . the backward probabilities @xmath100 , for @xmath9 in @xmath53 , are then set to @xmath25 . at time @xmath62 we set the pool state densities to @xmath101})p(x_{i+1}^{[\ell]}|x)\end{aligned}\ ] ] so that @xmath102 is constant for all @xmath30 ( see equation [ eq : beta ] ) . we then draw a sequence composed of these pool states as in the backward probability implementation of the embedded hmm method , with the @xmath102 s all set to @xmath25 . if the latent process is gaussian , and the latent state at time @xmath25 is sampled from the stationary distribution of the latent process , it is possible to update the latent variables by applying the forward scheme to the reversed sequence @xmath103 by making use of time reversibility , since @xmath104 is also sampled from the stationary distribution , and the latent process evolves backward in time according to the same transition density as it would going forward . we then use the forward pool state selection scheme along with a stochastic backward pass to sample a sequence @xmath105 , starting with @xmath75 and going to @xmath98 . it can sometimes be advantageous to alternate between using forward and backward ( or , alternatively , forward applied to the reversed sequence ) embedded hmm updates , since this can improve sampling of certain @xmath7 . the sequential pool state selection schemes use only part of the observed sequence in generating the pool states . by alternating update directions , the pool states can depend on different parts of the observed data , potentially allowing us to better cover the region where @xmath7 has high posterior density . for example , at time @xmath25 , the pool state density may disperse the pool states too widely , leading to poor sampling for @xmath75 , but sampling @xmath75 using a backwards scheme can be much better , since we are now using all of the data in the sequence when sampling pool states at time @xmath25 . to sample from @xmath106 or @xmath107 , we can use any markov transitions @xmath108 that leave this distribution invariant . the validity of the method does not depend on the markov transitions for sampling from @xmath106 or @xmath107 reaching equilibrium or even on them being ergodic . directly using these pool state densities in an mcmc routine leads to a computational cost per iteration that is proportional to @xmath15 , like in the original embedded hmm method , since at times @xmath70 we need at least @xmath13 updates to produce @xmath13 pool states , and the cost of computing an acceptance probability is proportional to @xmath13 . however , it is possible to reduce the cost per iteration of the embedded hmm method to be proportional to @xmath23 when we use @xmath106 or @xmath107 as the pool state densities . to do this , we start by thinking of the pool state densities at each time @xmath70 as marginal densities summing over the variable @xmath109 that indexes a pool state at the previous time . specifically , @xmath106 can be viewed as a marginal of the density @xmath110})\end{aligned}\ ] ] while @xmath107 is a marginal of the density @xmath111})p(x_{i+1}^{[\ell]}|x)\end{aligned}\ ] ] both of these densities are defined given a pool @xmath112 at time @xmath27 or pool @xmath113 at time @xmath93 . this technique is reminiscent of the auxiliary particle filter of pitt and shephard ( 1999 ) . we then use markov transitions , @xmath108 , to sample a set of values of @xmath9 and @xmath114 , with probabilities proportional to @xmath115 for the forward scheme , or probabilities proportional to @xmath116 for the backward scheme . the chain is started at @xmath9 set to the current @xmath7 , and the initial value of @xmath114 is chosen randomly with probabilities proportional to @xmath117})$ ] for the forward scheme or @xmath118})p(x_{i+1}^{[\ell]}|x_{i})$ ] for the backward scheme . this stochastic initialization of @xmath114 is needed to make the algorithm valid when we use @xmath115 or @xmath116 to generate the pool states . sampling values of @xmath9 and @xmath114 from @xmath115 or @xmath116 can be done by updating each of @xmath9 and @xmath114 separately , alternately sampling values of @xmath9 conditional on @xmath114 , and values of @xmath114 conditional on @xmath9 , or by updating @xmath9 and @xmath114 jointly , or by a combination of these . updating @xmath9 given @xmath114 can be done with any appropriate sampler , such as metropolis , or for a gaussian latent process we can use autoregressive updates , which we describe below . to update @xmath114 given @xmath9 , we can also use metropolis updates , proposing @xmath119 , with @xmath120 drawn from some proposal distribution on @xmath121 . alternatively , we can simply propose @xmath122 uniformly at random from @xmath123 . to jointly update @xmath9 and @xmath114 , we propose two novel updates , a `` shift '' update and a `` flip '' update . since these are also metropolis updates , using them together with metropolis or autoregressive updates , for each of @xmath9 and @xmath114 separately , allows embedded hmm updates to be performed in time proportional to @xmath23 . for sampling pool states in our embedded hmm mcmc schemes , as well as for comparison mcmc schemes , we will make use of neal s ( 1998 ) `` autoregressive '' metropolis - hastings update , which we review here . this update is designed to draw samples from a distribution of the form @xmath124 where @xmath125 is multivariate gaussian with mean @xmath126 and covariance @xmath127 and @xmath128 is typically a density for some observed data . this autoregressive update proceeds as follows . let @xmath13 be the lower triangular cholesky decomposition of @xmath127 , so @xmath129 , and @xmath54 be a vector of i.i.d . normal random variables with zero mean and identity covariance . let @xmath130 $ ] be a tuning parameter that determines the scale of the proposal . starting at @xmath131 , we propose @xmath132 because these autoregressive proposals are reversible with respect to @xmath125 , the proposal density and @xmath125 cancel in the metropolis - hastings acceptance ratio . this update is therefore accepted with probability @xmath133 note that for this update , the same value of @xmath134 is used for scaling along every dimension . it would be of independent interest to develop a version of this update where @xmath134 can be different for each dimension of @xmath131 . we can simultaneously update @xmath114 and @xmath9 at time @xmath70 by proposing to update @xmath135 to @xmath136 where @xmath122 is proposed in any valid way while @xmath16 is chosen in a way such that @xmath16 and @xmath137}$ ] are linked in the same way as @xmath9 and @xmath138}$ ] . the shift update makes it easier to generate a set of pool states at time @xmath12 with different predecessor states at time @xmath27 , helping to ensure that the pool states are well - dispersed . this update is accepted with the usual metropolis probability . for a concrete example we use later , suppose that the latent process is an autoregressive gaussian process of order @xmath25 , with the model being that @xmath139 . in this case , given @xmath122 , we propose @xmath140 } - x_{i-1}^{[\ell]})$ ] . this update is accepted with probability @xmath141 as a result of the transition densities in the acceptance ratio cancelling out , since @xmath142 } & = & x_{i } + \phi ( x_{i-1}^{[\ell ' ] } - x_{i-1}^{[\ell ] } ) - \phi x_{i-1}^{[\ell ' ] } \\ & = & x_{i } - \phi x_{i-1}^{[\ell]}\end{aligned}\ ] ] to be useful , shift updates normally need to be combined with other updates for generating pool states . when combining shift updates with other updates , tuning of acceptance rates for both updates needs to be done carefully in order to ensure that the shift updates actually improve sampling performance . in particular , if the pool states at time @xmath27 are spread out too widely , then the shift updates may have a low acceptance rate and not be very useful . therefore , jointly optimizing proposals for @xmath9 and for @xmath9 and @xmath114 may lead to a relatively high acceptance rate on updates of @xmath9 , in order to ensure that the acceptance rate for the shift updates is nt low . generating pool states locally can be helpful when applying embedded hmms to models with high - dimensional state spaces but it also makes sampling difficult if the posterior is multimodal . consider the case when the observation probability depends on @xmath143 instead of @xmath7 , so that many modes with different signs for some @xmath7 exist . we propose to handle this problem by adding an additional flip update that creates a `` mirror '' set of pool states , in which @xmath144 will be in the pool if @xmath7 is . by having a mirror set of pool states , we are able to flip large segments of the sequence in a single update , allowing efficient exploration of different posterior modes . to generate a mirror set of pool states , we must correctly use the flip updates when sampling the pool states . since we want each pool state to have a negative counterpart , we choose the number of pool states @xmath13 to be even . the chain used to sample pool states then alternates two types of updates , a usual update to generate a pool state and a flip update to generate its negated version . the usual update can be a combination of any updates , such as those we consider above . so that each state will have a flipped version , we start with a flip transition between @xmath145}$ ] and @xmath146}$ ] , a usual transition between @xmath146}$ ] and @xmath147}$ ] , and so on up to a flip transition between @xmath148}$ ] to @xmath149}$ ] . at time @xmath25 , we start with the current state @xmath75 and randomly assign it to some index @xmath150 in the chain used to generate pool states . then , starting at @xmath75 we generate pool states by reversing the markov chain transitions back to @xmath25 and going forward up to @xmath13 . each flip update is then a metropolis update proposing to generate a pool state @xmath151 given that the chain is at some pool state @xmath75 . note that if the observation probability depends on @xmath75 only through @xmath152 and @xmath24 is symmetric around zero then this update is always accepted . at time @xmath70 , a flip update proposes to update a pool state @xmath135 to @xmath153 such that @xmath137 } = -x_{i-1}^{[\ell]}$ ] . here , since the pool states at each time are generated by alternating flip and usual updates , starting with a flip update to @xmath34}$ ] , the proposal to move from @xmath114 to @xmath122 can be viewed as follows . suppose that instead of labelling our pool states from @xmath25 to @xmath13 we instead label them @xmath154 to @xmath14 . the pool states at times @xmath154 and @xmath25 , then @xmath155 and @xmath156 , and so on will then be flipped pairs , and the proposal to change @xmath114 to @xmath122 can be seen as proposing to flip the lower order bit in a binary representation of @xmath122 . for example , a proposal to move from @xmath157 to @xmath158 can be seen as proposing to change @xmath114 from @xmath159 to @xmath160 ( in binary ) . such a proposal will always be accepted assuming a transition density for which @xmath161 and an observation probability which depends on @xmath7 only via @xmath143 . the forward pool state selection scheme can be used to construct a sampler with properties similar to pgbs . this is done by using independence metropolis to sample values of @xmath9 and @xmath114 from @xmath115 . at time @xmath25 , we propose our pool states from @xmath24 . at times @xmath162 , we propose @xmath122 by selecting it uniformly at random from @xmath123 and we propose @xmath16 by sampling from @xmath163})$ ] . the proposals at all times @xmath12 are accepted with probability @xmath141 this sampler has computational cost proportional to @xmath90 per iteration , like pgbs . it is analogous to a pgbs sampler with importance densities @xmath164 and @xmath165 with the key difference between these two samplers being that pgbs uses importance weights @xmath28 on each particle , instead of an independence metropolis accept - reject step . we modify the original proof of neal ( 2003 ) , which assumes that the sets of pool states @xmath166 are selected independently at each time , to show the validity of our new sequential pool state selection scheme . another change in the proof is to account for generating the pool states by sampling them from @xmath115 or @xmath116 instead of @xmath106 or @xmath107 . this proof shows that the probability of starting at @xmath9 and moving to @xmath16 with given sets of pool states @xmath47 ( consisting of values of @xmath9 at each time @xmath12 ) , pool indices @xmath167 of @xmath7 , and pool indices @xmath168 of @xmath67 is the same as the probability of starting at @xmath16 and moving to @xmath9 with the same set of pool states @xmath47 , pool indices @xmath168 of @xmath67 , and pool indices @xmath167 of @xmath7 . this in turn implies , by summing / integrating over @xmath47 and @xmath167 , that the embedded hmm method with the sequential pool state scheme satisfies detailed balance with respect to @xmath169 , and hence leaves @xmath169 invariant . suppose we use the sequential forward scheme . the probability of starting at @xmath9 and moving to @xmath16 decomposes into the product of the probability of starting at @xmath9 , which is @xmath169 , the probability of choosing a set of pool state indices @xmath167 , which is @xmath170 , the probability of selecting the initial values of @xmath171 for the stochastic initialization step , the probability of selecting the sets of pool states @xmath47 , @xmath172 , and finally the probability of choosing @xmath16 . the probability of selecting given initial values for the links to previous states @xmath173 is @xmath174})}{\sum_{m=1}^{l}p(x_{i}|x_{i-1}^{[m]})}\end{aligned}\ ] ] the probability of choosing a given set of pool states is @xmath175 at time @xmath25 , we use a markov chain with invariant density @xmath176 to select pool states in @xmath65 . therefore @xmath177}|x_{1}^{[j-1]})\prod_{j = l_{1}-1}^{1}\tilde{r}_{1}(x_{1}^{[j]}|x_{1}^{[j+1 ] } ) \notag \\ & = & \prod_{j = l_{1}+1}^{l}r_{1}(x_{1}^{[j]}|x_{1}^{[j-1]})\prod_{j = l_{1}-1}^{1}r_{1}(x_{1}^{[j+1]}|x_{1}^{[j]})\frac{\kappa_{1}(x_{1}^{[j]})}{\kappa_{1}(x_{1}^{[j+1 ] } ) } \notag \\ & = & \prod_{j = l_{1}}^{l-1}r_{1}(x_{1}^{[j+1]}|x_{1}^{[j]})\prod_{j = l_{1}-1}^{1}r_{1}(x_{1}^{[j+1]}|x_{1}^{[j]})\frac{\kappa_{1}(x_{1}^{[j]})}{\kappa_{1}(x_{1}^{[j+1 ] } ) } \notag \\ & = & \frac{\kappa_{1}(x_{1}^{[1]})}{\kappa_{1}(x_{1}^{[l_{1}]})}\prod_{j=1}^{l-1}r_{1}(x_{1}^{[j+1]}|x_{1}^{[j]})\end{aligned}\ ] ] for times @xmath70 we use a markov chain with invariant density @xmath115 to sample a set of pool states , given @xmath112 . the chain is started at @xmath33 } = x_{i}$ ] and @xmath178 } = \ell_{i}$ ] . therefore @xmath179 } , \ell_{i}^{[j]}|x_{i}^{[j-1 ] } , \ell_{i}^{[j-1]})\prod_{j = l_{i}-1}^{1}\tilde{r}_{i}(x_{i}^{[j ] } , \ell_{i}^{[j]}|x_{i}^{[j+1 ] } , \ell_{i}^{[j+1 ] } ) \notag \\ & = & \prod_{j = l_{i}+1}^{l}r_{i}(x_{i}^{[j ] } , \ell_{i}^{[j]}|x_{i}^{[j-1 ] } , \ell_{i}^{[j-1]})\prod_{j = l_{i}-1}^{1}r_{i}(x_{1}^{[j+1 ] } , \ell_{i}^{[j+1]}|x_{i}^{[j ] } , \ell_{i}^{[j]})\frac{\lambda_{i}(x_{i}^{[j ] } , \ell_{i}^{[j]})}{\lambda_{i}(x_{1}^{[j+1 ] } , \ell_{i}^{[j+1 ] } ) } \notag \\ & = & \prod_{j = l_{i}}^{l-1}r_{i}(x_{i}^{[j+1 ] } , \ell_{i}^{[j+1]}|x_{i}^{[j ] } , \ell_{i}^{[j]})\prod_{j = l_{i}-1}^{1}r_{i}(x_{i}^{[j+1 ] } , \ell_{i}^{[j+1]}|x_{i}^{[j ] } , \ell_{i}^{[j]})\frac{\lambda_{i}(x_{i}^{[j ] } , \ell_{i}^{[j]})}{\lambda_{i}(x_{i}^{[j+1 ] } , \ell_{i}^{[j+1 ] } ) } \notag \\ & = & \frac{\lambda_{i}(x_{i}^{[1 ] } , \ell_{i}^{[1]})}{\lambda_{i}(x_{i}^{[l_{i } ] } , \ell_{i}^{[l_{i}]})}\prod_{j=1}^{l-1}r_{i}(x_{i}^{[j+1 ] } , \ell_{i}^{[j+1]}|x_{i}^{[j ] } , \ell_{i}^{[j]})\end{aligned}\ ] ] finally , we choose a new sequence @xmath16 amongst the collection of sequences consisting of the pool states with a backward pass . this is done by first choosing a pool state @xmath52 uniformly at random from @xmath53 . we then select the remaining states @xmath180}$ ] by selecting @xmath181 with probability @xmath182})}{\sum_{m=1}^{l}p(x_{i}'|x_{i-1}^{[m]})}\end{aligned}\ ] ] thus , the probability of starting at @xmath9 and going to @xmath16 , with given @xmath166 , @xmath183 and @xmath184 is @xmath185})}{\sum_{m=1}^{l}p(x_{i}|x_{i-1}^{[m ] } ) } \times \frac{\kappa_{1}(x_{1}^{[1]})}{\kappa_{1}(x_{1}^{[l_{1}]})}\prod_{j=1}^{l-1}r_{1}(x_{1}^{[j+1]}|x_{1}^{[j ] } ) \label{eq : proba } \\ & & \ \times \ \prod_{i=2}^{n}\biggl[\frac{\lambda_{i}(x_{i}^{[1 ] } , \ell_{i}^{[1]})}{\lambda_{i}(x_{i}^{[l_{i } ] } , \ell_{i}^{[l_{i}]})}\prod_{j=1}^{l-1}r_{i}(x_{i}^{[j+1 ] } , \ell_{i}^{[j+1]}|x_{i}^{[j ] } , \ell_{i}^{[j]})\biggr]\times \frac{1}{l } \times \prod_{i=2}^{n}\frac{p(x_{i}'|x_{i-1}^{[l_{i-1}']})}{\sum_{m=1}^{l}p(x_{i}'|x_{i-1}^{[m ] } ) } \notag \\ & & = \ \kappa_{1}(x_{1}^{[1]})\prod_{j=1}^{l-1}r_{1}(x_{1}^{[j+1]}|x_{1}^{[j ] } ) \times \ \prod_{i=2}^{n}\biggl[\lambda_{i}(x_{i}^{[1 ] } , \ell_{i}^{[1]})\prod_{j=1}^{l-1}r_{i}(x_{i}^{[j+1 ] } , \ell_{i}^{[j+1]}|x_{i}^{[j ] } , \ell_{i}^{[j]})\biggr]\notag \\ & & \ \times \ \frac{1}{l^{n+1 } } \times \frac{p(x|y)}{\kappa_{1}(x_{1})\prod_{i=2}^{n}\lambda_{i}(x_{i } , \ell_{i } ) } \times \prod_{i=2}^{n}\frac{p(x_{i}|x_{i-1}^{[\ell_{i}]})}{\sum_{m=1}^{l}p(x_{i}|x_{i-1}^{[m]})}\prod_{i=2}^{n}\frac{p(x_{i}'|x_{i-1}^{[l_{i-1}']})}{\sum_{m=1}^{l}p(x_{i}'|x_{i-1}^{[m ] } ) } \notag\end{aligned}\ ] ] here , we have @xmath186 } = x_{i-1}'$ ] . also @xmath187 and @xmath188})}{\sum_{x_{i } \in \mathcal{p}_{i } } \sum_{m=1}^{l}p(y_{i}|x_{i})p(x_{i}|x_{i-1}^{[m]})}\end{aligned}\ ] ] and @xmath189 therefore ( [ eq : proba ] ) can be simplified to @xmath190})\prod_{j=1}^{l-1}r_{1}(x_{1}^{[j+1]}|x_{1}^{[j ] } ) \times \ \prod_{i=2}^{n}\biggl[\lambda_{i}(x_{i}^{[1 ] } , \ell_{i}^{[1]})\prod_{j=1}^{l-1}r_{i}(x_{i}^{[j+1 ] } , \ell_{i}^{[j+1]}|x_{i}^{[j ] } , \ell_{i}^{[j]})\biggr ] \times \frac{1}{l^{n+1 } } \notag \\ & & \times \ \prod_{i=2}^{n}p(x_{i}|x_{i-1 } ) \times \prod_{i=2}^{n}p(x_{i}'|x_{i-1 } ' ) \times \prod_{i=2}^{n}\frac{1}{\sum_{m=1}^{l}p(x_{i}|x_{i-1}^{[m ] } ) } \times \prod_{i=2}^{n}\frac{1}{\sum_{m=1}^{l}p(x_{i}'|x_{i-1}^{[m ] } ) } \notag \\ & & \times \ \sum_{x_{1 } \in \mathcal{p}_{1}}p(x_{1})p(y_{1}|x_{1})\prod_{i=2}^{n}\sum_{x_{i } \in \mathcal{p}_{i } } \sum_{m=1}^{l}p(y_{i}|x_{i})p(x_{i}|x_{i-1}^{[m]})\end{aligned}\ ] ] the last factor in the product only depends on the selected set of pool states . by exchanging @xmath9 and @xmath16 we see that the probability of starting at @xmath16 and then going to @xmath9 , with given sets of pool states @xmath47 , pool indices @xmath167 of @xmath7 and pool indices @xmath168 of @xmath67 is the same . to demonstrate the performance of our new pool state scheme , we use two different state space models . the latent process for both models is a vector autoregressive process , with @xmath191 where @xmath192 and @xmath193 note that @xmath194 is the covariance of the stationary distribution for this process . for model 1 , the observations are given by @xmath195 for model 2 , the observations are given by @xmath196 for model 1 , we use a @xmath160-dimensional latent state and a sequence length of @xmath197 , setting parameter values to @xmath198 , and @xmath199 , @xmath200 , @xmath201 for @xmath202 , with @xmath203 . for model 2 , we increase the dimensionality of the latent space to @xmath204 and the sequence length to @xmath205 . we set @xmath198 and @xmath200 , @xmath206 for @xmath202 , with @xmath207 . we generated one random sequence from each model to test our samplers on . these observations from model 1 and model 2 are shown in figure [ fig : data ] . note that we are testing only sampling of the latent variables , with the parameters set to their true values . the code for all experiments in this paper is available at http://arxiv.org/abs/1602.06030 . 0.45 .,title="fig : " ] 0.45 .,title="fig : " ] a simple scheme for sampling the latent state sequence is to use metropolis - hastings updates that sample each @xmath7 in sequence , conditional on @xmath208 and the data , starting at time @xmath25 and going to time @xmath54 . we sample all dimensions of @xmath7 at once using autoregressive updates ( see section 5.4.3 ) . the conditional densities of the @xmath4 are @xmath209 the densities @xmath210 , and @xmath211 are all gaussian . the means and covariances for these densities can be derived by viewing @xmath212 or @xmath26 as a gaussian prior for @xmath7 and @xmath213 as a gaussian likelihood for @xmath7 . in particular , we have @xmath214 where @xmath215^{-1}[(\phi^{-1}\sigma\phi^{-1})^{-1}\phi^{-1}x_{2 } ] \\ & = & [ ( \phi^{-1}\sigma\phi^{-1})^{-1 } + \sigma_{\textnormal{init}}^{-1}]^{-1}[\sigma^{-1}(\phi x_{2 } ) ] \\ & = & [ \phi^{2 } + \sigma_{\textnormal{init}}^{-1}\sigma]^{-1}\phi x_{2 } \\ \sigma_{1 } & = & [ ( \phi^{-1}\sigma\phi^{-1})^{-1 } + \sigma_{\textnormal{init}}^{-1}]^{-1}\\ & = & [ \phi(\sigma^{-1}\phi ) + \sigma_{\textnormal{init}}^{-1}]^{-1 } \\ \\ \mu_{i } & = & [ ( \phi^{-1}\sigma\phi^{-1})^{-1 } + \sigma^{-1}]^{-1}[\sigma^{-1}\phi x_{i-1 } + ( \phi^{-1}\sigma\phi^{-1})^{-1}\phi^{-1}x_{i+1 } ] \\ & = & ( \phi^{-1}\sigma\phi^{-1})^{-1 } + \sigma^{-1}]^{-1}[\sigma^{-1}(\phi ( x_{i-1 } + x_{i+1 } ) ) ] \\ & = & [ \phi^{2 } + i]^{-1}\phi ( x_{i-1 } + x_{i+1 } ) \\ \sigma_{i } & = & [ ( \phi^{-1}\sigma\phi^{-1})^{-1 } + \sigma^{-1}]^{-1 } \\ & = & [ \phi(\sigma^{-1}\phi ) + \sigma^{-1}]^{-1 } \\ \\ \mu_{n } & = & \phi x_{n-1 } \\ \sigma_{n } & = & \sigma\end{aligned}\ ] ] to speed up the metropolis updates , we precompute and store the matrices @xmath216^{-1}\phi$ ] , @xmath217^{-1}\phi$ ] as well as the cholesky decompositions of the posterior covariances . in both of our test models , the posterior standard deviation of the latent variables @xmath218 varies depending on the value of the observed @xmath219 . to address this , we alternately use a larger or a smaller proposal scaling , @xmath134 , in the autoregressive update when performing an iteration of the metropolis sampler . we implement the pgbs method as described in section 5.3 , using the initial density @xmath24 and the transition densities @xmath26 as importance densities to generate particles . we combine pgbs updates with single - state metropolis updates from section 5.6.2 . this way , we combine the strengths of the two samplers in targeting different parts of the posterior distribution . in particular , we expect the metropolis updates to do better for the @xmath7 with highly informative @xmath29 , and the pgbs updates to do better for the @xmath7 where @xmath29 is not as informative . for model @xmath25 , we compared the embedded hmm sampler to the simple single - state metropolis sampler , to the pgbs sampler , and to the combination of pgbs with metropolis . for model 2 , we compared the embedded hmm sampler to the pgbs with metropolis sampler . for both models and all samplers , we ran the sampler five times using five different random number generator seeds . we implemented the samplers in matlab on a linux system with a 2.60 ghz intel i7 - 3720qm cpu . for the single - state metropolis sampler , we initialized all @xmath218 to @xmath154 . every iteration alternately used a scaling factor , @xmath134 , of either @xmath220 or @xmath221 , which resulted in an average acceptance rate of between @xmath222 and @xmath223 for the different @xmath7 over the sampler run . we ran the sampler for @xmath224 iterations , and prior to analysis , the resulting sample was thinned by a factor of @xmath160 , to @xmath225 . the thinning was done due to the difficulty of working with all samples at once , and after thinning the samples still possessed autocorrelation times significantly greater than @xmath25 . each of the @xmath226 samples took about @xmath227 seconds to draw . for the pgbs sampler and the sampler combining pgbs and metropolis updates , we also initialized all @xmath218 to @xmath154 . we used @xmath228 particles for the pgbs updates . for the metropolis updates , we alternated between scaling factors of @xmath220 and @xmath221 , which also gave acceptance rates between @xmath222 and @xmath223 . for the standalone pgbs sampler , we performed a total of @xmath229 iterations . each iteration produced two samples for a total of @xmath230 samples and consisted of a pgbs update using the forward sequence and a pgbs update using the reversed sequence . each sample took about @xmath231 seconds to draw . for the pgbs with metropolis sampler , we performed a total of @xmath232 iterations of the sampler . each iteration was used to produce four samples , for a total of @xmath233 samples , and consisted of a pgbs update using the forward sequence , ten metropolis updates ( of which only the value after the tenth update was retained ) , a pgbs update using the reversed sequence , and another ten metropolis updates , again only keeping the value after the tenth update . the average time to draw each of the @xmath233 samples was about @xmath234 seconds . for model 2 , we were unable to make the single - state metropolis sampler converge to anything resembling the actual posterior in a reasonable amount of time . in particular , we found that for @xmath218 sufficiently far from @xmath154 , the metropolis sampler tended to be stuck in a single mode , never visiting values with the opposite sign . for the pgbs with metropolis sampler , we set the initial values of @xmath235 to @xmath25 . we set the number of particles for pgbs to @xmath236 , which was nearly the maximum possible for the memory capacity of the computer we used . for the metropolis sampler , we alternated between scaling factors of @xmath237 and @xmath25 , which resulted in acceptance rates ranging between @xmath238 and @xmath239 . we performed a total of @xmath228 iterations of the sampler . as for model @xmath25 , each iteration produced four samples , for a total of @xmath240 samples , and consisted of a pgbs update with the forward sequence , fifty metropolis updates ( of which we only keep the value after the last one ) , a pgbs update using the reversed sequence , and another fifty metropolis updates ( again only keeping the last value ) . it took about @xmath241 seconds to draw each sample . for both model 1 and model 2 , we implemented the proposed embedded hmm method using the forward pool state selection scheme , alternating between updates that use the original and the reversed sequence . as for the baseline samplers , we ran the embedded hmm samplers five times for both models , using five different random number generator seeds . we generate pool states at time @xmath25 using autoregressive updates to sample from @xmath242 . at times @xmath243 , we sample each pool state from @xmath244 by combining an autoregressive and shift update . the autoregressive update proposes to only change @xmath9 , keeping the current @xmath245 fixed . the shift update samples both @xmath9 and @xmath245 , with a new @xmath245 proposed from a uniform@xmath123 distribution . for model 2 , we also add a flip update to generate a negated version of each pool state . note that the chain used to produce the pool states now uses a sequence of updates . therefore , if our forward transition first does an autoregressive update and then a shift update , the reverse transitions must first do a shift update and then an autoregressive update . as for the single - state metropolis updates , it is beneficial to use a different proposal scaling , @xmath134 , when generating each pool state at each time @xmath12 . this allows generation of sets of pool states which are more concentrated when @xmath29 is informative and more dispersed when @xmath29 holds little information . for model 1 , we initialized all @xmath218 to @xmath154 . we used @xmath246 pool states for the embedded hmm updates . for each metropolis update to sample a pool state , we used a different scaling @xmath134 , chosen at random from a @xmath247 distribution . the acceptance rates ranged between @xmath248 and @xmath249 for the metropolis updates and between @xmath250 and @xmath251 for the shift updates . we performed a total of @xmath252 iterations of the sampler , with each iteration consisting of an embedded hmm update using the forward sequence and an embedded hmm update using the reversed sequence , for a total of @xmath253 samples . each sample took about @xmath254 seconds to draw . for model 2 , we initialized the @xmath218 to @xmath25 . we used a total of @xmath255 pool states for the embedded hmm sampler ( i.e. @xmath256 positive - negative pairs due to flip updates ) . each metropolis update used to sample a pool state used a scaling , @xmath134 , randomly drawn from the @xmath257 distribution . the acceptance rates ranged between @xmath258 and @xmath223 for the metropolis updates and between @xmath250 and @xmath259 for the shift updates . we performed a total of @xmath252 iterations of the sampler , producing two samples per iteration with an embedded hmm update using the forward sequence and an embedded hmm update using the reversed sequence . each of the @xmath253 samples took about @xmath260 seconds to draw . as a way of comparing the performance of the two methods , we use an estimate of autocorrelation time for each of the latent variables @xmath235 . autocorrelation time is a measure of how many draws need to be made using the sampling chain to produce the equivalent of one independent sample . the autocorrelation time is defined as @xmath261 , where @xmath262 is the autocorrelation at lag @xmath120 it is commonly estimated as @xmath263 where @xmath264 are estimates of lag-@xmath120 autocorrelations and the cutoff point @xmath265 is chosen so that @xmath264 is negligibly different from @xmath154 for @xmath266 . here @xmath267 where @xmath268 is an estimate of the lag-@xmath120 autocovariance @xmath269 when estimating autocorrelation time , we remove the first @xmath270 of the sample as burn - in . then , to estimate @xmath268 , we first estimate autocovariances for each of the five runs , taking @xmath271 to be the overall mean over the five runs . we then average these five autocovariance estimates to produce @xmath268 . to speed up autocovariance computations , we use the fast fourier transform . the autocorrelation estimates are then adjusted for computation time , by multiplying the estimated autocorrelation time by the time it takes to draw a sample , to ensure that the samplers are compared fairly . 0.45 0.45 0.45 0.45 the computation time - adjusted autocorrelation estimates for model 1 , for all the latent variables , plotted over time , are presented in figure [ fig : acf ] . we found that the combination of single - state metropolis and pgbs works best for the unimodal model . the other samplers work reasonably well too . we note that the spike in autocorrelation time for the pgbs and to a lesser extent for the pgbs with metropolis sampler occurs at the point where the data is very informative . this in turn makes the use of the diffuse transition distribution the particles are drawn from inefficient and much of the sampling in that region is due to the metropolis updates . here , we also note that the computation time adjustment is sensitive to the particularities of the implementation , in this case done in matlab , where performance depends a lot on how well vectorization can be exploited . implementing the samplers in a different language might change the relative comparisons . we now look at how the samplers perform on the more challenging model 2 . we first did a preliminary check of whether the samplers do indeed explore the different modes of the distribution by looking at variables far apart in the sequence , where we expect to see four modes ( with all possible combinations of signs ) . this is indeed the case for both the pgbs with metropolis and embedded hmm samplers . 0.45 , title="fig : " ] 0.45 , title="fig : " ] next , we look at how efficiently the latent variables are sampled . of particular interest are the latent variables with well - separated modes , since sampling performance for such variables is illustrative of how well the samplers explore the different posterior modes . consider the variable @xmath272 , which has true value @xmath273 . figure [ fig : trace ] shows how the different samplers explore the two modes for this variable , with equal computation times used to produced the samples for the trace plots . we can see that the embedded hmm sampler with flip updates performs significantly better for sampling a variable with well - separated modes . experiments showed that the performance of the embedded hmm sampler on model 2 without flip updates is much worse . 0.45 , title="fig : " ] 0.45 , title="fig : " ] we can also look at the product of the two variables @xmath274 , with true value @xmath275 . the trace plot is given in figure [ fig : trace2 ] . in this case , we can see that the pgbs with metropolis sampler performs better . since the flip updates change the signs of all dimensions of @xmath7 at once , we do not expect them to be as useful for improving sampling of this function of state . the vastly greater number of particles used by pgbs , @xmath236 , versus @xmath255 for the embedded hmm method , works to the advantage of pgbs , and explains the performance difference . 0.45 0.45 looking at these results , we might expect that we can get a good sampler for both @xmath272 and @xmath274 by alternating embedded hmm and pgbs with metropolis updates . this is indeed the case , which can be seen in figure [ fig : trace3 ] . for producing these plots , we used an embedded hmm sampler with the same settings as in the experiment for model @xmath155 and a pgbs with metropolis sampler with @xmath276 particles and metropolis updates using the same settings as in the experiment for model @xmath155 . this example of model 2 demonstrates another advantage of the embedded hmm viewpoint , which is that it allows us to design updates for sampling pool states to handle certain properties of the density . this is arguably easier than designing importance densities in high dimensions . we have demonstrated that it is possible to use embedded hmm s to efficiently sample state sequences in models with higher dimensional state spaces . we have also shown how embedded hmms can improve sampling efficiency in an example model with a multimodal posterior , by introducing a new pool state selection scheme . there are several directions in which this research can be further developed . the most obvious extension is to treat the model parameters as unknown and add a step to sample parameters given a value of the latent state sequence . in the unknown parameter context , it would also be interesting to see how the proposed sequential pool state selection schemes can be used together with ensemble mcmc updates of shestopaloff and neal ( 2013 ) . for example , one approach is to have the pool state distribution depend on the average of the current and proposed parameter values in an ensemble metropolis update , as in shestopaloff and neal ( 2014 ) . one might also wonder whether it is possible to use the entire current state of @xmath9 in constructing the pool state density at a given time . it is not obvious how ( or if it is possible ) to overcome this limitation . for example , for the forward scheme , using the current value of the state sequence at some time @xmath277 to construct pool states at time @xmath12 means that the pool states at time @xmath120 will end up depending on the current value of @xmath278 , which would lead to an invalid sampler . at each time @xmath62 , the pool state generation procedure does not depend on the data after time @xmath12 , which may cause some difficulties in scaling this method further . on one hand , this allows for greater dispersion in the pool states than if we were to impose a constraint from the other direction as with the single - state metropolis method , potentially allowing us to make larger moves . on the other hand , the removal of this constraint also means that the pool states can become too dispersed . in higher dimensions , one way in which this can be controlled is by using a markov chain that samples pool states close to the current @xmath7 that is , a markov chain that is deliberately slowed down in order not to overdisperse the pool states , which could lead to a collection of sequences with low posterior density . we thank arnaud doucet for helpful comments . this research was supported by the natural sciences and engineering research council of canada . a. s. is in part funded by an nserc postgraduate scholarship . r. n. holds a canada research chair in statistics and machine learning . andrieu , c. , doucet , a. and holenstein , r. ( 2010 ) `` particle markov chain monte carlo methods '' , _ journal of the royal statistical society b _ , vol . 72 , pp . 269 - 342 . lindsten , f. ; schon , t.b . ( 2012 ) `` on the use of backward simulation in the particle gibbs sampler '' , in _ acoustics , speech and signal processing ( icassp ) , 2012 ieee international conference on _ , pp . 3845 - 3848 . neal , r. m. ( 2003 ) `` markov chain sampling for non - linear state space models using embedded hidden markov models '' , technical report no . 0304 , department of statistics , university of toronto , http://arxiv.org/abs/math/0305039 . neal , r. m. , beal , m. j. , and roweis , s. t. ( 2004 ) `` inferring state sequences for non - linear systems with embedded hidden markov models '' , in s. thrun , et al ( editors ) , _ advances in neural information processing systems 16 _ , mit press .

using a sample of 19,464 galaxies drawn from the deep2 galaxy redshift survey , we study the relationship between galaxy color and environment at . we find that the fraction of galaxies on the red sequence depends strongly on local environment out to , being larger in regions of greater galaxy density . at all epochs probed , we also find a small population of red , morphologically early type galaxies residing in regions of low measured overdensity . the observed correlations between the red fraction and local overdensity are highly significant , with the trend at detected at a greater than level . over the entire redshift regime studied , we find that the color density relation evolves continuously , with red galaxies more strongly favoring overdense regions at low relative to their red sequence counterparts at high redshift . at , the red fraction only weakly correlates with overdensity , implying that any color dependence to the clustering of galaxies at that epoch must be small . our findings add weight to existing evidence that the build up of galaxies on the red sequence has occurred preferentially in overdense environments ( i.e. , galaxy groups ) at . furthermore , we identify the epoch at which typical galaxies began quenching and moved onto the red sequence in significant number . the strength of the observed evolutionary trends at suggests that the correlations observed locally , such as the morphology density and color density relations , are the result of environment driven mechanisms ( i.e. , `` nurture '' ) and do not appear to have been imprinted ( by `` nature '' ) upon the galaxy population during their epoch of formation .

astro-ph0607512

the galaxy population both locally and out to @xmath9 is found to be effectively described as a combination of two distinct galaxy types : red , early type galaxies lacking much star formation and blue , late type galaxies with active star formation ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) . the spatial distribution of this bimodal galaxy population is frequently phrased today in terms of the so called morphology density or color density relation . as first quantified by @xcite , @xcite , and @xcite , the morphology density relation holds that star forming , disk dominated galaxies tend to reside in regions of lower galaxy density relative to those of red , elliptical galaxies . many physical mechanisms that could be responsible for this correlation between galaxy morphology , star formation history , and environment have been proposed ( see * ? ? ? * for a review of probable mechanisms ) . are the morphology - density and color density relations a result of environment driven evolution , or were these trends imprinted upon the galaxy population during their epoch of formation ? only through comprehensive studies of galaxy properties and environments , both locally and at high redshift , will we be able to understand the role of local density in determining the star formation histories and morphologies of galaxies . while the close relationship between galaxy type and density was primarily uncovered via the study of nearby clusters , recent work using the 2degree field galaxy redshift survey ( 2dfgrs , * ? ? ? * ; * ? ? ? * ) and the sloan digital sky survey ( sdss , * ? ? ? * ) has established that the connections between local environment and galaxy properties such as morphology , color , and luminosity extend over the full range of densities , from rich clusters to voids ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? furthermore , high resolution imaging and spectroscopic data in increasingly more distant clusters @xmath10 have shown that the trends observed locally persist to higher @xmath3 , at least in the highest density environments ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) . using a large sample of galaxies drawn from the deep2 galaxy redshift survey , @xcite extended the understanding of the color density relation at @xmath9 across the full range of environments , from voids to rich groups , showing that the correlation between galaxy color and mean overdensity found locally is in place , at least in a global sense , when the universe was half its present age . while the role of environment appears to have been very critical at @xmath9 and perhaps at earlier times , quantitative measures of the evolution of environmental influences on the galaxy population or of correlations such as the morphology density and color density relation are limited . comparisons of local results with studies of high redshift clusters have pointed towards significant evolution in the relationship between galaxy properties and local environment from @xmath9 to @xmath11 ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? while these results indicate an environment driven evolution in the galaxy population ( i.e. , pointing towards nurture versus nature as the origin of the galaxy bimodality ) , such work has been limited to the vicinity of rich clusters , and thus we know little about evolution in the relationship between galaxy type and density across the full scope of galaxy environments . furthermore , clusters include only a relatively small fraction of the total galaxy population at any epoch by number ( and an even smaller fraction by volume ) . thus , the evolution of the color density relation among the vast majority of the galaxy population remains unprobed . in large clusters , the physical mechanisms at work ( e.g. , galaxy harassment , ram pressure stripping , and global tidal interactions ) go beyond those acting in group sized systems and the field . results from the first comprehensive study of galaxy environment over a broad range of densities at high @xmath3 indicate that such cluster specific physical mechanisms can not explain the global color density relation as found at @xmath12 @xcite . accordingly , in looking for evolution in the relationship between the bimodal nature of galaxy properties and the local galaxy environment , we must turn our attention to the entire dynamic range of galaxy overdensities at high redshift . in this vein , recent work employing a sample of low and high redshift field galaxies from the vimos vlt deep survey ( vvds , * ? ? ? * ) has found a strong evolutionary trend in the color density relation for galaxies spanning the redshift range @xmath13 , with the color magnitude diagrams for galaxies at @xmath14 showing no significant dependence on environment @xcite . also working at high redshift , a study of the blue fraction ( that is , the fraction of galaxies that have blue color ) in galaxy groups and in the field population by @xcite finds instead that the field blue fraction significantly differs from that of the group population out to @xmath15 . in this paper , we use the large sample of high@xmath3 galaxies obtained by the deep2 survey to conduct a detailed study of the color density relation at @xmath0 . in 2 , we discuss the data sample employed along with our measurements of galaxy environments and colors . our main results regarding the relationship between color and environment are presented in 3 . finally , in 4 and 5 , we discuss our findings alongside other recent results and summarize our conclusions . throughout this paper , we assume a flat @xmath16cdm cosmology with @xmath17 , @xmath18 , @xmath19 , and @xmath20 . the deep2 galaxy redshift survey is a nearly completed @xmath21 project designed to study the galaxy population and large scale structure at @xmath12 @xcite . to date , the survey has targeted @xmath22 galaxies in the redshift range @xmath23 , down to a limiting magnitude of @xmath24 . currently , the survey covers @xmath25 square degrees of sky over four widely separated fields . in this paper , we utilize a sample of 32,002 galaxies with accurate redshifts ( quality @xmath26 or @xmath27 as defined by * ? ? ? * ) in the range @xmath0 and drawn from all four of the deep2 survey fields . while at @xmath28 the sample includes galaxies from each of the four survey fields , spanning a total of @xmath29 pointings @xcite , at @xmath30 our sample selection is limited to galaxies in the extended groth strip ( egs ) , where no color cut to pre select for @xmath31 galaxies was used @xcite . frame @xmath32 colors and absolute @xmath33band magnitudes , @xmath34 , are calculated from cfht @xmath35 photometry @xcite using the k correction procedure described in @xcite . all magnitudes discussed within this paper are given in ab magnitudes @xcite . for zero point conversions between ab and vega magnitudes , refer to table 1 of @xcite . for each galaxy in the data set , we compute the projected @xmath36nearest neighbor surface density @xmath37 about the galaxy , where the surface density depends on the projected distance to the @xmath36nearest neighbor , @xmath38 , as @xmath39 . in computing @xmath40 , a velocity window of @xmath41 is utilized to exclude foreground and background galaxies . in the tests of @xcite , this environment estimator proved to be the most robust indicator of local galaxy density for the deep2 survey . to correct for the redshift dependence of the sampling rate of the deep2 survey , each surface density value is divided by the median @xmath40 of galaxies at that redshift within a window of @xmath42 . ; correcting the measured surface densities in this manner converts the @xmath40 values into measures of overdensity relative to the median density ( given by the notation @xmath43 here ) and effectively accounts for redshift variations in the selection rate @xcite . finally , to minimize the effects of edges and holes in the survey geometry , we exclude all galaxies within @xmath44 comoving mpc of a survey boundary , reducing our sample from 32,002 to 19,464 galaxies . the redshift distribution of this subset is plotted in figure [ zdist ] . for complete details regarding the computation of the local environment measures , we direct the reader to @xcite . within the deep2 sample , the bimodality of galaxy colors in rest frame @xmath45 color is clearly visible ; out to @xmath46 , the galaxy color magnitude diagram exhibits a clear division into a relatively tight red sequence and a more diffuse `` blue cloud '' of galaxies @xcite . for this study , we compute the fraction of galaxies on the red sequence using the following color division , as defined by @xcite and illustrated as the dotted lines in figure [ magcut ] and the dashed lines in the lower panels of figure [ redfrac ] : @xmath47 the red fraction @xmath48 within a given redshift and environment range is given by the number of galaxies redward of this relation in @xmath45 color divided by the total number of galaxies within the same bin of redshift and environment . as discussed by @xcite , this division between the red sequence and blue cloud is derived from the work of @xcite , in which the color magnitude relation for early type ( i.e. , solely morphologically selected ) galaxies is measured . the results of this work and similar studies of clusters at @xmath49 show that ellipticals form a red sequence in color magnitude space with little scatter ( e.g. , * ? ? ? * ; * ? ? ? thus , dividing the sample according to the color cut presented in equation [ willmer_cut ] effectively selects the red , early type portion of the bimodal galaxy distribution . details regarding the possible evolution of this color division with redshift are highlighted by @xcite and examined in 3.1.1 of this work . because the deep2 survey extends over a broad redshift range , selecting galaxies according to a fixed apparent magnitude limit introduces differences in selection with redshift that depend in a nonnegligible way on galaxy color and luminosity . to study how these selection effects potentially influence our results , we identify several subsamples within the full catalog ( sample a ) of 19,464 deep2 galaxies spread across the redshift range @xmath0 ( cf . fig . 1 ) . the simplest selection method is to produce a volume limited subsample according to a strict cut in absolute magnitude . we create such a sample ( sample b ) by restricting to @xmath50 and requiring @xmath51 , the absolute magnitude to which deep2 is complete along both the red sequence and the blue cloud at @xmath52 ( cf . top of figure [ sample_select ] ) . producing a volume limited sample with a fixed absolute magnitude limit at all colors severely restricts either the redshift range probed or the number of galaxies selected at each redshift . here , we adopt a limiting redshift of @xmath52 in order to maintain enough galaxies ( 2,784 in sample b ) with which to accurately compute the red fraction over the full range of overdensities . given the large redshift range probed in our analysis , however , the number of galaxies with @xmath51 at low redshift @xmath53 is fairly small , and as such results at low @xmath3 using sample b are quite noisy . as discussed by @xcite , in studying the evolution of galaxy properties it is also possible to produce volume limited catalogs with a color dependent , absolute magnitude cut by defining a region of rest - frame color magnitude space that is uniformly sampled by the survey at all redshifts of interest . for the deep2 survey , such a selection cut is illustrated in the bottom of figure [ sample_select ] of this paper and figure 2 of @xcite and given by @xmath54,\ [ c(u - b ) + d]\right\ } , \end{array } \label{magcut}\ ] ] where @xmath55 is the limiting redshift beyond which the selected sample becomes incomplete , @xmath56 , @xmath57 , @xmath58 and @xmath59 are constants that are determined by the limit of the color magnitude distribution of the sample with redshift @xmath60 , and @xmath61 is a constant that allows for linear redshift evolution of the typical galaxy absolute magnitude @xmath62 . for the parameter @xmath61 , we adopt the @xcite value of @xmath63 , determined from a study of the @xmath33band galaxy luminosity function in the combo17 @xcite , deep1 @xcite , and deep2 surveys . by including this linear @xmath62 evolution in our selection cut , we are selecting a similar population of galaxies with respect to @xmath62 at all redshifts . adopting this approach with a limiting redshift of @xmath64 , we define a sample of 11,192 galaxies ( sample c ) over the redshift range @xmath65 that is volume limited relative to @xmath62 and selected according to this color dependent cut in @xmath66 . the values of the constants @xmath56 , @xmath57 , @xmath58 , and @xmath59 which define the color dependent selection are @xmath67 , @xmath68 , @xmath69 , and @xmath70 , respectively . for complete details of the selection method , we refer the reader to @xcite . as a final sample , we again select galaxies assuming linear evolution in @xmath62 , as for sample c. however , in determining the limiting absolute magnitude as a function of redshift , we do not apply a color dependent selection as utilized for sample c. instead , we restrict galaxies of all color to the selection limit , @xmath71 where the limiting redshift is chosen to be @xmath72 and @xmath73 defines the absolute magnitude at @xmath52 to which deep2 is complete along both the red sequence and the blue cloud ( cf . the top panel of fig . [ sample_select ] ) . due to the more severe , color independent cut in absolute magnitude , sample d totals only 2,150 galaxies , spanning the redshift range @xmath50 . in contrast , while sample b is similarly selected based just on absolute magnitude ( that is , independent of galaxy color ) , the magnitude cut in sample d evolves with redshift , staying fixed relative to @xmath74 . thus , sample b selects galaxies down to the same absolute magnitude at all redshifts , and sample d samples galaxies to the same depth in the luminosity function at each @xmath3 . a brief summary of the four galaxy samples utilized in this paper is provided in table [ sample_descript_tab ] . l l l l sample a & 19,464 & @xmath0 & + sample b & 2,784 & @xmath50 & + sample c & 11,192 & @xmath65 & + sample d & 2,150 & @xmath50 & + [ sample_descript_tab ] in order to test for possible systematic effects we employ a set of 12 mock galaxy catalogs based on those of @xcite . these catalogs are derived from @xmath75body simulations by populating dark matter halos with galaxies according to a halo occupation distribution ( hod ) function @xcite , which describes the probability distribution of the number of galaxies in a halo as a function of the host halo mass . the luminosities of galaxies are then assigned according to the conditional luminosity function ( clf ) formalism introduced by @xcite , which allows the galaxy luminosity function to be mass dependent as well . parameters for the hod and the clf are chosen to match the 2dfgrs luminosity function @xcite and two point correlation function @xcite . by assuming that the manner in which dark matter halos are populated with galaxies does not evolve from @xmath12 to @xmath11 @xcite save via an overall evolution in @xmath74 ( corresponding to @xmath76 here ) , mock catalogs can then be built using dark only simulation outputs at varying redshifts . the resulting simulated galaxy catalogs from @xcite are in excellent agreement with the lower redshift @xmath77 deep2 correlation function @xcite and the combo17 luminosity function @xcite . it is critical for this paper that we characterize any systematic effects which alter the relationship between galaxy color and observed overdensity as a function of redshift . therefore , we employ a modified version of the @xcite mock catalogs which incorporate galaxy colors ; these catalogs are described in detail by @xcite . to construct these new catalogs , we first measure the local overdensity of each object in the @xcite mock catalogs using the full , volume limited catalog . each galaxy is assigned a rest frame @xmath45 color drawn from the set of real deep2 galaxies at @xmath78 located in a corresponding bin of local galaxy overdensity and absolute @xmath33band magnitude , employing the environment measurements of @xcite . this produces mock catalogs which reproduce the observed relationships between galaxy color , luminosity , and environment in the deep2 data at @xmath78 , but do not allow those relationships to evolve with redshift unless the `` true '' distribution of galaxy overdensities evolves . after galaxy colors are assigned , apparent @xmath79 magnitudes are determined using the k correction methods applied for deep2 @xcite , so that red galaxies will be lost at the same luminosity at a given redshift as in the data . we then apply the standard deep2 target selection and slitmask making procedures @xcite to these mock catalogs , so that we may directly determine the effects of deep2 target selection algorithms on observed trends . these mock catalogs should not be a perfect representation of reality , as uncertainties in the observed environments will cause some objects to be assigned the color of a galaxy that is actually in a higher or lower density environment than was measured ; however , this effect is small ( i.e. , environment measurement errors at @xmath80 are significantly smaller than the bin sizes ) . nevertheless , they provide a robust test of whether we may falsely observe an evolution in the relationship between galaxy color and environment when in fact none exists . they also allow us to directly explore what effects the increased errors in environment measurements at higher redshifts ( where samples grow dilute ) may have on our results . we discuss these tests in 3.2 . figure [ redfrac ] shows the relationship between the red fraction @xmath48 and local galaxy overdensity for the full data set ( sample a ) in four distinct redshift bins spanning @xmath0 . we find that the red fraction exhibits a strong dependence on local environment in each redshift bin , such that @xmath81 is higher in regions of greater overdensity . fitting a linear relation to the trends in figure [ redfrac ] , we find that the slope of this color environment correlation evolves with @xmath3 , with the relative fraction of red galaxies in dense environments decreasing with lookback time ( cf . table [ fits_tab ] ) . over the full redshift range probed by the deep2 sample , the measured slope decreases from @xmath82 for @xmath30 to @xmath83 for @xmath84 , with the difference being significant at a greater than @xmath85@xmath86 level . the vertical error bars in figure [ redfrac ] indicate the poissonian uncertainty in each point . error estimates based on bootstrap and jackknife resampling amongst the 10 deep2 pointings used yielded comparable uncertainties ( within 1020% ) , suggesting that sample ( or `` cosmic '' ) variance is not the dominant source of uncertainty ; while cosmic variance will influence the distribution of environments in a given redshift bin , it should not strongly affect the relationship between galaxy color and density in a given environment bin . refer to 4.1 for further investigation of the role of cosmic variance in this analysis . all of the samples described in 2.3 yield similar results for the evolution of the color density relation ( cf . fig . [ redfrac_bcd ] and table [ fits_tab ] ) . while overall @xmath81 values by overdensity differ when either a color dependent or a color independent @xmath34 limit is applied to the full sample , in every case the color - environment relation weakens , but still is present , at higher @xmath3 . due to their small sample sizes , the effect is not statistically sigificant for samples b and d , however . c c c c c c c sample a & & & & + @xmath30 & 504 & 2319 & 0.111 & 0.208 & 0.018 & 0.010 + @xmath87 & 785 & 4419 & 0.081 & 0.175 & 0.012 & 0.007 + @xmath88 & 989 & 5357 & 0.080 & 0.182 & 0.012 & 0.006 + @xmath84 & 770 & 7369 & 0.047 & 0.104 & 0.007 & 0.004 + + sample b & & & & + @xmath30 & 48 & 76 & 0.240 & 0.546 & 0.230 & 0.140 + @xmath87 & 125 & 297 & 0.127 & 0.395 & 0.076 & 0.044 + @xmath89 & 464 & 1081 & 0.120 & 0.413 & 0.047 & 0.024 + + sample c & & & & + @xmath30 & 116 & 813 & 0.103 & 0.131 & 0.019 & 0.013 + @xmath87 & 207 & 2143 & 0.066 & 0.093 & 0.012 & 0.007 + @xmath88 & 282 & 2868 & 0.056 & 0.097 & 0.011 & 0.006 + @xmath90 & 372 & 5368 & 0.035 & 0.068 & 0.007 & 0.004 + + sample d & & & & + @xmath30 & 76 & 148 & 0.194 & 0.446 & 0.098 & 0.069 + @xmath87 & 102 & 237 & 0.150 & 0.405 & 0.086 & 0.049 + @xmath89 & 347 & 844 & 0.090 & 0.403 & 0.056 & 0.026 + [ fits_tab ] to further illustrate the evolution in the color density relation with @xmath3 , we present color magnitude diagrams in figure [ cmd_4panel ] divided by local environment and redshift . in this figure , the relationship between red fraction @xmath48 and galaxy overdensity is made directly discernable to the eye ; in high density regions , there is an increased proportion of sources on the red sequence relative to that observed in low density environments . while the results shown in figure [ redfrac ] show a clear signature of evolution in the color density relation over the redshift range @xmath91 , relatively large redshift bins are used , thereby coarsely sampling the redshift domain . to study the evolution of the @xmath81overdensity relationship with @xmath3 in more detail , we divide sample c into thirds according to overdensity and compute the red fraction as a function of redshift in a sliding bin of width @xmath92 for the galaxies in the high density @xmath93 and low density @xmath94 extremes of the overdensity distribution , with results shown in figure [ redfrac_withz ] . here , the high density and low density thirds are defined with respect to the overdensity distribution in each @xmath3 bin ; however , dividing the galaxy sample into high and low density regimes according to the overdensity distribution in the entire redshift range @xmath95 probed yields consistent results . as detailed in 2.3 , sample c is defined so as to be complete with respect to @xmath62 across the full redshift range , @xmath96 , establishing a sample uniformly selected at all redshifts studied . as illustrated in fig . [ redfrac_withz ] , the color density relation shows a continuous evolution with redshift from @xmath97 to @xmath98 , such that at redshifts approaching @xmath15 the red fraction in low and high density regions are statistically consistent with each other . extrapolating linear regression fits to the @xmath99 relations in high density and low density environments , we find convergence at a redshift of @xmath100 . we investigate this convergenece in more detail in the following subsections ( 3.1.2 and 3.1.3 ) . in our analyses , we have attempted to carefully account for selection effects due to the design parameters of the deep2 survey . for all galaxy samples but sample a , selection effects related to the survey s magnitude limit and @xmath79band selection , which bias the sample against faint and/or red galaxies at higher redshifts , shoud be minimal . the agreement amongst the highly disparate samples indicates that the results presented above are robust to such effects . these results are similar to the findings of @xcite , which investigates the evolution of the blue fraction among group and field galaxies at @xmath101 in deep2 . that work employed a very different technique for measuring galaxy environment and studied a somewhat different set of galaxy subsamples . the measurements of the evolution in the blue fraction presented by @xcite , however , are significantly noisier than those presented in this study and could be susceptible to biases if the group finding algorithm used performs differently at different @xmath3 . we note that while that work uses a different methodology for dividing galaxies into environment bins ( identifying members of galaxy groups rather than with a continuous measure of local galaxy density ) , both the @xcite results and those presented here are derived from subsamples of the deep2 survey ; hence any systematics affecting one may also affect the other . while the position of the bimodality in @xmath45 versus @xmath34 color magnitude space shows no significant dependence on redshift within the deep2 data , when studying the red fraction as a function of redshift , we must consider the possibility that there is evolution in the color of the red sequence population and thus in the position of the color bimodality . if the location of the color bimodality evolves with reshift , then the measured evolution in the red fraction in high density environments could simply result from galaxies moving ( e.g. , via passive evolution ) across our non evolving color division ( cf . equation [ willmer_cut ] ) . however , any affect related to possible evolution in the location of the @xmath45 bimodality would be minimized by the differential nature of our measurements . that is , the location of the color bimodality does not appear to depend on environment , and thus any evolution in the red fraction for the galaxy sample in high density regions should be mirrored in the @xmath102 for the sample of galaxies in intermediate and low density environments . furthermore , given that passive evolution will cause galaxies to move redward as they evolve and that the @xmath45 color division in equation [ willmer_cut ] is defined by a galaxy sample at @xmath103 @xcite , a non evolving color division will _ not _ increase the measured evolution in the color density relation due to an interloper population of blue galaxies on the red sequence at @xmath1 . instead , a non evolving cut would create a more stringent selection criterion at higher redshifts , limiting the number of galaxies residing in the trough between the red sequence and blue cloud ( also called the `` green valley '' ) that are counted as members of the red sequence , thereby actually producing weaker evolution in the color density relation . in the spirit of thoroughness , however , we test the effect of using an evolving color division on our results . allowing for passive evolution in the location of the color bimodality of @xmath104 magnitudes ( in @xmath45 ) per unit @xmath3 ( e.g. , * ? ? ? * ; * ? ? ? * ) , such that the division between red and blue galaxies moves blueward with lookback time , we find no significant change in our results . that is , employing a redshift dependent division between the blue cloud and red sequence , the relative red fraction in high and low density environments still shows strong evolution , consistent with no color density relation at @xmath105 . in analyzing the evolution of the red fraction @xmath48 as a function of redshift , we must consider the possible impact of redshift dependent selection effects in the deep2 survey . in particular , due to the finite amount of slit length real estate available on deimos slitmasks and the finite amount of observing time dedicated to the project , the deep2 survey only spectroscopically observes @xmath106 of the galaxies that meet the survey s target selection criteria . furthermore , about @xmath107 of those galaxies targeted for spectroscopy fail to yield a high quality ( quality @xmath108 ) redshift . initial follow up observations of sources for which deep2 fails to measure a redshift indicate that roughly half of all failures ( i.e. , @xmath109 of deep2 targets ) are at redshifts beyond the range probed by deep2 ( i.e. , @xmath110 ; c. steidel , private communication ) . additionally , redshift failures may result from poor observing conditions , data reductions errors , or instrumental effects . clearly important for this work is the fact that the deep2 redshift failure rate is correlated with observed galaxy color and magnitude , thus possibly introducing bogus trends in the measured @xmath99 relation . to understand the degree to which our results are impacted by redshift incompleteness , we employ the weighting scheme presented by @xcite . the derived weights account for variations in both redshift incompleteness and targeting rate as a function of apparent @xmath79 magnitude and apparent @xmath111 and @xmath112 colors , assuming that the redshift distribution for red @xmath113 failures is identical to that of the observed ( quality @xmath114 ) red galaxy sample and that the blue @xmath115 failures sit beyond the redshift range of the survey ( defined as the optimal model in * ? ? ? * ) . if we use these weights to compute @xmath99 for sample c ( as in fig . [ redfrac_withz ] ) , all changes in the measured trends with redshift are well within the measurement uncertainties . thus , biases related to redshift incompleteness should not introduce spurious results in our analysis . with and without weighting for incompleteness , we find a convergence of the red fraction in high and low density environments at @xmath4 ; extrapolating linear regression fits to the @xmath99 relations in high density and low density environments , we find convergence at a redshift of @xmath116 when weighting for incompleteness versus a convergence at @xmath100 without weighting . the mock galaxy catalogs described in 2.4 allow us to further test the degree to which our observations of the color density relation may be subject to systematic effects within deep2 ( e.g. , due to increased errors in environment measures as the sampling density drops with @xmath3 or due to the expected changes in the magnitude of peculiar velocities with @xmath3 compared to the 1000 @xmath117 window used ) . these catalogs should exhibit no evolution in the color density relation when measured with perfect information , by construction . therefore , when we apply our measurement techniques to subsets of the mock catalogs which emulate the observed samples , any apparent evolution in the color density relation will generally indicate the presence of observational biases . there is a notable exception to this . it is possible to select a subsample from the volume limited mock galaxy catalog ( i.e. , the catalog before applying deep2 target selection criteria or slitmask making algorithms ) using a color dependent , absolute magnitude limit which evolves with @xmath3 , as for our sample c above ; however , for the mock catalogs , @xmath118 is the appropriate evolution , as that value was used in their construction , rather than @xmath63 as used for sample c. we find that if we use the full volume limited mock catalog to determine local overdensities in real space , there is in fact evolution in the color density relation for such a sample ( cf . figure [ chromo_plots]a ) . this evolution does not reflect an observational selection effect , however . instead , it is a result of applying an absolute magnitude cut which changes with @xmath3 to mock catalogs which were constructed assuming rest frame color depends on @xmath34 in a way which is independent of @xmath3 ( cf.2.4 ) . because @xmath119 was brighter at higher @xmath3 , the galaxies included by deep2 tend to be brighter at higher @xmath3 ; and as may be seen in fig . [ sample_select ] and fig . [ cmd_4panel ] , brighter galaxies are more likely to be red . the color dependent , absolute magnitude limit exaggerates this effect . if instead we apply a color and redshift independent absolute magnitude limit to the volume limited mock sample , the mock catalogs show considerably weaker color density evolution , as we would expect ( cf . [ chromo_plots]b ) . for panels ( a ) and ( b ) of figure [ chromo_plots ] , we determine the `` true '' local galaxy environment according to the @xmath120nearest neighbor surface density ( converted into an overdensity , @xmath121 ) using the real space galaxy positions and the full volume limited galaxy population . although the mock catalogs exhibit a modest evolution in the color density relation for samples with an absolute magnitude limit that shifts with @xmath122 ( like our sample c ) , they still may be used to determine the net effects of deep2 sample selection on the observed color density relation . we do this by comparing @xmath99 in two different mock catalog samples . for the first , we do not apply deep2 target selection and slitmask making criteria , but only the color dependent absolute magnitude limit given by equation [ magcut ] with @xmath76 , and we use the best possible measure of environment in the catalogs ( based upon the real space overdensity of galaxies in the full , volume limited catalog , as employed by * ? ? ? * ) to divide the sample into environment bins ; see fig . [ chromo_plots]a . for the second , we apply the deep2 target selection and slitmask design algorithm to these mock catalogs , and use the overdensity measured from this `` observed '' sample to partition the sample ; results for this sample are shown in fig . [ chromo_plots]c . we find that in fact the deep2like sample exhibits both a smaller gap between @xmath99 for the extreme environment bins and a stronger apparent evolution over @xmath123 . this is consistent with a scenario where noise in environment measures causes objects to cross the boundary between the intermediate environment bin and the extremes , thereby artificially increasing the red fraction for the low density sample while decreasing @xmath81 for the high density sample . this effect should become worse at higher @xmath3 , as the sampling density decreases there , leading to greater environment errors . therefore , though we observe convergence of between the red fraction in low and high density environments at @xmath100 , in actuality convergence should not occur until a somewhat higher redshift . however , this test shows that observational selection effects can not explain the convergence in @xmath81 that we observe in high density and low density regions at @xmath15 . extrapolating linear regression fits to the @xmath99 relations in high density and low density environments , we find convergence at @xmath124 if we measure environment with the full sample ( fig . [ chromo_plots]a ) , or @xmath125 after we apply deep2 target selection procedures ( fig . [ chromo_plots]c ) . the mock catalogs provide clear evidence that the null hypothesis of no evolution in the color magnitude relation over @xmath123 can not hold . the relations presented in figure [ redfrac ] also show evidence of a trend with @xmath3 in the normalization of the @xmath81overdensity relation , such that the total fraction of red sequence galaxies averaged over all densities decreases with @xmath3 . this effect is dominated by two systematic trends in the data : ( 1 ) from @xmath11 to @xmath9 , the number density of red galaxies in the universe decreases by a factor of @xmath1263 ( e.g. , * ? ? ? * ; * ? ? ? * ) , and ( 2 ) at redshifts beyond @xmath127 , the red galaxy fraction decreases precipitously in the deep2 sample , due to the survey s @xmath79band magnitude limit @xcite . this latter effect is quite apparent in the final redshift bin in figure [ redfrac ] , where the normalization of the observed red fraction overdensity relation drops significantly ; the other volume limited samples studied here should not be affected by this , however ( cf . [ redfrac_bcd ] ) . in spite of these redshift dependent effects , we still find that at all redshifts probed in this paper some fraction of red galaxies populate very underdense environments . while uncertainties in the environment measurements will lead to some galaxies at intermediate densities being scattered into the low density third of the overdensity distribution , upper limits on this effect show that it can not account for all of the red galaxies found in low density environments at any redshift , as we now show . we proceed by assuming that the observed @xmath128 distribution of red galaxies matches the shape of the true distribution , an assumption that holds so long as the distribution is approximately linear over scales comparable to measurement errors ( as convolving a linear function with a gaussian leaves it unchanged ) . in actuality , e.g. if the @xmath128 distribution has some cutoff value at low overdensity , the observed distribution will have more objects below this cutoff than the true distribution ; thus the observed distribution provides an upper limit to the true contamination rate . we calculate this limit from the expectation value of the number of red objects that have true overdensities in the top two thirds of galaxies but measured overdensities in the lowest one third , minus the number of objects truly in the bottom one third but measured in the top two thirds . we assume that each overdensity measurement may be represented by a gaussian distribution in @xmath128 , with mean given by the measured value and standard deviation increasing linearly from @xmath129 at @xmath130 to @xmath131 at @xmath132 , based upon tests with the mock catalogs described in 2.4 . we find that the increasing uncertainties in environment measures combined with the weakening in the strength of the color density relation with lookback time yield a net contamination rate that is roughly constant with redshift @xmath133 . hence , the true red fraction in the extreme low density third of the sample is at least @xmath134 at all redshifts . similar studies at lower redshift ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) have shown that the existence of red galaxies in low density environments persists to @xmath11 . clearly , a population of red galaxies in low density environments exists at @xmath123 . however , galaxies may appear red either because they are true red sequence / early type galaxies or because of the presence of interstellar dust . based on past studies ( e.g. , * ? ? ? * ) , we might expect the former population to dominate the red sequence at lower redshift and the latter at higher @xmath3 . are the red sequence members in low density regions at low and high redshift comparable in terms of galaxy morphology ? out of the less than 40 red galaxies in underdense environments identified in deep2 for which _ hst_/acs imaging is available in the egs @xcite , an initial by eye inspection indicates that the majority of this population exhibits early type morphologies , with the remainder being reddened disk galaxies . while the sample imaged with _ hst_/acs is small , we find no significant trends of morphology with redshift ; for example , dusty disk galaxies do not increasingly dominate the red sequence in low density environments at higher @xmath3 . however , contamination by late type galaxies preferentially at higher @xmath3 can not be excluded as a possible explanation for the observational results presented here ; an analysis of galaxy morphologies by @xcite concludes that late type galaxies comprise 30% of the red sequence at @xmath135 , though they are only a minor contaminant at low redshift ( see also * ? ? ? such an increase in the contribution of dusty disks to the red sequence at @xmath1 could work to reduce the strength of an existing color density relation . are members of the red sequence in underdense regions the result of passive evolution , in which they consumed their gas supplies independent of environmental effects , or are they fossil groups that result from the merging of several smaller galaxies ? clearly , even in underdense environments galaxy mergers are capable of triggering events that reduce gas reservoirs and at least temporarily halt star formation as well as disrupt galactic disks , yielding merger remnants with surface brightness profiles and density distributions similar to those of early type galaxies @xcite . similarly , passive evolution when teamed with processes attributed to secular evolution such as bar instabilities could also explain the existence of red , morphologically early type galaxies in low density environments @xcite . in future work , we hope to explore the morphologies of red galaxies in low density regions within the egs in more detail , with a goal of differentiating between these two scenarios ; galactic bulges formed by bar instabilities tend to differ from those built via mergers in that the former exhibit more disk like properties such as flatter profiles and residual bars or spiral structure @xcite . from our study of 19,464 galaxies in the deep2 galaxy redshift survey , we conclude that the color density relation observed in the local universe is also seen at @xmath1 , with the fraction of red galaxies increasing with local galaxy overdensity at essentially all epochs studied @xmath136 and over all environments probed ( from voids to large groups ) . at all redshifts , however , there still exists a population of red galaxies in underdense environments . in addition , we find that the color density relation evolves with redshift , growing weaker with lookback time such that at @xmath4 the relationship may be nonexistent within the range of environments probed by the deep2 survey ( i.e. , not including massive clusters ) . when viewed in conjunction with the results from studies of the galaxy luminosity function at @xmath137 @xcite , our findings provide direct evidence that the red sequence is built up preferentially in overdense environments ( i.e. , galaxy groups , as deep2 does not sample rich clusters ) , thereby producing the observed increase in the slope of the red fraction versus overdensity relation at later time . these results are in agreement with the general picture painted by studies of the morphology density relation in galaxy clusters at @xmath137 . for example , building upon the work of @xcite , @xcite and @xcite find that the fraction of early type galaxies increases steadily with density in cluster environments out to @xmath9 , with the strength of the correlation weaker at @xmath9 than in local samples . corresponding work by @xcite also finds significant evolution in the morphology density relation within massive clusters at @xmath138 , with the early type fraction observed to steadily decline with increasing redshift . using data from the canada france hawaii telescope legacy survey ( cfhtls ) , @xcite similarly find that over the redshift range @xmath139 both the red fraction and early type fraction increase in high density regions with decreasing redshift . while the fraction of galaxies with early type morphologies in the @xcite sample is constant with @xmath3 in low density environments , they find that the red fraction @xmath140 steadily increases in the `` field '' from @xmath103 to @xmath11 , in constrast to our results which show no evolution in the red fraction in regions of low galaxy density at @xmath123 . the differences between our results and those of @xcite could result from the use of photometric redshifts to determine environments in that work ; as shown by @xcite , photometric redshifts can not cleanly discriminate the environments of galaxies , as even the smallest photometric redshift errors achieved are much larger than galaxy correlation lengths ( @xmath141 comoving mpc versus @xmath142 comoving mpc for @xmath143 ) . our findings also reproduce the general trend found by @xcite based on the vvds survey ; they , too , find that the color density relation weakens with redshift over the range @xmath144 . however , within their study the fraction of galaxies on the red sequence shows no significant dependence on overdensity at @xmath145 . in contrast , we find a highly significant relationship between red fraction and environment at @xmath146 , even for highly differing subsamples ; only at @xmath4 are our results consistent with density independence , as seen in fig.[redfrac_withz ] . when binning the deep2 data in the same redshift ranges as that of @xcite , the differences between the deep2 and vvds results are readily apparent ( see figure [ direct_comp ] of this work and figure 6 of * ? ? given this apparent contradiction for @xmath147 and that the work of @xcite employs the most analogous data set to that presented here , a more detailed examination is required . first of all , because rest frame galaxy colors ( which depend on photometry and coarsely on redshift ) are almost entirely independent of the environment measurements ( which depend upon angular positions and high precision redshifts ) , we consider it unlikely that the highly significant correlation between red fraction and environment , which we find at @xmath9 , is false . it persists in all four samples considered here , which should differ from each other by more than sample b should differ from the vvds volume limited sample . since there are few galaxies in the gap between the red and blue populations , modest differences in the definition of @xmath81 have minimal effect , as well , so this is unlikely to explain any differences . in an effort to conduct a more direct comparison between the vvds and deep2 results , we attempt to replicate as closely as possible those vvds subsamples for which we can establish a volume complete deep2 analogue and apply to them the same red fraction definition as employed by @xcite . while in this paper we generally define the red fraction , @xmath81 , according to a color division in @xmath45 versus @xmath34 color magnitude space , @xcite utilize a luminosity independent selection in @xmath148 . using the cfht / megacam @xmath149 and @xmath150 filter response , quantum efficiency , telescope throughput , and atmospheric extinction estimates , the k correction code ( _ kcorrect _ version v4_1_2 ) of @xcite , and our cfht @xmath35 photometry @xcite , we compute the rest frame @xmath148 color for each galaxy in the deep2 spectroscopic sample . @xcite divide samples according to johnson / cousins @xmath33band absolute magnitude in the ab system , taking @xmath20 ; this is identical to the @xmath34 used throughout this paper ( cf . we are therefore able to place the deep2 galaxies in the same color magnitude space as that of @xcite and define `` red '' according to the vvds definition ( @xmath151 ) . in order to match the vvds samples as closely as possible , we have constructed deep2 samples covering identical redshift regimes ( @xmath152 and @xmath153 ) and absolute magnitude limits ( @xmath154 ) as vvds subsamples studied by @xcite . because of the differences between the deep2 @xmath79-band and vvds @xmath155-band selections , it is possible to replicate only some of the vvds subsamples . for @xmath156 , we can construct volume limited deep2 samples down to @xmath157 , while for @xmath158 we are able to construct a volume limited sample matching the vvds @xmath159 data set , and a very nearly volume limited sample with @xmath160 . figures 7 and 8 of @xcite illustrate the red fraction versus overdensity trends for the corresponding samples drawn from vvds . as shown in figure [ ug_comp ] , the deep2 results for volume limited samples at @xmath161 and @xmath162 are consistent with the vvds results for similarly selected data sets . however , the errors on the vvds trends between red fraction and environment are significantly larger ( @xmath163 those for deep2 ) . to make this figure , we map the data points from figure 7 of @xcite onto the deep2 results by plotting the extreme - overdensity vvds points at the same abcissa values as the extreme - overdensity deep2 points . this is necessary because citetcucciati06 measures environments over much larger scales than we do in deep2 , and density contrasts on larger scales should be smaller . however , locally , at least , the same trends are found using environments measured on smaller and larger scales @xcite . no scaling , however , is applied to the vvds red fraction values , as plotted in figure [ ug_comp ] . the difference in normalization of @xmath164 between deep2 and vvds could be due to k correction errors in either sample , or , conceivably , due to a possible tendency of vvds to fail to obtain high confidence redshifts for red galaxies as often as blue , especially at @xmath135 . from this comparison to the @xcite results , we conclude that the color density relation at @xmath145 , as observed by deep2 , is generally consistent with the vvds measurements . however , due to their larger uncertainties ( likely principally due to their smaller sample size ) , the color density relation found here could not have been detected significantly by @xcite . given that a perfect match between our samples and techniques and those of @xcite is impossible , especially due to the overdensity mapping applied in figure [ ug_comp ] , we also consider other possible reasons for why a trend present in deep2 could be missed within the vvds data set . it is quite possible that fundamental differences in the samples observed could contribute to the differing conclusions at @xmath9 from the deep2 and vvds analyses . in particular , the vvds sample includes a subset of objects to @xmath165 , while deep2 samples down to a limit of @xmath24 . as a consequence , while the surveys have comparable depths for the bluest objects , vvds is deeper for red galaxies at @xmath1 , which have @xmath166@xmath167 . as a result , @xcite include fainter red galaxies in their @xmath81 measurements than we do in our standard samples ( in the volume limited subsamples , the respective depths are indentical , of course ) , and are able to study samples with color independent @xmath34 limits down to @xmath168 ( for @xmath158 ) . as shown in fig . [ ug_comp ] , however , volume limited deep2 samples matching vvds samples in luminosity range do exhibit a significant color density trend where the equivalent vvds samples do not . in addition to the @xcite sample being small ( 6,582 galaxies ) in comparison to our deep2 sample , yielding increased poisson errors , it is also limited to a single @xmath169 field , which increases the uncertainties due to sample ( or `` cosmic '' ) variance compared to our sample . the differences between our results and those of @xcite , however , are likely not attributable to cosmic variance alone . as shown in figure [ cvplot ] , the variation in the @xmath164overdensity relation from field to field within the deep2 survey is small , with each field yielding a color density relation consistent with that of the full sample . cosmic variance mostly changes the overall red fraction in a given field and the relative abundance of the different environments , but not the strength of a trend . in addition to smaller sample sizes , which yield larger errors in @xmath164 , there are other phenomena which could obscure a true correlation between red fraction and overdensity in vvds data preferentially at high redshift . one possibility is that environment errors in the @xcite sample are large compared to their environment bins at @xmath145 . we note that the 5%95% range of measured overdensities increases with @xmath3 in their data set , as seen in fig . 6 of @xcite ; this could reflect increasing errors in their environment measures ( in the deep2 data set , in contrast , the 5%95% range in measured overdensities is nearly independent of redshift ) . another difference is that @xcite use a large smoothing kernel ( @xmath170 mpc ) in measuring environments . estimating overdensities on such large scales can be problematic due to the small number of independent resolution elements across such a small field ( at @xmath12 , @xmath171 corresponds to @xmath172 comoving mpc ) and due to the large percentage of the sample which falls within one smoothing scale length of a survey boundary or edge ( @xmath173 of the vvds field by area at @xmath9 ) . differences between the true relationship connecting color and environment on smaller scales and 5 mpc scales ( as used by * ? ? ? * ) could also account for discrepancies between deep2 and vvds results , but at @xmath174 , at least , correlations between galaxy properties and environment measurements on @xmath175 and @xmath176 mpc scales appear to differ significantly only in noise properties @xcite . finally , another effect that could dilute the measurements of @xcite results from the fact that at @xmath135 the redshift regime where a lack of a color density relation is found in the @xcite data the vvds sample is dominated by lower quality vvds redshifts ( @xmath177 , * ? ? ? a number of tests have found that the @xmath177 redshifts have a nonnegligible error rate ( @xmath178 , o. ilbert , private communication ; c. wolf , private communication ; * ? ? ? * ) , which would cause sources to be included in the @xcite sample at @xmath1 incorrectly , with both erroneous k corrections and erroneous overdensity measurements . this contamination would dilute any true correlation between rest frame color and environment . in this paper , we use only higher confidence deep2 redshifts , which have an overall failure rate below 2% based upon tests with repeated observations @xcite . while the color density results of @xcite differ when compared to the trends found in the full deep2 sample , a comparison of similar samples ( cf . fig . [ ug_comp ] ) illustrates that the deep2 and vvds results are mostly in agreement where they overlap , given the significantly larger errors for the @xcite data set , which are likely due to the smaller vvds sample size . the various sources of error discussed above ( cf . 4.1 ) could also be at play , though the dominant uncertainty appears to be statistical in nature . overall , both studies independently find that the color density relation grows weaker with increasing redshift . although trends at @xmath153 are not significant for any of the vvds samples , at lower redshift @xcite conclude that the color density relation is stronger for brighter galaxies than faint ones . in contrast , we do not observe significant variations in the color density trend for samples differing in @xmath34 . that is , over the absolute magnitude ranges probed in fig . [ ug_comp ] , the slope of the @xmath164overdensity relation exhibits no statistically significant dependence on luminosity in our measurements . the volume limited subsamples in fig . [ ug_comp ] , however , are overlapping rather than independent in luminosity and only probe a limited range of @xmath34 . thus , the lack of significant luminosity dependence to the color density relation within deep2 is not a strong statement . as shown by @xcite , the dependence of mean environment on luminosity and color is effectively separable at @xmath9 , which implies that the dependence of the color density relation on luminosity should not be strong over the luminosity range covered by deep2 . within the context of the larger picture of galaxy evolution , as discussed in 4.2 , a luminosity dependence to the evolution of the color density relation may well exist . comparison of our results and those of clustering studies at @xmath9 ( e.g. , * ? ? ? * ; * ? ? ? * ) to clustering measurements of bright galaxies at @xmath179 ( e.g. , * ? ? ? * ) lead to similar conclusions . as explained by @xcite , such a result can be easily understood as an example of cosmic `` downsizing '' @xcite , where the cessation of star formation occurs first in high luminosity galaxies or high mass halos ( i.e. , high density environments ) . however , the methods used in this paper do not detect this trend within the deep2 data set . a quite striking result from our work is the difference between the evolution of the red fraction in under and overdense environments spanning the redshift range @xmath123 ; while @xmath81 in underdense environments remains roughly constant with @xmath3 in our sample , we find that the red fraction in dense environments decreases with @xmath3 , such that it is roughly equal to the fraction in underdense environments at @xmath98 . studying the properties of group galaxies in deep2 , @xcite find similar evolution in the color density relation from @xmath180 , with the fraction of blue galaxies in groups becoming comparable to that of the field at @xmath15 . given @xmath175 gyr to move from the blue cloud to the red sequence ( e.g. , * ? ? ? * ; * ? ? ? * ) , if the color density relation first arises at @xmath181 , then the physical mechanisms responsible must start to become operative at @xmath182 within a concordance cosmology . this observed evolution in the color density relation and the existence of an epoch at which environment dependent quenching initiates are both consistent with the current theoretical picture in which the conversion of blue galaxies into members of the red sequence occurs in dark matter halos with mass greater than some critical value @xmath183 . as halos pass this mass threshold in such hot flow / cold flow accretion models , the infalling cold gas supply to the central galactic disk is virial shocked and shut off such that the galaxy will quickly burn its remaining fuel and redden ( e.g. , * ? ? ? * ; * ? ? ? * ) . within the models of @xcite and others ( see also * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) , low energy agn activity is included to suppress the cooling of shocked gas and the recommencement of star formation . at yet higher redshifts @xmath184 , even in halos above this critical threshold , quenching does not occur since cooling is effectively able to remove the pressure support behind the virial shock @xcite . at later times , the evolution of the halo mass function @xcite in combination with the near redshift independence of the critical mass @xcite leads to a continual increase in the number of halos above the threshold mass . given the correlation between halo mass and environment in simulations ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) , this thereby predicts an evolution in the color density relation ( i.e. , becoming stronger ) at @xmath185 . in this picture , the color density or morphology density relation within cluster environments likely persists out to higher redshifts , as cluster galaxies reside in the most massive halos , those first to reach the critical mass at @xmath179 . as deep2 primarily samples the more common , less massive groups , the color density relation is still weak at somewhat later times , @xmath15 . in addition to predicting evolution of the color density relation at @xmath186 , these hot flow / cold flow accretion models also provide an additional mechanism by which to explain the presence of red galaxies residing in underdense regions ( cf . 3.2 ) . while passive evolution or merging of several smaller galaxies to form a fossil group are also viable physical mechanisms by which to create such galaxies , the tail ( at @xmath187 ) of the halo mass function in low density environments naturally leads to the existence of at least some true red sequence galaxies in voids . the measured evolution in the color density relation as presented here should be directly related to measurements of galaxy clustering at high redshift . we predict that the clustering of @xmath5 galaxies at @xmath188 should depend only weakly on color . studies of galaxy clustering by spectral type or by color at high redshift ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) have shown that blue , star forming galaxies at @xmath9 are less strongly clustered than their red counterparts . such clustering work , however , spans a broad redshift range ( extending to redshifts less than unity ) and the galaxy samples utilized are dominated in number by galaxies at @xmath158 , where we still find a significant color density relation . unfortunately , the deep2 sample includes a modest number of galaxies @xmath189 at @xmath190 , with very few on the red sequence ; we therefore lack the statistical power needed to compute correlation strengths for subsamples by galaxy color at such redshifts . at yet higher redshifts @xmath184 there are indications that clustering depends on color . studies of uv selected galaxies and red , near ir bright galaxies have found significant differences in clustering strengths depending on the sample selection , with the latter being more strongly clustered ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? furthermore , within near ir bright samples , the measured correlation length depends significantly on apparent color ( e.g. , @xmath191 or @xmath192 ) , such that blue near ir bright galaxies cluster like uv selected lyman break galaxies ( lbgs ) , while the red near ir bright galaxies exhibit a correlation length larger by a factor of roughly two @xcite . the red , near ir bright samples observed at early epochs , however , are likely the antecedents of the rarest , most massive red galaxies today and are not representative of the progenitors of deep2 galaxies ; their correlation length at @xmath193 already exceeds that of the brightest and reddest deep2 galaxy samples . presumably , if we were to study similarly extreme samples at @xmath15 ( e.g. , galaxies in massive clusters analogous to the candidates studied by * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) , we should find a significant color density relation . instead , the progenitors of deep2 galaxies were bluer and/or fainter at @xmath179 and resembled the lbgs more than those red galaxies which are most readily observed at high redshift . thus , our observations of the evolution of the red fraction in low density and high density environments do not contradict the current set of clustering measurements at higher redshift @xmath7 or observations of massive clusters at @xmath1 . like the cluster galaxies , the red , near ir selected , massive galaxies seen at high @xmath3 , given their observed clustering , are likely to reside in very massive halos , which should be the first halos to cross the threshold quenching mass . in fact , estimates of their star formation rates indicate that they are actively forming stars in large quantities @xcite ; hence , they do not appear to generally have been quenched at @xmath179 , though more recent observations of near ir selected galaxies at @xmath179 indicate relatively low star formation activity in at least a portion of the massive galaxy population at high redshift @xcite . in this paper , we present a detailed study of the evolution in the color density relation at @xmath0 . using a sample of galaxies drawn from the deep2 galaxy redshift survey , we estimate the local overdensity about each galaxy according to the projected @xmath36nearest neighbor surface density . from this , we measure the evolution of the red fraction with environment across time . our principal results are as follows : * we find that the color density relation observed locally still exists at @xmath1 ; the fraction of galaxies on the red sequence increases with local galaxy overdensity to nearly the redshift limits of the deep2 survey . * at all epochs probed @xmath194 , we find there exists a population of red , morphologically early type galaxies residing in the the most underdense environments . * the color density relation evolves with redshift , growing weaker with lookback time such that at @xmath4 there is no detectable dependence of galaxy color on local environment in the deep2 sample . * our results support a picture in which the red sequence grew preferentially in dense environments ( i.e. , galaxy groups ) at @xmath195 . clearly , the local environment plays an important role in `` nurturing '' galaxies , establishing the existence of correlations such as the morphology density and color density relation over cosmic time . the strength of evolutionary trends suggests that the correlations observed locally do not appear to have been imprinted ( by `` nature '' ) upon the galaxy population during their epoch of formation . * our findings imply that there should be little color dependence in the clustering of @xmath5 galaxies at @xmath4 . this work was supported in part by nsf grants ast0071048 ast0071198 , ast0507428 , ast0507483 . j.a.n . and a.l.c . acknowledge support by nasa through hubble fellowship grants hst hf01165.01a and hst hf01182.01a , respectively , awarded by the space telescope science institute , which is operated by aura inc . under nasa contract nas 526555 . s.m.f . would like to acknowledge the support of a visiting miller professorship at uc m.c.c . would like to thank greg wirth and all of the keck observatory staff for their help in the acquisition of the keck / deimos data . in addition , the authors thank olga cucciati , chris marinoni , and the vvds team for helpful discussions related to this work , for review of an early version of this text , and for providing their data points for comparison . we also thank the referee , felipe menanteau , for his insightful comments and suggestions for improving this work . we also wish to recognize and acknowledge the highly significant cultural role and reverence that the summit of mauna kea has always had within the indigenous hawaiian community . it is a privilege to be given the opportunity to conduct observations from this mountain .

we present new metallicity estimates for globular cluster ( gc ) candidates in the sd spiral ngc 300 , one of the nearest spiral galaxies outside the local group . we have obtained optical spectroscopy for 44 sculptor group gc candidates with the boller and chivens ( b&c ) spectrograph on the baade telescope at las campanas observatory . there are 2 gcs in ngc 253 and 12 objects in ngc 300 with globular - cluster - like spectral features , 9 of which have radial velocities above 0 km s . the remaining three , due to their radial velocities being below the expected 95% confidence limit for velocities of ngc 300 halo objects , are flagged as possible foreground stars . the non - clusterlike candidates included 13 stars , 15 galaxies , and an h region . one gc , four galaxies , two stars , and the h region from our sample were identified in archival hubble space telescope images . for the gcs , we measure spectral indices and estimate metallicities using an empirical calibration based on milky way gcs . the gcs of ngc 300 appear similar to those of the milky way . excluding possible stars and including clusters from the literature , the gc system ( gcs ) has a velocity dispersion of 68 km s , and has no clear evidence of rotation . the mean metallicity for our full cluster sample plus one literature object is [ fe / h ] = , lying above the relationship between mean gc metallicity and overall galaxy luminosity . excluding the three low - velocity candidates , we obtain a mean [ fe / h ] = , still higher than expected , raising the possibility of significant foreground star contamination even in this sample . visual confirmation of genuine gcs using high - resolution space - based imagery could greatly reduce the potential problem of interlopers in small samples of gcss in low - radial - velocity galaxies .

0810.4852

globular clusters ( gcs ) , relics of some of the earliest and/or most violent phases of star and galaxy formation , can be analyzed to understand how various types of galaxies formed , as described in the review by @xcite . two subpopulations of gcs , metal - rich ( red ) and metal - poor ( blue ) , exist in most early - type galaxies . both are thought to be old , but the red metal - rich gcs are thought to be slightly younger @xcite . large spiral galaxies , including the milky way @xcite and andromeda @xcite , also have metal - rich and metal - poor gc subpopulations . in smaller , later - type galaxies , however , often only a metal - poor population is seen @xcite . the three basic models for bimodal gc system ( gcs ) formation in early type galaxies are dwarf galaxy accretion @xcite , in situ formation @xcite , and gas - rich mergers @xcite . in the last scenario , the original spiral galaxies likely late - type spirals would provide the entire metal - poor gc population , and the metal - rich gc population would form in the merger . some evidence for gcs resulting from every one of these processes exists , but the properties of both blue and red gc populations have been found to scale with the mass of the host galaxy @xcite , indicating that the accretion model alone can not account for the majority of blue gcs in bright galaxies . in any case , at high redshift and with a hierarchical merging scenario as the main process of galaxy formation , the distinction between these models fades . the nearby sculptor group ( or filament ) is home to several late - type galaxies ( hubble types sc im ) , including ngc 253 ( the largest ) , ngc 300 , ngc 55 , and ngc 45 . at a distance of 1.9 mpc @xcite , ngc 300 is one of the nearest spiral galaxies outside of the local group . as a nearby example of a late - type , midsize spiral galaxy , it is especially useful for understanding the sparse and thus relatively poorly studied gcss of late - type galaxies . @xcite studied the star clusters in ngc 300 photometrically and identified 17 objects as gc candidates based on their size , shape , and color . @xcite performed photometry on fields in six sculptor group galaxies and identified gc candidates via shape and color , and then obtained spectra for 19 gcs . six of these spectroscopically confirmed clusters had high enough signal - to - noise spectra for metallicity determination via lick / ids indices . seven of the gcs olsen et al . spectroscopically confirmed were in ngc 300 , and two had spectroscopically determined metallicities . in this paper we present new metallicities derived from spectroscopy of gcs in sculptor group galaxies , primarily ngc 300 , using @xcite and @xcite as reference catalogs . we compare the metallicities of ngc 300 gcs to those of m31 and milky way clusters , and also determine ngc 300 s place on the galaxy luminosity - gcs metallicity relation of @xcite . we obtained spectra of 44 sculptor group cluster candidates with the boller & chivers ( b&c ) spectrograph on the baade telescope from 2002 november 6 to 2002 november 9 . we chose to observe all 17 gc candidates from @xcite , plus 25 @xcite ngc 300 gc candidates with no previous spectra ( except for ngc 300ax ) observed in order of increasing magnitude from brightest to faintest . we also observed the two brightest unconfirmed olsen et al . gc candidates in ngc 253 . nineteen of our 27 olsen et al . cluster candidates are among the 38 sculptor group gc candidates for which we also obtained _ jhk _ photometry with the panic camera on the baade 6.5 m telescope ( * ? ? ? * paper i ) . figure 1 shows the locations of the ngc 300 objects we observed with respect to a schematic of the ngc 300 disk . the advantages of using the long - slit b&c spectrograph rather than a fiber spectrograph are good coverage in blue optical wavelengths and better sky subtraction ( due to the sky background being determined at the same time and place as the object exposure ) . also , the sky at las campanas has significantly weaker telluric emission lines than less isolated telescope locations . our spectra were taken with the 600 l mm@xmath0 grating blazed at 5000 . the wavelength range is about 3700 - 6860 . peak count rates occurred between 5000 and 5500 for objects with typical gc colors . the dispersion of the spectra is @xmath31.6 pixel@xmath0 , and the resolution of the spectra is about 5 . the readout noise is 3.1 e@xmath4 . total exposure times were as low as 180 s for very bright objects and as high as 3600 s for faint objects , with 600 - 2400 s exposure times for most objects . exposure times under 600 s were usually done as single exposures , while longer exposures were usually broken up into 2 - 4 exposures of 600 - 900 s each . seeing ranged from 0.6 - 1.9@xmath5 over the course of obesrvations , with 1@xmath5 being typical , and the slit width used was 1@xmath5 . the median total signal - to - nose ratio ( all wavelengths combined , @xmath33700 - 6875 ) was about 20 per pixel for foreground stars and gc candidates , with values ranging from 4 to over 100 . if background galaxies , easily distinguished from these objects by their radial velocities , are included , the mean s / n per pixel is about 15 . bias subtraction ( with an averaged bias image ) , dark count correction , and flatfielding were done with iraf s ccdproc task , and spectra were extracted using iraf s apall task . other tasks in the specred and onedspec packages were used to combine spectra , calibrate wavelengths , flux - calibrate the spectra using standard star observations taken each night , and eliminate pollutants such as cosmic rays and improperly - subtracted telluric lines . velocities were determined with the xcsao task in the rvsao package @xcite in iraf using a list of standard sao templates useful for the expected features of gcs , background galaxies , and foreground stars . included are four templates derived from m31 gcs ( m31_a_temp , m31_f_temp , m31_k_temp , and fglotemp ) , three galaxy absorption templates ( fm32temp , habtemp90 , and fn4486btemp ) , two composite stellar absorption templates ( fabtemp97 and fallstars ) , a synthetic h region / galaxy emission template ( hemtemp0.0 ) , and a synthetic ca h&k absorption template ( hkabstemp ) . for each object , we list a heliocentric velocity determined from the template with the highest r value . objects labeled as clusters usually had spectra similar to g and early k giants . mid - type k stars ( @xmath3 k3-k5 ) typically differed visibly from gc candidates in having a redder spectral energy distribution and strong mg and ca42 features . in order to provide a systematic , quantitative classification scheme to remove k dwarfs from the pool of gc candidates , we focused on the strong ca42 feature and compared it to two other spectral features , g43 and h@xmath6 ( using the h@xmath6a index definition in @xcite ) . the difference between the ca42 and h@xmath6 indices was used by @xcite to distinguish between k dwarfs and gc candidates . in our own samples , the ca42 index also appeared strong in comparison to the g43 index in objects that had the spectral features of mid - k - type stars . to quantify the differences in these features between gcs and k dwarfs , we measured the ca42 , h@xmath6a , and g43 indices in the 41 @xcite milky way gc spectra and in 16 k dwarfs from the jones coudfeed spectral library described in @xcite . the k dwarfs range in type from k0 to k8 . figure 2 shows histograms of ca42-g43 and ca42-h@xmath6a ratios for the milky way gcs and jones coudfeed k dwarfs respectively . there is some overlap between the distributions of ca42-g43 for gcs and k dwarfs , particularly among early k - type stars ( k0-k2 ) , but this index appears useful for screening out mid - k - type stars in our sample . the ca42-h@xmath6a ratio seems more effective in discriminating between gcs and k dwarfs , with the only k dwarf overlapping the mw gc range of values being a k8 star , which in our ngc 300 sample would be easily distinguishable from a gc . the milky way gcs had a mean ca42-g43 value of -0.057 with a @xmath7 of 0.024 and a mean ca42-h@xmath6a value of 0.016 with a @xmath7 of 0.075 . for an object to be considered a gc candidate , we insisted that it have both ca42-g43 @xmath8 0.05 ( the value for ngc 300 - 05 which , despite having a ca42-g43 ratio more than 3@xmath7 above the mw gcs , has been visually confirmed as a gc via hst archive images ) and ca42-h@xmath6a @xmath8 0.241 ( @xmath9 above the milky way gc mean ) . this analysis excludes ngc 300cr and ngc 300df on the basis of ca42-h@xmath6a but not ca42-g43 ( which one might expect for k0-k2 dwarfs ) , and ngc 300ax , previously classified by @xcite as a gc , on the basis of ca42-g43 but not ca42-h@xmath6 . radial velocity can provide limited additional information on the likelihood of an object being a star or a gc . since the heliocentric radial velocity of ngc 300 is fairly low + 142 @xmath10 4 km s@xmath0 @xcite it is impossible to eliminate all stars and keep all gcs in the sample if a strict velocity cut is used to exclude candidates . hence , we can not automatically label objects with clusterlike spectra but very low radial velocities as definite stars . the velocity dispersion in ngc 300 s halo is estimated by @xcite to be 60 km s@xmath0 . if the milky way halo were non - rotating , the average expected radial velocity of a galactic star in the direction of ngc 300 would be about + 40 km s@xmath0 , and the velocity dispersion of the milky way halo is about 130 km s@xmath0 ( as viewed from the center of the milky way ) . so while there s a 95% chance that an ngc 300 halo object will have a radial velocity greater than 20 km s@xmath0 , milky way halo stars can very easily have velocities similar to those of genuine ngc 300 gcs . this means that most objects with negative radial velocities will be milky way stars , but those with positive radial velocities could easily be either foreground stars or ngc 300 clusters , and only spectral features or high - resolution imaging can distinguish them . on the basis of radial velocity , we flag three objects with spectral features similar to gc candidates - ngc 300 cm , ngc 300co , and ngc 300db - as possible stars . all have radial velocities less than 0 km s@xmath0 . figure 3 shows the correlations between our two indicators of star vs. gc status , ca42-g43 and ca42-h@xmath6a , and the radial velocities of foreground stars and gc candidates in ngc 300 . both trends are high in scatter , demonstrating the difficulty in measuring star vs. gc status on the basis of radial velocity . the correlation between ca42-h@xmath6a and radial velocity is somewhat stronger , probably due to its greater sensitivity to k0-k2 dwarfs compared to ca42-g43 . the best - fit velocity template also hints at whether an object is likely to be a star or a cluster . the templates we used included several templates developed using intermediate - age and old stellar populations ( m31_a_temp , m31_f_temp , m31_k_temp , fglotemp , and fm32temp ) , a couple templates suited for field stars ( fallstars , fabtemp97 ) , and a couple synthetic templates ( hemtemp0.0 for h emission and hkabstemp for ca h & k absorption ) . most gcs were best fit by the m31 f and k templates , while many foreground dwarfs were best fit by the stellar absorption templates . figure 4 shows sample spectra of an old gc , a typical foreground k star , and a background galaxy . the region shown in the plot covers wavelengths from 3700 to 5300 , containing key features such as cnb , ca 4227 , g4300 , h@xmath11 , mgb , and several fe features . one can see the k star s strong ca 4227 and mgb features , compared to the relatively weak ca 4227 and mgb and strong g4300 of the gc . one can also see that in the foreground star , the continuum on the blue side of the mg2 feature tends to be notably fainter than on the red side . the galaxy is recognizable by its visibly redshifted ca h&k lines . the names , heliocentric cross - correlation velocities , and object types ( background galaxy , foreground star , cluster candidate ) are shown in table 1 . the objects with number designations 1 - 17 are from kim et al . 2002 ; the rest are from olsen et al . 2004 . of the 17 kim et al . candidates , only five objects 2 , 3 , 4 , 5 , and 12 appear to be old gcs . note that object 3 is also ngc 300a in olsen et al.s catalog , and has been proven a gc and spectroscopically evaluated by olsen and collaborators . most of the remaining kim et al . objects were background galaxies , except for one h region ( object 6 ) and three probable ngc 300 stars ( ngc 300 - 09 , m - type ; ngc 300 - 10 , f - type ; and ngc 300 - 15 , a - type ) . since gcs in ngc 300 can easily be resolved with hst , we searched the hst acs and wfpc2 archives for images of our ngc 300 objects . the images were from programs go-8584 , go-9162 , go-9492 , and go-10915 . we found eight kim et al . objects in the archival hst images : objects 1 , 5 , 6 , 7 , 8 , 9 , 10 , and 11 . these eight objects , imaged in f555w for the acs images of objects 1 , 5 , 6 , 7 , and 11 and f606w for the wfpc2 images of objects 8 , 9 , and 10 , are shown in figure 5 . generally , their appearances match what we expect based on their spectra . objects 1 , 7 , and 11 are spiral galaxies ; object 8 is an early - type galaxy ; object 6 is an h complex ; object 5 appears to be a genuine gc ; and objects 9 and 10 , initially considered as possible open clusters , appear pointlike but non - saturated and without diffraction spikes . their @xcite v magnitudes are 19.73 for ngc 300 - 09 ( which has an m - type spectrum ) and 19.12 for ngc 300 - 10 ( which has an f - type spectrum ) , consistent with supergiant stars in ngc 300 . they both have many fainter stars surrounding them , so may also be open clusters or stellar blends . kim et al . selected their gc candidates by using a color cut ( 0.3 @xmath12 @xmath13 @xmath12 2.0 ) and morphological considerations , including visual inspection of the shapes and brightness profiles . they assigned three `` classes '' to the quality of gc candidates , class 1 being most likely to be gcs and class 3 being least likely . interestingly , class 2 yielded the highest fraction of objects that turned out to be gcs : 3 out of 7 had gc spectra , as compared to 1 out of 4 for class 1 and 1 out of 6 for class 3 . kim et al . reported using morphological considerations to determine the probability of an object being a gc . if they required that the clusters be resolved from the ground to be class i objects , they may have actually shifted the balance out of favor of true gcs and in favor of galaxies , since a typical gc at the distance of ngc 300 is close to 1@xmath5 in diameter , which is the limit imposed by typical ground - based seeing . of the olsen et al . candidates , 11 objects appeared to be stars , including ngc 300ax , which olsen et al . had identified and analyzed as a gc . its spectrum has wide spectral lines consistent with a dwarf of late k to early m type : a wide but shallow mgb ( 5180 ) feature and a wide and deep nad ( 5895 ) feature , as well as having a strong ca 4227 feature as compared to the g4300 feature much like other k dwarfs . in addition to this , 3 more gc candidates with ca42-g43 @xmath8 0.05 and ca42-h@xmath6a @xmath8 0.241 have been flagged as possible stars on the basis of radial velocities less than 0 km s@xmath0 . our gc velocities 9 highly probable clusters and 3 low - velocity possible clusters can be combined with seven previously observed gc velocities , five from @xcite and two from unpublished data contributed by one of us ( k. olsen ) , to obtain a gc velocity dispersion and simple mass estimate for ngc 300 . the velocities of the seven previously observed clusters are listed in table 2 . figure 6 shows the velocity histogram of all gc candidates , both previously observed by olsen et al . and observed in this paper , including those with v@xmath120 . if we include all highly probable and possible cluster candidates 19 objects in all we have a mean velocity of 79 km s@xmath0 and a velocity dispersion , corrected for the mean uncertainty of the velocities , of 92 km s@xmath0 . if we exclude the three v@xmath120 objects , we have 16 objects with a mean velocity of 107 km s@xmath0 and an uncertainty - corrected velocity dispersion of 68 km s@xmath0 . this latter dispersion is closer to the @xcite theoretical velocity dispersion estimate of 60 km s@xmath0 than is the velocity dispersion of all 19 objects . we can also investigate whether there is evidence for rotation among our gcs . figure 7 shows the velocity of cluster candidates as a function of position angle , along with a fit to the rotation of all candidates not labeled as possible stars and a representation of the @xcite h rotation curve ( with the velocity of 90 km s@xmath0 reduced by a factor of sin(i ) with i = 42.3 ) . we fit a function of the form @xmath14 to the velocities and position angles of the 16 cluster candidates with v@xmath150 , presumed to be the best cluster candidates . @xmath16 is the central velocity , @xmath17 is the projected rotational velocity ( the actual rotational velocity times sin(i ) ) , @xmath18 is the position angle of the cluster in degrees measured east from north , and @xmath19 is the position angle of the rotation axis . using a nonlinear least - squares fit with weights equal to @xmath20 , fitted to all v@xmath120 clusters , we found @xmath21 km s@xmath0 , @xmath22 km s@xmath0 , and @xmath23 . however , this rotation curve seems exceptionally strong and out of phase with the h rotation , and the uncertainty - corrected dispersion of the gc velocities minus their predicted rotational velocities is actually considerably worse than our previous result ( 80 km s@xmath0 ) . an unweighted fit to the same points gives a similarly unsatisfactory result : @xmath24 km s@xmath0 , @xmath25 km s@xmath0 , and @xmath26 , and still a worse velocity dispersion than without rotation ( 81 km s@xmath0 ) . we therefore conclude that there is no sign of rotation in our gc sample . we use the projected mass estimator @xcite to estimate the total mass of ngc 300 , both using the means of the total and `` good '' samples and the mean velocity of ngc 300 itself . the equation for the projected mass estimator is as follows : @xmath27 here we use @xmath28 ( for isotropic orbits , recommended by @xcite and @xmath29 . using the published mean velocity of ngc 300 itself gives @xmath30 for the total sample , and @xmath31 for the v@xmath150 subsample . the total sample reaches a maximum projected radius of 12.1 kpc , and the v@xmath150 sample reaches a maximum of 11.6 kpc . uncertainties were estimated using the bootstrap method with 1000 resamplings . @xcite , using h rotation curve data , calculate a mass of @xmath32 out to 10.6 kpc ; @xcite give mass - model - based estimates for the mass of ngc 300 ranging from @xmath33 . overall , our masses run large but are not outside the realm of possibility . we measured 26 spectral indices , including the 12 indices calibrated in @xcite using the @xcite bandpass definitions and the diagnostic h@xmath6a index with the @xcite bandpass definition . eight of the brodie and huchra indices are mostly similar to the lick / ids bandpass definitions , differing slightly from those of @xcite and @xcite . all spectral indices not in @xcite were measured using the bandpass definitions of @xcite . all bandpasses were shifted to the uncorrected , geocentric radial velocities of our objects . our indices are not true lick indices because our spectra have not been degraded to match the resolution of the lick star spectra , and we did not observe any lick standard stars to calibrate the fluxes and indices precisely . however , we did observe a template galaxy ( ngc 1052 ) that was also observed by @xcite , and found the overlapping indices to be similar to theirs . all indices are given in magnitudes , as in brodie and huchra , for easy comparison with their m31 spectral indices and index - metallicity calibration methods . converting to equivalent width requires multiplication by the bandwidth of the central index band shifted to the object s velocity . we determined metallicities using a method based on the @xcite prescription , but with new index - vs .- metallicity calibrations determined from the 41 milky way spectra from @xcite , degraded to match the 5 resolution of our spectra . we determined linear regressions for metallicity ( from @xcite for each milky way gc ) as a function of index strength for all our indices , as well as the `` range '' @xmath34 defined as in @xcite , and the @xmath35 representing the scatter about the regression line . we used the @xmath34 and @xmath35 values to construct weights similar to those in @xcite : @xmath36 with @xmath37 being the photometric uncertainty in the entire index , the quadrature sum of the uncertainties in each band . to calculate @xmath38 , the measure of our ability to measure the same index repeatedly for the same object over the course of different nights and conditions , we used spectral index measurements for the standard star ltt 9239 , which we observed each night . we modified the formula for the uncertainty in a band ( continuum or feature value for an index ) from @xcite to the following : @xmath39 where @xmath40 is the number of object counts per pixel , @xmath41 is the number of sky counts per pixel , and @xmath42 is the read noise . the weighted average and uncertainty in the weighted average were determined as in @xcite . our choice of indices to use in the weighted average was determined by considering the correlation coefficient in each milky way linear fit , the signal - to - noise ratio in the ngc 300 objects , and the degree of scatter in ngc 300 index - index relations . the final set of indices - mg2 , fe5270 , fe5335 , ca4227 , g4300 , @xmath6 , ca4455 , and cn2 - all had relatively high correlation coefficients in the milky way index - metallicity relations and relatively high signal - to - noise in typical ngc 300 clusters . table 3 lists measurements and uncertainties of the 12 indices calibrated in brodie and huchra and the h@xmath6a index used to help screen out foreground stars , and table 4 gives the measurements and uncertainties of the remaining 13 indices . table 5 lists the linear relationships between index and metallicity for the 8 indices we chose along with @xmath38 measurements based on the standard star ltt 9239 , and table 6 lists derived metallicities for our clusters , along with five spectroscopically determined metallicities for sculptor group gcs from @xcite . the highest @xmath38 for spectral features used to measure metallicity was 0.095 ( the @xmath6 index , in the region with the lowest signal - to - noise ratio and spanning a broad range of wavelengths ) and the lowest was 0.001 ( fe5270 , a narrow feature in the region with the highest signal - to - noise ratio ) . figures 8 - 12 compare various spectral indices to those of andromeda galaxy gcs @xcite and the schiavon et al . milky way gcs . figure 8 shows that the cnr index lies below typical values for andromeda and milky way clusters , while the cnb index , traditionally considered a more reliable indicator of actual cn content , looks about the same as the other glaaxies . figure 9 shows a close - up of the cnb region for averages of 10 milky way gcs and 10 ngc 300 gcs matched for fe52 strength . the cnb depths look fairly similar , although a tilt due to the high reddening of many of the mw gcs is visible . fe52 metallicities , as seen in figure 10 , look similar in all three galaxies . the g - band ( figure 11 ) seems about 0.02 mag high compared to the milky way and andromeda , although overall more similar to the milky way . and finally , h@xmath11 , the index that is sensitive to age and inversely correlated with metallicity , is shown in figure 12 . ngc 300 gcs show a mean and distribution more similar to the milky way than to andromeda , suggesting old clusters similar to those of the milky way . the deficit in cnr in our sample does not necessarily indicate a nitrogen deficiency in the gcs of ngc 300 , especially given the apparently normal cnb . andromeda gcs had once been thought to have nitrogen excesses compared to milky way gcs @xcite , but this was later challenged when the metal - rich milky way gcs were taken into account @xcite . instead , puzia et al . found that metal - rich gcs in both andromeda and the milky way were nitrogen - enhanced compared to the general old stellar populations of galaxies , which they propose may be due to low - mass gc stars accreting carbon and/or nitrogen from the agb winds of dying higher - mass stars in the cluster . a deficiency in cnr in our sample , therefore , may suggest a high rate of foreground star contamination especially among metal - rich objects , since stars should lack nitrogen compared to metal - rich gcs of similar metallicity . however , cnb is normal , and cnr is a weak feature that can easily be trumped by noise , so there may not be any cause for alarm . two of the four objects on the lower right of the cnr vs. mg2 plot ( figure 8(b ) ) , with mg2 indices above + 0.125 and lacking strong cnr , are among the low - velocity candidates . overall , the ngc 300 gcs seem similar to those of the milky way . figure 13 shows the metallicity distribution of the ngc 300 gcs as compared to the milky way , andromeda , and m33 , including ngc 300r from @xcite . with all candidates included , the distribution appears unimodal and rather metal - rich . with the three low - velocity objects excluded , the distribution appears to still be fairly metal - rich . the ngc 300 cluster sample is too small and possibly too contaminated to definitively determine bimodality or a lack thereof . an updated version of the @xcite relation between mean gcs metallicity and total galaxy luminosity was created using relatively recent published galaxy distance and spectroscopic gc metallicity information from the literature on several galaxies , along with contemporary 2mass k and @xcite b magnitudes found on the nasa / ipac extragalactic database ( ned ) . table 7 summarizes the data , gathered from a wide variety of studies of gcss using different selection criteria . the relationship between the mean gcs metallicity and absolute b magnitude is shown in figure 14 , and the relationship between mean gcs metallicity and absolute 2mass k magnitude is shown in figure 15 . ngc 300 , with its calculated mean gc metallicity of @xmath43 , lies above both the early - type and late - type galaxies on the plot . the high metallicity calculated for ngc 300 could be due to most of the `` higher metallicity '' cluster candidates being foreground stars , as suggested above in the discussion of cnr weakness . if the three clusters deemed to be possible foreground stars are eliminated , the mean metallicity of the remaining ten clusters ( our clusters plus ngc 300r ) is @xmath44 , which is still higher than expected for a galaxy the size of ngc 300 . besides possible foreground star contamination , color selection in the candidate samples may have created a bias in favor of metal - rich , intrinsically red clusters that drastically increases their relative numbers in our tiny sample . this is unlikely for the kim et al . clusters , but the @xcite color cuts were found to be @xmath30.2 mag redder than initially thought due to a calibration error . it is also worth noting that the spectroscopic metallicity estimates used in our metallicity - luminosity relations do not distinguish among the @xmath45 $ ] values that may differ among galaxy types . the gcs of ngc 300 appears to have features similar to that of the milky way . its place on the metallicity - luminosity relation is consistent with the positive correlation between galaxy luminosity and mean gc metallicity , although the average metallicity of our ngc 300 cluster sample is higher than expected . eliminating all ambiguous objects from our cluster sample gives a mean metallicity slightly more consistent with the milky way and andromeda . the gcs , excluding those deemed possible stars , have a velocity dispersion of 68 km s@xmath0 , and may be rotating , although the evidence does not strongly favor rotation . the high mean metallicity of our cluster sample may be primarily a result of the difficulty in screening out stars without the ability to use radial velocity cuts . the problem of mistaking foreground stars ( and giant stars within the target galaxy ) for gcs could easily be remedied with more space - based imaging of ngc 300 . gcs could then be identified visually or with more accurate size and shape determinations before spectroscopic study . we would like to thank the lco staff for helping us use the magellan telescopes and instruments . this work has been supported by the smithsonian institution and harvard college observatory . we also thank ricardo schiavon and collaborators for making their milky way spectra available to the public at http://www.noao.edu / ggclib/. ashman , k. m. , and zepf , s. e. 1992 , , 384 , 50 . barmby , p. , huchra , j. p. , brodie , j. p. , forbes , d. a. , schroder , l. l. , and grillmair , c. j. , 2000 , , 119 , 727 . beasley , m. a. , bridges , t. , peng , e. , harris , w. e. , harris , g. l. h. , forbes , d. a. , mackie , g. 2008 , , 386 , 1443 . beasley , m. a. , forbes , d. a. , brodie , j. p. , and kissler - patig , m. 2004 , , 347 , 1150 . bersier , d. 2000 , , 543 , l23 . brodie , j. p. , and huchra , j. p. 1990 , , 362 , 503 . brodie , j. p. , and huchra , j. p. 1991 , , 379 , 157 . brodie , j. p. , and strader , j. 2006 , , 44 , 193 . burstein , d. , faber , s. , gaskell , m. , and krumm , n. 1984 , , 287 , 586 . cappellari , m. et al . 2006 , , 366 , 1126 . carignan , c. and freeman , k. c. 1985 , , 294 , 494 . chandar , r. , whitmore , b. , and lee , m. g. 2004 , , 611 , 220 . chomiuk , l. , strader , j. , and brodie , j. 2008 , , 136 , 234 . cohen , j. g. , blakeslee , j. p. , and ct , , , 592 , 866 . ct , p. , marzke , r. o. , and west , m. j. 1998 , , 501 , 554 . de vaucouleurs , g. , de vaucouleurs , a. , corwin jr . , h. g. , buta , r. j. , paturel , g. , and fouqu , p. 1991 , third reference catalog of bright galaxies , version 3.9 ( new york : springer - verlag ) ferrarese , l. et al . 2000 , , 529 , 745 . forbes , d. a. , brodie , j. p. , and grillmair , c. j. 1997 , , 113 , 1652 freedman , w. 1990 , , 355 , l35 freedman , w. l. , hughes , s. m. , madore , b. f. , mould , j. r. , lee , m. g. , stetson , p. , kennicutt , r. c. , turner , a. , ferrarese , l. , ford , h. , graham , j. a. , hill , r. , hoessel , j. g. , huchra , j. , and illingworth , g. d. 1994 , , 427 , 628 . galleti , s. , bellazzini , m. , and ferraro , f. r. 2004 , a&a , 423 , 925 . gieren , w. , pietrzynski , g. , walker , a. , bresolin , f. , minniti , d. , kudritzki , r. , udalski , a. , soszinski , i. , fouqu , p. , storm , j. , and bono , g. 2004 , , 128 , 1167 . harris , w. e. 1996 , , 112 , 1487 . heisler , j. , tremaine , s. , and bachall , j. n. 1985 , , 298 , 8 . huchra , j. p. , brodie , j. p. , and kent , s. m. 1991 , , 370 , 495 . jarrett , t. h. , chester , t. , cutri , r. , schneider , s. e. , and huchra , j. p. 2003 , aj , 125 , 525 . jensen , j. b. , tonry , j. l. , barris , b. j. , thompson , r. i. , liu , m. c. , rieke , m. j. , ajhar , e. a. , and blakeslee , j. p. 2003 , , 583 , 712 . kang , a. , kim , j. , shin , i. , chun , s. , kim , h. , and sohn , y. 2007 , jass , 24 , 203 . kim , s.c . , sung , h. , and lee , m.g . 2002 , jkas , 35 , 9 . kissler - patig , m. , brodie , j. p. , schroder , l. l. , forbes , d. a. , grillmair , c. j. , and huchra , j. p. 1998 , , 115 , 105 . kundu , a. and whitmore , b. c. 1998 , , 116 , 2841 . kurtz , m. j. and mink , d. j. 1998 , , 110 , 934 . larsen , s. s. and brodie , j. p. 2002 , , 123 , 1488 . larsen , s. s. , brodie , j. p. , beasley , m. a. , and forbes , d. a. 2002 , , 124 , 828 . larsen , s. s. , brodie , j. p. , huchra , j. p. , forbes , d. a. , and grillmair , c. j. 2001 , , 121 , 2974 . lee , m. g. , kim , e. , and geisler , d. 1998 , , 115 , 947 . leitherer , c. et al . 1996 , pasp , 108 , 996 . mcconnachie a.w . , irwin m.j . , lewis g.f . , ibata , r.a . , chapman , s.c . ferguson a.m.n . , tanvir n. , 2004 , , 351 , l9 nantais , j. b. , huchra , j. p. , barmby , p. , olsen , k. a. g. , and jerret , t. h. 2006 , aj , 131 , 1416 . norris , m. a. , sharples , r. m. , bridges , t. , gebhart , k. , forbes , d. a. , proctor , r. , raul faifer , f. , carlos forte , j. , beasley , m. a. , zepf , s. e. , and hanes , d. a. 2008 , , 385 , 40 . olsen , k. a. g. , miller , b. w. , suntzeff , n. b. , schommer , r. a. , and bright , j. 2004 , , 127 , 2674 . perelmuter , j. , brodie , j. p. , and huchra , j. p. 1995 , , 110 , 620 . proctor , r. n. , forbes , d. a. , brodie , j. p. , and strader , j. 2008 , , 385 , 1709 . puche , d. , carignan , c. , and bosma , a. 1990 , , 100 , 1468 . puzia , t. h. , saglia , r. p. , kissler - patig , m. , maraston , c. , greggio , l. , renzini , a. , and ortolani , s. 2002 , a&a , 395 , 45 . rejkuba , m. 2004 , a&a , 413 , 903 . rhee , m. and chun , m. 1992 , jkas , 25 , 11 . schiavon , r. p. , rose , j. a. , courteau , s. , and macarthur , l.a . 2005 , apjs , 160 , 163 . schrder , l. l. , brodie , j. p. , kissler - patig , m. , huchra , j. p. , and phillips , a. c. 2002 , , 123 , 2473 . sharina , m. e. , afanasiev , v. l. , and puzia , t. h. 2006 , , 372 , 1259 . sikkema , g. , peletier , r. f. , carter , d. , valentijn , e. a. , and balcells , m. 2006 , a&a , 458 , 53 . strader , j. , brodie , j. p. , and forbes , d. a. 2004 , , 127 , 295 . strader , j. , brodie , j. p. , forbes , d. a. , beasley , m. a. , and huchra , j. p. 2003 , , 125 , 1291 . tonry j.l . , dressler a. , blakeslee j.p . , ajhar e. a. , fletcher a.b . , luppino g.a . , metzger m.r . , and moore c.b . 2001 , , 546 , 681 . trager , s. c. , worthey , g. , faber , s. m. , burstein , d. , and gonzalez , j. j. 1998 , apjs , 116 , 1 . worthey , g. , faber , s. , gonzalez , j. j. , and burstein , d. 1994 , apjs , 94 , 687 . zinn , r. 1985 , , 283 , 424 . ngc 253b & 00:46:50.70 & -25:26:38.90 & & 209 & 43 & 7.87 & m31_a_temp & gc + ngc 253ad & 00:47:31.91 & -25:26:42.53 & & 65 & 40 & 7.42 & fn4486btemp & gc + ngc 300cz & 00:53:26.93 & -37:41:55.54 & & 81 & 19 & 16.6 & fallstars & star + ngc 300cy & 00:53:27.89 & -37:55:21.48 & & 85377 & 24 & 6.73 & hemtemp0.0 & ( em.?)gal + ngc 300dt & 00:53:40.30 & -37:49:34.38 & & 127 & 40 & 8.68 & habtemp90 & gc + ngc 300db & 00:53:55.35 & -37:23:42.98 & & -123 & 46 & 5.95 & m31_f_temp & gc or star + ngc 300ds & 00:53:56.12 & -37:34:49.40 & & 169 & 21 & 13.27 & fabtemp97 & star + ngc 300dd & 00:53:56.47 & -37:58:46.59 & & 11483 & 29 & 4.26 & hemtemp0.0 & gal ? + ngc 300 cm & 00:54:05.51 & -37:42:06.11 & & -18 & 19 & 14.24 & m31_k_temp & gc or star + ngc 300 - 01 & 00:54:22.07 & -37:43:22.1 & 1 & 49374 & 34 & 8.78 & habtemp90 & gal * + ngc 300ck & 00:54:21.84 & -37:32:07.82 & & 21 & 13 & 19.27 & fallstars & gc + ngc 300cj & 00:54:24.13 & -37:25:46.17 & & 23 & 17 & 15.63 & m31_k_temp & gc + ngc 300 - 08 & 00:54:57.41 & -37:33:55.30 & 2 & 16797 & 22 & 14.43 & habtemp90 & gal * + ngc 300 - 02 & 00:54:35.80 & -37:43:05.60 & 2 & 59 & 21 & 15.11 & habtemp90 & gc + ngc 300 - 15 & 00:55:21.81 & -37:45:40.10 & 2 & 106 & 22 & 20.12 & m31_a_temp & star + ngc 300 - 03 & 00:54:40.02 & -37:49:18.71 & 3 & 78 & 28 & 10.04 & m31_f_temp & gc + ngc 300 - 04 & 00:54:45.62 & -37:43:41.00 & 2 & 161 & 28 & 9.14 & m31_f_temp & gc + ngc 300 - 14 & 00:55:17.67 & -37:36:57.90 & 2 & 16995 & 36 & 8.68 & fallstars & gal + ngc 300 - 05 & 00:54:49.35 & -37:43:30.50 & 2 & 120 & 35 & 8.08 & m31_f_temp & gc * + ngc 300 - 09 & 00:54:57.57 & -37:36:37.90 & 2 & 214 & 14 & 18.08 & fglotemp & star * + ngc 300 - 11 & 00:55:01.44 & -37:37:21.80 & 1 & 38350 & 17 & 8.83 & hemtemp0.0 & em.gal * + ngc 300cl & 00:54:53.85 & -37:23:57.44 & & 21 & 16 & 17.4 & fabtemp97 & star + ngc 300dc & 00:54:54.66 & -37:58:34.98 & & 68066 & 25 & 5.13 & hemtemp0.0 & em.gal + ngc 300df & 00:54:57.02 & -37:58:52.21 & & -16 & 48 & 5.76 & m31_k_temp & star + ngc 300 - 10 & 00:54:59.03 & -37:36:24.40 & 3 & 173 & 23 & 12.59 & m31_f_temp & star * + ngc 300ax & 00:55:01.49 & -37:47:36.40 & & -27 & 32 & 8.43 & m31_k_temp & star + ngc 300 - 07 & 00:54:53.22 & -37:43:11.1 & 1 & 43722 & 27 & 6.41 & hemtemp0.0 & em.gal * + ngc 300 - 12 & 00:55:11.03 & -37:36:46.10 & 1 & 177 & 18 & 15.19 & m31_f_temp & gc + ngc 300co & 00:55:19.49 & -37:29:38.73 & & -66 & 19 & 14.57 & m31_k_temp & gc or star + ngc 300 - 06 & 00:54:51.67 & -37:39:38.00 & 3 & 156 & 14 & 8.97 & hemtemp0.0 & hii * + ngc 300 - 17 & 00:55:38.79 & -37:46:30.70 & 3 & 16868 & 21 & 13.02 & fallstars & gal + ngc 300 - 16 & 00:55:27.56 & -37:46:09.40 & 3 & 16798 & 29 & 8.03 & m31_k_temp & gal + ngc 300 - 13 & 00:55:11.17 & -37:33:13.90 & 3 & 38182 & 20 & 6.19 & hemtemp0.0 & em.gal + ngc 300cr & 00:55:52.72 & -37:59:04.14 & & -23 & 15 & 17.05 & fallstars & star + ngc 300bd & 00:55:56.79 & -37:52:07.28 & & 68818 & 20 & 6.69 & hemtemp0.0 & gal + ngc 300cq & 00:56:07.09 & -37:57:02.64 & & -79 & 14 & 17.25 & fallstars & star + ngc 300cs & 00:56:07.62 & -37:57:23.07 & & 31 & 16 & 15.85 & fallstars & star + ngc 300da & 00:56:09.49 & -37:31:21.25 & & 49619 & 21 & 7.51 & hemtemp0.0 & em.gal + ngc 300cv & 00:56:13.97 & -37:34:16.27 & & 97 & 17 & 12.33 & fabtemp97 & star + ngc 300ct & 00:56:18.24 & -37:28:48.08 & & 4 & 15 & 17.59 & fabtemp97 & star + ngc 300cp & 00:56:19.03 & -37:29:17.23 & & 35 & 19 & 13.88 & m31_k_temp & gc + ngc 300cx & 00:56:20.30 & -37:59:11.68 & & 68618 & 53 & 5.97 & m31_f_temp & gal + ngc 300cn & 00:56:23.52 & -37:27:37.57 & & 49828 & 17 & 8.89 & hemtemp0.0 & em.gal + ngc 300cu & 00:56:24.62 & -37:52:56.29 & & 18 & 20 & 12.33 & fallstars & star + ngc 300ac & 00:54:02.48 & -37:44:31.64 & 82 & 45 & gc & olsen et al . 2004 + ngc 300ag & 00:54:23.22 & -37:59:17.67 & 120 & 38 & gc & olsen et al . 2004 + ngc 300am & 00:54:07.54 & -37:42:09.23 & 73 & 17 & gc & olsen ( unpublished ) + ngc 300ba & 00:54:05.49 & -37:23:11.24 & 118 & 60 & gc & olsen et al . 2004 + ngc 300 m & 00:54:07.04 & -37:41:10.35 & 18 & 13 & gc & olsen ( unpublished ) + ngc 300r & 00:56:23.13 & -37:33:26.98 & 248 & 15 & gc & olsen et al . 2004 + ngc 300s & 00:53:55.66 & -37:32:37.38 & 249 & 29 & gc & olsen et al . 2004 + ngc 253ad & 0.017 & 0.262 & 0.039 & 0.048 & 0.177 & 0.152 & 0.027 & 0.046 & 0.294 & 0.312 & -0.086 & 0.477 & -0.83 + @xmath7 & 0.060 & 0.071 & 0.037 & 0.022 & 0.025 & 0.043 & 0.035 & 0.041 & 0.120 & 0.121 & 0.069 & 0.031 & 0.066 + ngc 253b & -0.132 & 0.037 & 0.058 & 0.009 & 0.017 & 0.005 & -0.005 & 0.049 & 0.216 & 0.234 & -0.088 & 0.252 & 0.101 + @xmath7 & 0.042 & 0.053 & 0.033 & 0.020 & 0.022 & 0.037 & 0.033 & 0.042 & 0.071 & 0.067 & 0.050 & 0.019 & 0.045 + ngc 300 - 02 & -0.044 & 0.144 & 0.025 & 0.001 & 0.120 & 0.110 & 0.046 & 0.039 & 0.033 & 0.216 & 0.001 & 0.244 & 0.001 + @xmath7 & 0.011 & 0.015 & 0.010 & 0.006 & 0.007 & 0.012 & 0.010 & 0.013 & 0.020 & 0.019 & 0.015 & 0.005 & 0.012 + ngc 300 - 03 & -0.045 & 0.103 & 0.079 & -0.008 & 0.018 & 0.045 & -0.007 & 0.035 & 0.128 & 0.312 & -0.188 & 0.298 & 0.048 + @xmath7 & 0.026 & 0.031 & 0.020 & 0.013 & 0.014 & 0.023 & 0.021 & 0.029 & 0.047 & 0.044 & 0.035 & 0.012 & 0.028 + ngc 300 - 04 & -0.015 & 0.120 & 0.063 & 0.010 & 0.037 & 0.064 & 0.066 & 0.008 & 0.095 & 0.328 & -0.165 & 0.354 & 0.013 + @xmath7 & 0.031 & 0.037 & 0.023 & 0.014 & 0.015 & 0.026 & 0.023 & 0.030 & 0.058 & 0.057 & 0.040 & 0.015 & 0.034 + ngc 300 - 05 & -0.051 & 0.031 & 0.045 & 0.007 & 0.062 & 0.086 & 0.039 & -0.047 & -0.057 & 0.184 & -0.134 & 0.243 & 0.090 + @xmath7 & 0.036 & 0.044 & 0.029 & 0.017 & 0.019 & 0.033 & 0.029 & 0.043 & 0.068 & 0.062 & 0.051 & 0.016 & 0.038 + ngc 300 - 12 & -0.053 & 0.138 & 0.071 & 0.030 & 0.101 & 0.088 & 0.056 & 0.026 & 0.069 & 0.362 & -0.245 & 0.446 & 0.024 + @xmath7 & 0.018 & 0.023 & 0.013 & 0.008 & 0.009 & 0.015 & 0.013 & 0.016 & 0.036 & 0.034 & 0.026 & 0.009 & 0.020 + ngc 300cj & -0.058 & 0.225 & 0.054 & 0.012 & 0.136 & 0.136 & 0.051 & 0.049 & 0.175 & 0.399 & -0.187 & 0.536 & -0.043 + @xmath7 & 0.014 & 0.018 & 0.010 & 0.006 & 0.007 & 0.012 & 0.011 & 0.012 & 0.026 & 0.025 & 0.018 & 0.007 & 0.014 + ngc 300ck & -0.029 & 0.231 & 0.060 & 0.030 & 0.203 & 0.180 & 0.084 & 0.070 & 0.317 & 0.378 & -0.200 & 0.664 & -0.62 + @xmath7 & 0.016 & 0.021 & 0.012 & 0.007 & 0.008 & 0.014 & 0.012 & 0.013 & 0.032 & 0.029 & 0.023 & 0.008 & 0.017 + ngc 300 cm & -0.052 & 0.182 & 0.072 & 0.022 & 0.157 & 0.160 & 0.059 & 0.036 & 0.203 & 0.350 & -0.108 & 0.483 & -0.023 + @xmath7 & 0.018 & 0.023 & 0.014 & 0.008 & 0.009 & 0.016 & 0.014 & 0.016 & 0.034 & 0.033 & 0.023 & 0.009 & 0.019 + ngc 300co & 0.013 & 0.145 & 0.031 & 0.013 & 0.125 & 0.121 & 0.059 & 0.060 & 0.300 & 0.434 & -0.181 & 0.561 & -0.049 + @xmath7 & 0.019 & 0.024 & 0.014 & 0.009 & 0.010 & 0.017 & 0.014 & 0.016 & 0.037 & 0.035 & 0.025 & 0.009 & 0.020 + ngc 300cp & -0.047 & 0.202 & 0.090 & -0.001 & 0.152 & 0.166 & 0.071 & 0.027 & 0.142 & 0.375 & -0.082 & 0.460 & 0.005 + @xmath7 & 0.021 & 0.028 & 0.017 & 0.010 & 0.011 & 0.019 & 0.017 & 0.020 & 0.040 & 0.039 & 0.028 & 0.010 & 0.022 + ngc 300db & -0.071 & 0.262 & 0.092 & 0.002 & 0.078 & 0.114 & -0.007 & -0.026 & -0.010 & 0.268 & -0.022 & 0.226 & 0.002 + @xmath7 & 0.043 & 0.055 & 0.033 & 0.020 & 0.022 & 0.037 & 0.033 & 0.042 & 0.079 & 0.076 & 0.056 & 0.019 & 0.046 + ngc 300dt & -0.006 & 0.200 & 0.059 & 0.008 & 0.117 & 0.177 & 0.037 & 0.087 & 0.065 & 0.268 & -0.024 & 0.269 & -0.052 + @xmath7 & 0.034 & 0.044 & 0.023 & 0.013 & 0.015 & 0.026 & 0.022 & 0.028 & 0.069 & 0.065 & 0.046 & 0.017 & 0.037 + @xmath12ngc 300@xmath15 & -0.038 & 0.165 & 0.062 & 0.011 & 0.109 & 0.120 & 0.046 & 0.030 & 0.122 & 0.323 & -0.128 & 0.399 & -0.004 + @xmath7 & 0.007 & 0.009 & 0.005 & 0.003 & 0.004 & 0.006 & 0.005 & 0.007 & 0.013 & 0.012 & 0.009 & 0.003 & 0.026 + ngc 253ad & 0.041 & 0.102 & 0.088 & 0.048 & 0.056 & 0.060 & 0.052 & 0.036 & 0.036 & 0.058 & -0.014 & 0.008 & 0.028 + @xmath7 & 0.076 & 0.106 & 0.066 & 0.067 & 0.055 & 0.042 & 0.037 & 0.044 & 0.045 & 0.038 & 0.045 & 0.024 & 0.023 + ngc 253b & -0.104 & -0.048 & 0.001 & 0.049 & 0.075 & 0.067 & 0.001 & -0.029 & -0.029 & 0.001 & -0.021 & -0.010 & 0.008 + @xmath7 & 0.055 & 0.076 & 0.052 & 0.053 & 0.046 & 0.036 & 0.033 & 0.042 & 0.042 & 0.037 & 0.043 & 0.024 & 0.023 + ngc 300 - 02 & -0.030 & 0.066 & 0.067 & 0.032 & 0.046 & 0.013 & 0.032 & 0.047 & 0.047 & 0.018 & 0.022 & 0.009 & 0.012 + @xmath7 & 0.015 & 0.020 & 0.014 & 0.015 & 0.013 & 0.011 & 0.010 & 0.013 & 0.014 & 0.012 & 0.015 & 0.008 & 0.008 + ngc 300 - 03 & -0.021 & 0.072 & -0.020 & 0.023 & 0.081 & 0.024 & 0.046 & -0.008 & -0.008 & 0.019 & -0.019 & -0.032 & 0.007 + @xmath7 & 0.034 & 0.045 & 0.032 & 0.032 & 0.028 & 0.022 & 0.021 & 0.027 & 0.027 & 0.024 & 0.029 & 0.016 & 0.015 + ngc 300 - 04 & -0.026 & 0.072 & 0.060 & 0.028 & 0.069 & 0.044 & 0.065 & 0.052 & 0.052 & 0.038 & 0.024 & 0.007 & 0.011 + @xmath7 & 0.042 & 0.055 & 0.038 & 0.038 & 0.032 & 0.025 & 0.023 & 0.028 & 0.029 & 0.027 & 0.032 & 0.018 & 0.017 + ngc 300 - 05 & -0.052 & 0.081 & -0.004 & -0.045 & 0.003 & 0.012 & -0.017 & 0.031 & 0.031 & -0.025 & -0.047 & 0.005 & 0.016 + @xmath7 & 0.049 & 0.062 & 0.045 & 0.047 & 0.040 & 0.031 & 0.030 & 0.036 & 0.037 & 0.036 & 0.043 & 0.024 & 0.023 + ngc 300 - 12 & -0.023 & 0.102 & 0.064 & 0.030 & 0.069 & 0.027 & 0.036 & 0.040 & 0.040 & 0.023 & 0.022 & 0.042 & 0.064 + @xmath7 & 0.024 & 0.032 & 0.021 & 0.023 & 0.018 & 0.015 & 0.014 & 0.017 & 0.017 & 0.015 & 0.018 & 0.010 & 0.010 + ngc 300cj & -0.036 & 0.069 & 0.096 & 0.050 & 0.067 & 0.039 & 0.050 & 0.041 & 0.041 & 0.027 & 0.013 & -0.009 & 0.009 + @xmath7 & 0.018 & 0.024 & 0.017 & 0.017 & 0.014 & 0.012 & 0.011 & 0.013 & 0.014 & 0.012 & 0.014 & 0.008 & 0.008 + ngc 300ck & 0.011 & 0.147 & 0.120 & 0.069 & 0.072 & 0.050 & 0.068 & 0.077 & 0.077 & 0.016 & 0.030 & 0.001 & 0.017 + @xmath7 & 0.020 & 0.028 & 0.018 & 0.019 & 0.016 & 0.013 & 0.012 & 0.015 & 0.016 & 0.013 & 0.016 & 0.009 & 0.009 + ngc 300 cm & -0.026 & 0.144 & 0.083 & 0.071 & 0.053 & 0.004 & 0.026 & 0.052 & 0.052 & 0.001 & 0.030 & 0.020 & 0.005 + @xmath7 & 0.023 & 0.032 & 0.021 & 0.022 & 0.019 & 0.015 & 0.014 & 0.017 & 0.018 & 0.016 & 0.019 & 0.010 & 0.010 + ngc 300co & 0.039 & 0.124 & 0.039 & 0.041 & 0.049 & 0.012 & 0.044 & 0.039 & 0.039 & 0.023 & 0.019 & -0.001 & 0.008 + @xmath7 & 0.024 & 0.034 & 0.022 & 0.024 & 0.020 & 0.016 & 0.015 & 0.018 & 0.019 & 0.016 & 0.019 & 0.010 & 0.010 + ngc 300cp & -0.009 & 0.110 & 0.039 & 0.045 & 0.055 & 0.038 & 0.037 & 0.044 & 0.044 & 0.001 & 0.023 & -0.008 & 0.015 + @xmath7 & 0.027 & 0.038 & 0.026 & 0.027 & 0.022 & 0.018 & 0.017 & 0.021 & 0.022 & 0.019 & 0.023 & 0.012 & 0.012 + ngc 300db & -0.043 & 0.091 & -0.023 & -0.004 & -0.061 & 0.016 & 0.022 & 0.043 & 0.043 & -0.025 & -0.080 & -0.058 & -0.021 + @xmath7 & 0.056 & 0.074 & 0.055 & 0.057 & 0.048 & 0.037 & 0.034 & 0.041 & 0.042 & 0.038 & 0.046 & 0.024 & 0.025 + ngc 300dt & 0.000 & 0.131 & 0.134 & 0.088 & 0.063 & -0.006 & 0.037 & 0.009 & 0.009 & 0.025 & -0.048 & -0.021 & 0.002 + @xmath7 & 0.044 & 0.056 & 0.036 & 0.037 & 0.032 & 0.025 & 0.022 & 0.028 & 0.028 & 0.025 & 0.030 & 0.016 & 0.016 + @xmath12ngc 300@xmath15 & -0.018 & 0.101 & 0.055 & 0.036 & 0.047 & 0.023 & 0.037 & 0.039 & 0.037 & 0.012 & -0.001 & -0.004 & 0.012 + @xmath7 & 0.009 & 0.012 & 0.008 & 0.009 & 0.007 & 0.006 & 0.005 & 0.007 & 0.007 & 0.006 & 0.007 & 0.004 & 0.004 + ca4455 & 25.85 & -2.00 & 0.899 & 0.0387 & 0.006 & 0.003 + fe5270 & 26.50 & -2.31 & 0.888 & 0.0377 & 0.006 & 0.001 + mg2 & 9.83 & -1.89 & 0.886 & 0.1017 & 0.017 & 0.006 + @xmath6 & 4.12 & -2.68 & 0.884 & 0.2428 & 0.041 & 0.095 + fe5335 & 30.36 & -2.22 & 0.879 & 0.0329 & 0.006 & 0.005 + ca4227 & 18.65 & -2.09 & 0.878 & 0.0536 & 0.010 & 0.014 + cn2 & 9.45 & -1.27 & 0.852 & 0.1058 & 0.020 & 0.021 + g4300 & 10.37 & -2.24 & 0.844 & 0.0964 & 0.019 & 0.008 + ngc 55az & -1.76 & 0.43 & ols & 27 + ngc 253q & -1.07 & 0.30 & ols & 5 + ngc 253b & -1.96 & 0.54 & new & 26 + ngc 247a & -1.04 & 0.29 & ols & 64 + ngc 253a & -0.88 & 0.25 & ols & 46 + ngc 253ad & -0.52 & 0.56 & new & 57 + ngc 300dt & -0.89 & 0.66 & new & 13 + ngc 300db & -1.23 & 0.81 & new & 22 + ngc 300 cm & -0.54 & 0.36 & new & 33 + ngc 300ck & -0.07 & 0.25 & new & 42 + ngc 300cj & -0.71 & 0.35 & new & 45 + ngc 300 - 02 & -1.03 & 0.29 & new & n / a + ngc 300 - 03 & -1.61 & 0.47 & new & 48 + ngc 300 - 04 & -1.18 & 0.33 & new & n / a + ngc 300 - 05 & -1.57 & 0.52 & new & n / a + ngc 300 - 12 & -0.93 & 0.23 & new & n / a + ngc 300co & -0.67 & 0.29 & new & 69 + ngc 300cp & -0.59 & 0.29 & new & 116 + ngc 300r & -1.25 & 0.35 & ols & 121 + fornax & -1.71 & 0.08 & 9.28 & 0.136 & -11.38 & & & 32 & 5 & + m87 & -0.89 & 0.16 & 9.56 & 15.631 & -21.38 & 5.81 & -25.16 & 6 & 11 & 17 + m31 & -1.22 & 0.04 & 4.36 & 0.77 & -20.07 & 0.98 & -23.45 & 2 & 12 & 17 + mw & -1.31 & 0.05 & 5.52 & 0.76 & -18.9 & & & 16 & & + m81 & -1.19 & 0.14 & 7.89 & 3.6 & -19.89 & 3.83 & -23.95 & 29 , 26 & 13 & 17 + m33 & -1.55 & 0.37 & 6.27 & 0.847 & -18.37 & 4.1 & -20.54 & 6 & 14 & 17 + m49 & -0.56 & 0.3 & 9.37 & 16.368 & -21.7 & 5.4 & -25.67 & 9 & 11 & 17 + n 147 & -2.05 & 0.4 & 10.47 & 0.594 & -13.4 & 7.2 & -16.67 & 6 & 19 & 17 + n 205 & -1.40 & 0.15 & 8.92 & 0.752 & -15.46 & 5.59 & -18.79 & 6 & 30 & 17 + n 185 & -1.65 & 0.25 & 10.1 & 0.676 & -14.05 & 6.56 & -17.59 & 6 & 23 & 17 + m104 & -1.01 & 0.13 & 8.98 & 9.333 & -20.87 & 4.96 & -24.89 & 22 & 11 & 17 + n 5128 & -0.96 & 0.04 & 7.84 & 3.8 & -20.06 & 3.94 & -23.96 & 3 & 28 & 17 + n 2683 & -1.23 & 0.19 & 10.64 & 7.178 & -18.64 & 6.33 & -22.95 & 27 & 33 , 18 & 17 + n 7457 & -0.97 & 0.11 & 12.09 & 13.996 & -18.64 & 8.19 & -22.54 & 8 & 11 , 7 & 1 + n 1023 & -1.29 & 0.11 & 10.35 & 11.117 & -19.88 & 6.24 & -23.99 & 21 & 11 & 17 + n 1399 & -0.82 & 0.12 & 10 & 19.055 & -21.4 & 6.31 & -25.09 & 20 & 11 & 17 + n 3923 & -0.94 & 0.11 & 10.8 & 22.909 & -21.0 & 6.5 & -25.3 & 24 & 31 & 17 + n 524 & -0.97 & 0.12 & 11.3 & 28.2 & -20.95 & 7.16 & -25.09 & 4 & 10 & 1 + n 300 & -0.94 & 0.12 & 8.95 & 1.93 & -17.48 & 6.38 & -20.05 & 25 , this paper & 15 & 17 +

we give a common generalization of ( 1 ) hassett s weighted stable curves , and ( 2 ) hacking - keel - tevelev s stable hyperplane arrangements .

0806.0881

the moduli space @xmath0 of stable @xmath1-pointed rational curves has many generalizations , beginning of course with @xmath2 . for this paper , however , the following two generalizations will be important : 1 . hassett s moduli @xmath3 of weighted stable @xmath1-pointed curves @xcite , and 2 . hacking - keel - tevelev s moduli @xmath4 of stable hyperplane arrangements @xcite . a _ weight data _ , or simply a _ weight _ , @xmath5 is a collection of @xmath1 of rational ( or real ) numbers @xmath6 . we denote @xmath7 . a weighted stable curve of genus zero is a nodal curve @xmath8 whose dual graph is a tree , together with @xmath1 points @xmath9 satisfying two conditions : 1 . ( on singularities ) @xmath10 the nodes , and whenever some points @xmath11 coincide , one has @xmath12 . 2 . ( numerical ) @xmath13 is ample . in plain words , this means that for every irreducible component @xmath14 of @xmath15 , one has @xmath16 the space @xmath3 is the fine moduli space for flat families of such weighted curves ; it is smooth and projective . a _ stable pair _ @xmath17 is a natural higher - dimensional analogue of the above notion . it consists of a connected equidimensional projective variety @xmath15 together with @xmath1 weil divisors @xmath18 satisfying the following conditions ( see @xcite for more details ) : 1 . ( on singularities ) @xmath15 is reduced , and the pair @xmath19 is slc ( semi log canonical ) , and 2 . ( numerical ) @xmath20 is ample . in @xcite the authors construct a projective scheme , which we will denote @xmath4 , together with a flat family @xmath21 such that every geometric fiber @xmath22 is a stable pair in the above definition , with all coefficients @xmath23 . over an open ( but not dense in general ) subset @xmath24 this gives the universal family of @xmath1 hyperplanes @xmath18 on a projective space @xmath25 such that @xmath18 are in general position . the construction originates in @xcite , see also @xcite . more generally , let @xmath5 be a weight , and @xmath26 be @xmath1 hyperplanes in @xmath27 . then the pair @xmath28 is 1 . _ lc ( log canonical ) _ if for each intersection @xmath29 of codimension @xmath30 , one has @xmath31 , and 2 . _ klt ( kawamata log terminal ) _ if the inequalities are strict , in particular all @xmath32 . ( this is consistent with the standard definitions of the minimal model program . ) the pair @xmath28 is stable in the above definition iff it is lc ( slc being an analog of lc for possibly nonnormal pairs ) and @xmath33 . we call such pairs _ weighted hyperplane arrangements _ , or simply _ lc hyperplane arrangements_. one easily constructs a fine moduli space @xmath34 for them ; it is smooth , of dimension @xmath35 , and usually not complete ( but see theorem [ thm : small - weights ] for the exceptions ) . throughout the paper , we work over an arbitrary commutative ring @xmath36 with identity . the main results of this paper are the three theorems below and the detailed description of the weighted stable hyperplane arrangements given in section [ sec : defin - moduli - univ ] . [ thm : existence ] for each @xmath37 and a rational weight @xmath38 with @xmath39 , there exists a projective scheme @xmath40 together with a locally free ( in particular , flat ) family @xmath41 such that : 1 . every geometric fiber of @xmath42 is an @xmath43-dimensional variety @xmath15 together with @xmath1 weil divisors @xmath18 such that the pair @xmath44 is stable . 2 . for distinct geometric points of @xmath45 , the fibers are non - isomorphic . 3 . over an open ( but not dense in general ) subset @xmath46 , @xmath42 coincides with the universal family of weighted hyperplane arrangements . for every positive integer @xmath47 such that all @xmath48 , the sheaf @xmath49 is relatively ample and free over @xmath50 . the fibers of @xmath42 will be called _ weighted stable hyperplane arrangements _ , or simply _ slc hyperplane arrangements_. as one has @xmath51 , in particular , each of @xmath40 provides a moduli compactification of the moduli space of generic hyperplane arrangements . [ defn : chambers ] we define the _ weight domain _ @xmath52 and a subdivision of it into locally closed _ chambers _ , denoted @xmath53 , by the hyperplanes @xmath54 for all @xmath55 and @xmath56 , and by the faces @xmath23 . we introduce a partial order on the points of @xmath57 : @xmath58 if for all @xmath59 one has @xmath60 , with at least one strict inequality . we will frequently assume @xmath61 fixed and drop it from the notation . [ exmp:3-degs ] ( @xmath62 , @xmath63 ) consider a 1-parameter family of 5 lines on @xmath64 in general position such that in the limit @xmath65 meet at a point @xmath66 , and @xmath67 meet at a point @xmath68 . this is not allowed by the lc singularity condition if @xmath69 or @xmath70 . since the spaces @xmath71 are proper , there is always a stable pair limit , but its shape depends on the weight : 1 . the variety is @xmath73 , where @xmath74 is the blowup of @xmath64 at @xmath66 and @xmath75 , and @xmath74 is glued along the exceptional @xmath76 s to @xmath77 , @xmath78 . the divisor @xmath79 has three irreducible components , each of @xmath80 has two . @xmath81 are contained in @xmath82 , and @xmath83 in @xmath84 . all five divisors are cartier . 2 . @xmath85 . the variety @xmath86 is obtained from @xmath87 by contracting the @xmath88-curve @xmath89 . the image of @xmath74 is @xmath90 , the divisors @xmath81 restricted to @xmath91 are fibers of a ruling , and @xmath83 restricted to it are fibers of the second ruling . the divisor @xmath79 intersects @xmath91 at one point , and so is not @xmath92-cartier . 3 . @xmath93 . the variety @xmath94 is obtained from @xmath87 by contracting @xmath95 . note that @xmath96 and @xmath97 , we have natural morphisms @xmath98 , and the first of these morphisms is birational . [ thm : red - moduli ] 1 . ( same chamber ) for @xmath99 lying in the same chamber , one has @xmath100 and @xmath101 . in particular , the divisors @xmath102 are @xmath92-cartier . 2 . for @xmath103 , there are natural reduction morphisms @xmath104^{\pi_{\w,\w ' } } \ar[d ] & ( \cx,\cb_i)_{\w ' } \ar[d ] \\ \om_{\w } \ar[r]^{\rho_{\w,\w ' } } & \om_{\w ' } } \ ] ] one has @xmath105 for any @xmath47 such that all @xmath106 . ( specializing up ) for @xmath107 with @xmath108 , @xmath109 is an isomorphism , and on the fibers @xmath110 is a birational contraction restricting to an isomorphism @xmath111 . 4 . for any @xmath58 , there is a natural reduction morphism @xmath112 on the fibers , the rational map @xmath113 is a sequence of log crepant contractions and log crepant birational extractions . further , @xmath114 is the log canonical model for the pair @xmath115 . [ thm : small - weights ] let @xmath116 be a weight with @xmath117 ( lying on the boundary of @xmath118 ) which belongs to the closure of a unique chamber @xmath119 . then @xmath120 is the git quotient for the line bundle , resp . linearization corresponding to @xmath121 . for _ any _ boundary weight ( i.e. with @xmath122 ) , we can formally define @xmath123 to be the above git quotient . over an open and dense subset @xmath124 it gives the moduli of lc hyperplane arrangements on @xmath27 such that @xmath125 . for @xmath121 as in the theorem , one has @xmath126 . we work over an arbitrary commutative base ring @xmath36 with identity , without the noetherian assumption , and indeed can work over any base scheme . @xmath127 will denote an @xmath36-algebra , and @xmath128 an @xmath36-algebra which is an algebraically closed field . the tilde will be used to denote affine schemes @xmath129 , cones @xmath130 , etc . , which are cones over the corresponding projective schemes @xmath15 , polytopes @xmath131 , etc . it may help the reader to grasp some combinatorial aspects of this paper with the following general outline . the ( unweighted ) stable hyperplane arrangements are described by matroid tilings of the hypersimplex @xmath132 . their weighted counterparts are described by partial tilings of @xmath132 as viewed through a smaller `` window '' @xmath133 ; the window must be completely covered . another key idea is the git interpretation of the weight @xmath134 explained in section [ sec : git ] . we begin with some general definitions and then specialize them to the case of grassmannians . [ setup ] we fix two lattices @xmath135 and @xmath136 , a homomorphism @xmath137 , and a homomorphism @xmath138 , such that @xmath139 for all @xmath140 . associated to this data are affine @xmath141 and projective @xmath142 spaces over @xmath36 and linear actions of split tori @xmath143 on @xmath144 and of @xmath145 on @xmath146 . let @xmath131 be the lattice polytope that is the convex hull of @xmath147 , and @xmath130 be the corresponding cone in @xmath148 . we also fix a @xmath136-graded ideal @xmath149\subset\ca[z_1,\dotsc , z_n]$ ] such that the quotient is a locally free ( i.e. projective ) @xmath36-module . hence , @xmath150 is a @xmath151-invariant closed subscheme . let @xmath152 be the corresponding @xmath153-invariant closed subscheme . for a geometric point @xmath154 , the closure of the orbit @xmath155 is a possibly nonnormal toric subvariety of @xmath156 . it corresponds to a lattice polytope @xmath157 which we will call the _ @xmath158-polytope _ or the _ moment polytope of @xmath159_. ( indeed , when @xmath160 , @xmath157 is the moment polytope of @xmath155 , as defined in symplectic geometry . ) a character @xmath161 is in the cone @xmath162 iff there exists a monomial @xmath163 such that @xmath164 and @xmath165 . a _ @xmath158-tiling _ @xmath166 is a face - fitting subdivision of @xmath131 into @xmath158-polytopes . we fix several faces @xmath167 , @xmath168 , of @xmath131 . each of them is defined by the inequality @xmath169 for a unique @xmath170-primitive linear function @xmath171 . in a completely parallel fashion with our grassmannian setup , an element @xmath172 , @xmath173 , is called a weight . for each weight we define a subpolytope @xmath174 the _ weight domain _ @xmath175 is the set of the weights for which @xmath176 is nonempty and maximal - dimensional . a _ weighted @xmath158-polytope _ is a polytope of the form @xmath177 for some @xmath158-polytope @xmath157 , called _ the parent _ of @xmath178 , such that @xmath179 . a _ weighted @xmath158-tiling _ @xmath180 is a face - fitting tiling of @xmath133 by weighed @xmath158-polytopes . the _ partial _ cover @xmath166 of @xmath131 by the parent polytopes is called _ the parent cover _ of @xmath180 . [ defn : z - chamber - decomp ] the @xmath158-chamber decomposition of @xmath118 is defined as follows : @xmath99 lie in the same chamber if for every @xmath158-polytope @xmath157 , one has @xmath181 . consequently , weighted @xmath158-tilings of @xmath176 and @xmath182 are in a bijection . we now specialize these definitions to the case of the grassmannians . the polytope @xmath131 in this case is called the _ hypersimplex _ and the @xmath158-polytopes are called _ matroid polytopes_. for the unweighted version , these notions were introduced in @xcite . let @xmath183 be the grassmannian of @xmath184-planes in a fixed affine space @xmath185 , together with its plcker embedding into @xmath186 , where @xmath187 . let @xmath188 be the affine cone . it is defined by the classical quadratic plcker relations . for @xmath189 , the plucker coordinate @xmath190 has character @xmath191 with @xmath184 ones in the places @xmath192 and with @xmath193 zeros elsewhere . the convex hull of these @xmath194 points is called the _ hypersimplex _ @xmath132 . alternatively , it can be described as follows : @xmath195 it has @xmath196 faces @xmath197 and @xmath198 , isomorphic to @xmath199 and @xmath200 respectively . we fix the lattice @xmath201 in @xmath148 consisting of the vectors @xmath202 such that @xmath203 is divisible by @xmath184 and a homomorphism @xmath204 so that the characters of the plcker coordinates @xmath190 generate @xmath205 and have degree 1 . a _ matroid polytope _ @xmath206 is the polytope corresponding to the toric variety @xmath207 for some geometric point @xmath208\in\gr(r , n)(k)$ ] . ( theorem [ thm : matroid - polytopes](1 ) implies that this projective toric variety and the corresponding affine variety are normal , unlike the case of general @xmath158 ) . the equations of the coordinate hyperplanes restricted to @xmath209 give @xmath1 vectors @xmath210 , what is called a _ realizable matroid_. then @xmath211 is a vertex of @xmath212 iff @xmath213 form a basis of @xmath214 . alternatively , @xmath212 can be described inside @xmath132 by the inequalities @xmath215 , for all @xmath216 . one can also describe the matroid polytopes in terms of hyperplane arrangements . let @xmath217 be the corresponding projective space and assume that it is not contained in any of the @xmath1 coordinate hyperplanes @xmath218 ( i.e. all @xmath219 on @xmath220 ) ; let @xmath221 be @xmath222 . then @xmath223 is a vertex of @xmath212 iff @xmath224 is a point . alternatively , @xmath212 can be described inside @xmath132 by the inequalities @xmath225 for all @xmath56 . note that the matroid polytope in this case is not contained in any of the faces @xmath198 . a _ matroid tiling _ @xmath166 of @xmath132 is a face - fitting subdivision @xmath226 of @xmath132 into matroid polytopes . matroid polytopes form a very particular class of lattice polytopes , with many properties not shared by general lattice polytopes . some of their properties can be summarized as follows : [ thm : matroid - polytopes ] 1 . every matroid polytope is generating , i.e. its integral points generate the group of integral points of @xmath227 . moreover , the semigroup of integral points in @xmath162 is generated by the vertices of @xmath157 . a matroid polytope of codimension @xmath228 is in a canonical way the product of @xmath229 maximal - dimensional matroid polytopes for smaller @xmath37 . so , one has @xmath230 and @xmath231 , and @xmath232 , where @xmath233 is a maximal - dimensional matroid polytope . we now introduce the weighted versions of these notions . let @xmath234 be a weight . weighted hypersimplex _ is the polytope given by @xmath235 similarly , we also have definitions of a _ weighted matroid polytope _ , a _ weighted matroid tiling @xmath180 _ , and _ the parent cover _ of @xmath180 . [ qu : extend - to - parent ] can every parent cover be extended to a _ complete cover _ of @xmath132 ? for @xmath236 the answer is easily seen to be `` yes '' . for @xmath237 we expect the answer to be `` no '' , following the general philosophy of `` mnev s universality theorem '' ( cf . * thm.i.14 ) which shows that matroid geometry can be arbitrarily complicated . [ thm : chamber - decs - coincide ] the chamber decomposition of @xmath57 defined in [ defn : z - chamber - decomp ] coincides with that of definition [ defn : chambers ] . starting with @xmath238 and then varying the weight @xmath5 , the matroid decompositions of @xmath176 and @xmath182 may possibly change if for some matroid polytope @xmath239 , a vertex of @xmath176 would lie on a face of @xmath157 that @xmath240 did not belong to . every vertex of @xmath176 is of the following form : @xmath241 or @xmath242 , for all but possibly one @xmath243 . every face of a matroid polytope lies in the intersection of hyperplanes @xmath244 for some @xmath245 and @xmath246 . possibly after replacing @xmath247 by its complement , we can assume that @xmath248 . then for some @xmath249 we get @xmath250 . if @xmath5 belongs to a face of @xmath157 that @xmath240 did not belong to , then we get a new equation of this form . so @xmath5 lies in a different , smaller chamber . @xcite gives three different interpretations of matroid polytopes . here , we add another one . [ thm : matroid - lc ] the matroid polytope @xmath212 is the set of points @xmath251 such that the pair @xmath252 is lc and @xmath253 ; the interior @xmath254 is the set of points such that @xmath252 is klt and @xmath253 . indeed , the defining inequalities @xmath255 of @xmath212 also happen to be the conditions for the pair @xmath256 to be lc . similarly with the strict inequalities and klt . we do nt even have to assume that @xmath257 for this theorem : the pair @xmath252 can only be lc if @xmath258 , otherwise @xmath259 is not a divisor . and indeed if @xmath260 then @xmath261 , so the theorem still holds . let @xmath152 be a projective scheme locally free over @xmath36 and invariant under the @xmath153-action , and @xmath262 be its affine cone , with the @xmath151-action , as in our general setup [ setup ] . two moduli spaces of varieties with torus action will be relevant for this paper . 1 . the toric hilbert scheme @xmath263 , constructed in @xcite . the moduli space @xmath264 of finite @xmath153-equivariant maps @xmath265 of stable toric varieties @xmath266 over @xmath158 , constructed in @xcite , see also @xcite . this is the equivariant multiplicity - free version of the moduli space of branchvarieties @xcite . both of these moduli spaces are projective schemes . both are available in much more general settings ; we will only need the simplest versions . for an @xmath36-algebra @xmath127 , @xmath267 is the set of closed @xmath151-invariant subschemes @xmath268 which are _ multiplicity - free : _ for every @xmath269 the graded piece @xmath270_x$ ] is a locally free rank-1 @xmath127-module if @xmath271 and is @xmath272 otherwise . a geometric fiber @xmath273 need not be reduced . @xmath274 is a union of possibly non - normal toric varieties glued along torus orbits in a fairly complicated way . in contrast , @xmath275 is the set of locally free proper families @xmath266 over @xmath276 together with a finite @xmath153-equivariant morphism @xmath277 such that every geometric fiber @xmath278 is a projective _ stable toric variety . _ in the ring @xmath279 , for each @xmath269 , the @xmath280-graded piece is a locally free rank-1 @xmath127-module if @xmath271 and is @xmath272 otherwise . a stable toric variety is a _ reduced _ variety glued from _ normal _ toric varieties along orbits in a very simple way , so that the result is seminormal . the price for such niceness is that @xmath281 is a finite morphism rather than a closed embedding . the projective stable toric variety @xmath278 comes with the polarization @xmath282 . for each irreducible component of @xmath278 , this gives a lattice polytope @xmath283 , and together they give a tiling @xmath284 describing the gluing in a rather precise way . if the cone semigroups @xmath285 are generated in degree 1 then @xmath286 is a closed embedding and gives a point of @xmath263 . in general , @xmath264 is only a coarse moduli space since a finite map @xmath286 may have automorphisms ( deck transformations ) . however , it is a fine moduli space over an open subscheme where @xmath286 is birational to its image ( on every irreducible component ) . finally , we note how these moduli spaces change if we replace @xmath146 by a @xmath287-tuple veronese embedding , which means replacing the ring @xmath288 $ ] by the subring generated by monomials of degrees divisible by @xmath287 . the answer for @xmath289 is very non - obvious , and sometimes they indeed change . the moduli space @xmath290 , however , does not change . indeed , the scheme @xmath158 does not change , and neither does the @xmath153-action . therefore , @xmath290 can be defined as easily for a _ rational polytope _ @xmath131 : it can always be rescaled to make it integral . we now apply the general theory of the previous section to the grassmannians . let @xmath291 be the grassmannian with its plcker embedding into @xmath292 , @xmath293 , and @xmath294 be its affine cone . hence , @xmath295 $ ] is generated by the @xmath194 plcker coordinates @xmath296 , modulo the usual quadratic relations . the corresponding polytope is precisely the hypersimplex @xmath132 , and the polytopes @xmath283 appearing in the constructions of the previous section are the matroid polytopes . @xmath297 a closed @xmath151-invariant multiplicity - free subscheme @xmath298 gives a matroid subdivision of @xmath132 . [ thm : matroid - polytopes](1 ) implies that @xmath278 is reduced and is a stable toric variety , so we are in the `` nice case '' . note that @xmath299 , so @xmath266 is not the required stable variety @xmath15 . instead , it should be thought of as the _ log albanese variety _ @xmath300 . @xmath301 is the intersection of @xmath266 with the subvariety in @xmath183 defined by @xmath302 @xmath303 is a schubert variety , isomorphic to @xmath304 . one easily shows that @xmath305 is a regular codimension @xmath306 embedding ( not @xmath307-equivariant ) , the zero set of a section of the tautological bundle @xmath308 on @xmath309 . @xmath15 does not contain any @xmath153-orbits . this implies that @xmath301 is a regular codimension @xmath306 embedding as well , the zero section of the bundle @xmath310 with @xmath311 . on the other hand , by @xcite , the pair @xmath312 has semi log canonical singularities and @xmath313 , where @xmath314 are the divisors corresponding the faces @xmath315 of @xmath132 . in addition , @xmath316 . denoting @xmath317 and combining this together gives the following : [ thm : hkt - main ] 1 . there exists a smooth morphism @xmath318 whose image is the open subset @xmath319 swept by the @xmath153-orbits of @xmath301 ; this is compatible with the divisors @xmath18 and @xmath320 . 2 . consequently , @xmath22 and @xmath321 are isomorphic locally in smooth topology ; in particular , the pair @xmath22 is slc . 3 . @xmath322 ; and so is ample . the poset of the stratification of @xmath15 defined by the irreducible components , their intersections , and the divisors @xmath18 coincides with the poset of the stratification on @xmath323 defined by the subdivision @xmath166 . here we define certain projective schemes @xmath324 and describe their basic properties . we begin with the elementary case which already contains the pertinent combinatorics of the general situation . let @xmath325 be a lattice polytope , and @xmath266 be the correponding projective toric scheme over @xmath36 ( a toric variety when working over a field @xmath30 ) , together with an ample @xmath153-linearized ample invertible sheaf @xmath326 . let @xmath47 be a positive integer , @xmath327 a rational polytope , and @xmath328 be the corresponding @xmath92-polarization . let us fix several faces @xmath167 of this polytope . each of them is given by a linear equation @xmath329 , where @xmath330 is an integral primitive linear function , so that @xmath157 lies in the half space @xmath331 . each of these faces corresponds to a divisor @xmath18 on @xmath266 . now consider the polytope @xmath332 obtained by replacing the inequalities @xmath331 with @xmath333 for some @xmath334 . here , we denote @xmath38 and @xmath335 , so that @xmath336 . note that for some @xmath240 one may have @xmath337 . one says that two polytopes are _ normally equivalent _ if their normal fans coincide , in other words , they define the same toric variety ( with possibly different @xmath92-polarizations ) . the following elementary lemma is well - known , and we omit the proof . [ lem : toric - ex ] 1 . @xmath338 and @xmath332 are normally equivalent iff @xmath240 belongs to the interior @xmath119 of a certain rational polytope . 2 . if @xmath339 then there exists a natural morphism @xmath340 ; it is birational if @xmath341 . 3 . one has @xmath342 . thus , any positive multiple of this @xmath92-line bundle that is integral , is semiample . we note that this is precisely the kind of data that appears in theorem [ thm : red - moduli ] . now consider a projective scheme @xmath343 as in the general setup [ setup ] , @xmath344 $ ] . for each weight @xmath345 , let @xmath346\subset \ca[\wz]$ ] be the subalgebra generated by the monomials whose characters lie in @xmath347 . we define @xmath348 $ ] . [ thm : properties - of - zbeta ] 1 . every @xmath349-polytope is the intersection of a @xmath158-polytope with @xmath176 . 2 . there exists a rational map @xmath350 . it is regular on the open subset of points of @xmath158 whose moment polytopes intersect @xmath176 . @xmath158 and @xmath349 share an open subset @xmath351 whose points are the points with moment polytopes intersecting @xmath352 . 4 . there exists a chamber decomposition of the weight domain into finitely many interiors of polytopes , with the following properties : 1 . if @xmath353 belong to the same chamber , then @xmath354 . 2 . if @xmath355 then there exists a proper morphism @xmath356 . 5 . further , assume that for any @xmath36-field @xmath30 , the corresponding variety @xmath357 is integral , normal , and its monomials span the whole @xmath130 . then @xmath350 is a birational map , and in the previous statement one can take the @xmath158-chamber decomposition defined in [ defn : z - chamber - decomp ] . we reduce the proof to the elementary case [ lem : toric - ex ] of toric schemes , as follows . the homomorphism @xmath137 in the setup [ setup ] gives a surjective map of polytopes @xmath358 , where @xmath359 is a simplex with @xmath194 vertices . the preimage of @xmath176 is a certain subpolytope @xmath360 . monomials of high enough degree @xmath287 generate the subalgebra @xmath346^{(d)}$ ] , and this gives the embedding of @xmath349 into the toric scheme corresponding to the polytope @xmath361 . now the properties ( 14 ) are elementary for the ambient toric schemes , and hence they also hold for the subschemes @xmath349 . to prove ( 5 ) , note that that ( 2 ) and ( 3 ) together imply that @xmath350 is a birational map . let @xmath353 belong to the same @xmath158-chamber . then on every geometric fiber we get a birational morphism @xmath362 . since the @xmath349-polytopes and @xmath363-polytopes are the same , the @xmath153-orbits of @xmath364 are in a dimension - preserving bijection , and so @xmath365 is finite . since @xmath366 is normal , @xmath365 is an isomorphism by the main zariski theorem . since @xmath349 and @xmath363 are free over @xmath36 , and @xmath367 is an isomorphism fiberwise , it is an isomorphism . we now specialize to the case of grassmanians . thus , for every weight @xmath368 we get a projective scheme @xmath324 , which we will call the _ weighted grassmannian _ , and the collection @xmath369 satisfies the conclusions of theorem [ thm : properties - of - zbeta ] , where the chamber decomposition is the one defined in definition [ defn : chambers ] . a key role in our definition of weighted stable hyperplane arrangements will be played by the geometric invariant theory of the universal family @xmath370 . let us first review the basics relevant to our case . let @xmath152 be as in the setup [ setup ] . then we have an action of @xmath371 on @xmath158 and an action of @xmath151 on each @xmath372 , @xmath373 . the character group of @xmath153 is @xmath374 . a @xmath153-linearization of @xmath372 is an extension of the @xmath153-action from @xmath158 to @xmath372 . it can be given by assigning to a monomial @xmath375 of degree @xmath376 an element @xmath377 so that @xmath378 . this is equivalent to choosing an element @xmath379 of degree @xmath287 , so that @xmath380 . we can also take @xmath345 to be of arbitrary positive degree , and subtract the unique element of degree @xmath287 on the line @xmath381 . then for every @xmath382 that is a multiple of @xmath287 , the element of degree @xmath382 on this ray also describes the induced linearization of @xmath383 . given a @xmath153-linearized ample sheaf @xmath384 , one considers the ring of sections @xmath385 . this ring was already graded by @xmath136 by the setup . the linearization provides a new grading by @xmath386 . the git quotient @xmath387 is defined to be @xmath388 , where the latter denotes the elements of degree 0 in the @xmath389-grading . in the original @xmath136-grading , this means that we consider the elements spanned by the monomials whose character in @xmath136 lies on the line @xmath381 . note as well that replacing @xmath390 by a positive power does not change @xmath387 . hence , the input for this construction is a weight @xmath345 up to a multiple , and an ample invertible sheaf @xmath390 up to a multiple . applied to the grassmannian @xmath391 and the plcker line bundle @xmath392 , this means that every weight @xmath368 gives a linearization and a git quotient @xmath393 . the quotients do not respect the chamber structure of @xmath57 , however . _ our key observation now is that the chamber structure describes not the git quotients of @xmath394 but those of the universal family @xmath395 over it . _ let @xmath396 be the universal family of linear spaces @xmath397 . each of the @xmath1 hyperplanes @xmath398 in @xmath399 defines a hyperplane @xmath400 , unless @xmath260 . the natural ample invertible sheaves on @xmath395 are @xmath401 for @xmath402 . the total ring of global sections of all of them is @xmath403 } { ( \text{pl\"ucker relations on } p_i,\ r_j ) } \ ] ] where for each @xmath404 , @xmath405 . the @xmath136-character of each @xmath406 is @xmath407 , the @xmath408-th coordinate vector in @xmath136 , and for @xmath190 it is @xmath409 . hence , the @xmath170-degrees of @xmath406 and @xmath190 are @xmath410 and @xmath184 respectively , differing slightly from [ setup ] . we choose : 1 . the ample @xmath92-line bundle @xmath411 with @xmath412 , or any actual ample invertible sheaf for a multiple @xmath413 such that @xmath414 is integral . + note that if @xmath415 then @xmath416 . 2 . the @xmath153-linearization corresponding to @xmath345 . we denote the corresponding git quotient @xmath417 by @xmath418 . [ lem : gr - e ] @xmath418 is a closed subscheme of @xmath324 . suppose that @xmath414 is integral , and restrict to the subalgebra @xmath419^{(m)}$ ] whose homogeneous elements have degrees divisible by @xmath47 . a monomial in @xmath190 can be complemented to a monomial in @xmath420 whose character is proportional to @xmath345 exactly when its character , divided by the number of @xmath190 s , lies in @xmath132 and in the cone @xmath421 . the intersection of these two sets is precisely @xmath176 . hence , we have a surjective homomorphism @xmath419^{(m)}\to r^{(m)}\lw$ ] sending a monomial @xmath422 to its complement @xmath423 . this gives the closed embedding . if @xmath424 then every monomial in @xmath190 can be complemented , and @xmath425 is the zero set of the equations @xmath426 . thus , @xmath427 is the same as @xmath303 that appeared in section [ sec : unweighted - case ] . git gives the description of @xmath428 in terms of orbits for each geometric fiber . to recall , there are two open subsets in @xmath395 : 1 . the set @xmath429 of semistable points @xmath159 for which there exists a section @xmath430 2 . the set @xmath431 of ( properly ) stable points whose orbit in @xmath429 is closed and the stabilizer is finite ; in our case trivial . ( this set was denoted by @xmath432 in @xcite . we use the currently prevalent notation . ) then we have a surjective morphism @xmath433 , the action is free on @xmath431 and @xmath434 is a geometric quotient . points of @xmath429 have the same image iff the closures of their orbits intersect . among such orbits , there exists a unique closed one . for the torus action , the hilbert - mumford s criterion for ( semi)stability takes an especially simple form . the following criterion is well - known ( e.g. , cf . @xcite ) : 1 . @xmath435 @xmath345 belongs to the moment polytope of @xmath159 . @xmath436 @xmath345 lies in the interior of the moment polytope of @xmath159 and the latter is maximal - dimensional . the moment polytope here lies in @xmath437 which we shift so that it lies in the hyperplane @xmath438 in @xmath148 . for our choice @xmath439 of an ample @xmath92-line bundle , the moment polytope of the point @xmath440\in u$ ] is : @xmath441 1 . @xmath212 , as before , is the matroid polytope of @xmath208 $ ] , and 2 . denoting @xmath442 , @xmath443 @xmath444 is the face of @xmath176 where @xmath445 for @xmath446 . 1 . @xmath447 2 . @xmath448 and @xmath449 spans @xmath437 . \(1 ) @xmath450 . the intersection of the latter polytope with @xmath131 is @xmath444 . \(2 ) the point is stable iff we can replace @xmath345 with any nearby @xmath451 . this means that @xmath452 and @xmath449 spans @xmath437 . [ thm : git - lc - klt ] 1 . if @xmath453 or @xmath454 for some @xmath408 then no @xmath455 is @xmath345-semistable . 2 . suppose @xmath456 . then @xmath457 is lc at @xmath159 . 3 . suppose @xmath458 . then @xmath459 is klt at @xmath159 . \(1 ) if @xmath453 then @xmath460 by the lemma . if @xmath454 then @xmath461 , and @xmath462 , so they do not intersect . \(2 ) suppose @xmath463 take @xmath116 in this intersection . by theorem [ thm : matroid - lc ] the pair @xmath464 is lc . since one has @xmath465 near @xmath159 , the latter divisor is lc as well . vice versa , assume that @xmath466 is lc at @xmath159 . by assumption , there exists @xmath467 . if @xmath468 then we are going to construct another @xmath469 . if @xmath212 is maximal - dimensional and @xmath470 then begin by increasing @xmath330 for @xmath446 until we get to @xmath445 while decreasing @xmath330 with @xmath471 and keeping @xmath472 . this is possible to do since @xmath473 . by doing this , we either achieve the required @xmath474 or get to a lower - dimensional matroid polytope @xmath475 . but by theorem [ thm : matroid - polytopes ] @xmath475 is the product of maximal - dimensional polytopes for lower @xmath476 . we finish by induction on @xmath184 . \(3 ) is proved by the same argument using the second part of theorem [ thm : matroid - lc ] . denote by @xmath477 the projective toric scheme over @xmath36 ( a toric variety when working over @xmath30 ) corresponding to the polytope @xmath176 . in particular , @xmath478 is the toric variety corresponding to the hypersimplex @xmath132 . the morphism @xmath479 factors through @xmath480 . consider @xmath481 with the very ample sheaf @xmath411 , @xmath482 . the rational map @xmath483 is given by the monomials @xmath484 whose character is proportional to @xmath345 and , when normalized , the @xmath422-part belongs to @xmath176 and the @xmath485-part belongs to @xmath486 , which is just another copy of @xmath176 , reflected . this rational map is regular on the open subset where at least one of these monomials , considered as a section of @xmath487 , is nonzero . but the ring generated by these monomials contains @xmath488 , so this open subset contains @xmath429 . recall from section [ sec : weight - grassm ] that we denoted by @xmath489 an open subset of @xmath324 and @xmath394 that corresponds to @xmath352 . @xmath490 will denote the pullback of @xmath491 under the open inclusion @xmath492 . [ thm : free - git - quot ] 1 . the @xmath153-action on @xmath493 is free . 2 . the geometric quotient @xmath494 is projective and comes with a semiample invertible sheaf defining a proper birational morphism @xmath495 , that is an isomorphism over @xmath496 . \(1 ) a point @xmath497 corresponds to a linear space @xmath498 whose matroid polytope intersects @xmath352 . take @xmath455 such that @xmath499 . then by theorem [ thm : git - lc - klt](1 ) the pair @xmath500 is lc at @xmath159 . if we take a nearby @xmath501 with @xmath502 then @xmath500 will be klt . then by [ thm : git - lc - klt](2 ) we have @xmath503 . hence , @xmath504 , the action is free , and the quotient is projective . by removing @xmath505 , we changed the equivalence relation on @xmath493 : for some of the orbits in @xmath493 their closures in @xmath429 intersect , and so they map to the same point of @xmath428 . the criterion of theorem [ thm : git - lc - klt ] implies that every closed orbit in @xmath429 is contained in the closure of an orbit of @xmath493 . hence , @xmath506 is surjective . it is given by the pullback of an ample invertible sheaf @xmath507 on @xmath388 . this morphism is an isomorphism on the open subset @xmath508 which contains @xmath509 . in the case of @xmath424 our construction is different from that of @xcite . to explain it succinctly , @xcite proceeds `` horizontally '' , while we proceed `` vertically '' . the points in @xmath510 are : the points @xmath511 and the points @xmath455 such that @xmath512 for some @xmath408 , satisfying the conditions of theorem [ thm : git - lc - klt ] . so the action of @xmath153 on @xmath513 is not free . there are several ways to restrict it to a subset with a free action : 1 . `` horizontally '' , by removing the points @xmath511 . the remaining set then is @xmath514 , where @xmath515 . this is the choice of @xcite . 2 . `` vertically '' , by removing points with @xmath512 our choice . ( over @xmath128 ) for each stable toric variety @xmath516 over @xmath324 , we define the corresponding _ weighted stable hyperplane arrangement _ as @xmath517 we also define divisors @xmath518 and then @xmath519 [ thm : x - xhat - y ] 1 . the @xmath153-action on the restriction to @xmath520 is free . the geometric quotient @xmath521 by this free action is projective and comes with a semiample invertible sheaf defining a proper birational morphism @xmath522 , an isomorphism on the complements of @xmath523 , @xmath18 . 2 . @xmath524 . @xmath15 is reduced and @xmath18 are reduced weil divisors on @xmath15 . \(1 ) follows immediately from theorem [ thm : free - git - quot ] and ( 2 ) from lemma [ lem : gr - e ] by functoriality of git quotients . ( 3 ) follows since git quotients of reduced schemes are reduced . [ exmp : xy - for - ha ] let @xmath500 be an lc hyperplane arrangement . then @xmath525 . the weighted moment polytope of @xmath526 is therefore @xmath176 itself . the normalization @xmath266 of the closure of the orbit @xmath527 is a toric variety and it comes with a finite morphism to @xmath324 . the pullback of @xmath489 under @xmath516 is simply the orbit @xmath527 , isomorphic to @xmath153 . then @xmath528 , and the quotient @xmath529 is @xmath220 itself , together with the original divisors @xmath18 . since @xmath530 , the morphism @xmath522 must be an isomorphism . hence , every lc hyperplane arrangement appears as a particular case of our construction . [ thm : wsha - sings ] 1 . any weighted stable hyperplane arrangement @xmath44 is a stable pair , i.e. it has slc singularities and @xmath531 is an ample @xmath92-divisor . 2 . @xmath529 is gorenstein . 3 . @xmath15 is cohen - macaulay , and @xmath532 is gorenstein . 4 . assume @xmath533 . then @xmath534 is the restriction under @xmath535 of the ample invertible sheaf @xmath536 on @xmath266 corresponding to the polytope @xmath537 . \(1 ) by theorems [ thm : matroid - lc ] and [ thm : git - lc - klt ] , @xmath538 is a family of open lc subsets of hyperplane arrangements . hence , @xmath539 is smooth , and there exists a finite sequence of blowups of @xmath399 giving a simultaneous resolution of singularities of @xmath540 . on the other hand , a stable toric variety @xmath266 together with its boundary is slc by @xcite ; and the boundary is contained in @xmath541 so can be omitted . the stable toric variety is cohen - macaulay , and its interior is gorenstein by the stanley - reisner theory because the topological space @xmath176 is a smooth manifold with boundary . therefore , the pullback @xmath542 , together with the boundary , has slc singularities , and it is gorenstein . then the geometric quotient @xmath543 by the free @xmath153-action is gorenstein , giving ( 2 ) . now let @xmath544 be such that @xmath414 is integral , and let @xmath545 , @xmath546 , @xmath547 be the invertible sheaves on @xmath209 , @xmath529 , @xmath15 given by the git construction : @xmath545 is the pullback of @xmath548 , sections of @xmath545 descend to sections of @xmath546 and @xmath547 , @xmath546 is semiample and defines the contraction @xmath522 , @xmath547 is ample . we observe that by construction one has @xmath549 . this implies that @xmath550 and that @xmath551 is slc . since @xmath552 , @xmath553 is gorenstein . @xmath15 is cohen - macaulay because it is the result of a log crepant contraction isomorphic outside of @xmath554 and there exists a positive combination of @xmath18 which is cartier . let @xmath536 be the ( integral ) ample invertible sheaf corresponding to the polytope @xmath537 . then by the same argument as in theorem [ lem : gr - e ] sections of @xmath536 restrict to sections of @xmath547 . this gives ( 3 ) . let @xmath555 be a stable toric variety over @xmath324 , and @xmath556 be the corresponding cover of @xmath352 . each of these polytopes has a parent , so that @xmath557 . we denote by @xmath558 $ ] the corresponding projective toric variety . we also denote by @xmath559 the simplex @xmath560 . the corresponding to it toric variety is @xmath399 . if @xmath561 is maximal - dimensional then toric geometry gives a natural birational map @xmath562:\bp^{n-1}\dashrightarrow y[p_{v,\wei}]$ ] , an isomorphism on the torus @xmath563 . now let @xmath561 be a weighted matroid polytope of codimension @xmath228 , and let @xmath212 be its parent , a matroid polytope . recall from theorem [ thm : matroid - polytopes ] that @xmath564 , the product of maximal - dimensional polytopes for a subdivision @xmath565 . then we have a natural rational map @xmath566 which on an open subset is the quotient by a free action by @xmath567 . in this case , we denote by @xmath562:\bp^{n-1}\dashrightarrow y[p_{v,\wei}]$ ] the latter rational map followed by the birational map @xmath568 $ ] corresponding to polytopes @xmath569 and @xmath561 . [ thm : wsha - detailed - desc ] the stratification of @xmath15 into irreducible components and their intersections is in a dimension - preserving , with a shift by @xmath306 , bijection with the tiling @xmath570 of @xmath352 . moreover , the closure of the stratum corresponding to @xmath561 is the closure of the image of @xmath571 under @xmath562:\bp^{n-1}\dashrightarrow y[\pwei]$ ] . this rational map is regular on the open subset of @xmath220 where @xmath572 is lc . the image of this regular set gives a locally closed stratum in @xmath15 corresponding to @xmath561 . each of the polytopes corresponds to an arrangement @xmath573 so that the moment polytope @xmath212 intersects @xmath352 . then we simply follow the @xmath153-orbit of @xmath220 in @xmath489 , to the pullback in @xmath574 , to the quotient in @xmath15 , then back to the irreducible component of @xmath575 $ ] under the inclusion @xmath301 . when @xmath561 is maximal - dimensional , the orbit @xmath527 in @xmath489 has trivial stabilizer . hence , under the quotient by the free action by @xmath153 , the open subset of lc points on @xmath576 is preserved ; then part is contracted by a birational morphism . if @xmath576 has codimension @xmath228 then the stabilizer of @xmath527 in @xmath489 is @xmath567 . then the quotient factors through the quotient of an open subset of @xmath399 by @xmath577 . we are now ready to define the moduli space and the universal family of pairs over it . [ defn : my - moduli ] @xmath578 @xmath579 is a fine moduli space . indeed , every @xmath153-orbit in @xmath489 is also an orbit in @xmath394 . if its stabilizer is finite then it is in fact trivial . therefore , every irreducible component of a stable toric variety @xmath516 maps to its image birationally , and the automorphism group of @xmath580 is trivial . then @xmath581 is a fine moduli space . [ defn : my - family ] the family of weighted stable hyperplanes arrangements @xmath582 is the git quotient of the pullback of @xmath583 by the universal family of stable toric varieties @xmath584 @xmath585 is a locally free ( in particular , flat ) morphism . the families @xmath586 and @xmath587 are locally free , i.e. locally they are given by locally free modules . this implies that the pullback @xmath588 is locally free . algebraically , the git quotient is constructed by taking the degree-0 component in an algebra . thus , this subalgebra is a direct summand , and a direct summand of a locally free module is locally free ( by kaplansky s theorem @xcite , over any ring a module is locally free iff it is projective ) . the parts ( 1 ) and ( 3 ) were established in the previous section . the subset @xmath589 is the open subset of @xmath579 where the stable toric varieties are irreducible , cf . example [ exmp : xy - for - ha ] . the sheaf @xmath590 is free over @xmath50 because by theorem [ thm : wsha - sings ] it is the restriction of the invertible ample sheaf @xmath591 from the universal family of stable toric varieties that corresponds to the lattice polytope @xmath537 , and @xmath591 is free : it s sections give the finite morphism to @xmath324 . the remaining part ( 2 ) is proved in the reconstruction theorem [ thm : reconstruction ] below . \(1 ) for @xmath353 in the same chamber , we have @xmath592 by theorem [ thm : properties - of - zbeta ] , applied to grassmannians . also the conditions for git ( semi)stability are the same . so the moduli and the families are the same . \(2 ) if @xmath593 , we have a reduction morphism @xmath594 again by theorem [ thm : properties - of - zbeta ] . the third application of the same theorem gives the reduction morphism between the stable toric varieties @xmath595 over @xmath596 . finally , this gives in a canonical way the reduction morphisms between the pullbacks of the universal families and their git quotients . each @xmath597 is log crepant . that is because the morphism on the ambient stable toric varieties is given by pullback of @xmath598 ( cf . [ lem : toric - ex ] , [ thm : properties - of - zbeta ] ) , and @xmath599 on @xmath266 restricts to @xmath13 on @xmath15 by theorem [ thm : wsha - sings ] . \(3 ) when specializing up , the morphism @xmath600 is an isomorphism . indeed , a stable toric variety @xmath555 is uniquely determined by its restriction @xmath601 to @xmath489 : @xmath266 is the partial normalization at the boundary of the closure of @xmath601 . but @xmath602 in this case . additionally , when specializing up , the morphism @xmath603 is simply our morphism @xmath522 , so by theorem [ thm : x - xhat - y ] it is birational and an isomorphism outside @xmath604 . \(4 ) is an immediate consequence of the parts ( 1,2,3 ) . note that if the source of the git quotient were fixed , with only the line bundle and the polarization changing , the statement would be an application of the well - known theory of variation of git quotients @xcite . in this case @xmath605 , the @xmath153-action on it is free , and every stable toric variety over @xmath324 is the closure of a unique @xmath153-orbit . hence , @xmath606 . the equivalence of the git quotients @xmath607 and @xmath608 is well - known , see , e.g. , @xcite . intuitively , the contribution @xmath609 to the polarization @xmath439 in this case approaches zero and only the quotient of @xmath324 remains . [ thm : reconstruction ] the stable toric variety @xmath555 can be uniquely reconstructed from @xmath551 . let @xmath575 $ ] be an irreducible component of a stable toric variety @xmath516 , as in theorem [ thm : wsha - detailed - desc ] , and @xmath610 \subset y[\pwei]$ ] be the corresponding irreducible component of @xmath15 . we first show that @xmath575 $ ] can be reconstructed from @xmath610 $ ] intrinsically . indeed , the boundary of @xmath610 $ ] in @xmath15 is labelled by the divisors @xmath18 , some of them coinciding . then the defining inequalities of @xmath561 can be read off this configuration : every missing intersection @xmath611 of codimension @xmath30 gives the inequality @xmath612 . this recovers the polytope @xmath561 . then the embedding @xmath610\to y[\pwei]$ ] is recovered as follows . for every @xmath47 such that @xmath414 is integral , every integral point @xmath613 gives a section of the sheaf @xmath614 } ( m(k_x+\sum b_ib_i ) ) $ ] . namely , it is a unique up to a constant section vanishing at @xmath18 to the order @xmath330 . the collection of these sections gives the embedding @xmath610\to y[\pwei]$ ] defined up to @xmath1 choices of multiplicative constants , one for each @xmath18 , i.e. up to the action of @xmath151 . finally , @xmath220 is recovered from the image of @xmath610\to y[\pwei]$ ] by applying the inverse of the rational map @xmath562 $ ] of theorem [ thm : wsha - detailed - desc ] . then the orbit @xmath527 in @xmath489 gives the morphism @xmath575\to \gr\lw$ ] . the whole stable toric variety @xmath516 is recovered this way by looking at all maximal - dimensional polytopes @xmath561 . we note that for @xmath424 this proof is very different from the one given in @xcite , which does not extend to the weighted case . @xmath615 , @xmath424 . consider the subdivision of @xmath616 , the octahedron on the @xmath617 vertices @xmath618 , into two pyramids : @xmath619 missing the vertex 34 , and @xmath620 , missing the vertex @xmath621 . @xmath619 corresponds to the configuration of 4 points in @xmath76 for which the plcker coordinate @xmath622 , i.e. @xmath623 . this polytope is given by the inequality @xmath624 , which is precisely the lc condition for this configuration . similarly , for @xmath620 one has @xmath625 . on the intersection @xmath626 one has @xmath625 , @xmath623 , and the defining inequalities become @xmath627 , @xmath628 , i.e. @xmath629 . the irreducible component @xmath630 then is the closure of the image of @xmath631 , so isomorphic to @xmath76 ; and similarly for @xmath632 . the intersection @xmath633 is the quotient @xmath634 , so a point . so @xmath15 is a union of two @xmath76 s intersecting at a point . @xmath615 , @xmath635 . consider the trivial subdivision of @xmath176 , with just the polytope itself . the points @xmath636 and @xmath637 may or may not coincide , depending on whether the parent polytope is the pyramid @xmath620 from the previous example , or @xmath616 , otherwise the points are pairwise distinct . @xmath638 . say , @xmath625 . then @xmath301 intersects the stratum corresponding to the edge @xmath639 of @xmath176 , at a point @xmath640 . in this case , the @xmath153-translates of @xmath15 do not sweep out an open subset of @xmath266 , and this is very different from the unweighted situation of section [ sec : unweighted - case ] . if we consider the git quotient of the pullback family over the whole @xmath266 ( not just @xmath641 as in our construction ) , then on the boundary some fibers to the git quotient are modelled on the curve @xmath642 , which is a transversal slice of @xmath266 at the point @xmath640 . then @xmath648 corresponds to the configuration of @xmath649 lines such that @xmath650 is a point , @xmath651 is a point , and otherwise generic . the matroid polytope @xmath212 is obtained from @xmath644 by cutting two corners , and the intersection @xmath652 has codimension 2 , not 1 as might be expected : it is @xmath653 , so the corresponding face gets contracted . as in theorem [ thm : wsha - detailed - desc ] , the irreducible component @xmath74 is the image of @xmath64 under @xmath654 $ ] . on @xmath64 it blows up two points and contracts the strict preimage of @xmath79 . the configuration @xmath655 is lc outside of two points , so the divisor @xmath79 is present on @xmath529 ; it is contracted by the log crepant morphism @xmath522 . then @xmath668 . for @xmath236 this was established in @xcite . for the general case , we first observe that theorem [ thm : small - weights ] applies in this case after replacing @xmath410 with @xmath669 for some @xmath670 , and so @xmath50 a moduli space of lc hyperplane arrangements . the lc condition implies that the @xmath43 hyperplanes with weight @xmath410 must intersect normally . restricting to an intersection to any @xmath671 of these hyperplanes , a line , gives the @xmath236 situation , for the weight @xmath672 , and the moduli space for this is @xmath673 . each of the hyperplanes with weight @xmath674 is uniquely determined by the intersections with these @xmath43 lines , and all of these configurations are lc . so @xmath675 . for any @xmath184 , the closure of @xmath679 in @xmath50 is the toric variety for the fiber polytope @xmath680 , where @xmath681 is the simplex with @xmath184 vertices and side 1 , and @xmath682 is @xmath681 dilated by @xmath193 . this moduli space also has an interpretation as the moduli space of stable toric pairs @xmath683 , as in @xcite but with @xmath193 divisors instead of one . explaining this in detail would take quite some space , and is better done elsewhere . m. brion and c. procesi , _ action dun tore dans une varit projective _ , operator algebras , unitary representations , enveloping algebras , and invariant theory ( paris , 1989 ) , progr . 92 , birkhuser boston , boston , ma , 1990 , pp . 509539 . d. mumford , j. fogarty , and f. kirwan , _ geometric invariant theory _ , third ed . , ergebnisse der mathematik und ihrer grenzgebiete ( 2 ) [ results in mathematics and related areas ( 2 ) ] , vol . 34 , springer - verlag , berlin , 1994 .

transport and the spread of heat in hamiltonian one dimensional ( 1d ) momentum conserving nonlinear systems is commonly thought to proceed anomalously . notable exceptions , however , do exist of which the coupled rotator model is a prominent case . therefore , the quest arises to identify the origin of manifest anomalous energy and momentum transport in those low dimensional systems . we develop the theory for both , the statistical densities for momentum- and energy - spread and particularly its momentum-/heat - diffusion behavior , as well as its corresponding momentum / heat transport features . we demonstrate that the second temporal derivative of the mean squared deviation of the momentum spread is proportional to the equilibrium correlation of the total momentum flux . subtracting the part which corresponds to a ballistic momentum spread relates ( via this integrated , subleading momentum flux correlation ) to an effective viscosity , or equivalently , to the underlying momentum diffusivity . we next put forward the intriguing hypothesis : normal spread of this so adjusted excess momentum density causes normal energy spread and alike normal heat transport ( fourier law ) . its corollary being that an anomalous , superdiffusive broadening of this adjusted excess momentum density in turn implies an anomalous energy spread and correspondingly anomalous , superdiffusive heat transport . this hypothesis is successfully corroborated within extensive molecular dynamics simulations over large extended time scales . our numerical validation of the hypothesis involves four distinct archetype classes of nonlinear pair - interaction potentials : ( i ) a globally bounded pair interaction ( the noted coupled rotator model ) , ( ii ) unbounded interactions acting at large distances ( the coupled rotator model amended with harmonic pair interactions ) , ( iii ) the case of a hard point gas with unbounded square well interactions and ( iv ) a pair interaction potential being unbounded at short distances while displaying an asymptotic free part ( lennard - jones model ) . we compare our findings with recent predictions obtained from nonlinear fluctuating hydrodynamics theory .

1407.1161

the investigation of heat conduction in low dimensional nonlinear lattices has attracted ever increasing attention in the statistical physics community @xcite . although early relevant work @xcite can be traced back to @xmath0 , an increased activity has spurred since the discovery of anomalous heat conduction occurring in one dimensional ( 1d ) momentum - conserving fermi - pasta - ulam ( fpu)-@xmath1 lattices @xcite in 1997 . in those low dimensional study cases the thermal conductivity @xmath2 of the fpu-@xmath1 lattice was found to diverge with the lattice size @xmath3 as @xmath4 , with @xmath5 . this finding consequently yields a system - size dependent thermal conductivity , thus breaking fourier s law of heat conduction . similar anomalous heat conduction behavior has also been identified for other archetype 1d momentum - conserving stylized nonlinear systems , such as the 1d diatomic toda lattices @xcite , and , importantly , has been predicted to occur in momentum - conserving physical materials , such as in carbon nanotubes @xcite , silicon nanowires @xcite and in polymer chains @xcite . experimentally , the breakdown of fourier s law has presently been confirmed for 1d carbon nanotubes and boron - nitride nanotubes @xcite and in 2d suspended graphene @xcite . on the other hand , the low ( 1d , 2d ) spatial dimension alone is not the sole feature that determines whether the validity of fourier s law holds up . for example , normal heat conduction obeying fourier s law has been established beyond doubt for 1d nonlinear frenkel - kontorova ( fk ) @xcite lattices and @xmath6 lattices @xcite . for those nonlinear lattice systems the total momentum is not conserved , being due to the presence of the on - site potentials . these numerical results for 1d lattices led to a conjecture that the property of momentum - conservation in low dimensional systems might be at the origin to give rise to anomalous heat conduction for 1d and 2d nonlinear lattices , e.g. see @xcite . it then later came as a surprise that contradictory results emerged for other stylized momentum - conserved nonlinear 1d lattices , exhibiting saturated thermal conductivities such as the rotator model @xcite and a momentum - conserving variation of the ding - a - ling model @xcite . giardin and kurchan also provided a family of models with or without momentum - conservation which , however , all obey fourier s law @xcite . therefore this situation gives rise to the dilemma of what physics is at the root for the occurrence of the breakdown of the fourier behavior in 1d nonlinear lattices @xcite . most recently , relying on numerical simulations , savin and kosevich @xcite showed that thermal conduction obeys fourier s law for 1d momentum - conserving lattices with a 1d lennard - jones interaction , a morse interaction , and as well a coulomb - like interaction . those numerical findings let them to conclude ( we think erroneously , see in sect . @xmath7 below , and , as well , in ref . @xcite ) that normal heat conduction emerges for momentum - conserving lattices whenever the pair interaction potentials are asymptotically free at large interaction distances . in this work , we focus on heat transport in 1d momentum - conserving nonlinear lattices from another aspect , namely , the _ diffusive spread _ of energy and momentum . it is acknowledged that there exists a profound connection between heat conduction and heat diffusion within the region where fourier s law is valid . for example , take the normal heat conduction in 1d cases : fourier s law states that @xmath8 , where @xmath9 denotes the local heat flux and @xmath10 is the nonequilibrium temperature gradient . if we combine this with local energy conservation ; i.e. , @xmath11 and , additionally , use the relation between the local energy density @xmath12 and the temperature @xmath13 , i.e. , @xmath14 ( with @xmath15 being the volumetric specific heat ) , then the familiar heat diffusion equation @xmath16 can be derived . the normal heat diffusivity equals @xmath17 . microscopically , normal heat diffusion can be characterized by the mean square displacement of the corresponding helfand moment @xcite , which then connects to normal heat conductivity via the green - kubo formula . the efforts trying to bridge heat conduction and diffusion beyond the normal case have only been put forward in the recent decade @xcite . remarkably , it is only recently that a general and rigorous connection between heat conduction and heat diffusion has been established from first principles @xcite : it is shown that in the linear response regime , the evolution of the second order time - derivative of the mean squared deviation ( msd ) of a general energy diffusion process is determined by the equilibrium heat flux autocorrelation function of the system the central quantity that enters the green - kubo formula for the thermal heat conductivity . the key ingredient for obtaining this msd of the energy spread relies on the energy - energy correlation function @xmath18 @xcite , as rigorously shown in recent work @xcite . this thermal equilibrium excess energy - energy correlation indeed is the fundamental quantity that determines the behavior of nonequilibrium heat diffusion , as well as the nonequilibrium heat conduction in a regime not too far displaced from thermal equilibrium . thus , using the energy - energy correlation function , we can conveniently identify whether the heat diffusion in a nonlinear lattice occurs normal or anomalous . with this present study we aim to shed more light on the conundrum that underpins anomalous heat transport in 1d nonlinear lattices . in doing so we study with molecular dynamics ( md ) simulations four different nonlinear 1d momentum - conserving nonlinear lattices . the 1-st one is the 1d coupled rotator lattice which has a bounded interaction potential ; i.e. , the potential is bounded in configuration space and therefore the motion of the particles are not confined . the 2-nd test case studies an unbounded harmonic interaction potential in combination with the coupled rotator interaction potential . the 3-rd test case is the hard point gas model with alternating masses subject to infinite square well pair interactions . this model is believed to show good mixing properties and therefore fast convergence features . as yet a 4-th 1d nonlinear system we complement the rotator model with a lennard - jones 1d - interaction potential , being unbounded at short interaction distances while being free at large interaction distances . this latter model thus allows for bond dissociation at large interaction distances . for all these test beds the correlation functions for the local excess energy deviations as well as the local excess momentum are calculated via extensive equilibrium numerical md - simulations . our studies corroborate the result that normal heat diffusion is found for the coupled rotator lattice . we also demonstrate that in addition to normal heat diffusion the overall dynamics is accompanied by a normal momentum diffusion . we then elucidate that these two features imply that the system dynamics is ruled by the emergence of a finite momentum diffusivity . this observation therefore insinuates that the 1d rotator model physically mimics a fluid behavior . in clear contrast , we find that anomalous heat diffusion occurs for momentum - conserving nonlinear 1d lattices which contain an unbounded interaction potential , as it is the case also with nonlinear fpu - lattices , the hard point gas and also the lennard - jones case . the anomalous heat diffusion and corresponding anomalous heat conductivity behavior is shown to be accompanied in all those test cases with the momentum excess density to undergo anomalous superdiffusion . this latter feature causes a divergent effective viscosity , thus mimicking physically a solid - like behavior . the present study is organized as follows . in section 2 , we briefly review the state of the art of the theory for excess energy diffusion and then develop the theory describing the diffusion of excess momentum . in section 3 , we put forward our hypothesis for the occurrence of normal / anomalous heat transport . this hypothesis is tested thoroughly in section 4 . we start out by performing numerical studies on an overall bounded interaction potential , namely the coupled rotator model . this is then followed by studying a variant of this rotator model by complementing it with unbounded harmonic pair interactions . in addition we discuss the cases with a hard point gas and a lennard - jones pair interaction . these detailed numerical md studies for these four nonlinear lattice systems support the fact that it is not the mere presence or absence of the symmetry of momentum conservation but rather the presence or absence of a fluid - like behavior , as characterized with normal spread of the momentum excess density , which we speculate to be at the source for the validity or the breakdown of fourier s law behavior . for the prior known cases with the dynamics subjected in addition to nonlinear on - site potentials the momentum conservation is broken : the emergence of fourier s law in this latter situation is then ruled by nonlinear scattering processes which provide a finite mean free path behavior for the heat transfer @xcite . additional conclusions and remaining open issues are presented with section 5 . let us consider systems with a momentum - conserving , homogeneous 1d nonlinear hamiltonian lattice dynamics with nearest neighbor interactions . their hamiltonian can be cast in the general form @xmath19\equiv\sum_{i } h_i \;,\ ] ] where the set @xmath20 denote the momenta of particles of identical masses @xmath21 . the set @xmath22 are the displacements from the equilibrium position for the @xmath23-th atom with @xmath24 , where an odd value of @xmath3 is assumed for the sake of convenience . the part @xmath25 is the interaction potential between neighboring sites @xmath23 and @xmath26 . with @xmath27 we formally denote the local energy at site @xmath23 . moreover , throughout our numerical analysis we shall make use of periodic boundary conditions ; i.e. , we set @xmath28 and @xmath29 . the center of mass velocity of the system is chosen at rest ; i.e. @xmath30 . note also that we use here strictly hamiltonian lattice systems which contain no stochastic interaction parts of a spatial or a temporal nature . we start out with the description of heat diffusion in a discrete 1d lattice following ref . @xcite . in doing so , we introduce the energy - energy correlation function , reading : @xmath31 where @xmath32 and @xmath33 denotes the ensemble average over canonical thermal equilibrium at a temperature @xmath13 and @xmath15 is the specific heat per particle . given this autocorrelation function of energy fluctuations , one can evaluate the time evolution of the excess energy distribution @xmath34 starting out from an initial , near thermal equilibrium state , characterized by the initial excess energy perturbation @xmath35 . we consider the case of a localized , small initial excess energy perturbation at the central site , i.e. , @xmath36 . we can then use linear response theory for the excess energy distribution @xmath34 to obtain @xcite : @xmath37 this excess energy distribution remains normalized at all later times @xmath38 , being due to the conservation of energy . the commonly used quantity which quantifies the speed of _ heat diffusion _ is the msd @xmath39 of the excess energy distribution . for a discrete 1d lattice with @xmath3 sites one thus obtains with @xmath40 @xmath41 this msd has been shown to obey the salient second order differential equation @xcite ; i.e. , @xmath42 where @xmath43 denotes the equilibrium autocorrelation function of total heat flux defined as @xmath44 wherein @xmath45 is the local heat flux . note that this correlation @xmath43 is just what enters the green - kubo formula for thermal conductivity @xcite , being written as @xmath46 . this is so because here with @xmath47 and @xmath48 , as the equilibrium average obeys @xmath49 . moreover , @xmath50 contains no energy current stemming from transporting charge in an electromagnetic field or an energy current stemming from a particle concentration gradient . @xmath43 is the quantity that enters the well - known green - kubo expression for the thermal conductivity @xmath2 . for normal heat flow it explicitly reads , @xmath51 . the relation in ( [ heat - theory ] ) connects heat conduction with heat diffusion in a rigorous way . as a consequence , the investigation of heat conduction can equivalently be obtained from studying heat diffusion . the most important quantity is the energy fluctuation autocorrelation function @xmath52 in eq . ( [ e - corr ] ) ; it encodes all the necessary information about heat diffusion and heat conduction . as one can defer from eq . ( [ heat - dif ] ) and eq . ( [ heat - theory ] ) , the energy - energy correlation function @xmath53 determines the dynamical behavior of the msd of heat diffusion as well as the autocorrelation function of total heat flux @xmath43 . as an example take the fpu-@xmath1 model which displays anomalous heat diffusion : there , the energy autocorrelation @xmath52 follows a levy walk distribution , being quite distinct from a normal gaussian distribution in the long time limit @xcite . this statistics then gives rise to a superdiffusive behavior for the energy spread , reading @xmath54 the corresponding , formally diverging anomalous thermal conductivity can be extracted to read @xcite @xmath55 here , @xmath56 with @xmath3 chosen sufficiently large presents the characteristic time - scale of heat diffusion . the quantity @xmath57 refers to the speed of sound for inherent renormalized phonons @xcite . the scheme for the excess energy heat diffusion can likewise be generalized for the problem of corresponding diffusion of excess momentum . for a nonlinear lattices with a hamiltonian in eq . ( [ ham ] ) , the translational invariance of the hamiltonian necessarily indicates that the total momentum @xmath58 is conserved ; i.e. , we have @xmath59 by observing that @xmath60 . using an analogous reasoning as put forward with the preceding subsection for heat diffusion we can define the autocorrelation function for the excess momentum fluctuation @xcite , reading explicitly : @xmath61 where @xmath62 , observing that @xmath63 in thermal equilibrium . following the reasoning of the previous subsection we next demonstrate that this momentum - momentum autocorrelation function describes , within linear response theory , the diffusion of momentum along the lattice . to elucidate this issue we consider alike a lattice in thermal equilibrium at temperature @xmath13 . we apply a small kick of short duration to the @xmath9-th particle . the kick occurs with a constant impulse @xmath64 , yielding a force kick at site @xmath9 as @xmath65 upon integrating the equation of motion from the moment immediately before the kick ( denoted as @xmath66 ) to the moment immediately after the kick ( denoted as @xmath67 ) , we find that the sole effect of this kick is to change the momentum of the @xmath9th particle by an amount @xmath64 . the momenta of all other particles , as well as the position of all particles remain unchanged . formally , this is recast as @xmath68 @xmath69 the full time evolution of the momenta and positions is not analytically accessible for non - integrable nonlinear lattice systems . however , given that @xmath64 is small , the validity regime of linear response is obeyed . the explicit response can be obtained by referring to canonical linear response theory for an isolated system @xcite . specifically , we assume that the system has been prepared in the infinite past , @xmath70 , with the canonical distribution @xmath71 ; \quad z=\int d\gamma \exp[-\beta_t h]\;,\ ] ] where @xmath72 and @xmath73 . with a time dependent force @xmath74 applied to the @xmath9th particle , the total hamiltonian reads @xmath75 . with the system dynamics being closed , the evolution of the phase space distribution is governed by the liouville equation @xmath76 where @xmath77 denotes the poisson bracket . the linear response solution can be readily obtained up to the first order of @xmath78 , yielding @xmath79 the operator @xmath80 is the liouville operator for the original , unperturbed system , i.e. @xmath81 for any quantity @xmath82 . therefore , in presence of the kick - force the thermally averaged particle momenta read for @xmath83 @xmath84 for @xmath67 , it reduces to @xmath85 due to equipartition @xmath86 , which is consistent with eq . ( [ eq : dp ] ) . the conservation of total momentum implies that , @xmath87 , is conserved as well . evaluating this sum at time @xmath88 yields @xmath89 for all later times @xmath38 . the excess momentum density function @xmath90 therefore assumes the form @xmath91 which remains normalized in the course of time @xmath83 . the density @xmath90 is , however , not necessarily semi - positive everywhere ; i.e. it again does not present a manifest probability density for all later times @xmath38 . with time evolving , we notice that the excess momentum autocorrelation eq . ( [ p - corr ] ) describes the spread of the momentum distribution after the initial kick has occurred . as can be observed below , for increasing times @xmath38 the quantity @xmath92 decreases ( at least for some finite time ) . this implies the decrease of the momentum of the @xmath9th particle . the lost momentum is transferred to its neighbors . this feature physically mimics a viscous behavior . let us next assume that the kick is applied to the center particle ; i.e. we explicitly set @xmath93 . similarly ( [ heat - dif ] ) , we define the msd of the excess _ momentum _ @xmath94 for a discrete lattice as @xmath95 because of the conservation of total momentum , in analogy to the energy continuity relation , we may define a `` momentum flux '' @xmath96 via the local momentum continuity relation . to see this , we write down the newtonian equation of motion for the @xmath23th particle , reading @xmath97 by defining the momentum flux as @xmath98 , we obtain a discrete form of the momentum continuity relation , reading @xmath99 note that the momentum flux @xmath100 is actually the force exerted on particle @xmath23 from particle @xmath101 . its ensemble average @xmath102 yields the average internal pressure . following the strategy used for heat diffusion , one can derive a corresponding relation for the second time derivative @xmath94 . it reads : @xmath103 here , the centered autocorrelation function of the momentum flux is given by @xmath104 it should be observed that here the momentum flux @xmath105 , unlike for energy flux , can not be replaced with @xmath106 itself . this is so because the equilibrium average is typically non - vanishing with @xmath107 , where @xmath108 denotes a possibly non - vanishing internal equilibrium pressure in cases where the interaction potential is not symmetric . the presence of a finite , isothermal sound speed @xmath57 may imply that the momentum spread contains a ballistic component . spreading then occurs into the positive and negative directions with velocity @xmath57 , with the two centers of equal weight @xmath109 moving at velocities @xmath110 @xcite . we hence must subtract this trivial ballistic part @xmath111 for the weighted @xmath112 one - sided spread in configuration space . the effective bulk viscosity @xmath113 is thus given as an integration over this _ subleading _ excess momentum correlation @xmath114 over time in terms of a green - kubo formula @xcite , reading @xmath115 in case that the momentum diffusion occurs normal one can invoke the concept of a finite momentum diffusivity by defining , upon use of eqs . ( [ eq : relation ] , [ eta ] ) : @xmath116 therefore , for the discrete lattices discussed here , this so introduced viscosity @xmath113 precisely equals the momentum diffusivity times the atom mass , namely @xmath117 given a situation where the excess momentum density spreads not normally the limit in eq . ( [ eq : d_p ] ) no longer exits . the integration in eq . ( [ eta ] ) formally diverges , thus leading to an infinite viscosity . in the context of this work we find that such an infinite viscosity indicates a manifest solid - like behavior . in distinct contrast , however , a result with a finite effective viscosity indicates an effective fluid - like behavior . the general folklore in the field of anomalous heat conduction @xcite is that in momentum - conserving 1d nonlinear lattices one encounters an anomalous heat conductance behavior . the case with the rotator model , however , presents an eminent exception . so what is the physical mechanism which can explain such exceptions ? an observation is that in all those presently known cases exhibiting anomalous 1d heat conductance the interaction potential has been of unbounded nature at large interaction distances . the known exceptions , predominantly the well studied case with the rotator model , do not possess such unbounded pair interactions at long distances . obviously the form of the overall interaction does matter for the violation of fourier s law . one may speculate that the emergence of the anomalous behavior is rooted in the form of an excess momentum density dynamics that behaves solid - like in the sense that the momentum diffusion does not support a finite effective viscosity in the spirit defined above . in contrast , a fourier - like behavior may become possible if the inherent momentum dynamics is more fluid - like , consequently possessing a finite effective momentum diffusivity . an appealing conjecture therefore is that it is the physics of momentum diffusion which rules whether heat transport occurs normal or anomalous . in short , we next test with different models the following hypothesis : + ( i ) _ heat transport in nonlinear 1d momentum - conserving hamiltonian lattice systems occurs normal whenever the spread of the profile of the excess momentum density , upon subtracting a possibly present leading ballistic part , is normal . _ + ( ii ) _ the corollary being that heat transport occurs anomalous whenever this so adjusted , subleading momentum excess density spreads superdiffusive_. + if this hypothesis holds true it is expected to hold vice versa , i.e. , with heat / momentum substituted by momentum / energy . we next test this so stated hypothesis numerically with four classes of nonlinear hamiltonian lattice dynamics . the numerical procedure used and the details of scaling of parameters and dimensionless units are deferred to the appendix . as a first test bed for the above hypothesis we scrutinize the normal heat transport behavior in a nonlinear , momentum - conserving 1d occurring with the coupled rotator lattice . throughout the remaining we shall use hamiltonian lattice models with corresponding dimensionless units @xcite . the hamiltonian for the coupled rotator lattice dynamics reads @xmath118\big ) \;.\ ] ] notably , here the nonlinear , momentum - conserving interaction potential is bounded for all arguments via the cosine function . the local energy density is @xmath119 $ ] . without loss of generality , we consider the initial distribution of the excess energy or momentum to be a kronecker - delta function in the lattice center . the autocorrelation functions @xmath52 and @xmath120 for energy and momentum are defined according to eqs . ( [ e - corr ] ) and ( [ p - corr ] ) . thus , the temporo - spatial behavior of @xmath52 and @xmath120 describe the dynamics of energy and momentum diffusion starting out from the central position . with the interaction potential being symmetric there is vanishing internal pressure . and the momentum autocorrelation @xmath121 , respectively . the correlation times are @xmath122 ( dotted blue ) , @xmath123 ( dashed red ) , and @xmath124 ( solid green ) . lower panels ( c ) and ( d ) : the mean squared deviation ( msd ) of the energy @xmath39 and the momentum @xmath94 , respectively . a perfect linear time dependence of the msd can be clearly detected for both , the energy and the momentum . the lattice size is chosen @xmath125 and the temperature is @xmath126 . , title="fig:",scaledwidth=40.0% ] and the momentum autocorrelation @xmath121 , respectively . the correlation times are @xmath122 ( dotted blue ) , @xmath123 ( dashed red ) , and @xmath124 ( solid green ) . lower panels ( c ) and ( d ) : the mean squared deviation ( msd ) of the energy @xmath39 and the momentum @xmath94 , respectively . a perfect linear time dependence of the msd can be clearly detected for both , the energy and the momentum . the lattice size is chosen @xmath125 and the temperature is @xmath126 . , title="fig:",scaledwidth=41.0% ] and the momentum autocorrelation @xmath121 , respectively . the correlation times are @xmath122 ( dotted blue ) , @xmath123 ( dashed red ) , and @xmath124 ( solid green ) . lower panels ( c ) and ( d ) : the mean squared deviation ( msd ) of the energy @xmath39 and the momentum @xmath94 , respectively . a perfect linear time dependence of the msd can be clearly detected for both , the energy and the momentum . the lattice size is chosen @xmath125 and the temperature is @xmath126 . , title="fig:",scaledwidth=41.0% ] and the momentum autocorrelation @xmath121 , respectively . the correlation times are @xmath122 ( dotted blue ) , @xmath123 ( dashed red ) , and @xmath124 ( solid green ) . lower panels ( c ) and ( d ) : the mean squared deviation ( msd ) of the energy @xmath39 and the momentum @xmath94 , respectively . a perfect linear time dependence of the msd can be clearly detected for both , the energy and the momentum . the lattice size is chosen @xmath125 and the temperature is @xmath126 . , title="fig:",scaledwidth=42.0% ] in fig . [ corr - con ] ( a ) , we depict the correlation functions @xmath52 for the energy diffusion versus evolving relative time span @xmath38 . for sufficiently large times @xmath38 we observe that the energy autocorrelation function @xmath52 evolves very closely into a gaussian distribution function ( but still spatially bounded with the causal cone , as determined by a finite speed of sound ) ; i.e. , its profile is perfectly well given by @xmath127 with @xmath128 denoting the diffusion constant for heat diffusion . as a result , the msd of heat diffusion @xmath39 then depicts at for sufficiently long time @xmath38 a linear dependence in time @xmath38 , being the hall mark for normal diffusion . in summary , normal diffusion for heat is accurately corroborated numerically with the findings depicted with fig . [ corr - con ] ( c ) . @xmath129 accordingly , heat diffusion theory in @xcite for normal diffusion of heat @xmath130 implies that the heat conduction behavior is normal as well , with the heat conductivity given by @xmath131 . this gaussian behavior for @xmath52 with its corresponding linear time - dependence of the msd for heat diffusion @xmath132 has been observed previously in nonlinear 1d lattices which explicitly do break momentum conservation by including an on - site potential . for example , this is so for the case of 1d lattices with a @xmath6 on - site potential @xcite . in the latter case it is agreed among all practitioners that normal heat conduction occurs beyond any doubt @xcite . the situation with momentum - conserving 1d - coupled rotator lattices , however , is far from being settled in the literature @xcite . here the possibility for a diverging thermal conductivity in the thermodynamic limit is still considered as an option by some practitioners . the present state of the art is nonconclusive although prior extensive numerical simulations , using either the green - kubo method or the non - equilibrium molecular dynamics ( nemd ) method , both seem to indicate that the thermal conductivity is size - independent @xcite . the source of the ongoing dispute is that the numerical results stemming either from the green - kubo method and/or the nemd method , all performed for finite lattice sizes , may possibly not be consistent with manifest asymptotic results in the thermodynamical limit . in contrast , as we emphasized with the previous section , the energy autocorrelation function @xmath52 constitutes a fundamental , detailed measure yielding information well beyond the msd of energy spread @xmath39 @xcite . this is so because of its equivalence with the green - kubo formula , which derives from the salient relation detailed with eq . ( [ heat - theory ] ) . put differently , the temporal - spatial distribution of @xmath52 yields improved , more detailed insight as compared to a method that merely evaluates via md directly the green - kubo integral expression . next we study the diffusion of the excess momentum via the momentum autocorrelation function @xmath120 . our findings are depicted with fig . [ corr - con ] ( b ) . one finds that not only does the energy diffusion obey a gaussian behavior , but also the momentum diffusion occurs gaussian within our explored large regimes of correlation time spans @xmath38 . this behavior of @xmath120 in this coupled rotator lattice possessing a bounded interaction potential is therefore very distinct from the behavior of the @xmath120 occurring in the momentum - conserving in fpu-@xmath1 lattice @xcite . our msd of the excess momentum @xmath94 nicely follows a perfect linear time dependence , as can be deduced from fig . [ corr - con ] ( d ) . according to eq . ( [ eta ] ) , the viscosity @xmath113 for this coupled rotator 1d lattice is therefore finite . put differently , it exhibits a fluidic - like characteristics referred to in the previous section . in distinct contrast , the effective viscosity @xmath113 for the fpu-@xmath1 lattice is diverging towards infinity in the thermodynamic limit ; thus displaying the solid - like characteristics , as discussed in section 2 , cf . ( [ eta ] ) . in contrast to the case of the fpu-@xmath1 lattice with three local conservation laws , here the angle @xmath133 is not conserved . thus , only two local conservation laws for momentum and energy are present , but none for the stretch ( or mass ) . nonlinear fluctuating hydrodynamics theory then predicts a central , diffusive spreading for momentum @xcite without opposite moving side - peaks ; this being in full agreement with our findings . the investigation of the _ momentum _ diffusion behavior in this coupled rotator lattice ( for a preliminary account see in the arxiv @xcite ) has inspired renewed attention from other groups as well @xcite . for the rotator with unbounded interaction potentials . the correlation time are @xmath122 ( blue ) , @xmath124 ( red ) , and @xmath134 ( green ) . ( b ) : the msd of the energy spread @xmath39 . the time dependence ceases to be linear for the energy diffusion . the solid blue power law lines serve as a guide to the eye for the power law like behavior of the data in the large time regime . the parameters used in the numerical simulations are @xmath135 and @xmath136 . the calculated equilibrium temperature is at @xmath137.,title="fig:",scaledwidth=40.0% ] for the rotator with unbounded interaction potentials . the correlation time are @xmath122 ( blue ) , @xmath124 ( red ) , and @xmath134 ( green ) . ( b ) : the msd of the energy spread @xmath39 . the time dependence ceases to be linear for the energy diffusion . the solid blue power law lines serve as a guide to the eye for the power law like behavior of the data in the large time regime . the parameters used in the numerical simulations are @xmath135 and @xmath136 . the calculated equilibrium temperature is at @xmath137.,title="fig:",scaledwidth=40.5% ] in testing our hypothesis further we next amend the rotator coupling by adding an additional unbounded , but symmetric harmonic interaction potential . this transforms the original coupled rotator 1d lattice with bounded interaction into a momentum - conserving 1d lattice with a vanishing internal pressure , but now with an unbounded pair interaction , being provided by the harmonic contribution . the hamiltonian for this so amended coupled rotator model reads : @xmath138 + \frac{k}{2}(q_{i+1 } - q_{i})^2\right ) \,,\end{aligned}\ ] ] where @xmath139 denotes the strength of the harmonic interaction . the total momentum is still conserved . using the same numerical procedure we numerically study the heat and momentum diffusion for this set up . in fig . [ corr - noncon - energy ] ( a ) , the energy autocorrelation function @xmath52 at different correlation times is shown . the finite broadened side peaks exhibited by @xmath52 imply that heat conduction no longer proceeds normal ; instead an anomalous , faster than linear superdiffusive time dependence of the msd of the energy spread @xmath39 is depicted with fig . [ corr - noncon - energy ] ( b ) . this numerically confirms that heat conduction in this unbounded 1d lattice is rendered anomalous . our numerical fit exhibits this superdiffusive heat spreading , growing as @xmath140 . notably , this superdiffusion exponent , @xmath141 , for the amended rotator model is consistent with a previous result of @xmath141 for the fpu-@xmath1 lattice @xcite . both , the amended rotator model and the fpu-@xmath1 lattice dynamics dwell a symmetric potential with a corresponding internal vanishing pressure . we emphasize that the energy autocorrelation function @xmath52 is directly connected with the transport coefficient of thermal conductivity @xcite . in the recent developed _ nonlinear fluctuation hydrodynamic theory _ ( nfht ) @xcite , three normal modes , including one central heat mode @xmath142 and two opposite moving sound modes @xmath143 are obtained upon expanding the three euler equations up to second order only @xcite . whether such a minimal modification is sufficient to model the transport features is still under debate . in particular , it remains to be shown whether this approximate procedure yields in fact a sufficiently good approximation of the true dynamical transport behavior . in this spirit we hope that our present work sheds more light onto this still open question . according to spohn @xcite , the energy autocorrelation function @xmath144 can be decomposed into the three normal modes as @xmath145 . the prefactors @xmath146 and @xmath147 are model dependent and usually depend on temperature . for example , it is obtained that @xmath148 and @xmath149 for the fpu-@xmath1 lattice at @xmath150 @xcite . in this case , the energy autocorrelation function @xmath144 and the heat mode @xmath142 are equivalent , except for a different value for the prefactor . therefore , the msd obtained from the energy autocorrelation function @xmath144 and of the central heat mode @xmath142 should follow the same time dependence . however , nfht predicts an exponent of @xmath151 for the heat mode in lattices with symmetrical potential at zero pressure @xcite . this prediction for @xmath151 , although quite close , distinctly differs nevertheless from our finding here that @xmath141 . this value @xmath141 agrees , as mentioned above , also with the prior results for the fpu-@xmath1 lattice dynamics @xcite . this discrepancy between the numerical results and the nfht may originate from an apparent inconsistent assumption employed in ref . @xcite : namely , in ref . @xcite , it is assumed that all the three peaks of the normal modes have a width much less than @xmath152 , where @xmath57 denotes the sound velocity . using this assumption , one employs the decoupling that the product @xmath153 for large @xmath38 . imposing such zero overlap one proceeds in deriving that the diffusion of the sound modes occurs normal while the diffusion of heat mode is superdiffusive with an exponent of @xmath154 . note however that here the width of the heat mode ( @xmath155 ) exceeds @xmath152 in the asymptotic large time limit , apparently thus contradicting the assumption made . the question then arises whether this anomalous heat transport behavior is also reflected by the behavior for momentum diffusion . the numerically evaluated momentum autocorrelation function @xmath120 _ vs. _ the lattice site is depicted in fig . [ corr - noncon - momentum ] ( a ) for different correlation times . the solely present two side peaks move outwards with a constant sound velocity @xmath57 , giving rise to a ballistic diffusion behavior for the momentum autocorrelation function @xmath120 with the leading term proportional to @xmath156 . the true diffusion behavior of momentum is reflected by the subleading term or the self - diffusion of the side peaks themselves @xcite . the best way to illustrate this momentum behavior of self - diffusion is to present the decay of the height of the side peaks as a function of time . for a normal diffusion behavior this decay of the height of the peaks must follow an inverse square root law , being proportional to @xmath157 . a decay faster than @xmath158 does manifest itself as a non - diffusive , superdiffusive behavior . indeed this feature is corroborated numerically with a behavior for the decay of the central height of the peak(s ) of @xmath120 , which is found to be proportional to @xmath159 . this can be detected clearly from fig . [ corr - noncon - momentum ] ( b ) . in order to double - check this non - diffusive behavior of the momentum self - diffusion , we plot the rescaled side peaks of @xmath160 in a co - moving frame at the sound velocity @xmath57 for different times : in fig . [ corr - noncon - momentum ] ( c ) with @xmath161 ( diffusive ) and in ( d ) with @xmath162 ( non - diffusive ) . it is fair to say that the value @xmath162 fits much better the data . this in turn indicates that the self - diffusion behavior of the momentum is non - diffusive for the symmetrically amended rotator model at zero pressure . for this model , the momentum autocorrelation function @xmath120 coincides with the two sound normal modes @xmath163 defined in nfht . however , our numerical results of @xmath162 again deviates from the prediction that @xmath161 from nfht @xcite . for the rotator with unbounded interaction potentials . the correlation time are @xmath122 ( blue ) , @xmath124 ( red ) , and @xmath134 ( green ) . each has two symmetric side peaks moving outside with a constant sound velocity @xmath57 . ( b ) : the decay of the height of the side peak of @xmath90 . the solid blue power law lines with the dependence of @xmath164 is the best fit for the data from @xmath165 to @xmath166 . ( c ) the rescaled plot of the side peaks of @xmath90 with the exponent of @xmath167 in the moving frame of sound velocity @xmath57 at @xmath168 and @xmath134 . ( d ) the rescaled plot of the side peaks of @xmath90 with the exponent of @xmath169 in the moving frame of sound velocity @xmath57 at @xmath168 and @xmath134 . the parameters used in the numerical simulations are @xmath135 and @xmath136 . the calculated equilibrium temperature is at @xmath170.,title="fig:",scaledwidth=40.0% ] for the rotator with unbounded interaction potentials . the correlation time are @xmath122 ( blue ) , @xmath124 ( red ) , and @xmath134 ( green ) . each has two symmetric side peaks moving outside with a constant sound velocity @xmath57 . ( b ) : the decay of the height of the side peak of @xmath90 . the solid blue power law lines with the dependence of @xmath164 is the best fit for the data from @xmath165 to @xmath166 . ( c ) the rescaled plot of the side peaks of @xmath90 with the exponent of @xmath167 in the moving frame of sound velocity @xmath57 at @xmath168 and @xmath134 . ( d ) the rescaled plot of the side peaks of @xmath90 with the exponent of @xmath169 in the moving frame of sound velocity @xmath57 at @xmath168 and @xmath134 . the parameters used in the numerical simulations are @xmath135 and @xmath136 . the calculated equilibrium temperature is at @xmath170.,title="fig:",scaledwidth=42.0% ] for the rotator with unbounded interaction potentials . the correlation time are @xmath122 ( blue ) , @xmath124 ( red ) , and @xmath134 ( green ) . each has two symmetric side peaks moving outside with a constant sound velocity @xmath57 . ( b ) : the decay of the height of the side peak of @xmath90 . the solid blue power law lines with the dependence of @xmath164 is the best fit for the data from @xmath165 to @xmath166 . ( c ) the rescaled plot of the side peaks of @xmath90 with the exponent of @xmath167 in the moving frame of sound velocity @xmath57 at @xmath168 and @xmath134 . ( d ) the rescaled plot of the side peaks of @xmath90 with the exponent of @xmath169 in the moving frame of sound velocity @xmath57 at @xmath168 and @xmath134 . the parameters used in the numerical simulations are @xmath135 and @xmath136 . the calculated equilibrium temperature is at @xmath170.,title="fig:",scaledwidth=40.0% ] for the rotator with unbounded interaction potentials . the correlation time are @xmath122 ( blue ) , @xmath124 ( red ) , and @xmath134 ( green ) . each has two symmetric side peaks moving outside with a constant sound velocity @xmath57 . ( b ) : the decay of the height of the side peak of @xmath90 . the solid blue power law lines with the dependence of @xmath164 is the best fit for the data from @xmath165 to @xmath166 . ( c ) the rescaled plot of the side peaks of @xmath90 with the exponent of @xmath167 in the moving frame of sound velocity @xmath57 at @xmath168 and @xmath134 . ( d ) the rescaled plot of the side peaks of @xmath90 with the exponent of @xmath169 in the moving frame of sound velocity @xmath57 at @xmath168 and @xmath134 . the parameters used in the numerical simulations are @xmath135 and @xmath136 . the calculated equilibrium temperature is at @xmath170.,title="fig:",scaledwidth=40.6% ] the hard point gas model mimics a sort of idealized fluid with unbounded interactions strength . the hamiltonian of a one - dimensional hard point gas model can be expressed as @xcite : @xmath171 where the setup of masses @xmath172 for even @xmath23 and @xmath173 for odd @xmath23 , see in ref . this choice converts this model into a non - integrable dynamics with strong mixing properties . the latter aspect is advantageous when it comes to the convergence issues at long times and large sizes in md simulations . the symmetric square - well interaction potential reads @xcite @xmath174 because each unit cell contains two particles , the local energy @xmath175 and the momentum @xmath176 used for calculation need to be redefined as @xmath177 and the local momentum as @xmath178 where the number of unit cells amounts to half of the total particles . . the correlation times are @xmath165 ( blue ) , @xmath179 ( red ) , and @xmath134 ( green ) . ( b ) : the anomalous msd of the energy spread @xmath39 . the solid blue power law lines serve as a guide to the eye for the power law like behavior of the data in the asymptotic large time regime . the parameters used in the numerical simulations are identical to the choice made in ref . @xcite with a total number of particles @xmath180.,title="fig:",scaledwidth=42.0% ] . the correlation times are @xmath165 ( blue ) , @xmath179 ( red ) , and @xmath134 ( green ) . ( b ) : the anomalous msd of the energy spread @xmath39 . the solid blue power law lines serve as a guide to the eye for the power law like behavior of the data in the asymptotic large time regime . the parameters used in the numerical simulations are identical to the choice made in ref . @xcite with a total number of particles @xmath180.,title="fig:",scaledwidth=40.0% ] according to nfht @xcite , this hard point gas model with a square well interaction potential and alternating masses can be classified into the same class as the fpu-@xmath1 lattice , and alike the amended coupled rotator model . in this model , the energy and momentum autocorrelation functions @xmath52 and @xmath120 coincide with the heat mode @xmath181 and sound modes @xmath163 in the nfht , respectively . the nfht predicts that the energy diffusion is levy walk superdiffusive with @xmath182 , whereas the self - diffusion of momentum is predicted within nfht to be normal diffusive . in fig . [ corr - hpg - energy ] ( a ) , we depict the energy autocorrelation function @xmath52 at different times . compared with the amended rotator model and the fpu-@xmath1 lattice , the two side peaks are much smaller , although still not vanishing ( being only barely visible in fig . 4 ( a ) ) . the msd of the energy spread is plotted in fig . [ corr - hpg - energy ] ( b ) , yielding a superdiffusive behavior with @xmath183 . as for the fpu-@xmath1 lattice and our amended coupled rotator model result our finding distinctly deviates from the nfht prediction ; it is however consistent with our numerical results of amended rotator model as well as the previously studied fpu-@xmath1 lattice , which all yield numerically an exponent @xmath184 . this again may indicate that nfh - theory is quite good , although not sufficiently accurate enough to account for the full nonlinear dynamics at work . of greater concern are the deviations for momentum spread which theory predicts to be normal but which seemingly does not fit our numerical results . the momentum autocorrelation functions @xmath120 at different times are depicted in fig . [ corr - hpg - momentum ] ( a ) . here we find results that are quite similar to the amended coupled rotator model : the two side sound peaks move in opposite direction with a constant sound speed @xmath57 . to explore the momentum self - diffusion behavior in greater detail , we closely investigate the decay of the central height of the two side peaks , see in fig . [ corr - hpg - momentum ] ( b ) . this decay of the height of the peak are best fitted with a decay law proportional to @xmath185 . being different from the normal diffusive scaling @xmath157 this indicates a non - diffusive behavior for the momentum spread . the rescaled momentum excess density @xmath160 in the co - moving frame of the sound velocity of the center of the side peaks are plotted in fig . [ corr - hpg - momentum ] ( c ) with ( i ) @xmath161 ( normal diffusion ) and also ( d ) with ( ii ) @xmath186 ( anomalous superdiffusion ) . most importantly , the curves with @xmath186 fit convincingly better with the numerical data . this feature therefore reconfirms ( contrary to the nfht prediction @xcite ) anomalous momentum spread for the hard point gas with a square well interaction potential . . the correlation times are @xmath165 ( blue ) , @xmath179 ( red ) , and @xmath134 ( green ) . each has two symmetric side peaks moving in opposite direction with a constant sound velocity @xmath57 . ( b ) : the decay of the height of the side peaks of @xmath90 . the solid blue power law lines depict a decay law proportional to @xmath187 as the best fit for the data from @xmath165 to @xmath166 . ( c ) the rescaled plot of the side peaks of @xmath90 with the exponent @xmath167 in the co - moving frame of the sound speed @xmath57 at @xmath188 and @xmath134 . ( d ) the rescaled plot of the side peaks of @xmath90 with the exponent of @xmath189 in the moving frame of sound velocity @xmath57 at @xmath188 and @xmath134 . the parameters used in the numerical simulations are the same as in ref . @xcite with @xmath180.,title="fig:",scaledwidth=40.0% ] . the correlation times are @xmath165 ( blue ) , @xmath179 ( red ) , and @xmath134 ( green ) . each has two symmetric side peaks moving in opposite direction with a constant sound velocity @xmath57 . ( b ) : the decay of the height of the side peaks of @xmath90 . the solid blue power law lines depict a decay law proportional to @xmath187 as the best fit for the data from @xmath165 to @xmath166 . ( c ) the rescaled plot of the side peaks of @xmath90 with the exponent @xmath167 in the co - moving frame of the sound speed @xmath57 at @xmath188 and @xmath134 . ( d ) the rescaled plot of the side peaks of @xmath90 with the exponent of @xmath189 in the moving frame of sound velocity @xmath57 at @xmath188 and @xmath134 . the parameters used in the numerical simulations are the same as in ref . @xcite with @xmath180.,title="fig:",scaledwidth=40.6% ] . the correlation times are @xmath165 ( blue ) , @xmath179 ( red ) , and @xmath134 ( green ) . each has two symmetric side peaks moving in opposite direction with a constant sound velocity @xmath57 . ( b ) : the decay of the height of the side peaks of @xmath90 . the solid blue power law lines depict a decay law proportional to @xmath187 as the best fit for the data from @xmath165 to @xmath166 . ( c ) the rescaled plot of the side peaks of @xmath90 with the exponent @xmath167 in the co - moving frame of the sound speed @xmath57 at @xmath188 and @xmath134 . ( d ) the rescaled plot of the side peaks of @xmath90 with the exponent of @xmath189 in the moving frame of sound velocity @xmath57 at @xmath188 and @xmath134 . the parameters used in the numerical simulations are the same as in ref . @xcite with @xmath180.,title="fig:",scaledwidth=39.0% ] . the correlation times are @xmath165 ( blue ) , @xmath179 ( red ) , and @xmath134 ( green ) . each has two symmetric side peaks moving in opposite direction with a constant sound velocity @xmath57 . ( b ) : the decay of the height of the side peaks of @xmath90 . the solid blue power law lines depict a decay law proportional to @xmath187 as the best fit for the data from @xmath165 to @xmath166 . ( c ) the rescaled plot of the side peaks of @xmath90 with the exponent @xmath167 in the co - moving frame of the sound speed @xmath57 at @xmath188 and @xmath134 . ( d ) the rescaled plot of the side peaks of @xmath90 with the exponent of @xmath189 in the moving frame of sound velocity @xmath57 at @xmath188 and @xmath134 . the parameters used in the numerical simulations are the same as in ref . @xcite with @xmath180.,title="fig:",scaledwidth=38.0% ] inspecting the preceding three test model cases one is led to speculate that it may well be the unbounded part of the interaction potential that is at the cause for a normal heat and momentum transport behavior in nonlinear 1d momentum - conserving lattices . such a reasoning has obtained support in view of the recent numerical studies by savin and kosevich @xcite which numerically find that heat conductivity remains finite in 1d interaction potentials possessing a regime that allows for dissociation at asymptotic large interaction distances as it occurs , for example , with the lennard - jones 1d case . if so , then for our hypothesis to hold up we should find that in this case the subleading momentum self - diffusion behavior should emerge normal . , @xmath121 for the case with a lennard - jones interaction potential . the correlation times are @xmath190 ( blue),@xmath134 ( red ) , and @xmath191 ( green ) . ( b ) : the msd of the energy spread @xmath39 . ( d ) : the decay of the height of the side peaks of @xmath90 . in both situations ( b ) and ( d ) , the solid blue power law lines serve as a guide to the eye for the data in the large time regime . the parameters in the numerical simulations are for @xmath192 , @xmath193 and @xmath194 , which are the same parameters as used in savin and kosevich s paper @xcite . the calculated equilibrium temperature is at @xmath195.,title="fig:",scaledwidth=38.5% ] , @xmath121 for the case with a lennard - jones interaction potential . the correlation times are @xmath190 ( blue),@xmath134 ( red ) , and @xmath191 ( green ) . ( b ) : the msd of the energy spread @xmath39 . ( d ) : the decay of the height of the side peaks of @xmath90 . in both situations ( b ) and ( d ) , the solid blue power law lines serve as a guide to the eye for the data in the large time regime . the parameters in the numerical simulations are for @xmath192 , @xmath193 and @xmath194 , which are the same parameters as used in savin and kosevich s paper @xcite . the calculated equilibrium temperature is at @xmath195.,title="fig:",scaledwidth=40.9% ] , @xmath121 for the case with a lennard - jones interaction potential . the correlation times are @xmath190 ( blue),@xmath134 ( red ) , and @xmath191 ( green ) . ( b ) : the msd of the energy spread @xmath39 . ( d ) : the decay of the height of the side peaks of @xmath90 . in both situations ( b ) and ( d ) , the solid blue power law lines serve as a guide to the eye for the data in the large time regime . the parameters in the numerical simulations are for @xmath192 , @xmath193 and @xmath194 , which are the same parameters as used in savin and kosevich s paper @xcite . the calculated equilibrium temperature is at @xmath195.,title="fig:",scaledwidth=40.5% ] , @xmath121 for the case with a lennard - jones interaction potential . the correlation times are @xmath190 ( blue),@xmath134 ( red ) , and @xmath191 ( green ) . ( b ) : the msd of the energy spread @xmath39 . ( d ) : the decay of the height of the side peaks of @xmath90 . in both situations ( b ) and ( d ) , the solid blue power law lines serve as a guide to the eye for the data in the large time regime . the parameters in the numerical simulations are for @xmath192 , @xmath193 and @xmath194 , which are the same parameters as used in savin and kosevich s paper @xcite . the calculated equilibrium temperature is at @xmath195.,title="fig:",scaledwidth=42.6% ] using the same numerical schemes as for the foregoing three lattice cases we next test our hypothesis for a lennard - jones setup . the corresponding hamiltonian is given by @xmath196 , \ ] ] using the same parameters as in savin and kosevich s paper ; i.e. , @xmath193 and a binding energy @xmath194 @xcite . here , the pair interaction potential is unbounded at short interaction distances but becomes free at large interaction distances , allowing dissociation . due to this asymmetry in the interaction potential the internal pressure @xmath197 assumes a finite value . the autocorrelation functions @xmath52 and @xmath120 for energy and momentum are defined as before with eqs . ( [ e - corr ] ) and ( [ p - corr ] ) , respectively . in fig . [ corr - nonconlj ] ( a ) , we depict the correlation functions @xmath52 for the energy diffusion versus the correlation time @xmath38 . for sufficient large times @xmath38 we observe that the energy autocorrelation function @xmath52 evolves with two broadened side peaks , being rather distinct from a normal , gaussian - like energy distribution spreading . consequently , the corresponding energy msd is therefore not normal , i.e. it is not proportional to time @xmath38 . in fact it assumes at long times a power - law like behavior , being below an overall ballistic spreading , cf . [ corr - nonconlj ] ( b ) . let us next also study the momentum spread for this test case . in fig . [ corr - nonconlj ] ( c ) , the momentum autocorrelation function @xmath120 at different times are shown . the decay of the height of the side peaks are also depicted with fig . [ corr - nonconlj ] ( d ) . we detect numerically a behavior for the decay of the peak heights proportional to @xmath198 . in perfect agreement with our stated hypothesis , we thus find as well a non - diffusive momentum self - diffusion for this forth test case . our findings not only contradict the recent results reported with @xcite , predicting therein a normal behavior for heat transport , but as well make evident that it is _ not _ necessarily the shape of the interaction potential which rules whether transport proceeds normal or anomalous . the objective of studying energy and momentum transport in low - dimensional systems has recently attracted renewed interest in view of profound advances in theory , namely ( i ) the derivation of new transport relations @xcite and ( ii ) new insight into scaling behaviors @xcite . apart from the role of energy spread and energy transport also the problem of associated momentum spread and momentum transport gained recent attention @xcite . despite this recent progress many open problems remain and the regime of validity of approximate theory predictions , most prominently for the appealing nonlinear fluctuating hydrodynamics theory @xcite , is still under active debate . with this work we studied transport and diffusion characteristics of different classes of momentum - conserving nonlinear @xmath199 hamiltonian dynamics for both , heat and momentum . using recent results of ref . @xcite we started out showing that for energy diffusion there exists a close relationship between the behavior of excess energy diffusion and the overall conductivity behavior for thermal heat transport . this relationship has then been generalized alike for the case of momentum diffusion in 1d nonlinear lattices . for the subleading part of momentum spread beyond its possible ballistic transport yields a diffusivity which relates to the time derivative of the asymptotic msd for excess momentum , see in eq . ( [ eq : d_p ] ) . the consideration of momentum spread offers the possibility to quantify an effective viscosity , being proportional to the momentum diffusivity , eq . ( [ eq : eta - d_p ] ) . for normal momentum diffusion this effective viscosity is finite while it diverges with increasing time @xmath38 if the intrinsic momentum diffusion occurs superdiffusive . a main open problem in this field is the question when and under what conditions the energy and momentum transport deviate from normal . put differently , when is transport and diffusive spreading occurring anomalously in low dimensional nonlinear hamiltonian systems . in this context the authors here put forward their speculative hypothesis that normal ( anomalous ) heat transport has its origin in normal ( anomalous ) momentum spread , and vice versa . having no proof available for this hypothesis we tested the claim by investigating numerically four different nonlinear model systems of momentum conserving nonlinear dynamics that are expected to belong to different classes for their energy / momentum transport characteristics . these were ( i ) the coupled rotator dynamics , ( ii ) its generalization involving the addition of unbounded harmonic interactions , ( iii ) the hard point gas and ( iv ) a case with an asymptotic free dissociation regime ( lennard - jones interaction potential ) . as a main finding from these extensive numerical simulations we can assess that our so stated hypothesis does hold up . this encouraging positive result , however , does not assure that it is fundamentally correct , as we have tested only a finite sample of nonlinear hamiltonian models . moreover , one may argue fairly that any numerical verification lacks a profound analytical foundation . particularly , the question remains whether the numerical findings still hold true in the extreme asymptotic regime of time @xmath200 , being beyond any numerical accessibility at this time . it can be convincingly stated , however , that the mere conservation of momentum in 1d hamiltonian systems does generally not imply anomalous transport . our simulations also shed new light on the question of whether the recent nfht @xcite is accurate enough to predict the scaling regimes for energy and momentum transport . as mentioned , this theory is approximative in that it is based on a expansion of the euler equations to second order only . in addition , it involves further approximations such as a decoupling of different modes at large times , which seemingly can not be convincingly justified in presence of anomalous , superdiffusive energy transport . nevertheless , this theory admittedly is the best available at present times . its scaling prediction for energy transport in models with symmetric unbounded interaction potentials yields an exponent @xmath201 ; this being quite close , but still distinctly different from our numerical value that @xmath202 . even more interesting is the prediction of nfht that momentum spread should occur normal in these cases , thus violating our stated hypothesis . our precise numerics shows however that such a normal momentum diffusion behavior does not fit with our numerical findings . this has been shown with the non - diffusive decay characteristics of the central peaks of the two opposite moving two side peaks in the excess momentum density function . this deviation is additionally substantiated with the failure of a collapse of the data for an assumed normal diffusion in the co - moving frame of sound propagation . the behavior rather fits beautifully , however , with a collapse using _ anomalous _ momentum diffusion ; thereby corroborating our stated hypothesis . in this context we may point out that similar deviations from a normal diffusive scaling for the sound mode are present in the numerics performed by the advocates of nfht : upon inspecting fig . @xcite one detects a similar failure of a diffusive collapse . the numerically established failure here of a diffusive collapse for the case of the fully chaotic hard point gas is particularly trustworthy as we profit from underlying fast numerical converge features . an interesting question for future studies is whether the criterion can be extended to anomalous / normal heat flow occurring in two - dimensional momentum - conserving nonlinear lattice systems . typically , the anomalous heat conductance then tends to diverge in system size logarithmically @xcite . last but not least , the discussed complexity of normal versus anomalous heat and momentum transport in low dimensions might possibly be put to constructive use when designing 1d low dimensional devices for function , such as it is the case for the timely topic of `` phononics '' @xcite . the authors much appreciate those many mutually stimulating and insightful correspondences with professor herbert spohn . we also appreciate the useful correspondence with professor abhishek dhar and his collaborators . the numerical calculations were carried out at shanghai supercomputer center , which has been supported by the nsf china with grant no . 11334007 ( b.l . ) . this work has been supported by the nsf china with grant no . 11334007 ( y.l . , b.l . ) , the nsf china with grant no . 11205114 ( n.l . ) , the program for new century excellent talents of the ministry of education of china with grant no . ncet-12 - 0409 ( n.l . ) , the shanghai rising - star program with grant no . 13qa1403600 ( n.l . ) , the nsf china with grant no . 11347216 ( y.l . ) and tongji university grant no . 2013kj025 ( y.l . ) . for the investigation of the dynamics of 1d nonlinear lattice models , dimensionless units have been applied throughout as a convenient tool . as discussed in ref . @xcite , the setup of dimensionless units is model dependent . we will elaborate below the details of the used dimensionless units for the 1d nonlinear lattice models considered in this work . the dimensional hamiltonian of coupled rotator model can be expressed as @xmath203\right)\;,\ ] ] where @xmath20 and @xmath22 denote the dimensional momentum and displacement from equilibrium position for @xmath23-th atom . @xmath21 denotes the atom mass and @xmath146 is the lattice constant . the parameter @xmath204 , possing the dimension of energy , represents the coupling strength of the neighboring rotators . for this coupled rotator model , one can introduce the dimensionless variables by measuring lengths in units of @xmath205 $ ] , energies in units of @xmath206 $ ] , masses in units of @xmath207 $ ] , momenta in units of @xmath208 $ ] , time in units of @xmath209 $ ] . the temperature will be measured in units of @xmath210 $ ] where @xmath211 is the boltzmann constant . if we implement the following substitutions : @xmath212,\,\,p_i\rightarrow p_i[(vm)^{1/2}],\,\,q_i\rightarrow q_i[a/(2\pi)]\;.\ ] ] the hamiltonian of eq . ( [ dim - ham - cr ] ) can be transformed into the dimensionless one of eq . ( [ ham - con ] ) . the dimensional hamiltonian of amended rotator model is @xmath213+\frac{k_0}{2}(q_{i+1}-q_i)^2\right)\;,\ ] ] where @xmath214 denotes the extra coupling strength between neighboring atoms . the dimensionless units setup is the same as that for coupled rotator model . applying the same transformation of eq . ( [ trans - rotator ] ) , eq . ( [ dim - ham - ar ] ) can be transformed into the dimensionless hamiltonian of eq . ( [ ham - ar ] ) with the dimensionless @xmath215 . the dimensional hamiltonian of hard point gas model is @xmath216 where @xmath217 is the mass for @xmath23-th particle and the square well potential can be described as @xmath218 with @xmath146 denoting the average distance between neighboring particles . the alternating masses are introduced by setting particle masses @xmath219 for an even number of @xmath23 and @xmath220 for an odd number of @xmath23 . for this hard point gas model , one can introduce the dimensionless variables by measuring lengths in units of @xmath221 $ ] , masses in units of @xmath222 $ ] . since there is no characteristic potential energy for this model , its dynamics is essentially the same for any energy scale . one can arbitrarily choose an energy scale @xmath223 as the reference energy and the energies can be measured in units of @xmath224 $ ] . as a result , the momenta can be measured in units of @xmath225 $ ] and the time can be measured in units of @xmath226 $ ] . the temperature can also be measured in units of @xmath227 $ ] . in our study we used the same parameters as used in ref . @xcite . the dimensional lennard - jones model has the following hamiltonian @xmath228\;,\ ] ] where @xmath21 is the atom mass and @xmath146 is the lattice constant . @xmath229 denotes the binding energy and @xmath230 is a dimensionless parameter . @xmath231 is yet another dimensionless parameter . for this lennard - jones model , one can introduce the dimensionless variables by measuring lengths in units of @xmath221 $ ] , masses in units of @xmath207 $ ] , energies in units of @xmath232 $ ] , momenta in units of @xmath233 $ ] , time in units of @xmath234 $ ] . the temperature will be measured in units of @xmath235 $ ] . if we implement the following substitutions @xmath236,\,\,p_i\rightarrow p_i[(\varepsilon_0 m)^{1/2}],\,\,q_i\rightarrow q_i[a]\;.\ ] ] the hamiltonian of eq . ( [ dim - ham - lj ] ) can then be transformed into the dimensionless hamiltonian of eq . ( [ ham - lj ] ) . in order to obtain precise numerical results , we employ md simulations for an isolated system evolving with the corresponding liouvillian over large , extended time spans and used throughout periodic boundary conditions . the method to obtain the correlation functions is adopted from ref . the equations of motions are integrated with a fourth order symplectic algorithm @xcite .

we report a study of the intensity and time dependence of scintillation produced by weak-particle sources in superfluid helium in the presence of an electric field ( kv / cm ) in the temperature range of 0.2 k to 1.1 k at the saturated vapor pressure . both the prompt and the delayed components of the scintillation exhibit a reduction in intensity with the application of an electric field . the reduction in the intensity of the prompt component is well approximated by a linear dependence on the electric field strength with a reduction of 15% at 45 kv / cm . when analyzed using the kramers theory of columnar recombination , this electric field dependence leads to the conclusion that roughly 40% of the scintillation results from species formed from atoms originally promoted to excited states and 60% from excimers created by ionization and subsequent recombination with the charges initially having a cylindrical gaussian distribution about the track of 60 nm radius . the intensity of the delayed component of the scintillation has a stronger dependence on the electric field strength and on temperature . the implications of these data on the mechanisms affecting scintillation in liquid helium are discussed .

1110.0570

the phenomenon of liquid helium ( lhe ) scintillation due to passage of charged particles was discovered in the late 1950 s @xcite . since then rather extensive studies have been conducted , motivated both by its intrinsic interest ( including an interest in illuminating the behavior of ions and neutrals in superfluid helium ) and by the application of liquid helium as a particle detector @xcite . ( for a brief review of early work , see ref . @xcite . a more recent review can be found in the introduction of ref . @xcite . ) the following picture has emerged from these studies as the process for liquid helium scintillation production ( see e.g. refs . @xcite ) : when a charged particle passes through liquid helium , it deposits energy to the medium , part of which goes to ionization , creating electron - ion pairs along its track . the deposited energy also goes to promoting atoms to excited states . the @xmath2 value , the average energy required to produce one electron - ion pair , in liquid helium is @xmath3 ev @xcite . electrons and ions then thermalize with the liquid helium . the electron subsequently forms a `` bubble '' in the liquid , pushing away surrounding helium atoms as a consequence of pauli exclusion . the he@xmath4 ion , on the other hand , forms a `` helium snowball '' attracting surrounding helium atoms . the bubbles and snowballs then recombine quickly ( @xmath5 s ) forming excited helium molecules ( excimers ) . the molecules are formed in the triplet and singlet states . the lowest - energy singlet state molecule , he@xmath6(a@xmath7 ) , radiatively decays in less than 10@xmath8 s to the ( unbound ) ground state , emitting a @xmath9 ev extreme - ultraviolet ( euv ) photon , contributing to the prompt component of scintillation . excited atoms in singlet states also contribute to the prompt scintillation signal . the triplet state molecule , he@xmath6(a@xmath10 ) , on the other hand , has a lifetime of @xmath11 s in liquid helium @xcite . in a high excitation density environment , however , the triplet state excimers can also be destroyed through the penning ionization process @xmath12 or @xmath13 if a singlet excimer is formed as a result of penning ionization after the first @xmath14 s , then it makes a contribution to the delayed scintillation component . the study of liquid helium scintillation has been conducted using radioactive sources , such as @xmath15 , conversion electron , and @xmath0 sources as well as electron beams . there were also experiments that studied liquid helium scintillation induced by the products of the @xmath16he(@xmath17,@xmath18)@xmath16h reaction @xcite . various characteristics of the scintillation light depend on the type of ionizing particles employed . when a @xmath15 or conversion electron source is used , the scintillation light yield is known not to show a strong temperature dependence , whereas when the scintillation is induced by @xmath0 particles , the scintillation yield does exhibit a temperature dependence . in addition , a difference in time dependence of the scintillation signal has been observed between @xmath0-induced scintillation and @xmath15-induced scintillation @xcite : @xmath0-induced scintillation light shows a long lived component that decays with a @xmath19 dependence following the prompt pulse . this slow component shows up as a series of single photoelectron ( pe ) pulses following the prompt pulse in an experiment in which the euv scintillation is wavelength - shifted and detected using a photomultiplier ( pmt ) . ( thus these single pe pulses are called `` afterpulses '' . ) a typical pmt oscilloscope trace from our experiment described in this paper is shown in fig . [ fig : pmtsignal ] . one distinct difference between @xmath0 tracks and electron tracks is the ionization density . for example , a 5.5-mev @xmath0 particle deposits energy into superfluid liquid helium ( @xmath20 g / cm@xmath16 ) at a rate of @xmath22 ev/@xmath23 m on average along its 0.27 mm long track ev/@xmath23 m and occurs when the remaining kinetic energy is 0.65 mev . ] , whereas for a 500 kev electron , the average energy deposition rate is 40 ev/@xmath23 m @xcite . with @xmath24 ev , it follows that ionization events are separated by an average distance of @xmath25 nm and @xmath26 nm in @xmath0 and electron tracks , respectively . with an average separation of several tens of nanometers between the electron and its parent ion @xcite , ionization events in @xmath0 tracks overlap with each other whereas electron - ion pairs are well separated in electron tracks . as we will discuss later , many of the differences in the characteristics between @xmath0-induced scintillation and electron - induced scintillation can be attributed to the difference in the ionization density . although the amount of data on lhe scintillation induced by protons and tritons from the @xmath16he(@xmath17,@xmath18)@xmath16h reaction is rather limited , it shows qualitatively similar characteristics to @xmath0-induced scintillation because of the high ionization density of the proton and triton tracks . in establishing the picture described above , a study of scintillation in the presence of an electric field played an important role . the effect of an electric field on the lhe scintillation was first studied by hereford and moss @xcite . in this study , the authors measured both the intensity of @xmath0-particle - induced scintillation and the ionization current extracted from the @xmath0 tracks in the presence of an electric field ( 9 to 43 kv / cm ) in the temperature range of 1.23 to 4.2 k. the scintillation signal was integrated over @xmath27 @xmath23s . they observed a decrease in the scintillation intensity and a corresponding increase in the ionization current with an increasing electric field particle . ] . we have extended their work and have studied the effect of an application of an electric field on both the prompt scintillation and afterpulses in the temperature range of 0.2 k to 1.1 k with an electric field up to @xmath28 kv / cm . we measured the time dependence of the scintillation signal up to 14 @xmath23s . the research reported here is directed primarily towards an understanding of the physical processes underlying the electric field dependence of the scintillation yield of liquid helium . it provides information on the dynamics of the charged particles generated by the passage of an alpha particle through helium and the interaction of molecular species produced on recombination . however , the work is motivated by the desire to use helium scintillation in the presence of a high electric field in the nedm experiment @xcite . the nedm experiment , currently being developed to be mounted at the spallation neutron source at the oak ridge national laboratory , will search for the neutron electric dipole moment ( edm ) using a method proposed by golub and lamoreaux @xcite . in this experiment , spin polarized ultra cold neutrons will be produced and stored in a volume filled with superfluid helium , to which spin polarized @xmath16he atoms will be added as cohabiting magnetometer . lhe scintillation produced by the reaction products of the spin - dependent @xmath16he(@xmath17,@xmath18)@xmath16h reaction will be used to analyze the spin of the neutrons . the signature of a non - zero edm is a shift in the neutron precession frequency upon application of an electric field . therefore , the effects of an electric field on lhe scintillation produced by heavy particles such as protons , tritons , and @xmath0 particles are of particular interest . an earlier study performed by part of the nedm collaboration on lhe scintillation produced by the products of neutron capture on @xmath16he was reported in ref . the experiment was performed using a leiden cryogenics model cf-600 cryogen - free dilution refrigerator @xcite , which had a measured cooling power of 1.8 mw at 0.2 k. a measurement cell , which housed the electrodes immersed in lhe , was mounted on the mixing chamber of the dilution refrigerator ( dr ) . an @xmath29am @xmath0 source was electroplated on the ground electrode . @xmath0 particles from the source induced scintillation light in the volume of lhe between the electrodes . the euv scintillation light was converted to blue light by a wavelength shifter , then was guided through a light guide to the viewport at the bottom of the measurement cell and to a pmt mounted outside the measurement cell in vacuum . a description of each component is given below . a schematic of the measurement cell and the pmt is shown in fig . [ fig : apparatus ] . the measurement cell was made of a stainless steel cross with 4 - 5/8 and 4 - 3/4 conflat flanges attached to its ends . the volume of the lhe inside the cell was approximately 600 ml . at the top of the cell was a heat exchanger made of stack of gold - coated copper plates , which provided cooling from the mixing chamber to the lhe inside the cell . the electrodes were made of stainless steel . the ground electrode was 19.05 mm in diameter and 6.35 mm in thickness , and had an edge rounded with a radius of 3.175 mm . the high voltage ( hv ) electrode was 19.05 mm in diameter and 12.7 mm in thickness , and had an edge rounded with a radius of 6.35 mm . the gap between the electrodes was approximately 4 mm . the hv electrode was mounted on a commercially made hv vacuum feedthrough . the hv was supplied by a glassman high voltage inc . model eh30r3 power supply . a thin wall stainless steel tubing was used as the hv wire inside the cryostat to reduce the heat load . a hv resistor ( @xmath30 g@xmath31 or @xmath30 m@xmath31 ) needed to be inserted in series in the hv cable , in order to reduce the noise on the pmt signal induced by the ripple of the hv power supply for the hv electrode . the capacitance of the hv cable and this resistor formed a low pass filter , which reduced the effect of the ripple to a sufficiently low level . approximately 300 bq of @xmath29am was electroplated on the ground electrode . the activity had a diameter of 6 mm . because the range of 5.5 mev @xmath0 particles in superfluid liquid helium is only @xmath32 mm , the ionization due to @xmath0 particles occurred in a region very close to the surface of the ground electrode . the electric field at the location of the ionization was uniform to 5% . larger electrodes and a smaller gap would have given a more uniform electric field at the location of the ionization . however , that would have , at the same time , significantly reduced the fraction of euv photons detected and hence the detection efficiency by limiting the solid angle subtended at the location of the ionization by the wavelength - shifter - coated surface . the size and shape of the electrodes were chosen as a compromise between electric field uniformity and the detection efficiency . the electrodes were enclosed in a sleeve made of g10 . this was to prevent possible hv breakdowns due to the hv electrode seeing structures on the inner walls of the stainless steel measurement cell . a sapphire viewport with a view diameter of 49.3 mm was mounted at the bottom of the measurement cell . an acrylic light guide with a diameter of 49.3 mm and a length of 30.5 mm was mounted between the viewport and the g10 sleeve . the g10 sleeve had a 48 mm diameter hole on the side to allow the scintillation light to reach the light guide . the top surface of the light guide was coated with polystyrene doped with tetraphenyl butadiene ( tpb ) , which is known to have good euv - visible conversion properties ( see ref . @xcite and references therein ) . a hamamatsu r7725 pmt , modified for cryogenic use with a pt underlayer @xcite , was mounted underneath the viewport in vacuum . a fraction of the @xmath0 particle induced euv light emitted from the region between the electrodes was converted to blue light at the top surface of the light guide . the blue light was then guided towards the pmt through the light guide , transmitted through the sapphire viewport and then detected by the pmt . the fraction of the euv light that reached the tpb coated surface of the light guide was about 5% ( determined by the solid angle subtended by the coated surface of the light guide at the location of scintillation events ) . the overall detection efficiency of the apparatus (= the number of detected pes per emitted euv photon ) was estimated to be @xmath33 from the solid angle mentioned above ( 5% ) , the euv to visible conversion efficiency ( @xmath34% @xcite ) , the estimated transport efficiency of the visible light ( @xmath35% ) , and the pmt s quantum efficiency ( 18% @xcite ) . the observed number of pes in the prompt pulse in the absence of an electric field @xmath36 ( see sec . [ sec : adcanalysis ] ) was consistent with the expectation based on the known scintillation efficiency (= the number of euv photons in the prompt pulse per energy deposition ) @xcite and the estimated detection efficiency . the pmt was thermally anchored to the 3 k plate of the dr , which was cooled by a pulse - tube refrigerator , in order to reduce heat load , both radiative and conductive , to the measurement cell . the base circuit adopted the split design described in ref . @xcite with some modifications . the resistor chain was heat - sunk on the 3 k plate . in order to reduce the heat load from the joule heating of the resistor chain , a resistor value 10 times larger than the manufacturer recommended value was used . because of the low event rate ( @xmath37 s@xmath38 ) , this did not affect the performance of the pmt . the rest of the base circuit , which consisted mostly of capacitors , was mounted at the back of the pmt . previously , this pmt model had been demonstrated to function properly at temperatures down to @xmath39 k @xcite . the present work showed that it functions properly at @xmath40 k. there were two ruo@xmath6 temperature sensors mounted on the g10 sleeve to monitor the temperature of the lhe inside the measurement cell . there were two lhe level sensors inside the measurement cell . they were each made of two concentric cylinders that formed a capacitor . one was mounted in such way that it surrounded the heat exchanger and the other surrounded the light guide , monitoring the lhe level at the top and the bottom of the measurement cell . in addition there was a lhe pressure sensor monitoring the lhe pressure inside the measurement cell . a silicon diode sensor was mounted on the pmt as well as on the pmt heat shield to monitor the temperature of the pmt and the heat shield . a schematic of the helium gas / liquid flow in the system is shown in fig . [ fig : flowdiagram ] . in filling the system , clean helium gas taken from boiloff from a dewar was further cleaned by a ln@xmath6 cold trap and then was introduced into the system . the helium was first condensed by a liquefier ( a box filled with sintered metal ) mounted at the 3 k plate , then flowed through a capillary tube with an i d of 0.6 mm and a length of 1.8 m which was thermally anchored at the still , the 50 mk plate , and at the mixing chamber plate to the measurement cell . the cell was filled with the dr running , which kept the cell temperature at @xmath41 k during the fill . it took approximately 24 hours to fill the 600 ml cell in this way . data on the pmt signal were collected using a camac and nim based data acquisition ( daq ) system . the schematic of the daq system is shown in fig . [ fig : daqdiagram ] . the trigger was generated by a discriminator whose threshold was set to a level corresponding to a fraction of a pe , which allowed us to monitor the pmt gain using single pe events from the dark noise . for each event , the size of the prompt signal was recorded using a lecroy 2249a charge - sensitive analog - to - digital converter ( adc ) with a 80 ns - long gate . the length of the gate was chosen to capture all of the prompt part of the signal ( see fig . [ fig : pmtsignal ] ) . the pmt signal was amplified and then was split into two , one of which was further amplified with a gain of @xmath42 before being fed to the adc . the high - gain signal ( adc ch1 in fig . [ fig : daqdiagram ] ) was used to monitor the pmt gain using the single pe events from the dark noise . the low - gain signal ( adc ch0 in fig . [ fig : daqdiagram ] ) was used to measure the size of the signal generated by @xmath0 particles . randomly generated pulses were mixed into the trigger in order to monitor the adc pedestal . a typical low - gain adc spectrum is shown in fig . [ fig : adc ] . the intensity of the delayed component was measured by recording the timing of each afterpulse ( see fig . [ fig : pmtsignal ] ) with respect to the prompt pulse using a lecroy 2277 multi - hit tdc . this tdc can record up to 16 hits within a 16 @xmath23s time window . typical time spectra thus obtained are shown in fig . [ fig : tdc ] for two temperatures . also a typical distribution of the number of afterpulses per event is shown in fig . [ fig : tdc_mult ] . s is due to afterpulses caused by helium contamination in the pmt.,width=302 ] the cell was filled with lhe following the procedure described in sec . [ sec : gashandling ] . once the cell was filled and a desirable measurement cell temperature was reached , the prompt scintillation yield and the time spectrum of the afterpulses were measured as the function of the applied hv . when increasing or decreasing the hv , the pmt was kept turned off to protect the pmt from being exposed to bright flashes of light due to hv breakdowns , which occurred occasionally when changing the hv setting ; we had found that exposing the pmt to flashes of light from hv breakdowns decreased the pmt gain and the change appeared to be permanent . the hv electrode was biased with a negative voltage . for each hv and temperature setting , data were acquired for approximately 5 minutes , which corresponds to @xmath43 prompt scintillation events . each of such a continuous data taking period is called a `` run '' . data were taken at five different temperature settings . the average temperature and the 1-@xmath44 variation in temperature for each of these settings were @xmath45 mk , @xmath46 mk , @xmath47 mk , @xmath48 mk , and @xmath49 mk . in the remainder of this paper , for conciseness we refer to these temperatures by their `` nominal '' values : 0.2 k , 0.3 k , 0.4 k , 1.0 k , and 1.1 k. in addition , data were taken at temperatures up to 2 k with no hv applied . the gap in the data between 0.4 k and 1.0 k was due to a thermal short that developed in the system after the data were taken for 0.4 k , 0.2 k , and 0.3 k. the thermal short did not allow the system to run below 1.0 k. as mentioned earlier , a hv resistor was inserted in the hv supply line to suppress noise due to the ripple on the hv . most of the data were taken with a 1 g@xmath31 resistor . the results did not change when a 1 m@xmath31 resistor was used , confirming that the resistance between the two electrodes was significantly larger than 1 g@xmath31 and the voltage across the electrode gap was indeed what was supplied by the hv power supply . the effect of the electric field on the prompt scintillation yield manifests itself as a shift in the @xmath0 peak in the low - gain adc spectrum . in order to remove effects due to possible changes in the pmt gain , the horizontal scale of the spectrum needs to be properly calibrated in terms of the number of pes . as noted earlier , the gain of the pmt was continuously monitored throughout the experiment by the high - gain adc that recorded single pe events from the dark noise . in order to calibrate the horizontal scale of the low - gain adc spectrum using the pmt gain information obtained by the high - gain adc , it is necessary to determine the gain of the amplifier . this was achieved by determining the slope of the correlation between the high - gain and low - gain adc spectra . determination of the pmt gain and the amplifier gain was performed for each run . the resulting low - gain adc spectrum was fit to determine the location of the @xmath0 peak using the following model function ( a poisson distribution representing the distribution of the number of pes convoluted with a gaussian distribution representing the pmt response ) : @xmath50 where @xmath51 is the overall normalization , @xmath52 a variable corresponding to the adc channel , @xmath23 the mean number of photoelectrons , and @xmath53 the gain of the system ( the adc channel corresponding to one pe ) . @xmath54 is given by @xmath55 where @xmath56 is the width of the response function of the pmt for single pe events and @xmath57 is the width of the adc pedestal . @xmath56 and @xmath57 were determined by fitting to the single pe and pedestal peaks respectively , prior to fitting eq . ( [ eq : adcfit ] ) to the @xmath0 peak . @xmath51 and @xmath23 were varied in the @xmath0 peak fit . the shape of the @xmath0 peak in the low - gain adc spectrum is well described by the model function in eq . ( [ eq : adcfit ] ) ( see fig . [ fig : adcfit ] ) . peak in the low - gain adc spectrum fit with the model function in eq . ( [ eq : adcfit]).,width=340 ] figure [ fig : pevstemp ] shows the dependence of the mean number of pes observed in the prompt pulse ( @xmath58 ) on the temperature with no electric field . shown in fig . [ fig : pevshv ] is @xmath58 plotted against the strength of the electric field for different temperatures . the prompt scintillation yield decreases by about 15% with an electric field of 45 kv / cm . the effect of the electric field on the prompt scintillation has little temperature dependence . ) vs the temperature.,width=340 ] ) vs the electric field strength.,width=340 ] our data shown in figs . [ fig : pevstemp ] and [ fig : pevshv ] are qualitatively consistent with what was observed in the earlier work by hereford and coworkers @xcite . ( compare , for example , fig . [ fig : pevstemp ] of this work with fig . 2 of ref . @xcite and fig . [ fig : pevshv ] of this work with fig . 3 of ref . @xcite . ) note , however , that a direct comparison between our data and the data in refs . @xcite and @xcite can not be made because of the difference in the integration time for the prompt pulse . the mean number of afterpulses per prompt pulse observed in the first 14 @xmath23s was determined by fitting a poisson distribution to the observed distribution of the number of afterpulses per event ( fig . [ fig : tdc_mult ] ) . the mean number of afterpulses per prompt pulse thus obtained ( @xmath59 ) is plotted against the temperature in fig . [ fig : apvst ] . our results , showing a reduction in the afterpulse intensity at lowered temperatures , are consistent with refs . @xcite and @xcite . ( although ref . @xcite only shows a plot of the `` pulse height '' which is the signal integrated for the first 1 @xmath23s , and the total intensity plotted against the temperature , the temperature dependence of the afterpulse intensity can be inferred from the data . ) s ( @xmath59 ) plotted as a function of the temperature for zero electric field.,width=340 ] figures [ fig : apvse ] and [ fig : apperpe ] show the mean number of the afterpulses ( @xmath59 ) and the mean number of the afterpulses normalized to the mean number of pes observed in the prompt pulse ( @xmath60 ) plotted against the strength of the electric field , respectively . we see from the figures that the afterpulse intensity decreases with an increasing electric field strength ( fig . [ fig : apvse ] ) and that the afterpulse intensity is more strongly affected by the electric field than the prompt scintillation ( fig . [ fig : apperpe ] ) . also , in contrast to the prompt scintillation , the effect of the electric field on the afterpulse intensity exhibits some temperature dependence . s ( @xmath59 ) plotted against the strength of the electric field.,width=340 ] ) plotted against the strength of the electric field.,width=340 ] what we have observed from our data can be summarized as follows : + 1 ) the prompt scintillation yield is reduced by 15% at @xmath28 kv / cm . + 2 ) the reduction in the prompt scintillation yield is approximately linear in the strength of the electric field . + 3 ) the effect of the electric field on the prompt scintillation yield has a very weak temperature dependence in the temperature range of 0.2 k to 1.1 k. + 4 ) the electric field has a stronger effect on the intensity of the delayed component than it has on the intensity of the prompt pulse . + 5 ) the observed temperature dependence of the prompt and afterpulse intensities at zero electric field is in qualitative agreement with previous work . + below we discuss the implication of these observations . the prompt scintillation signal is the result of the radiative decay of singlet he@xmath61 atoms in excited states and of he@xmath62 excimers also in singlet states . an @xmath0 particle produces these excimers and atoms either by ionization followed by recombination or by promoting an atom directly to an excited state without ionization . the ratio of the number of direct excitations to ionizations has been predicted in helium to be 0.45:1 @xcite . ( although this calculation was performed with 100 kev electrons as the primary particle , the expectation is that it holds for @xmath0 particles as well @xcite . ) among the atoms promoted directly to excited states , the ratio of singlets to triplets was calculated to be 5:1 . on the other hand , the ratio of singlets to triplets formed by recombination of ionized atoms is expected to be the statistical value of 1:3 for @xmath0 particles ( not the case when electrons are the primary ionizing particle @xcite ) . scintillation from atoms in excited states that have previously not undergone ionization is not affected by an electric field whereas the number of those that have been ionized is decreased in an electric field because of charge separation . were there no other complicating factors , the electric field - dependent fraction , @xmath52 , of the prompt scintillation would be @xmath63 . however , a number of phenomena can influence this fraction , most likely increasing its value . \1 ) the prompt signal for @xmath0 particles is substantially quenched @xcite by the nonradiative destruction of singlet excimers by the penning process [ eq . ( [ eq : penning2 ] ) ] but by the same mechanism it may be slightly enhanced with the destruction of triplet excimers and the subsequent recombination of the resulting ions into singlet states . excited state singlet atoms may also nonradiatively decay upon interaction with other species of excitations . ( nonradiative destruction of singlet species is discussed in sec . [ sec : quenching ] . ) \2 ) singlet helium atoms in excited states with principal quantum number , @xmath17 , of 3 or greater can autoionize by the hornbeck - molnar process @xcite @xmath64 since the binding energy of he@xmath65 is @xmath25 ev , which is greater than the energy to ionize a he(@xmath66 ) atom . based on the oscillator strengths for the transitions between the ground state and the various excited states of helium @xcite slightly more than one third of the atoms promoted to excited states will have a principal quantum number of 3 or greater , the other two thirds having @xmath67 . if all atoms with @xmath66 were to undergo autoionization , then the electric field - dependent fraction of the prompt scintillation would be increased to @xmath68 . various theoretical models have been developed that describe electron - ion recombination in an ionization track in fluid media : jaffe s columnar theory of recombination @xcite describes the case in which a dense plasma of positive and negative ions is formed along the ionization tracks while onsager s theory @xcite describes the case where each electron - ion pair is spatially separated and recombination occurs between an electron and its parent ion ( geminate recombination ) . for the highly ionizing track of an @xmath0 particle in lhe , the jaffe model would appear the more applicable . in the columnar theory , the rate of recombination is governed by the following equations : @xmath69 and @xmath70 where @xmath71 and @xmath72 are the densities of the snowballs and bubbles , @xmath73 and @xmath74 are the mobilities , @xmath75 and @xmath76 are the diffusion coefficients , and @xmath0 is the recombination coefficient . what this model describes are clouds of negative and position ions being pulled away from each other by the electric field while at the same time recombining at a rate proportional to the product of their densities . also , the clouds expand radially due to the diffusive motion of the charges . both the reduction in the scintillation yield and the generation of an ionization current result from the suppression of recombination due to the presence of an electric field . the motion and recombination of electrons and ions described by these equations occur on a time scale of @xmath5 s after an @xmath0 particle is stopped in helium . the processes associated with the decay and destruction of excimers created on recombination , processes that are not influenced by the electric field and occur on time scales of @xmath77 to @xmath78 s , are discussed separately in later sections . in looking for a solution to these rate equations jaffe assumed the diffusion terms were larger than those associated with recombination and hence , along with the electric field term , were the principal cause of the change in densities of the charges . this is a valid approach for ionization in gases . on the other hand , kramers @xcite , in attempting to explain the electric field dependence of the ionization currents from @xmath0 particles stopped in liquid helium measured by gerritsen @xcite , pointed out that in a dense fluid the diffusivity of the ions is small and that recombination has a much larger influence on the time dependence of the charge density . gerritsen s data for the electric field dependence of the ionization current can be fit approximately by the kramers theory with a cylindrical gaussian charge distribution having the same density and spatial distribution for @xmath71 and @xmath72 @xmath79 where @xmath80 is the initial number of positive ion snowballs or electron bubbles per unit length along the track . the diffusion coefficient was assumed to be zero . the relationship between the recombination coefficient and the mobilities is given by the langevin relation ( see below ) . the best fit of kramers theory to gerritsen s data with the one adjustable parameter , @xmath81 nm , is shown in fig . [ fig : gerritsen ] . the ionization current is normalized to the saturation current.,width=340 ] several comments regarding this result are relevant to understanding the scintillation and ionization currents of lhe . \1 ) the ionization current data contains contributions that are not included in the jaffe - kramers theory , which describes the ionization current due to charge carriers that escaped the initial recombination particle . ] . however , at electric field strengths below that corresponding to the saturation current , charge carriers that initially recombine can contribute to the ionization current through processes such as penning ionization [ eqs . ( [ eq : penning1 ] ) and ( [ eq : penning2 ] ) ] . a crude estimate indicates that such contributions could be larger than 10% . \2 ) the langevin relation between recombination coefficient and mobility is @xmath82 . in the approximation that the mobilities of the two species are the same , the electric field dependence of the current is independent of mobility , as it enters the recombination and electric field terms in the same manner . at the temperature within the track of approximately 2 k ( see below ) , the mobilities of the positive and negative ions are approximately the same @xcite , having the value of @xmath83 m@xmath84/(v@xmath85s ) . hence the recombination coefficient is @xmath86 m@xmath16/s . \3 ) from the einstein relation @xmath87 , the diffusion coefficient is calculated to be @xmath88 m@xmath84/s . since the initial charge densities along the track are the order of @xmath89 m@xmath90 , diffusion plays little role in affecting the density distribution at early times when recombination is dominant . at longer times when the concentrations have decreased substantially , diffusion can influence the charge separation . \4 ) the assumption that the positive ions and the electron bubbles initially have the same spatial distribution is unlikely to be correct . the ionization events from the @xmath0 particle or from secondary electrons with energy sufficient to create further ionizations occur in a small cylinder about the track of less than 10 nm . he ions do not move appreciably in the time it takes for the electrons to thermalize and form a bubble . however , once the energy of secondary electrons drops below 20 ev , the first excited state of he , the only process by which they can lose energy is elastic scattering from helium atoms . since the fraction of energy an electron loses in such a collision is very small , being dependent on the ratio of the mass of the electron to that of the atom , the order of @xmath91 collisions are required to decrease an initial energy of 10 ev to below 0.1 ev thought to be necessary for bubble formation . from the cross section for elastic scattering , which varies from @xmath92 cm@xmath84 between 1 and 10 ev , and the liquid density , the mean free path is roughly @xmath94 nm and the mean distance for diffusion , proportional to the square root of the number of scatters , is the order of 100 nm @xcite . but as mu@xmath95 oz _ et al . _ @xcite point out , if the initial positive and negative charge distributions are gaussian with very different radii , then the radial field arising from the different charge distributions can be large , easily in the range of 100 to 1000 kv / cm for an @xmath0 track in helium assuming 10 nm for the radius of the positive ions and 100 nm for the electrons . the two distinct distributions will therefore rapidly merge to form a more uniform distribution assumed in the jaffe and kramers theories . the positive ion snowballs having a mobility comparable to that of the electron bubble at the initial temperature of @xmath25 k along the track ( see below ) , will expand outward and the electrons contract inward . \5 ) a significant fraction of the secondary electron produced by an @xmath0 particle have energies above the he ionization energy of 24.6 ev and are therefore capable of creating further ionization . at low energies the probability that a secondary electron will have a particular energy @xmath96 decreases relatively slowly with increasing energy , varying as @xmath97 @xcite . since the cross section for ionization or excitation by an electron in the range of 25 to several hundred ev is the order of @xmath98 cm@xmath84 , these ionizations / excitations would occur a considerable distance from the track were it not for elastic scattering . the elastic scattering cross section for an 100 ev electron is @xmath99 cm@xmath84 so that energetic electrons will undergo many random scatters , on average remaining in the vicinity of the @xmath0 track . such electron may have the effect of modifying the charge distribution but not to the extent that a gaussian distribution is not a reasonable approximation . the measured electric field dependence of the prompt scintillation can be fit well using the kramers theory for a range of different gaussian widths , @xmath100 . the value of @xmath100 depends on the choice of the fraction , @xmath52 , of scintillation resulting from species created by ionization , which can be affected by the electric field . the results are shown in fig . [ fig : promptfit ] . figure [ fig : b_vs_x ] shows the @xmath101 per degree of freedom of the fit of kramers theory to the prompt scintillation data plotted against @xmath52 and @xmath100 . if @xmath52 is taken to be 0.95 then the best fit is for @xmath102 nm , while for @xmath103 , @xmath104 nm . in the kramers theory @xcite , the fraction of ions that recombine depends on a single parameter @xmath105 . that is , increasing the radius of the track @xmath100 has the same effect as increasing the electric field @xmath96 . from this it is expected that a smaller @xmath52 would require a larger @xmath100 . our results ( figs . [ fig : promptfit ] and [ fig : b_vs_x ] ) are consistent with this expectation . . the curves are calculated using eq.(9 ) of ref . @xcite.,width=340 ] per degree of freedom of the fit of kramers theory to the prompt scintillation data plotted against @xmath52 and @xmath100.,width=302 ] a few comments about the relationship between ionization current and the effect of an electric field on the prompt scintillation intensity are appropriate . these two types of data reflect two different aspects of the same phenomenon , and as such it should be possible to construct a model that describe both data consistently . however , such an attempt is complicated by a number of factors . 1 ) as mentioned earlier , the ionization current has a contribution from charge carriers that initially recombine . 2 ) there is uncertainty in the value of @xmath52 . 3 ) even with a short integration time ( in our case @xmath106 ns ) , there is inevitably a contribution to the prompt signal from the delayed component , which has a different electric field dependence ( see sec iv.c ) . because of these complications we have not attempted a unified analysis of our data and gerritsen s ionization current data . while we note that , as seen in figs . [ fig : promptfit ] and [ fig : b_vs_x ] , there is considerable range in the values of @xmath52 and @xmath100 that fit the data , the choice of @xmath107 and @xmath108 nm is consistent both with our estimate of the fraction @xmath52 and the best fit of kramers theory to gerritsen s data . as seen in fig . [ fig : promptfit ] , the effect of the electric field on the prompt scintillation has little temperature dependence . this is attributed to three causes : 1 ) the effect of the electric field on recombination is nearly independent of the mobility of the ions as discussed earlier . 2 ) the initial temperature of the track is in the vicinity of 2 k , irrespective of the temperature of the bulk liquid ( see below ) , resulting in the same mobility of ions in the track irrespective of the bulk liquid temperature . 3 ) the characteristic time of recombination @xmath109 , where @xmath110 is the initial ionization density , is shorter than the time it takes the @xmath0 track to thermalize with the bulk liquid @xmath111 , where @xmath112 is the sound velocity ( @xmath113 m / s ) . indeed , with @xmath114 nm , which gives @xmath115 m@xmath90 , the recombination time is @xmath116 s as compared to the thermalization time @xmath117 s. the fraction of energy of an @xmath0 particle that is quickly down converted to excitations of the superfluid , phonons and rotons , is sufficient to heat a column of liquid a few tens of nanometers radius along the track to the vicinity of 2 k irrespective of how low the temperature of the bulk liquid @xcite . since the density of liquid helium changes by less that 0.5% below 2 k , the initial environment in which the electrons , ions , excimers and excited state atoms reside , on their creation , is independent of the temperature of the bulk liquid . therefore , the temperature dependence of the prompt scintillation , illustrated in fig . [ fig : pevstemp ] , must result from the manner in which those entities responsible for the scintillation are influenced by the expansion of the roton / phonon cloud ( with velocity of @xmath118 m / s ) taking place at times the order of @xmath119 s. below a temperature of 0.6 k the density of thermal excitations in the bulk liquid is so low that the hot columnar cloud expands into essentially a vacuum . but at higher temperatures the thermal excitations in the bulk become sufficiently dense to scatter and slow the radially expanding cloud . as discussed below , the @xmath19 component of the afterpulse intensity also decreases with decreasing temperature . the temperature dependence of both the prompt and afterpulse signals is attributed to the more rapid decrease of the triplet excimer density due to expansion , leading to a reduced production of singlets via the penning process . hereford and coworkers @xcite , in modeling the temperature dependence of the scintillation they observed , argued that the increased diffusion at low temperatures results increased quenching , that is , the enhanced nonradiative destruction of the singlet species . however , their argument was not informed by the fact that the roton and phonon densities in the vicinity of the track is essentially independent of the temperature of the bulk liquid . following the work of mckinsey _ et al . _ @xcite the time dependence of the afterpulse scintillation for times between 0.4 @xmath23s and 14 @xmath23s has been characterized using the function @xmath120 @xmath121 , @xmath122 , @xmath123 , and @xmath124 were varied in the fit . for convenience , we define @xmath125 , @xmath126 , and @xmath127 to be the contributions to the number of afterpulses from the first , second , and third terms in eq . [ eq : apspectrum ] , respectively , that is @xmath128 @xmath129 and @xmath130 @xmath125 , @xmath126 , and @xmath127 are plotted as a function of the electric field for different temperatures in figs . [ fig : apexp ] , [ fig : aptinv ] , and [ fig : apconst ] respectively . in figs . [ fig : apexp ] and [ fig : aptinv ] , the 1.0 k data points for electric fields between 10 kv / cm and 20 kv / cm seem to show a larger scatter . we are not able to offer an explanation . the value of @xmath121 that we obtained was @xmath131 @xmath23s , slightly shorter than the value quoted in ref . @xcite . our results show that it is nearly independent of the temperature and the electric field . also in fig . [ fig : atinvvst ] , @xmath126 is plotted as a function of the temperature for zero electric field . , the number of afterpulses from the exponential component ( @xmath132 ) in the afterpulse time spectrum plotted against the strength of the electric field.,width=340 ] , the number of afterpulses from the @xmath133 component in the afterpulse time spectrum plotted against the strength of the electric field.,width=340 ] , the number of afterpulses from the constant component in the afterpulse time spectrum plotted against the strength of the electric field.,width=340 ] , the number of afterpulses from the @xmath133 component in the afterpulse time spectrum plotted as a function of the temperature for zero electric field.,width=340 ] at times longer than @xmath14 s all singlet state excimers and excited atoms ( except , perhaps , for the first excited state he(2@xmath134s ) , which in vacuum has a lifetime of 19 ms @xcite ) , will have long decayed . any photons produced at a later time must result from triplet states . mckinsey _ et al . _ @xcite hypothesized that the afterpulse term dependent exponentially on time was the result of he(2@xmath134s ) atoms forming excimers in a@xmath7 states from which they radiatively decay to the dissociated ground state of the helium pair . given that the lifetime of the triplet state he(2@xmath16s ) has been determined in the liquid to be 15 @xmath23s @xcite , this appears to be a reasonable hypothesis . furthermore , there appears to be little or no temperature dependence to the exponentially time - dependent term , as expected for such a source . on the other hand , one would not expect that a signal arising from a neutral atom to exhibit a dependence on electric field as we have observed ( see fig . [ fig : apexp ] ) , unless it originated through recombination . since the probability that ions form excimers prior to recombination is high , atomic he(2@xmath134s ) appears more likely to be produced by decay from a higher lying he(n@xmath134p ) state created de novo by electromagnetic excitation from the ground state than through a cascade from an atom formed by recombination . we are unable to offer an explanation of this exponentially time - dependent component of the afterpulse scintillation . keto _ et al . _ @xcite , using a beam of 160 kev electrons as the primary ionizing source , measured the long - time afterglow in the infrared spectra associated with a number of transitions between excited states of excimers and atoms to have a time dependence that varied as @xmath19 . they ascribed this time dependence to the production of the excited states by the penning ionization of triplet excimers , since the bimolecular process , @xmath135 results in @xmath17 proportional to @xmath19 for @xmath136 . a more detailed description of the time dependence of the singlet excimer density is expressed by eq . ( [ eq : quench ] ) in the next section . a similar relation can be given for the triplet density , the essentials of which for the purposes of this discussion are contained in eq . ( [ eq : birate ] ) . the problem of describing the time dependence of the afterpulse scintillation from an @xmath0 particle differs from that in the work of keto _ et al . _ @xcite since in one case the density of excimers is uniform whereas in the other the excimer cloud is expanding about the @xmath0 track . king and voltz @xcite developed a theoretical model of afterpulses that includes the spatial evolution of the excimer density with time . the description involves the kinetics of the diffusion of the triplets at times well after the prompt scintillation has decayed . if the triplets maintain a gaussian distribution as they expand by diffusion , then at long times the afterpulsing decreases approximately inversely with time . et al . _ @xcite cited this model by way of explaining the origin of the @xmath19 component of the afterpulses . at shorter times the king model predicts a time variation that is considerably faster than @xmath19 , but it does not provide an explanation of the exponential dependence on time , as experimentally observed . the amplitude of the @xmath19 afterpulse component exhibits interesting temperature and electric field dependences . firstly , the number of afterpulses varies much more with temperature than the number of photons in the prompt scintillation . secondly , the number of afterpulses decreases more rapidly with increasing electric field than the prompt signal . as with the slight variation of the prompt signal with temperature below 1 k , the rapid variation of the @xmath19 afterpulse component with temperature can be understood qualitatively as being the result of the change in propagation of the triplet excimers away from the @xmath0 track . while the temperature within the core of the track is not sensitive to the ambient temperature of the liquid , the thermal excitations in the surrounding liquid can affect the rate at which the excimers expand away from the track . at high temperatures , the excimers diffuse slowly because of scattering from phonons and rotons . at low temperatures the excimers propagate ballistically into the cold liquid in the absence of scattering . the excimer density drops much more rapidly at low temperatures and as a consequence the rate of generation of singlet species by the penning process is strongly temperature dependent . the strong electric field dependence of the @xmath19 component of the afterpulse results from the two separate ionization steps that are involved in the production of these photons . firstly , the initial creation of triplet excimers by recombination is affected by the electric field in the same manner as is the production of singlet excimers responsible for the prompt signal . secondly , the recombination of the electron - ion pair produced by the penning process involving the triplet excimers is also influenced by the electric field . those recombinations that result in singlet species are responsible for the @xmath19 component . since the delayed generation of the electron - ion pair occurs in isolation from other charges , the recombination is geminate . the onsager theory of geminate recombination @xcite is not an appropriate approach to calculating the effect of electric field on the separation of charges created via penning ionization in liquid helium at low temperature . that theory assumes diffusion to be the dominant process affecting the charge motion in the presence of the field . at low field and high temperatures where the ion mobilities are low the motion of the charges is governed by viscous drag . in this case the charges move along field lines , and whether two charges recombine or not is dependent only on the magnitude and direction of their initial separation with respect to an applied electric field . if @xmath137 the initial separation and @xmath138 is the angle between @xmath137 and the applied electric field @xmath96 , the charges will recombine if @xmath139 geminate recombination in lhe under this condition was studied in ref . @xcite using a @xmath15 source at 1.5 k. however , at 1 k where the mobility of ions is the order of @xmath140 m@xmath84/v@xmath85s the inertial term in the equation of motion becomes important at modest fields . in the limit of high mobility charges will recombine if @xmath141 independent of orientation of the charges with respect to direction of the field . if this expression were used in conjunction with a gaussian distribution of charge separations to fit the field dependence of the @xmath19 component of the afterpulse signal , the width @xmath100 of the distribution would be unreasonably small , much less than the 60 nm determined for the distribution of electrons about an alpha track . since the average energy of electrons created by the penning process responsible for the @xmath19 afterpulse component is comparable to the average energy of secondary electrons from an alpha , the inclusion of the inertial term in describing the ion motion does not provide a satisfactory explanation to the field dependence . at low temperatures and at zero pressure both positive ion snowballs and electron bubbles moving in liquid helium create quantized vortex rings @xcite when their velocities reach the order of 60 m / s . since ion mobilities are the order of @xmath140 m@xmath142v@xmath85s , at 1 k , fields of less than 1 kv / cm are sufficient for vortex creation . at low fields the charges remain attached to the rings that they have produced , but in very high fields , greater than 25 kv / cm at low temperatures , it has been found @xcite that electrons no longer remain attached to vortices but create and shed vortex rings as they move through the superfluid . we have not attempted to apply the theory of vortex creation by moving ions @xcite to explain the field dependence of the @xmath19 component of the afterpulses . the energy a charge acquires in moving in the field is transferred to the vortex , which increases its diameter and decreases its velocity . qualitatively , the vortices introduce an enormous drag on the motion of the ions , which has the effect of retarding their separation by the applied field . presumably vortices are the reason the field dependence of the afterpulses produced by penning ionization is not stronger than observed . the constant component of the afterpulses , independent of time , is most naturally associated with triplet excimers whose density is constant and are uniformly distributed throughout the volume of the cell , the rate of destruction being balanced by the production from the 300 bq @xmath0 source . destruction occurs in part by the spontaneous radiative decay of triplet excimers to the dissociated ground state and in part by the penning process . the most notable features of the data seen in fig . [ fig : apconst ] are the apparent weak field dependence of this afterpulse component at 1 k and the substantial decrease in intensity as the temperature is lower from 1 k to 0.4 k. these two properties are naturally explained by the scintillation being principally due to the spontaneous radiative decay of the excimers , the @xmath143 state having a lifetime of 13 s in the bulk @xcite . while the electric field is not expected to affect this process , it does lead to a decreased recombination and density of triplets with increasing field . below about 0.6 k the mean free path of the excimers becomes long , comparable to the dimensions of the helium cell , so that nonradiative destruction of the excimers on solid surfaces decreases the photon yield at low temperatures . one observation remains unexplained . hereford and coworkers @xcite noted that the total scintillation below 0.5 k depended on the geometry of their cell , the smaller cell having somewhat greater light output . the scintillation intensity decreased with decreasing temperature to a minimum at 0.4 k and then increased on going from 0.4 k to 0.2 k. we have observed similar behavior of the time - independent component of the afterpulse signal at low temperatures ( see figs . [ fig : apvst ] and [ fig : atinvvst ] ) . as discussed in the introduction , in the nedm experiment , lhe scintillation produced by the products of the neutron capture reaction on @xmath16he will be used as the neutron spin analyzer . in this reaction , the @xmath144 value of 760 kev is shared between the two reaction products , a proton and a triton ; the proton and the triton are emitted with kinetic energies of 570 kev and 190 kev respectively . the range of a 570-kev proton in superfluid helium is 0.06 mm , corresponding to an average energy deposition rate of @xmath145 ev/@xmath23 m @xcite . a 190-kev triton travels 0.018 mm in superfluid helium depositing energy at a rate of @xmath146 ev/@xmath23 m before it comes to rest @xcite . in both cases , the average energy deposition rate and hence the number of ionizations per unit length along the track ( @xmath80 ) are almost exactly half of that for a 5.5-mev @xmath0 particle ( @xmath147 ev/@xmath23 m ) . the radius of the ionization track @xmath100 is expected to be the same for @xmath0 particles and the products of neutron capture reaction on @xmath16he because the track radius is determined by the diffusion of the electrons . therefore , the ionization density inside the tracks created by the reaction products of the @xmath16he(@xmath17,@xmath18)@xmath16h reaction is also almost exactly half of that in the track created by a 5.5-mev @xmath0 particle . the lower number of ionizations per unit length along the track ( and hence the lower ionization density ) causes lhe scintillation produced by the products of the @xmath16he(@xmath17,@xmath18)@xmath16h reaction to differ from @xmath0-induced lhe scintillation in two aspects : 1 ) the lower ionization density results in reduced quenching , thereby a larger fraction of the deposited energy being emitted as prompt scintillation compared to @xmath0-induced scintillation . 2 ) as noted earlier in sec . [ sec : promptdiscussion ] , in the kramers theory the fraction of ions that recombine depends on a single parameter @xmath105 . therefore , the lower number of ionizations per unit length along the track means that the effect of an electric field is larger . the effect of the quenching by the nonradiative destruction of singlet species by the penning process can be described by the following equation : @xmath148 where @xmath149 and @xmath150 are the densities of the singlet and triplet species , respectively , @xmath151 is the coefficient for bimolecular decay involving singlet species , @xmath152 is the coefficient for bimolecular decay of the triplet species , @xmath153 is the diffusion coefficient of the singlet species , and @xmath121 is the radiative lifetime of the singlet species . @xmath154 , @xmath155 , and @xmath156 are numerical factors that come from the fact that in the penning process for every two excimers ( or excited atoms ) destroyed , a new one is formed , one - quarter of the time in the singlet state and three - quarters of the time in the triplet state . the diffusion term can be neglected for the conditions we are interested in , namely , the first @xmath78 s of the hot track . also the term describing the penning ionization of two triplet states can be neglected since @xmath151 is sufficiently larger than @xmath152 ( see below ) . noting that @xmath150 changes with time more slowly than @xmath149 , we solve eq . ( [ eq : quench ] ) for two limiting cases : 1 ) @xmath150 changes with time keeping the @xmath157 ratio constant , that is @xmath158 , where @xmath159 is a time independent constant . 2 ) @xmath150 does not vary with time , that is @xmath160 , where @xmath161 is the initial triplet density . solving eq . ( [ eq : quench ] ) yields an analytical solution of the same form for both cases . integrating the last term of the right hand side of eq . ( [ eq : quench ] ) over time gives the fraction , @xmath162 , of the singlets that contribute to the prompt scintillation escaping the bimolecular annihilation . @xmath163 where @xmath164 is the initial singlet density and @xmath165 . for the case in which @xmath158 , @xmath166 and @xmath167 are given by @xmath168 and @xmath169 . for the case where @xmath170 , @xmath166 and @xmath167 are given by @xmath171 and @xmath172 . for @xmath0-induced lhe scintillation , the fraction of deposited energy that is emitted as prompt scintillation was found to be 10% @xcite . on the other hand , following a similar discussion to the one in the beginning of sec . [ sec : promptdiscussion ] , we expect 23% of the deposited energy to be emitted as prompt scintillation in the absence of quenching , giving @xmath173 for @xmath0-induced lhe scintillation . solving eq . ( [ eq : fs ] ) for @xmath174 gives @xmath175 for @xmath0-induced scintillation . the numerical value of @xmath176 thus obtained allows us to make a crude estimate of the coefficient for bimolecular decay of the singlet species . for @xmath177 , with @xmath178 cm@xmath90 and @xmath180 s , @xmath175 gives @xmath181 cm@xmath16/s for the singlet excimers . this value for @xmath151 is about an order of magnitude larger than the value measured for the coefficient for bimolecular decay of the triplet excimers @xcite . the difference could be attributed to the suppression of this process for the triplets states due to the electron spin flip required for the triplet excimer to go to the dissociated ground state . similar differences have been observed in the deexcitation rates of singlet and triplet states of helium atoms when encountered by other atoms and compounds @xcite . assuming that the quenching fraction is the same for excimers and excited atoms , setting @xmath182 in eq . ( [ eq : fs ] ) yields @xmath183 for lhe scintillation induced by the @xmath16he(@xmath17,@xmath18)@xmath16h reaction . it follows that the fraction of deposited energy emitted as prompt scintillation is expected to be 13% for lhe scintillation produced by the products of the @xmath16he(@xmath17,@xmath18)@xmath16h . note that this result is somewhat independent of the details of the model . as long as the differential equation for @xmath149 has the form @xmath184 , eq . ( [ eq : fs ] ) holds and we obtain the same value for @xmath162 for lhe scintillation induced by the @xmath16he(@xmath17,@xmath18)@xmath16h reaction . with the model for quenching discussed above and kramers theory , we can make a prediction on the number of the prompt euv photons emitted when a neutron is captured by a @xmath16he atom in superfluid lhe . in fig . [ fig : neuv_3he_capture ] the predicted number of the prompt euv photons for @xmath16he(@xmath17,@xmath18)@xmath16h is plotted as a function of the electric field . he(@xmath17,@xmath18)@xmath16h reaction with @xmath185 and @xmath104 nm.,width=340 ] habicht @xcite in his thesis reports measurements of the scintillation resulting from a number of different ionization sources including the @xmath16he(@xmath17,@xmath18)@xmath16h reaction in liquid helium at zero field and 1.8 k. however , it is difficult to make a comparison with his results given the lack of specificity of the experimental parameters such as geometry , solid angle , etc . the prompt scintillation signal from @xmath0 particles stopped in helium exhibits a 15% reduction in an electric field of 45 kv / cm . this field dependence is consistent with the current versus field measurements of gerritsen @xcite and kramers analysis of columnar recombination . we conclude using kramers theory that roughly 40% of the scintillation results from species formed from atoms originally promoted to excited states by the @xmath0 particle and 60% from excimers created by ionization and subsequent recombination , with the electrons initially having a cylindrical gaussian distribution about the @xmath0 track of 60 nm . the delayed scintillation signal , the time dependence of which is decomposed in the manner suggested by mckinsey _ et al . _ @xcite , exhibits stronger field and temperature dependences than does the prompt scintillation . the stronger field dependence is the consequence of the fact that the slow component of afterpulses are from triplet excimers undergoing the penning ionization process followed by recombination forming singlet excimers . as such , it receives the effect of the electric field twice , once at the initial recombination producing triplet excimers , and once at the recombination following the penning process . the temperature dependence involves the diffusion of the excimers away from the @xmath0 track into the surrounding bulk liquid . the authors are grateful for the help and advice provided by the following individuals at various stages of this work : e. bond , m. fanning , s. currie , d. g. haase , g. frossati , r. golub , w. guo , e. korobkina , h .- o . meyer , c. nelson , j. self , b. lozowski , t. rinckel , p. childress , d. baxter , w. m. snow , and a. edwards . this work was supported by the us department of energy , the national science foundation , and the doe mie no.71re . g. archibald , j. boissevain , r. golub , c. r. gould , m. e. mayden , e. korobkina , w. s. wilburn , and j. zhou , in _ proceedings of the 24th international conference on low temperature physics ( lt24 ) _ , edited by y. takano , s. p. hershfield , s. o. hill , p. j. hirschfeld , and a. m. goldman , aip conf . proc . * 850 * , 143 ( 2006 ) . m. j. berger , j. s. coursey , m. a. zucker , and j. chang , _ estar , pstar , and astar : computer program for calculating stopping - power and range tables for electrons , protons and helium ions _ , national institute of standards and technology , gaithersburg , md ( 2005 ) , http://physics.nist.gov/star [ originally published as : m. j. berger , nistir 4999 , national institute of standards and technology , gaithersburg md ( 1993 ) ] .

we describe a new , direct method for determining the opacity of foreground galaxies which does not require any _ a priori _ assumptions about the spatial distribution or the reddening law of the obscuring material . the method is to measure the colors and counts of background galaxies which can be identified through the foreground system . the method is calibrated , and the effects of confusion and obscuration are decoupled by adding various versions of a suitable deep reference frame containing only field galaxies with known properties into the image of the foreground galaxy , and analyzing these `` synthetic field '' images in the same way as the real images . we test the method on hst wfpc2 archived images of two galaxies which are quite different : ngc 4536 is a large sc spiral , and ngc 3664 is a small magellanic irregular . the reference frames are taken from the hubble deep field . from the background galaxy counts , ngc 4536 shows an extinction 1 mag in the northwestern arm region , and lower than 0.5 mag in the corresponding interarm region ( no correction for inclination has been attempted ) . however , from the galaxy colors , the same reddening of is observed in both the arm and the interarm regions . in the interarm region , the combination of extinction and reddening can be explained by a diffuse component with a galactic reddening law ( ) . in the spiral arm , however , the same diffuse , low opacity component seems to coexist with regions of much higher opacity . since the exposures are shorter the results for ngc 3664 are less clear , but also appear to be consistent with a two component distribution . 2h 3coco 3nh # 110 2 3 # 1 # 2 # 3 # 4 # 1 , # 2 , # 3 , # 4 # 1#2=.45 = .45 # 1 = 0.85 # 1#2#3#4#5#6#7 to#2 ' '' '' = 2 = 2

astro-ph9806001

the issue of whether galaxy disks are significantly opaque is a lively topic of current research and debate ( e.g. , davies & burstein 1995 , and papers therein ) , and the conclusions of different workers often differ widely . statistical evidence that most large presented by valentijn ( 1990 ) based on a diameter - selected sample of galaxies from the eso survey . valentijn s result indicated that the bright galaxies are substantially opaque right out to the 25 @xmath9 mag arcsec@xmath10 isophote . this conclusion has been challenged by burstein and others , who initially questioned the selection effects inherent in the sample used by valentijn , and have maintained that very little extinction is present . more detailed studies on individual galaxies have generally concluded in favor of at least some extinction ( e.g. , studies of multicolor optical and infrared images by block et al . 1994 , and peletier et al . there is also some indication that even galaxy halos may contain dust ; zaritski ( 1994 ) compared faint galaxies in fields bordering nearby spirals ( ngc 2835 and ngc 3521 ) and in control fields further away on the sky , and found a small difference in the average colors . however , from analysis of surface photometry on a sample of 1700 galaxies , giovanelli et al . ( 1995 ) have recently concluded that less luminous galaxies are completely transparent , and that even the most luminous spirals suffer from only small amounts of extinction ( 0.1 mag ) at an isophotal radius of 23.5 mag arcsec@xmath10 in @xmath11 . tests of the opacity of spiral disks which rely on intrinsic properties of the galaxies ( surface brightness , inclination angle , isophotal profiles ) are strongly dependent on the necessary assumption of just how the dust is distributed within the galaxy ( witt , thronson , & capuano 1992 ; witt & gordon 1996 ) . for instance , layered models with a physically thin but optically thick dust layer behave like optically thin disks ( disney , davies , & phillips 1989 ) . in the case of edge - on galaxies , surface photometry can help determine the geometrical dust distribution ( kylafis & bahcall 1987 ) , but in more general cases the results are inconclusive . the use of stars or star clusters embedded in the galaxy can only tell us something about the remaining opacity between us and the star , and is clearly biased to low opacities . hii regions have also been used ; they have the same bias , but in addition those bright enough to be easily observed may alter the surrounding ism so drastically that they are poor test objects ( e.g. , allen et al . 1997 ) . a more direct approach has been proposed by white & keel ( 1992 ) , and followed up by white , keel , & conselice ( 1996 ) and berlind et al . they have looked at partially - overlapping galaxy pairs , the first two studies at optical wavelengths , the last one in both the optical and the near - infrared . assuming that the background galaxy isophotes are smooth and/or symmetric , they use the unobscured part of the background galaxy to calculate the amount of background light that is lost in crossing the foreground disk . all three studies have found higher extinction in the arms than in the interarm regions , and extinction curves flatter than galactic , consistent with a patchy dust distribution in the foreground galaxy . the use of background illuminating objects has the advantage that no assumptions need to be made about the relative distribution of dust and light in the foreground galaxy . also , the overlapping - galaxy method in particular provides results with good spatial resolution . however , it is necessary to assume that the light distribution in the background galaxy image is smooth and centrally symmetric . in addition , there is an uncertain contribution of scattered light from the background galaxy into the line of sight . but perhaps the biggest disadvantage is that there are only about two dozen galaxy pairs which offer the right geometry for the method to be applied ( white et al . 1996 ) . a potentially more general way to establish the opacity of a galaxy is to look for effects on the counts , colors , and morphologies of distant field galaxies seen through the disk of the foreground galaxy . this conceptually simple approach unfortunately suffers from an equally simple problem ; the general distribution of disk stars in the foreground galaxy reduces the contrast on background galaxy images , and foreground star clusters and hii regions add strongly to the confusion . high sensitivity is required in order to reach a surface density of background galaxies which is large enough to derive a statistically significant result . high angular resolution and color information are required to assist with the separation and identification of background galaxy images from objects in the foreground galaxy . hubble space telescope ( hst ) wide - field planetary camera 2 ( wfpc2 ) images offer a significant advantage over images from ground - based telescopes in their greatly improved resolution . this reduces the perturbation of faint galaxy images by nearby foreground objects , and facilitates the separation of stellar objects from galaxies . color information is also important ; a foreground compact hii region will be blue compared to the image of a distant galaxy , whether or not there is any amount of obscuration . once galaxies have been identified , their counts and colors can be compared with counts and colors of galaxies in reference fields not covered by a foreground galaxy . the hst wfpc2 archive contains many images of background galaxy fields and of fields covering parts of nearby galaxies . as an example of the data available for this study , and an illustration of the analysis problems which must be overcome before quantitative results can be obtained , we show in the left panel of figure 1 one wf frame ( wf3 , 13 @xmath12 13 in size ) from a typical reference field ( the hubble deep field ) , and in the right panel a similar - sized portion of a deep exposure on ngc 4536 . we shall describe both of these images in more detail in following sections of this paper . for the moment it is sufficient to note that the differences in the counts and colors of background galaxies visible in the two panels contain the information we seek on the opacity of the foreground galaxy . only a few background galaxies can be discerned visually with any certainty in the right panel . however , before we can conclude that the foreground galaxy is quite opaque , we obviously need to understand quantitatively how the data analysis methods we use to determine the counts and colors of the background galaxies are affected by structural features in the foreground galaxy . in this paper we describe a new method , the _ synthetic field method _ , which permits us to quantify and calibrate the effects of crowding and confusion and to determine the accuracy with which statements can be made about the total opacity and reddening of an average line of sight through the foreground galaxy . these synthetic fields are obtained by adding the hdf reference field images into the foreground galaxy image ; the combined image is then processed in the same way as the original foreground galaxy image alone . the procedure is repeated with different values of extinction applied to the reference field , and a plot drawn of the final number counts in each simulation versus the assumed extinction in order to ascertain the `` best '' fit to the actual counts of real background galaxies . can be estimated separately from the biases owing to confusion and crowding in the foreground galaxy field . finally , a comparison between the average colors of , respectively , the control hdf galaxies and the real background galaxies will allow an estimate of the degree of reddening . compared to the foreground galaxy fields we analyze , the hdf frames are virtually noiseless ; synthetic fields can therefore be constructed by adding suitably attenuated versions of them directly into the foreground galaxy images , without degrading the sensitivity of the data . figure [ histograms ] shows 4 cumulative histograms of galaxy counts in @xmath13 images which illustrate the steps in the _ synthetic field method _ : histogram _ a _ is for field galaxies in a version of the hdf to which noise has been added in order to approximate the shorter integration time used for the exposures of ngc 4536 , as well as the higher sky background at the position of the spiral ; _ b _ is for hdf `` control '' galaxies in an un - extincted ngc 4536 simulation ; _ c _ is for hdf galaxies in a simulation with @xmath14 0.6 mag of extinction at @xmath11 control field has been attenuated to mimic @xmath14 0.6 mag of extinction . ] ; and finally _ d _ is for real background galaxies in the ngc 4536 field . from the figure , we would conclude that there is about 1 mag of extinction at @xmath11 averaged over the wf2 and wf3 fields in this region of the foreground galaxy , and about 0.1 mag of average extinction at @xmath11 in the region of ngc 4536 covered by the wf4 field . clearly , one major limitation of our method is that the results are only statistical in nature , i.e. , the method does not give an answer for the extinction and reddening along any _ specific _ line of sight through the foreground galaxy , but only a statistical estimate of these quantities averaged over some substantial fraction of the foreground galaxy image . the accuracy of the method is limited not only by the counting statistics of background galaxies , but also by the intrinsic inhomogeneities in their spatial distribution . these uncertainties must also be quantified in order to provide a convincing result . in addition , there are several smaller selection effects which plague the analysis and which will be discussed later ; for instance , the fact that field galaxies also show a mild color magnitude dependence ( tyson 1988 ; williams et al . 1996 ) must also be taken into account . another limitation is that , while the numbers of galaxies probe the whole disk ( there is useful information in both detecting or not detecting a background galaxy ) , the conclusions from the colors of the galaxies will be biased in favor of regions of lower opacity ( because we need to detect the galaxies in order to measure there colors ) . therefore , it is possible to determine an extinction law by combining the number - count and color information only when we can safely assume that nowhere in the region of interest is the extinction so high that we can not sample it with background galaxies ( see appendix ) . in this paper we present a description of the _ synthetic field method _ through a detailed application to deep wfpc2 images of the sc spiral ngc 4536 . in order to examine the range of applicability of the method , we also analyze shallow images of the sbm magellanic dwarf irregular ngc 3664 ( ddo95 ) . we find substantial opacity in both galaxies , although the precision of the result is low for ngc 3664 , a consequence of the shorter exposure time used for the observations of this galaxy . in future papers we intend to apply the method to images of other nearby galaxies drawn from the hst archives . we begin in [ reffield ] with a discussion of the reference field and the processing necessary in order to render it useful for our present purposes . next in [ n4536 ] we describe the data on the foregound galaxy ngc 4536 which we have chosen to illustrate the method . section [ photometry ] describes the methods we use to identify and measure the magnitudes of field galaxies . section [ results ] presents the synthetic field simulations and the results on extinction through the disk of ngc 4536 . section [ n3664 ] presents an abbreviated discussion applying the method to ngc 3664 . in [ errors ] we examine the various sources of errors and biases we have been able to identify . finally , we summarize and discuss our results on the extinction and reddening through these two galaxies in [ summary ] . the choice of the reference field is clearly of prime importance to this method . for simplicity , images of this field ought to be available with the same camera and range of colors as the images of the foreground galaxy , and ought to be entirely representative of the sky in areas free of large nearby foreground galaxies . there are by now many fields observed with the wfpc2 which could satisfy these criteria , including the medium deep survey fields ( typically exposures of 5000 s in the f606w and f814w passbands ; griffiths et al . 1994 ; ratnatunga , griffiths , & ostrander 1998 ) and the field around the weak radio - galaxy 53w002 ( 24 orbits in 3 colors ; e.g. , pascarelle et al . 1996a , b ) . however , the _ hubble deep field _ ( hdf ; 150 orbits in 4 colors , williams et al . 1996 ) offers the deepest set of images . its properties are also presently the object of intense scrutiny by many workers , and detailed analyses of the counts , colors , and redshifts of the galaxies visible in it are becoming available ( e.g. , cohen et al . 1996 ; gwyn & hartwick 1996 ; villumsen , freudling , & da costa 1997 ; colley et al . 1996 ; madau et al . 1996 ) . the question as to what extent our chosen reference field is `` representative '' can therefore be answered with considerable certainty . for example , at @xmath15 26 ( appropriate to the present work ) , the number - counts of the hdf galaxies are known to be complete , and the galaxy number - magnitude relation is similar to that of other fields ( williams et al . the hdf therefore provides an excellent reference field for our purposes . we used the version 2 release wfc images of the hdf ( williams et al . 1996 ) retrieved from the stsci / hst archive . this release of the hdf was made from dithered exposures using the `` drizzling '' method ( fruchter & hook 1997 ) . there are 3 contiguous wfc fields in this data set ( and one pc field which we do not use here ) . each wfc field was recorded in the 4 photometric bands designated f300w , f450w , f606w , and f814w ( cf . the hst data handbook ) ; we have not used the images in the uv band f336w . several processing steps were carried out on the hdf data set in order to render it compatible with the images of ngc 4536 and ngc 3664 . first , the hdf images were all re - binned to their original pixel size of 01 since this is the scale of the ( un - dithered ) exposures of ngc 4536 and ngc 3664 ( and most other exposures of nearby galaxies in the hst archives ) . second , since there is no hdf image in the f555w passband used for the foreground galaxies , we created one by interpolating linearly between the existing f450w and f606w images , taking into account the differing responses of the three filters . this procedure will produce the right result for galaxies that have a linear spectrum between 4520 @xmath16 and 5935 @xmath16 , and an error in the flux at 5400 @xmath16 proportional to the deviation of a galaxy spectrum from linear in this wavelength range . for redshifts lower than @xmath17 = 0.13 ( v @xmath14 40000 km s@xmath18 ) , the 4000 @xmath16 break is blueward of the interpolation range , and the spectra should be fairly linear . however , a significant fraction of the hdf galaxies recovered in the simulations have @xmath19 . most of the recovered hdf galaxies have an observed ( isophotal ) @xmath20 brighter than 25 ( figure [ fhdfi ] ) . from the estimates of villumsen et al . ( 1997 ) for @xmath21 magnitude limits , and assuming @xmath22 ( smail et al . 1995 ) , the recovered hdf galaxies should be at redshifts @xmath23 . these galaxies are also mostly brighter than @xmath24 = 26 ( figure [ fhdfv ] ) . the majority of hdf galaxies brighter than @xmath25 = 26 are bluer than 1.5 in ( @xmath26 ) ( williams et al . 1996 , their figure 32 ) . these estimates allow us to model the error in our interpolated f555w image by using stellar population synthesis models ( babul & ferguson 1996 ) in order to produce spectra of galaxies with different star formation rates . from the results for the synthetic spectra , galaxies with @xmath27 1 and ( @xmath26 ) @xmath28 1.5 could suffer from a systematic shift of @xmath14 -0.03 mag in @xmath24 . the error will be in the sense of _ very _ slightly overestimating the reddening of the real galaxies relative to the simulations . however , the dispersion in ( @xmath29 ) is @xmath14 0.08 mag , of the same order as the dispersion of the mean ( @xmath30 ) color between the 3 hdf wfc fields . we conclude that the error introduced into the interpolated f555w image is negligible . the hdf images at the full sensitivity of the original observations are added directly to the images of the foreground galaxies in order to create the synthetic fields . the signal - to - noise ratio of the hdf is so high that this simple addition results in an image which retains the noise level of the foreground galaxy image . before the addition , the hdf can be attenuated and reddened by any amount in order to create a set of synthetic images in which one can search for and measure `` real '' background galaxies as well as the hdf `` control '' galaxies . in order to provide a visual demonstration of how many galaxies in a `` representative '' field are lost to _ both _ crowding _ and _ extinction , we have also created a special version of the hdf , simply by adding an appropriate amount of noise to it . we refer to this as the `` degraded '' hdf . after adding the noise , its signal - to - noise ratio is the same as that of the images of the foreground galaxy , which have a shorter exposure time and a higher sky background . the wf3 frame from the `` degraded '' hdf image is shown in figure 1 ( left panel ) . we produced this color image by combining together the @xmath31 frame as blue , the @xmath13 frame as red , and an average of the two as green . the image is displayed in a square root scale which favors the dynamic range of the ngc 4536 color image ; this latter image is shown in the right panel of figure 1 . the `` degraded '' hdf represents the number of galaxies that would be recovered in the _ total absence of any foreground galaxy _ , and hence without confusion with globular clusters , hii regions , and stars , but at the same position in the sky and with the same exposure time as the images of ngc 4536 . the differences in the numbers of background gal- axies which can be recognized between the two frames in figure 1 is largely due to crowding and confusion , as will be seen in [ completeness ] from the comparison between the `` degraded '' hdf and the synthetic images without extinction . these data , however , also imply a significant amount of extinction , as we shall subsequently show . the power of the synthetic field method resides in its ability to separate the effects of extinction from those of confusion and crowding ; such a large number of galaxies is lost to crowding only , and it has such significant effects on the photometry ( [ photerr ] ) , that it would be impossible to do this project without accounting properly for it . the images of ngc 4536 were originally acquired with the hubble space telescope ( hst ) wide - field planetary camera 2 ( wfpc2 ; holtzman et al . 1995 ) for the distance scale key project ; they have already been published by saha et al . ( 1996 ) , who give a very detailed account of the observations and the pipeline calibration of the data . the wfpc2 field is shown in figure [ fn4536 ] , overlaid on an stsci digitized sky survey image of the galaxy . we do not use the pc frame ( wf1 ) in this work . ngc 4536 is a bright spiral galaxy sbc(rs)i - ii ( de vaucouleurs et al . 1991 ) , with @xmath32 = 379 , in the direction of virgo , at a distance of 16.2 mpc ( saha et al . this yields a spatial resolution of @xmath14 8 pc per wf pixel . the galactic extinction towards ngc 4536 is @xmath33 = 0.00 ( de vaucouleurs et al . 1991 ) . there are 34 individual exposures in the f555w passband , taken at 17 discrete epochs between 1994 june 3 and 1994 august 9 ; in the f814w passband , 10 images were taken at 5 discrete epochs , over the same two - month period . for each filter , two individual exposures 2000 seconds long were taken back - to - back in each epoch , to allow for cosmic ray ( cr ) removal by an anti - coincidence technique described in anderson et al . 1995 and in saha et al . 1996 . we began working on the images as retrieved from the archive , i.e. , bias - subtracted and flat - fielded . the values of `` warm '' pixels ( identified in the calibration pipeline as having a moderately high dark count during the time of observation ) were repaired by replacing the measured dark current with the interpolated dark value between two bracketing dark frames ( voit 1997 ) . we then removed the cosmic rays in each image using the `` cr - split '' procedure ( anderson et al . 1995 ) . since we are not looking for time variability , we combined all the data in each passband into a single deep image . relative shifts between the different epochs were found with a cross - correlation technique . the images at all epochs were nominally taken with the same telescope roll - angle ; however , given that distinguishing between point and extended objects is crucial for our project , we also used this technique to check that in fact there were no unintended relative rotations between the frames . finally , the images were corrected for geometric distortion ( trauger et al . 1995 ) , aligned , and co - added using the `` drizzle '' variable - pixel linear reconstruction algorithm ( fruchter & hook 1997 ) . the resulting images have total exposure times of , respectively , @xmath34 seconds in the f555w passband , and @xmath35 seconds in the f814w passband . we used sextractor ( bertin & arnouts 1996 ) to perform the @xmath13 and @xmath36 photometry on all objects , up to the @xmath37 arcsec@xmath10 isophote in @xmath13 . the parameters for the search and photometry were chosen to optimize the background determination , as judged from the lack of `` holes '' ( oversubtracted regions ) in the background - subtracted ngc 4536 images . likewise , the threshold surface brightness was selected so as not to `` find '' noise . we performed the search and fixed the threshold in the f814w passband not only because of the actual selection bias in favor of red objects ( see below [ colbias ] ) , but also because our photometry was limited by the shorter exposure time of the @xmath13 frame . from the list of all objects found by sextractor , we selected the background galaxies by visual inspection of the images . we used the morphology , the fuzziness and , especially in crowded regions , the color . in crowded regions , we selected preferentially fuzzy objects with @xmath38 . our search and selection criteria mean that , in principle , from the number - counts we will measure the extinction at @xmath13 . figure [ cbgi4536 ] shows the cumulative histograms of real galaxies found in the background of ngc 4536 , in the f814w passband . the left panel of this figure displays the results for the wf2 and wf3 fields combined ( the numbers are normalized to the area of one wf field ) ; the right panel shows the results for the wf4 image . we analyzed the wf2 and wf3 images of ngc 4536 together , given that they span most of the northwestern arm of the galaxy . in contrast , the wf4 field covers the northern interarm region and a small area that appears to be beyond the visible disk of the galaxy . we found a total of , respectively , 14 and 16 real background galaxies in the wf2 and wf3 fields , and 35 galaxies in the wf4 field . the difference between the number - counts in this figure and the number - counts of the hdf galaxies ( figure [ cumi ] ) is caused by the combined effects of systematics , crowding and , finally , what we are interested in : extinction . synthetic fields were produced with varying + amounts of ( spatially uniform ) attenuation and reddening ( table 2 ) . the idea is to attenuate the hdf fields until one recovers the same number of simulated and real background galaxies . this attenuation , derived from the number - counts , should correspond to the average extinction through the disk of ngc 4536 to an uncertainty given by the poisson and clustering errors . in practice , what we measure is : @xmath39 @xmath40 is the normalization , in this case the number of hdf galaxies recovered in the synthetic fields without attenuation , and @xmath41 parametrizes the dependence of the number vs. extinction curve on object selection biases ( [ ext4536 ] ) . the reddening suffered by the background galaxies is found by comparing their average color to the average color of the hdf galaxies recovered from the synthetic fields . because galaxies have a color magnitude dependence , one must be careful to perform this comparison with simulations that have the same extinction as that measured ( from the counts ) for the real galaxies or that have the same magnitude limit as the extinction - corrected limit of the real galaxies . the selection biases of the real and synthetic samples of galaxies , and especially the color selection bias ( [ colbias ] ) , should also be the same for the comparison to be meaningful . to produce the synthetic fields , we : ( 1 ) scaled the hdf wfc frames to the exposure times of the ngc 4536 mosaics ; ( 2 ) attenuated the hdf frames to mimic different amounts of attenuation and reddening ; ( 3 ) added them in directly to the images of the spiral . to improve the statistics , at each attenuation / reddening we added the three hdf wfc frames , one at a time , to each one of the ngc 4536 wfc frames . what we call one simulation includes the results of repeating the procedure with 3 synthetic fields , one for each hdf wfc frames . as in the case of the real background galaxies , the photometry on the synthetic frames was done with sextractor , and the threshold surface brightness was fixed in the @xmath11 image . the identification of control galaxies was done , like that of real background galaxies , by visual inspection of each synthetic frame ( as opposed to , for example , through cross - correlation with a list of known hdf galaxies ) . systematics and confusion should thereby affect synthetic and real galaxies in the same way . since we did not know _ a priori _ the extinction law for the disk , we made simulations with both galactic reddening ( fitzpatrick 1986 ) and completely grey extinction ( i.e. , the hdf was attenuated equally in both the f555w and the f814w passbands ) . we hypothesize that the extinction through galaxy disks will lie somewhere in between these two limiting cases . another reason to have the two sets of simulations is that they should establish a relative scale that takes into account all possible systematic errors , against which to measure the attenuation and the reddening of the background galaxies . we also constructed synthetic field simulations with a `` mixed '' extinction law ( the hdf @xmath36 frame was more attenuated than the @xmath13 frame , but less than for galactic reddening ) , to check that the color scale established by the two limiting cases was roughly linear . we covered a large range in attenuation , from no attenuation at all up to 2.6 mag in both passbands . since ngc 4536 is inclined at @xmath42 to the line of sight ( de vaucouleurs et al . 1991 ) , we performed these highly - attenuated simulations anticipating the possibility of a face - on optical depth of @xmath43 = 1 , both with grey and with galactic reddening extinction curves . table 2 shows the input parameters of the simulations , as well as the number and average output color of the hdf galaxies recovered from the ngc 4536 frames ; a line indicating the actual ngc 4536 observations is included for comparison . to assess the effects of crowding alone , we compared the `` degraded '' hdf frames ( [ synthetic ] ) to the simulation without extinction , i.e. , a synthetic field where the hdf was added to the ngc 4536 frames without any attenuation . figures [ ihist ] and [ vhist ] compare the average numbers , of hdf galaxies only , recovered from the ngc 4536 frames in the simulation , against the average number per @xmath20 and @xmath44 bin , respectively , of galaxies in the `` degraded '' hdf . the effect of crowding on the number - counts is , indeed , dramatic . given the isophotal @xmath13 brightness limit , while the hdf galaxies in the `` degraded '' frame are complete to at least @xmath20 = 25 , when added to the ngc 4536 they are complete only to @xmath20 = 24 in the wf4 frame and , optimistically , to @xmath20 = 23 in the wf2 and wf3 frames . this is the range where galaxy clustering is most noticeable ( [ clusterr ] ) . on the other hand , we were able to detect galaxies up to an integrated ( isophotal ) magnitude of @xmath20 = 26 in all frames . the @xmath36 histogram shows that we were able to detect many more blue , faint galaxies in the wf4 ngc 4536 field . the histograms of the recovered synthetic galaxies , however , might suffer from two competing effects : the confusion of faint wings with the foreground disk , and the contamination of the photometric aperture with foreground objects . as described below ( [ bgerr ] ) , when using isophotal magnitudes galaxies are measured , on average , 0.1 mag too bright in the f814w bandpass , and up to 0.6 mag in the f555w filter , depending both on the crowding and on the faintness of the object . fortunately , our investigation does not require absolute or total measurements of the galaxy brightnesses , as long as everything is compared in the same _ relative _ scale provided by the simulations . an absolute measurement of the total magnitude limit is needed only to estimate the expected clustering error . for this purpose , we correct the @xmath20 completeness limit for the faulty background subtraction , and we also assume that total magnitudes are 0.3 brighter than isophotal . after an additional correction for the extinction ( [ ext4536 ] ) , we set our completeness limits at @xmath20 = 23.7 in the wf4 field , and at @xmath20 = 21.7 in the wf2 and wf3 frames . figures [ simreali ] , [ simrealv ] , and [ simrealcol ] compare the number of real galaxies to the number of hdf `` control '' galaxies recovered in the simulations without attenuation , per @xmath13 ( isophotal ) magnitude bin , @xmath36 ( isophotal ) magnitude bin , and color bin , respectively . the effects of extinction are readily noticeable in the wf2 and wf3 fields , both from the reduction in the number - counts and from the shift to fainter magnitudes of the real galaxies . for the wf4 field , any effects are much less apparent . the reddening , though , is harder to discern from figure [ simrealcol ] , even for the more extincted wf2 and wf3 images . the relative photometric error for the individual galaxies was estimated by comparing the three measurements we had of each one of the simulated objects in the synthetic fields without attenuation ( each hdf frame was added to each ngc 4536 frame ) . therefore , it includes the effect of differential crowding , and amounts to 0.16 mag at @xmath13 ( isophotal ) and 0.26 mag at @xmath36 ( isophotal ) . however , the error in ( @xmath45 ) for each galaxy is only 0.10 mag , given that the color has been measured using a small , fixed aperture of 05 diameter , precisely to minimize the impact of crowding ( see [ bgerr ] ) . figure [ fext4536 ] shows the comparison between the number of real galaxies ( horizontal lines ) and the average number of `` control '' galaxies recovered in the simulations at each given @xmath13 attenuation . the top panel displays the results for the wf2 and wf3 fields ; the bottom panel , those for the wf4 image . the solid triangles are for simulations with galactic reddening ; the open triangles , for synthetic fields with grey extinction . the different shapes of the curves for both types of extinction illustrate the dependence of the number - counts on the colors of the galaxies . the actual measurement of the extinction @xmath46 was performed by linearly interpolating between the simulations bracketing the real galaxies . the error bars include poisson and clustering errors ( [ clusterr ] ) . we find for the arm region of ngc 4536 ( wf2 and wf3 fields ) an absorption of @xmath47 mag in the case of grey extinction , and @xmath48 mag in the case of a galactic reddening law . without the contribution for clustering , the purely poisson error is @xmath49 mag . for the interarm and outside area ( wf4 ) , the results are , respectively , @xmath50 mag ( @xmath51 mag without clustering error ) for grey extinction , and @xmath52 0.52 mag ( @xmath53 mag without clustering error ) for galactic reddening . the @xmath54 error of 0.37 ( 0.21 ) mag is already included ; the figure shows that the difference between the attenuation inferred from the two models increases with extinction . the difference is in the sense that we expect ; it is due to our color selection bias , which favors the detection of faint galaxies ( statistically blue ) when they are reddened ( [ colbias ] ) . the difference , however , is smaller than the errors . for the background galaxies , we measure a mean color of @xmath55 in the arm , and of @xmath56 in the interarm and `` outside '' region . figure [ fred4536 ] shows the comparison of the color of the background galaxies to those of the simulations . we remind the reader that this is the appropriate comparison , rather than a confrontation between the two samples ( i.e. , wf2+wf3 and wf4 ) , given their different magnitude limits and all the systematics that come into play in the presence of severe crowding . we also point out that the purpose of this figure is to compare the color of the background galaxies to the synthetic fields closest in extinction and object selection biases ; the locus of the real galaxies in the plot does not necessarily imply an extinction law for the obscuring material in the foreground disk , unless we can safely assume that the whole region is sampled by galaxies whose colors we can measure ( [ discussion ] and appendix ) . the left panel shows the results for the wf2 and wf3 fields ; the right panel , those for wf4 . the simulations with galactic reddening ( open triangles ) do not follow exactly the theoretical extinction law ( solid line ) for large absorptions , neither do the simulations with grey extinction ( filled triangles ) follow a constant line in color . the departure of the grey simulations from the zero - reddening line is due to our selection for red galaxies [ colbias ] , and to the color magnitude relation of galaxies . as we go to larger extinctions , given our detection magnitude limit we are left with statistically brighter galaxies , which are also intrinsically redder . the deviation of the reddened simulations from the theoretical galactic reddening line is due to the systematic error in the background subtraction in crowded areas described in [ bgerr ] . in the case of wf4 , even if it is less crowded , we still see the effect because our magnitude limit is fainter , and faint galaxies are proportionally more affected by the problem . this is also the cause of the relative shift between the scales of the two panels . the wf2 and wf3 images of ngc 4536 are more crowded than the wf4 field ; statistically , _ all _ galaxies in the former fields have been measured relatively bluer , producing the downward shift of the whole color vs. extinction plot . all these systematic effects will affect the real galaxies in the same way and should not bias the estimate of the reddening , as long as the scale defined by the simulations with the two limiting extinction laws is linear . to check that this is the case we have performed simulations with a `` mixed '' extinction law light of each individual galaxy is more attenuated than the @xmath11 light , but less than in the galactic reddening case ; this type of extinction can be caused by an unresolved clumpy dust distribution ( witt & gordon 1996 ) ] . the output of the `` mixed '' simulations is shown in figure [ fmix4536 ] . when compared with their input line , the simulations with a `` mixed '' reddening law seem more affected by the selection bias that pushes the grey simulations redward than by the background subtraction that pushes the galactic reddening simulations blueward . the simulations , however , follow the input line within the errors , which shows that the color scale is , indeed , roughly linear . the background galaxies behind the northwestern arm of ngc 4536 ( wf2 and wf3 fields , figure [ fred4536 ] , left ) fall in between the lines for the completely grey extinction and the galactic reddening law in the the color vs. extinction plot . since , unfortunately , the color line of the grey simulations shows a kink precisely at the location of the real background galaxies , to find the reddening we will rather use the color obtained by interpolating between the grey simulations with @xmath46 = 0.58 mag and @xmath46 = 1.51 mag , at the extinctions ( inferred from the grey and galactic reddening models , respectively ) of the background galaxies . the interpolation yields @xmath28(@xmath57)@xmath58 at @xmath46 = 0.74 mag , and @xmath28(@xmath57)@xmath59 at @xmath46 = 1.07 mag . therefore , the background galaxies suffer a reddening of @xmath60 mag ( grey extinction model ) or @xmath61 mag ( galactic reddening law ) . the error in the reddening , already included , is 0.18 mag . the background galaxies in the interarm region ( wf4 , figure [ fred4536 ] , right ) , however , seem to follow the galactic reddening line . comparing their color to the mean color of the simulation without extinction , they exhibit a reddening of @xmath62 mag , regardless of the attenuation model used . ngc 3664 was observed with the wfpc2 in 1996 march 3 . the total exposure times are 600 seconds in the f555w passband , 800 seconds in the f814w passband , and 1600 seconds in the f439w passband . ngc 3664 is a magellanic peculiar ( sbm(s)iv - v , de vaucouleurs et al . 1991 ) . from the recession velocities ( @xmath63 ) cited in this reference for ngc 3664 and ngc 4536 , and using the distance to ngc 4536 above ( [ obs4536 ] ) , ngc 3664 is at a distance of 11.9 mpc . the data will then have a spatial resolution of @xmath14 6 pc per wf pixel . we assume the foreground galactic absorption is @xmath33 = 0.12 mag ( de vaucouleurs et al . 1991 ) . figure [ fn3664 ] shows the wfpc2 field , overlaid on an stsci digitized sky survey image of the galaxy . once again , we do not use the pc ( wf1 ) frame . ngc 3664 covers only 45% of the three wfc chips ; this decreases the number of galaxies expected to be seen through the disk , but provides a fairly large surrounding field for additional comparison . from its morphology , ngc 3664 is believed to be seen face - on . partly because of the shallowness of the data , partly because the work on ngc 3664 started out in a relatively independent way , and also owing to a desire to explore diverse lines of attack , the analysis of the ngc 3664 data differs in the details from that of ngc 4536 . the main differences are : ( 1 ) the subtraction of the available ( and properly scaled ) @xmath64 image from the @xmath11 image to facilitate the search for relatively red objects ( [ f3664 ] ) ; ( 2 ) the use of the rieke & lebofsky ( 1985 ) galactic reddening curve ; ( 3 ) the semi - automatic identification of background galaxies , both real and simulated ( [ hdf3664 ] ) . to improve the statistics , we did more simulations at smaller extinction intervals ( table 3 ) . we did not try to fix the warm pixels in the raw images , and cosmic rays were iteratively removed based on the deviation of individual pixel values from the mean . finally , we adjusted the parameters of sextractor to minimize the scatter of the difference between the measured brightnesses of the hdf galaxies before and after adding them to the ngc 3664 fields . for the same reason , in what follows we use the `` automatic '' aperture magnitudes calculated by sextractor , rather than the isophotal magnitudes . to remove most of the bright , blue star - forming regions in the disk of ngc 3664 , we subtracted a properly - scaled f439w image from the f814w image . the scaling was done arbitrarily , to remove this blue component without creating substantial `` holes '' in the residual image . this method is reasonable , because we can not or can only barely see the background galaxies in the f439w passband . the disadvantage of this procedure is that it increases the noise . however , since background galaxies are identified primarily through their relatively red appearance , the disadvantage is far outweighed by removing the bright , relatively blue , foreground in the first step . the search for galaxies with sextractor was performed in this artificial , blue - subtracted image . the photometry of the identified galaxies , however , was done in the original f555w and f814w frames . table 3 shows the input parameters of the simulations , and the number and average output color of the hdf galaxies recovered behind and surrounding ngc 3664 ; a line indicating the actual ngc 3664 observations is included for comparison . as opposed to what we did in the case of ngc 4536 , the selection of galaxies out of the object catalogs produced by sextractor was done in a semi - automatic fashion . for example , we only selected objects redder than ( @xmath65 ) = 0.17 mag ( in the catalog , i.e. , several tenths of mag redder in actuality in crowded regions , see [ bgerr ] ) . other selection criteria were a ( sextractor ) star - galaxy separation parameter `` class_star '' @xmath28 0.8 ( the values go from 0.0 for an unmistakable galaxy to 1.0 for a perfect star ) ; a fwhm of a gaussian fit to the object profile @xmath66 02 ; and an error in the `` automatic '' aperture magnitude @xmath28 0.17 mag . in bright foreground star forming regions , the search and measurement of background galaxies is virtually impossible . therefore , we excluded from the search regions of ngc 3664 with a surface brightness higher than 21 mag arcsec@xmath10 ( wf3 ) and 20.7 mag arcsec@xmath10 ( wf4 ) , respectively . we still had to exclude by hand saturated stars that were included in the automatic selection of galaxies . because of the difficulties in identifying galaxies reliably , we decided to exclude from the sample behind the disk of ngc 3664 all candidates fainter than @xmath20 = 22.8 . this reduced our sample of real galaxies found behind the projected disk of ngc 3664 to 8 objects . to the same limiting magnitude of @xmath20 = 22.8 , in the simulation without extinction we recovered an average of 17 hdf galaxies within the disk . the discrepancy must be caused by extinction . for the areas outside the visible disk of the galaxy , we kept galaxies up to an integrated magnitude of @xmath20 = 24 . we found a total of 38 background galaxies in the surrounding field , vs. an average of 37 control galaxies . we find for the disk of ngc 3664 ( figure [ fext3664 ] , _ top _ ) an absorption of @xmath67 mag in the case of grey absorption , and @xmath68 mag in the case of a galactic reddening law . the purely poisson error is @xmath69 mag . for the surrounding area ( figure [ fext3664 ] , _ bottom _ ) the results are , respectively , @xmath52 0.31 mag ( @xmath28 0.17 mag without clustering error ) for grey extinction and @xmath52 0.30 mag ( @xmath28 0.16 mag without clustering error ) for galactic reddening law . ( the results have been corrected for foreground extinction in the milky way . ) table 1 lists the relevant parameters for the calculation of the error in the extinction . for the background galaxies , we measure a mean color of @xmath28(@xmath57)@xmath70 behind the disk of ngc 3664 , and of @xmath28(@xmath57)@xmath71 in the surrounding field . figure [ fred3664 ] shows how we compare the color and extinction of the background galaxies to the simulations . the left panel exhibits the results for the field covered by the disk of ngc 3664 . as in the case of ngc 4536 , the background galaxies behind the disk of ngc 3664 lie in between the lines for the grey and galactic reddening simulations . to find the reddening , we compare the mean color of the background galaxies to the mean color of the grey simulation with the closest amount of extinction , i.e. , the one with @xmath46 = @xmath72 = 0.95 mag . the output mean color of this simulation is @xmath28(@xmath57)@xmath73 . therefore , the background galaxies suffer a reddening of @xmath74 mag . the right panel of figure [ fred3664 ] shows the results for the galaxies recovered outside the visible disk of ngc 3664 . the dashed line follows the color of the grey simulations after correcting for incompleteness : a shift was added to the average color of the recovered galaxies as a function of increasing extinction , so that it would follow the same behavior as the average color of the ( un - added ) hdf galaxies with a progressively brighter magnitude limit . although we prefer to use the color without correction , because otherwise the small numbers make the error in the color very large , the line does indicate that we are seeing an incompleteness effect . conversely , in the case of the simulations with galactic reddening , the line after correction ( not plotted ) basically follows the output mean color of the recovered galaxies . this confirms that , given our color selection bias , we can still find relatively faint galaxies , even in the presence of extinction , as long as they are reddened . the result for the background galaxies outside the visible disk of ngc 3664 is that their number and color are consistent with no attenuation . because spiral disks are relatively blue , red background galaxies are more readily detected behind a foreground disk . as mentioned above , in crowded regions of ngc 4536 we selected in practice for red galaxies [ i.e. , with @xmath38 ] , to be able to distinguish them from the stellar content and hii regions ( which are predominantly bluer ) in the foreground galaxy . this bias will be even more noticeable in grey simulations , where ( given a certain attenuation at f555w ) the f814w passband will be more attenuated than in the corresponding simulation with galactic reddening . as a result , more faint blue galaxies will be discarded in the grey simulation than in the reddened one , the mean color measured for the outcome of the grey simulation will be redder than otherwise , and the measured color shift between it and a reddened simulation will be smaller ( bluer ) than the input . the color magnitude relation of galaxies ( intrinsically brighter galaxies tend to be redder ) will have an effect in the same direction : at higher extinctions , the statistically brighter galaxies will be detected , and the average color of the sample will be redder . as it searches for and measures objects , sextractor determines the background with a sigma - clipping algorithm , plus a modified mode in crowded areas . in very crowded fields , however , this results in an underestimation of the actual background when using large apertures ( larger than 06 in radius for the two galaxies investigated here ) . the background determination algorithm has the purpose , precisely , of removing bright spots ( stars , clusters or hii regions ) from the measurement of the background , so that faint , extended objects can be identified . on the other hand , the larger the photometric aperture , the higher the probability that these same bright spots in the foreground galaxy disk will lie within the aperture , and will be included in the measurement of the flux of the object of interest , in this case , the background galaxy . the difference between the true and the measured background is equal to the difference between the average of the background and its mode . the result of the wrong background determination is equivalent to adding this difference to the galaxy luminosity measurement . we find that , on average , galaxies are measured 0.1 mag brighter in the f814w bandpass simulations when using isophotal magnitudes . in the f555w passband , the shifts depend on the crowding . for ngc 4536 , they go from @xmath76 in the wf4 frame to 0.6 mag for many galaxies fainter than @xmath77 in the wf2 frame , the most crowded of the three wfc frames . in short , fainter galaxies will be measured up to 0.5 mag too blue , if using isophotal apertures . as with the other errors discussed above , this is a systematic error that will affect both the simulations and the real galaxies , and therefore will not affect their mean relative reddening . however , for individual galaxies , the fainter the galaxy the larger the error will be ( in magnitudes ) . hence , this effect will increase the error in the estimate of the mean background galaxy color . we have reduced this problem by determining the galaxy colors from a fixed aperture of 05 ( 5 pixel ) diameter . the use of this aperture size decreased the shift between the input and the measured color in the simulations by a tenth of a magnitude , and the dispersion in the shift by up to a tenth of a magnitude as well . the error introduced by not measuring possible color gradients over larger background galaxy radii should be much smaller than this error , especially when the measurement of a color gradient depends so critically on an accurate estimation of the background . much more importantly , by choosing an aperture to perform the color measurement , we are choosing the scale over which we are averaging the effects of the obscuring material ( [ discussion ] ) . at the distance of ngc 4536 , 05 is @xmath78 pc ; for ngc 3664 , the linear size of the aperture is @xmath79 pc . the systematic errors discussed above in [ colbias ] and [ bgerr ] will affect the simulations and the real galaxies alike , and hence will not have an effect on the reddening determination . the only error that could have an impact on the determination of the reddening is the interpolation error in the hdf @xmath36 images ( [ hdfdata ] ) . however , the mean expected shift of @xmath14 -0.03 mag between the interpolated and the real @xmath44 of the hdf galaxies is smaller than ( 1 ) the dispersion of the shift ; ( 2 ) the error in the mean colors of the real galaxies and of the sets of output simulated galaxies ( [ photerr ] ) ; and ( 3 ) the dispersion in the measurements of individual galaxy colors([com4536 ] ) . we conclude that the error introduced by the interpolation will have a negligible effect over both the background galaxy detection and the reddening determination . to estimate the contribution from galaxy clustering to the error in the number - counts , we take the two - point correlation function @xmath80 from roche et al . ( 1993 ) or from brainerd et al . ( 1995 ) , depending on the completeness magnitude limit of the data . we assume @xmath81 ( smail et al . 1995 ) , since these references list @xmath80 for limits in the @xmath21-band . table 1 shows , for each disk ( ngc 4536 and ngc 3664 ) and field , the angular size @xmath82 of the field , the assumed ( apparent ) @xmath21 magnitude completeness limit , @xmath80 , the number of background galaxies , the derived @xmath46 ( reddened and grey ) extinction , the total error in the extinction , the error contributed by the poisson error in the number - counts , and the error owing to clustering errors . to estimate how the simulations contribute to the clustering error , rather than presenting data for each of the two simulations used in the calculation of the extinction ( the simulations with numbers of recovered galaxies immediately above and below the number of real galaxies ) , we list data for only one fiducial simulation with the same number ( per wfc field ) of recovered hdf galaxies as the real background galaxies . even though the 3 hdf wfc fields have been added one at a time to each of the wfc fields of the investigated galaxies , in reality they are of course contiguous , and have to be treated as one single field for the calculation of the clustering error . the errors contributed by the simulations to the final estimation of the extinction will be @xmath83 the numbers given for the fiducial simulation . except for ngc 4536 , wf4 , all the other values of @xmath80 ( and therefore of the clustering errors ) are slightly overestimated , both because we take the correlation function for a square with an area equal to the area under consideration ( not a square ) , and because roche et al . ( 1993 ) include galaxies with @xmath84 in their calculation of the correlation function ; these galaxies add very significantly to @xmath80 , but there are none in our sample of background objects ( we detect no more than a couple in the simulations without extinction ) . conversely , in the case of the ngc 4536 wf2 and wf3 fields ( analyzed in a combined fashion ) , there is a correlated error in the simulations that we have not included , and that is due to the fact that we count the same hdf fields twice . if the color distribution of galaxies was gaussian and we were sampling it in a systematic way , we could confidently equate the error in the mean color to its measured dispersion divided by @xmath85 , where @xmath86 is the number of galaxies used for the color measurement . however , neither of these assumptions is true . the color distribution of galaxies depends on their limiting magnitude and has an intrinsic dispersion that is not gaussian . moreover , in our case we have to infer their mean color from the few galaxies that we can find in the presence of crowding . crowding will have three adverse effects on the color measurement . the color - magnitude distribution of galaxies resembles an inverted triangle ( cf . for example williams et al . [ ( 1 ) crowding will push the completeness limit up to brighter magnitudes , where the intrinsic color dispersion of galaxies is largest . ( 2 ) it will reduce the number of galaxies available for the analysis in a dramatic way ( figure [ ihist ] ) . ( 3 ) crowding will deplete galaxies in an inhomogeneous fashion : in combination with clustering , it will determine the random sample of background galaxies from which we will try to infer the mean color of _ all _ background galaxies . to check our estimate of the error in the color measurement , we have looked at the dispersion in the mean color of 100 random samples of @xmath86 galaxies ( where @xmath86 is equal to the number of background galaxies ) taken from the hdf galaxies , to our magnitude limit . we have repeated the procedure with the simulations , and we have found that the predicted error is of the same order as the error found by assuming a gaussian distribution . most of the time , the error from bootstrap statistics ( babu & feigelson 1996 ) is slightly smaller than the gaussian error , probably because the latter is calculated from the real sample and therefore includes the effect of clustering and other systematics . in the case of ngc 4536 , we adopt the larger of the two error estimates for each simulation , as well as for the real galaxies . in the case of ngc 3664 , we use the gaussian uncertainty . summarizing the results ( errors in parenthesis are poisson only , no clustering ) : the northwestern arm of ngc 4536 shows an attenuation of @xmath47 ( @xmath87 0.26 ) mag ( grey extinction ) or @xmath48 ( @xmath87 0.26 ) mag ( galactic reddening law ) , and a reddening of @xmath88 mag ( grey extinction ) or @xmath89 mag ( galactic reddening law ) . the @xmath54 error of 0.18 mag is already included in the color uncertainty . for the northern interarm and outside regions , the results are , respectively , @xmath52 0.46 ( @xmath28 0.33 ) mag ( grey extinction ) or @xmath52 0.52 ( @xmath28 0.39 ) ( galactic reddening ) , and @xmath90 mag . the extinction values for the interarm region already include a 1@xmath91 error of @xmath920.37 ( 0.21 ) mag . the poisson errors in parenthesis would be adequate for the extinction if the galaxy density was the same behind ngc 4536 as it is in the hdf . to summarize the results : the disk of ngc 3664 shows an attenuation of @xmath68 ( @xmath87 0.49 ) mag ( galactic reddening law ) or @xmath67 ( @xmath87 0.49 ) mag ( grey extinction ) , and a reddening of @xmath74 mag . the poisson errors in parenthesis would be adequate for the extinction if the galaxy density was the same behind ngc 3664 as it is in the hdf . as it turns out , in the surrounding field of ngc 3664 , this is precisely the case . we have used background galaxy counts and colors , for the first time , to measure the extinction and reddening through two very different galaxy disks . the images also have very different available exposure times , which allowed us to test the limits of the method . \(1 ) in the case of spiral disks , the background galaxy approach is mostly limited by crowding and confusion in the foreground disk , even more than by the background galaxy clustering . we have developed and calibrated the `` synthetic field '' method to decouple the effects of crowding from those of extinction . the method comprises the production of artificial frames by adding suitable and properly scaled reference fields , in this case the hdf , directly into the spiral galaxy images . the reference frames are attenuated to mimic different amounts of extinction and reddening ; the idea is to attenuate the control field until one recovers the same number of simulated galaxies as there are real background galaxies . to ensure that systematics and confusion affect synthetic and real galaxies in the same way , both real and control galaxies are identified in the same fashion ( i.e. , the known positions of hdf galaxies are never used ) . the attenuation derived from the number - counts should correspond to the average extinction through the disk of the galaxy . finally , a comparison between the average colors of , respectively , the control hdf galaxies and the real background galaxies will allow an estimate of the degree of reddening . \(2 ) we tried slightly different lines of attack with both galaxies , but found them to be fairly equivalent . when a @xmath9 image is available , like for ngc 3664 , its subtraction from the @xmath11 image does seem to facilitate the search for galaxies , at least in frames with short exposure times . it would also be desireable to automate as much as possible the final identification of real and synthetic galaxies . unfortunately , a fully automatic identification might be hard to achieve in frames as crowded as the ones we are studying . \(4 ) there is more extinction in the northwestern arm of ngc 4536 than in the northern interarm region of the galaxy . the reddenings , however , are comparable . we have mentioned before that we derive both parameters independently : the extinction from the galaxy number - counts , and the reddening from the background galaxy photometry . the method we use to measure the extinction will take into account galaxies that are so attenuated that we can not see them , while we measure the reddening from galaxies that are attenuated only to a point where we can still observe them . also , although we average the color excess over a number of background galaxies , the color of each one of these galaxies is measured in an aperture with a linear size of @xmath14 40 pc at the distance of ngc 4536 . it is possible that at this scale the dust distribution is genuinely smooth ( keel & white 1997 ) . in view of all the above , in principle we prefer to present the results obtained from the number - counts and from the colors independently , and not in the form of a reddening law . the number - counts measure the total average extinction through a whole region ; the colors measure the extinction owing to a diffuse component . on the other hand , if we assume that nowhere in the region of interest the extinction is so high that we can not sample it with background galaxies , it is possible to determine an extinction law from the two values ( appendix ) . this seems to be a safe assumption for the northern interarm region of ngc 4536 , and there both the extinction and the reddening are consistent with a galactic reddening law . the similar reddening but significantly higher extinction in the northwestern arm can be explained by the combination of a diffuse obscuring component with regions of higher opacity . if , in spite of the caveats , we use the extinction and the reddening measured in the northwestern arm to derive a reddening law , then the extinction there is greyer than galactic and can be caused by unresolved clumps ( witt & gordon 1996 ) , in this case at the scale of @xmath14 40 pc . \(5 ) for the interarm region of ngc 4536 , the extinction can be deprojected to a face - on @xmath52 0.22 mag . for the northerm arm of ngc 4536 , however , given the patchy dust distribution , it is not possible to derive a number for the face - on extinction in a model - independent way . we can , however , assume that the reddening of the background galaxies behind the arm is caused by a smooth component and follows a galactic extinction law . in that case , the reddening implies a face - on extinction through the diffuse component in the arm of @xmath52 0.16 mag . \(6 ) the usefulness of the background galaxy counts and colors to measure foreground extinction has been recently confirmed by their application to the field of the gamma - ray burster grb970228 ( fruchter et al . 1998 ) . in that analysis , we used four different reference frames to measure the extinction _ in the f606w passband _ : the hdf ; the weak radio galaxy 53w002 field ; and two fields located , respectively , at j(2000 ) 15:58:49.8 , 42:05:23 , and at j(2000 ) 14:17:43.63 , 52:28:41.2 ( westphal , kristian , & groth 1994 ) . since the grb970228 field is not crowded , it was not necessary to produce synthetic fields " . under those circumstances , and assuming a galactic reddening law , both counts and colors yield the same result ( respectively , [email protected] mag , and [email protected] mag ) . the average of these two measurements of the galactic extinction towards the gamma - ray burster ( [email protected] mag ) is consistent with the value of 0.72 + @xmath870.24 mag obtained by burstein & heiles ( 1982 ) , and with the extinction of [email protected] measured from maps of dust ir emission ( schlegel , finkbeiner , & davis 1997 ) . the marginal systematic difference in the measurements can be explained by our better spatial resolution ( roughly 1@xmath93 , vs. 61 of schlegel et al . and 06 of burstein & heiles ) , since the burster is located at a position ( @xmath94 = 188.913 , @xmath95 ) where the milky way exhibits a steep gradient in extinction ( cf . burstein & heiles 1982 ) . \(7 ) from the cepheids detected in the same ngc 4536 data , saha et al . ( 1994 ) derive a mean extinction of [email protected] mag in kron - cousins @xmath11 is directly comparable with the extinction in @xmath13 . ] . likewise , they estimate the @xmath11 differential extinction between the arm and the interarm regions to be [email protected] mag . these numbers imply for the arm of ngc 4536 an extinction @xmath460.2 mag , much lower than what we find from the galaxy number - counts . more importantly , the detected cepheids span a range of at least 2 magnitudes in brightness ; saha and co - workers had the capability to measure an extinction much higher than 0.2 mag . the fact that they did not implies that the cepheids detected in the arm lie preferentially either in areas of low extinction or on top of the obscuring material . together with the extinction derived from the galaxy number - counts , the cepheid reddening is consistent with the existence of regions of significant opacity in the arm of ngc 4536 . on the other hand , the extinction measured from the cepheids agrees with the one we find for the interarm region and , from the galaxy colors , for the diffuse component in the arm ( @xmath280.52 and @xmath280.42 mag , respectively , for a galactic reddening model ) . lastly , since the cepheids are embedded in the disk , statistically they should suffer only about half of the total extinction , which background illuminating objects can probe ; our measurements and those by saha and collaborators are consistent with this picture . \(8 ) that in the case of spiral galaxy disks we are totally dominated by crowding is demonstrated by the fact that there is less than 30 percent improvement in the extinction errors for ngc 4536 , relative to ngc 3664 , even though we have 5 times better signal - to - noise ratio in @xmath13 and more than 10 times better signal - to - noise ratio in @xmath36 . likewise , the color errors are only 20 percent better for ngc 4536 . perhaps even more telling , the accuracy of both the extinction and the color measurements is the same in the less obscured regions of the galaxies , despite the enormous difference in exposure times . we can both reduce the contribution to the relative errors from the control fields , and perhaps even measure the clustering error directly , by using an increasing number of them . and since the dominating source of error is crowding , it certainly will not matter if we use reference fields with shorter exposure times than the hdf to produce the `` synthetic fields '' . however , even if we manage to determine very accurately the mean and the dispersion of the number of galaxies in a typical wfpc2 reference field , we will still be left with the larger source of error : the studied galaxies themselves , with their `` single realization '' background fields . if ngc 4536 and ngc 3664 are representative , the extinction errors will go from 0.3 mag in galaxies with exposures of several @xmath12 10@xmath96 seconds , to 0.5 mag in galaxies with shallow exposures of a few @xmath12 10@xmath97 seconds ; the error in the color will amount to @xmath14 0.1 mag in all cases . in the future we will try to reduce this source of error by combining together the results from background galaxies behind spirals of similar hubble type , inclination , and distance to the milky way . gonzlez is grateful to jay anderson for his cosmic - ray removal program , to abi saha for his advice on how to deal with the images of ngc 4536 , and to stefano casertano and andy fruchter for useful discussions on galaxy clustering and bootstrap statistics . larry petro kindly confirmed that the images of ngc 4536 were not affected by scattered light . we want to thank m. dahlem and s. m. baggett for providing the image of ngc 3664 . b. dirsch is especially grateful to s. m. baggett for her help in the reduction of the wfpc2 images of ngc 3664 . we thank the referee , w.e . keel , whose comments helped us improve the presentation of the manuscript . support for this work was provided partly by nasa through grant ar-06400.01 - 95a and partly by the director s discretionary research fund at the space telescope science institute . let us consider the case where each of the background galaxies used to derive the `` average '' extinction through the disk suffers from a different amount of galactic reddening . our method will derive a reddening law that can be described by the following equation : splitting the summation in equation ( 4 ) into two summations , one running over the areas with a given attenuation @xmath110 , the other running over all galaxies in each of these areas , equation ( 4 ) becomes : equation ( 7 ) shows that the extinction obtained from the number - counts is a linear average _ over the whole area considered , and as such takes into account real background galaxies that we can not see _ , while the color excess involves the average of the absorption in magnitudes , _ and only over background galaxies that we actually see_. strictly , the measured @xmath116 and the intrinsic @xmath117 will be equal only if @xmath46 is the same for all galaxies . but , mainly , it is possible to measure a much higher extinction from the number - counts than the extinction implied by the reddening .

we study the problem of resonant tunneling through a quantum dot in a spinful luttinger liquid . for a range of repulsive interactions , we find that for symmetric barriers there exist resonances with a universal peak conductance that are controlled by a non - trivial intermediate fixed point . this fixed point is also a quantum critical point separating symmetry - protected topological phases . by tuning the system through resonance , all symmetry protected topological phases can be accessed . for a particular interaction strength with luttinger parameters and , we show that the problem is equivalent to a two channel kondo problem( cft ) . at the toulouse limit , both problems can be mapped to a quantum brownian motion model on a kagome lattice , which in turn is related to the quantum brownian motion on a honeycomb lattice and the three - channel kondo problem( cft ) . level - rank duality in the quantum brownian motion model relating cft to cft is also explored . utilizing the boundary conformal field theory , the on - resonance conductance of our resonant tunneling problem is calculated as well as the scaling dimension of the leading relevant operator . this allows us to compute the scaling behavior of the resonance line - shape as a function of temperature .

1604.08280

symmetry and topology are two foundational principles that shape our understanding of matter . in the last decade , our understanding of their interplay has led to dramatic progress in our understanding of topological electronic phases . a hallmark of this development is the topological insulator . insulating states with time reversal symmetry fall into two distinct topological phases that are separated by a topological quantum critical point@xcite . for non - interacting systems described by the band theory , our understanding of such topological phenomena is highly developed , and there are many well understood examples of topological states protected by different types of symmetry , such as time - reversal symmetry@xcite , particle - hole symmetry@xcite , spin rotation symmetry@xcite , as well as crystal symmetries@xcite . a current frontier is to develop a similar understanding of symmetry protected topological phenomena in strongly interacting systems@xcite . since the general many - body problem is notoriously difficult , a promising approach is to consider the simplest version of a strongly interacting symmetry protected topological state : one which occurs in a @xmath9 dimensional quantum impurity problem . the archetypal quantum impurity problem is the kondo problem@xcite , along with its multichannel variants@xcite . previous works have explored the kondo physics in closely related problems , such as resonant tunneling in non - abelian quantum hall states coupled to a quantum dot@xcite , fractional quantum hall / normal - metal junctions in the strongly coupling regime@xcite and resonant tunneling through a weak link in an interacting one dimensional electron gas - or a luttinger liquid @xcite . in this paper , we revisit the resonant tunneling problem in a luttinger liquid . this problem was studied extensively in the 1990s@xcite , where it was found that for spinless electrons with repulsive interactions ( described by a luttinger parameter @xmath10 with @xmath11 ) an arbitrarily weak barrier leads to an insulating behavior in the limit of zero temperature . however , for @xmath12 it is possible , by tuning two parameters , to achieve a resonance with perfect conductance at zero temperature . at small but finite temperature , the line shape of the resonance is described by a universal crossover scaling function that connects two renormalization group fixed points : the perfectly transmitting ( small barrier ) fixed point and the perfectly reflecting ( large barrier ) fixed point . it was further observed that for symmetric barriers , a perfect resonance could be achieved by tuning only a _ single _ parameter . here we observe that this is an example of symmetry protected topological critical point separating two topologically distinct symmetry protected insulating states . in the presence of inversion symmetry a one dimensional insulator is characterized by a quantized polarization , which takes two values : @xmath13 mod @xmath14 or @xmath15 mod @xmath14@xcite . likewise , for our resonant tunneling problem , we can define a polarization @xmath16 mod @xmath14 to distinguish different symmetry protected topological phases . @xmath16 is the number of charges transferred across the infinite barrier in the large - barrier limit . without an inversion symmetric barrier , @xmath16 can take any continous value . with an inversion symmetric barrier , in the large barrier limit , @xmath17 or @xmath18 mod @xmath14 , which characterizes two insulating phases . these insulating states are topologically distinct : one can not go smoothly from one phase to the other without going through a topological quantum critical point - the perfectly transmitting fixed point . for @xmath12 , this critical point has only a single relevant operator in the presence of inversion symmetry . for the special value @xmath19 , this fixed point can also be identified with the non - fermi liquid fixed point of the two - channel kondo problem , described by a @xmath20 conformal field theory@xcite . armed with this insight we consider resonances in a spinful luttinger liquid , which will lead us to a class of symmetry protected resonance fixed points that was not studied in detail in the early work . a spinful luttinger liquid is characterized by two luttinger parameters @xmath21 and @xmath22 , with @xmath5 spin symmetry fixing @xmath23 . and @xmath22 is set to @xmath24 for noninteracting electrons in spinful luttinger liquid in ref . ] as shown in ref . , the system can achieve perfect resonance by tuning a single parameter for @xmath25 . this resonance , which is controlled by the perfectly transmitting fixed point , corresponds to a transition between insulating phases characterized by a polarization @xmath26 . this polarization reflects whether or not a pair of electrons with opposite spins is transferred across the infinite barrier in the large - barrier limit . with inversion and time - reversal symmetry , the only possible values of @xmath26 are @xmath27 or @xmath14 mod @xmath28 . when @xmath29 , the perfectly transmitting fixed point becomes unstable even on resonance . in that case a new kind of insulating phase emerges characterized by @xmath30 mod @xmath28 . even though , like the other two insulating phases , the new phase is charge insulating , with time - reversal symmetry , the spin degree of freedom in this phase is not completed locked due to the fact that an unpaired spin can be transferred across resulting in a finite conductance for spin . transitions between these insulating states are governed by a quantum critical point that can not be described by a free luttinger liquid fixed point . rather , it is an _ intermediate _ fixed point@xcite , which could only be described in certain perturbative limits . here we will show that like the spinless case there is a special value of @xmath31 for which the nontrivial fixed point maps to a two - channel @xmath3 kondo problem , described by a @xmath4 conformal field theory@xcite . this analysis allows us to compute the nontrivial on - resonance conductance , as well as the scaling behavior of the width of the resonance as a function of temperature , which is determined by the dimension of the leading relevant operators at the nontrivial fixed point . quasiparticle tunneling processes in a @xmath32 fractional quantum hall system . b ) two backscattering processes in a spinful luttinger liquid.,width=336 ] we also note that the special point @xmath33 of the 1d spinful luttinger liquid model is also of direct relevance to a corresponding resonant tunneling problem between edge states in the fractional quantum hall effect , for which the luttinger parameter is not an interaction dependent quantity . specifically , at filling @xmath32 , disorder is predicted to lead to an edge state that has an _ upstream _ neutral mode with an emergent @xmath5 symmetry@xcite . further calculations@xcite show that the backscattering terms of an electron in a luttinger liquid can be identified as the tunneling terms of an @xmath34 quasiparticle in the fractional quantum hall system ( fig . 1 ) . in this case , we show that the problem of resonant tunneling through symmetric barriers is controlled by the @xmath4 fixed point . insight into the relationships between the resonant tunneling problem and the kondo problem is provided by mapping both problems to a quantum brownian motion model@xcite . in the case of a spinful luttinger liquid , the single impurity problem maps to a quantum brownian motion on a two dimensional lattice . we will argue that both the resonant tunneling problem and the two - channel @xmath3 kondo problem are described by the quantum brownian motion on a _ kagome _ lattice , when they are tuned to an appropriate toulouse limit@xcite . we will show that this , in turn is closely related to the quantum brownian motion on a _ honeycomb _ lattice , which was shown earlier to be related to the three - channel @xmath5 kondo problem . we will argue that this quantum brownian motion picture provides a new insight into the level - rank duality that relates the @xmath4 and @xmath6 conformal field theories . the paper is organized as follows . in section ii , we review symmetry protected topological phases for both the spinless and spinful luttinger liquid . in section iii , we analyze our resonant tunneling problem which maps to a quantum brownian motion model on a kagome lattice at the toulouse limit . from the quantum brownian motion model , an intermediate fixed point is identified . then we show how our resonant tunneling problem maps to a two - channel @xmath3 kondo problem which comes handy for later analysis of the same fixed point . in section iv , utilizing the boundary conformal field theory , we calculate the on - resonance conductance and by identifying the knob " controlling resonance we determine the critical exponent determining the scaling of the resonance line - shape with temperature . in section v , we show that the quantum brownian motion on both the honeycomb lattice and the kagome lattice flows to the same fixed point characterized by its mobility which manifests the so called level - rank duality . we also point out some generalizations . let us first take a look at the resonant tunneling problem in a spinless luttinger liquid@xcite . introduced by haldane@xcite , for spinless electrons , we can represent electrons in terms of two bosonic fields @xmath35 and @xmath36 with the following commutation relation @xmath37=i \pi \delta(x - x').\ ] ] the fermionic charge operators can then be bosonized as @xmath38 where @xmath39 is the fermi momentum and the effective hamiltonian density may be written as @xmath40\ ] ] where @xmath41 is the sound velocity and @xmath10 is the luttinger parameter characterizing strength of interaction , with @xmath42 corresponding to noninteracting fermions . the euclidean action is @xmath43 . integrating our either @xmath35 or @xmath36 , we have two equivalent dual representations . at the small - barrier limit , integrating over @xmath36 gives @xmath44.\ ] ] this representation of the action is particularly convenient . scattering of electrons from coupling to a potential @xmath45 adds @xmath46 to the hamiltonian . assuming @xmath47 varies slowly on the scale of the potential and @xmath45 is nonzero only near @xmath48 , integrating out fluctuations in @xmath47 away from zero , the effective action becomes @xmath49 plus an extra term corresponding to the effect of the potential : @xmath50 where @xmath51 are fourier components of @xmath45 at momenta @xmath52 and @xmath53 is the matsubara frequency . the extra term serves as the effective weak pinning potential for our resonant tunneling problem and we denote it as @xmath54 $ ] . to leading order in the backscattering , the rg flow equations are @xmath55 notice that for @xmath12 , the only relevant perturbation is the backscattering term at @xmath56 : @xmath57 in general , the system achieves resonance by tuning the two coefficients @xmath58 and @xmath59 . with inversion symmetric barrier ( @xmath60 ) , @xmath61 is a real number and therefore only one parameter needs to be tuned . at the opposite limit - the large - barrier limit , it is not hard to see that a convenient representation of the action should be in @xmath36 variables since @xmath35s must be locked in the minimum of cosine potential and can not change continuously . any perturbation away from this limit can be represented as hopping processes of electrons across the infinite barrier . we denote the strength of hopping as @xmath62 . is the gate voltage on the quantum dot one can tune to achieve resonance at @xmath63.,width=288 ] from fig . 2 , we see there are two inversion symmetry protected topological insulating phases for @xmath12 when the system flows into the large - barrier limit and they are separated by the perfect transmitting fixed point . if we choose the center of inversion as our origin , then for one insulating phase ( @xmath64 ) , with the infinite barrier , we must have @xmath54 $ ] pinned in the potential minimum which is at @xmath65 . this corresponds to a polarization @xmath66 mod @xmath14 . the other insulating phase ( @xmath67 ) must have its minimum of the potential pinned at @xmath68 and results in a polarization @xmath17 mod @xmath14 . the two phases are topologically robust since the transition between them is only possible by tuning the system through resonance(tuning @xmath61 through 0 ) . on resonance , we know the system is perfectly transmitting and electrons move freely . therefore , our transition is analogous to the transition from a topological insulator to an ordinary insulator@xcite . both transitions go from one insulating phase to another insulating phase via a conducting state . of course the conducting state for topological / ordinary insulator transition refers to the familiar band gap closure . the resonant problem in a spinless luttinger liquid provides the simplest example of a @xmath69 system with symmetry protected topological phases . the paradigm here is to recognize the perfect resonance fixed point as the symmetry - protected quantum critical point separating symmetry - protected topological phases . adopting this new interpretation , let us now move on to the resonant problem in a spinful luttinger liquid . for electrons with spin , we have two luttinger parameters , the dimensionless conductance @xmath21 and the dimensionless spin conductance " @xmath22 which describes the spin - current response to a magnetic field . for each spin @xmath70 , there are two bosonic fields @xmath71 . it is convenient to separate them into charge and spin degrees of freedom : @xmath72 in the small - barrier limit , there are two competing perturbation terms in the action which are most relevant for @xmath73 and @xmath74@xcite @xmath75 and @xmath76 where @xmath77 is the process that backscatters an electron and @xmath78 is the process that backscatters an up - spin and also a down - spin electron . these two perturbation terms combined is the effective weak pinning potential @xmath79 of the spinful resonant tunneling problem and their flow equations are given as @xmath80 notice that the @xmath81 process is relevant only for @xmath29 . there is another process @xmath82 which corresponds to backscattering of an up spin and a down spin electron incidenting from opposite directions(net charge momentum unchanged ) . if this process is relevant , it could pin @xmath83 to the minimum of potential . however , in the range of luttinger parameters of our discussion , this process will always be irrelevant.@xcite and @xmath2 . the top(bottom ) line represents small(large ) barrier limits . arrows represent rg flows and the solid dot represents the perfectly transmitting fixed point . at the large - barrier limit , two inversion and time - reversal symmetry protected insulating phases emerge shown in a ) and b ) represented in their cosine potential configuration . the dashed line indicates the center of inversion . @xmath84 is the gate voltage on the quantum dot one can tune to achieve resonance at @xmath63.,width=288 ] now , when @xmath85 and @xmath2 , the @xmath86 process will be irrelevant . two topologically distinct insulating phases separated by a perfectly transmitting fixed point again emerge as shown in fig . 3 . however , this time , they are protected by both inversion symmetry and time - reversal symmetry with the potential minimum pinned at @xmath87 for @xmath88 and @xmath89 for @xmath90 as the system flows into the large - barrier limit . they are characterized by @xmath91 or @xmath27 mod @xmath28 respectively . the transition between these two symmetry protected topological phases are achieved by tuning the system to resonance(tuning @xmath77 through @xmath27 ) . and @xmath2 , the perfectly transmitting fixed point(the black solid dot ) becomes unstable and opens up to the intermediate fixed point(the red solid dots ) as indicated by the dashed arrows . under rg flows(indicated by arrows ) , three inversion and time - reversal symmetry protected phases emerge shown in a ) , b ) and c ) represented in their cosine potential configuration . @xmath84 is the gate voltage on the quantum dot one can tune to achieve resonance at @xmath92.,width=336 ] a more interesting intermediate fixed point can be found if we make interactions in charge sector more repulsive ( @xmath93 ) while keep spin symmetry ( @xmath2 ) . the @xmath94 ( due to inversion symmetry , the @xmath95 term is eliminated ) process is now relevant , and we have to take it into account alongside the @xmath77 process . there are two different situations depending on the sign of @xmath61 . when @xmath64 , the minimum of the potential @xmath96 is pinned to @xmath97 . the @xmath77 process is thus eliminated and the @xmath61 process dominates and grows to infinity under rg flows . the previous perfectly transmitting fixed point becomes unstable in this occasion and flows into an intermediate fixed point . a new symmetry protected topological phase emerges as shown in fig . 4 . note that at this new phase we are free to change @xmath83 since the @xmath98 process is still irrelevant . thus , it is a charge insulating phase with a finite spin conductance characterized by @xmath30 mod @xmath28 . on the other hand , when @xmath67 , the minimum of the potential is pinned to either @xmath99 or @xmath100 . in this case , since the @xmath77 process is also present and its magnitude grows to infinity under rg flows , ( @xmath101 , @xmath102 ) will be locked to either @xmath103 or @xmath104 depending on the sign of @xmath77 as before . this gives us two charge and spin insulating phases . all three symmetry protected topological phases can be accessed by adjusting a single parameter - the ratio @xmath105 . a previous study of this intermediate fixed point can be found in ref . . it was shown that this fixed point becomes perturbatively accessible from the perfectly transmitting fixed point with an @xmath106-expansion near critical values of luttinger parameters @xmath107 and @xmath108 at the small - barrier limit . this is due to the fact that the perfectly transmitting fixed point becomes unstable at the aforementioned values of luttinger parameters . unfortunately , for our @xmath5 symmetric case with @xmath74 , this intermediate fixed point is not perturbatively accessible using the @xmath106-expansion method . however , for @xmath1 , an exact description can be obtained by using boundary conformal field theory as shown in section slowromancap4@. in this section , we will further develop our understanding of the resonant tunneling problem in spinful luttinger liquid and explore the connection between our resonant tunneling problem and other quantum impurity problems . first , the simpler spinless resonant tunneling problem@xcite is reviewed to pave the way for understanding the more complicated spinful case . then , we perform renormalization group calculations for our spinful resonant tunneling problem . at the toulouse limit , our resonant tunneling problem is nothing but a quantum brownian motion model on a kagome lattice . at both small ( small @xmath109 ) and large ( small @xmath62 ) barrier limits of the quantum brownian motion model , the system flows to an intermediate fixed point . to obtain an exact description of this fixed point , we map our resonant tunneling problem to a two - channel kondo problem with @xmath3 impurity spin@xcite . -@xmath110 plane . a ) @xmath111 , b ) @xmath112.,width=336 ] again we start with the spinless resonant tunneling problem . taking the large - barrier limit , if the capacity on the quantum dot is small , a large charging energy fixes the number of charge on the dot and transmissions through the dot are suppressed . by tuning the gate voltage , the chemical potential on the dot can be adjusted and resonant tunneling can be achieved . a theoretical model takes a double - barrier structure , it is a wire with two @xmath113-functions on it separated by a quantum dot with size @xmath114 . on the dot , a gate voltage @xmath84 is assigned . we denote @xmath115 as the number of electrons tunneling through the corresponding barrier . we also define @xmath116 as the number of electrons transferred across two barriers and @xmath117 as the number of electrons on the dot . then , the action has deep minima when @xmath118 is an integer ( see fig . 5a ) . since for infinite large barriers @xmath35 fields are pinned at minima , it is more convenient to use the @xmath36 representation(@xmath119 are dual bosonic fields of @xmath120 following from the standard bosonization terminology ) . in this case , the partition function describes instantons connecting these degenerate minima . the hopping processes of instantons correspond physically to electrons hopping on or off the quantum dot . the partition function can be analyzed in the coulomb - gas representation in powers of the tunneling amplitude @xmath62 , @xmath121 where @xmath122 are quantum states on the dot labeling the number of electrons on the dot . @xmath123 is the renormalization constant and is initially set to be @xmath124 . its value flows under renormalization group . integrating out bosonic fields @xmath36 mediates a logarithmic interaction between charges " in the coulomb - gas representation . the charges " correspond to physical hopping processes and we have two kind of charges " in our problem : hopping electrons on and off the dot . after integration , the partition function is in the following form : @xmath125 where @xmath126 denotes the charge transferred to the right in a hopping event and @xmath127 denotes the change in charge on the dot . due to the discreteness of charge on the dot which can only change by 1 , @xmath128 must alternate whereas @xmath129 can have any ordering . tuning into resonance , the system renormalizes according to the rg flow equations@xcite @xmath130 . \end{aligned}\]]during the process , electrons on the dot can virtually tunnel back to the leads reducing the average charge on the dot @xmath110 . on resonance , the hopping directions along @xmath110 collapse(@xmath112 ) which renders @xmath131 into precisely a half integer(fig . this reduction of dimensionality from two to one at the toulouse limit greatly simplifies the problem . a similar simplification will arise in the more complicated spinful problem discussed below . it is also worth mentioning that at @xmath19 , the spinless resonant tunneling problem can be mapped to a two - channel kondo problem with @xmath5 impurity spin@xcite . examining the spinful case , at the large - barrier limit , for each barrier , there are bosonic fields @xmath132 as defined in eq . we can reorganize @xmath132 fields into the following @xmath133 the superscript @xmath134 and @xmath135 denote physical quantities transferred across two barriers or changed in the dot respectively . the subscript @xmath134 and @xmath135 denote charge or spin respectively . now , the action @xmath136(@xmath96 is a periodic potential possesses lattice symmetry shown in fig . 6 ) will have deep minima whenever @xmath137(the number of electrons with up spin transferred over two barriers ) or @xmath138 ( the number of electrons with down spin transferred over two barriers ) is an integer(fig . 6 ) . we can adopt the same coulomb - gas representation of the partition function to describe our resonant tunneling problem . in our case there are three tunneling processes . the processes in which a spin up or down electron hops on or off the quantum dot has a tunneling amplitude @xmath139 or @xmath140 . the other processes in which both the spin of an electron on the lead and that of an electron on the quantum dot are flipped has a tunneling amplitude @xmath141 . when @xmath33 and @xmath74 , the three tunneling processes have the same tunneling amplitude @xmath142(fig . 6 ) . expanding the partition function in powers of @xmath141 and @xmath143 , we arrive at @xmath144}\\ & + t_{h_\uparrow } \ket{\uparrow } \bra{0 } e^{-i[a(\frac{1}{2\sqrt{g_{\rho } } } \varphi_{\rho}^{\rho}+\frac{1}{2\sqrt{g_{\sigma } } } \varphi_{\sigma}^{\rho})+(\frac{k_{\rho}}{2\sqrt { g_{\rho } } } \varphi^{\sigma}_{\rho}+\frac{k_{\sigma}}{2\sqrt { g_{\sigma } } } \varphi^{\sigma}_{\sigma})]}\\ & + t_{h_{\downarrow } } \ket{\downarrow } \bra{0 } e^{-i[a(\frac{1}{2\sqrt{g_{\rho } } } \varphi_{\rho}^{\rho}-\frac{1}{2\sqrt{g_{\sigma } } } \varphi_{\sigma}^{\rho})+(\frac{k_{\rho}}{2\sqrt { g_{\rho } } } \varphi^{\sigma}_{\rho}-\frac{k_{\sigma}}{2\sqrt { g_{\sigma } } } \varphi^{\sigma}_{\sigma})]}\\ & + \text{h.c.})^n\rangle\\ & = \sum_n \sum_{a=\pm } \int \frac{d^n\tau_j}{\tau_c } \langle ( \sum_{k=\sigma , h_{\uparrow},h_{\downarrow } } t_k \delta^{k+ } e^{-i ( a\vec{h_{k+ } } \cdot \vec{\varphi^{\rho}}+\vec{h_{k+ } } \textbf{k } \vec{\varphi^{\sigma}})}\\ & + \text{h.c.})^n \rangle \end{aligned}\ ] ] where @xmath145 have been relabeled as @xmath146 and exponents are shortened as dot products of vectors @xmath147 @xmath148 @xmath149 and @xmath150 . for @xmath33 and @xmath74 , we have @xmath151 which is initially set to @xmath124 and @xmath152 . integrating out bosonic @xmath36 fields mediates logarithmic interactions between charges " in the following form : @xmath153 let us explain this coulomb - gas model further . because of the extra degree of freedom from spin , instantons now move in a four - dimensional space coordinated by @xmath154 with discrete values . charges " here are again different physical processes . since there are six physical hopping processes : hopping on or off either an up or down electron to the dot and flipping the spin on the dot , relations among all possible processes for a single time step constitute a triangle(see fig . charges " are now vectors of the triangle instead of scalars and their physical relevance are encoded in their length depending on luttinger parameters . three charges " are given in eq . ( 19)-(21 ) in @xmath155 coordinates characterizing the change of both spin and charge on the quantum dot like @xmath128 in the spinless case . at each time @xmath156 , it has to alternate among all three possible states on the quantum dot . the other three charges " are @xmath157 in @xmath158 coordinates analogous to @xmath129 , characterizing both spin and charge transferred across two barriers with no restriction of alternation . to put it more visually , imagine that we have two kinds of hopping on the four - dimensional lattice space , those perpendicular to hyper - surfaces with @xmath159 coordinates held fixed and those parallel to hyper - surfaces with @xmath160 coordinates held fixed . for the former case , instantons hop along @xmath161s in the two dimensional sublattice with @xmath160 and for the latter , instantons hop along @xmath157s in the two dimensional sublattice with @xmath159 . here , @xmath162 since at each lattice site there are two corresponding vectors with opposite directions and instantons are free to choose one . conservation of spin and charge in our resonant tunneling problem poses two constraints @xmath163 where @xmath164 are the charge and spin on the quantum dot . thus , different values of @xmath165 label hyper - surfaces in which tunneling processes take place . however , since spin or charge on the dot has to alternate among the three possible occupation states ( @xmath166 ) , following the same reasoning as the spinless case , directions for tunneling between different hyper - surfaces along @xmath155 will get renormalized and eventually leads to the four- dimensional lattice collapsing into the two - dimensional lattice shown in fig . 6 with a change of basis to @xmath167 . ) . the circle denotes the fixed point.,width=288 ] a detailed rg calculation in appendix a gives the following flow equations with the corresponding flow diagram fig . the toulouse limit is along the line with @xmath112 . @xmath168 this toulouse limit(@xmath112 ) of our resonant tunneling problem , where instantons are confined in a two dimensional sublattice with coordinates @xmath159 and directions along the other two bosonic fields decouple , is identical to a quantum brownian motion model on a _ kagome _ lattice . more rigorously , at the toulouse limit , the action for our resonant tunneling problem is @xmath169 . \end{aligned}\]]if we do the following mapping @xmath170 then this is precisely the action of a quantum brownian motion model tunneling on a kagome lattice in the large - barrier limit @xmath171.\ ] ] where @xmath172 is the amplitude of hopping between minima connected by a lattice vector @xmath173 , @xmath174 is the position of particle in the momentum space with a potential possessing the symmetry of the reciprocal lattice and @xmath175 describe lattice sites shown in fig . the quantum brownian motion model was originally proposed as a theoretical model for heavy charged particle in a metal@xcite . although the applicability of this model to its original proposed problem is questioned@xcite , the model is later shown to be relevant to quantum impurity problems . it describes a brownian particle moving in a lattice with a periodic potential . the coupling of the potential to the particle generates a frictional force which acts as dissipative bath . there are two perturbatively accessible limits to analyze the effect of the periodic potential in a quantum brownian motion model . in the small @xmath109 limit for which the barrier is small , the action is @xmath176-\int \frac{d\tau}{\tau_c}\sum_{\vec{g } } v_{\vec{g } } e^{i 2 \pi \vec{g } \cdot \vec{l}}\ ] ] where @xmath177 is the dissipative kinetic energy and the latter integral represents the energy of the periodic potential . in the integrand the periodic potential amplitude at the particle trajectory @xmath178 is written in sums of fourier components @xmath179(@xmath180 is the reciprocal lattice vector ) . under rg calculations in the leading order , the flow equation depends on the length of the reciprocal lattice vector @xmath181 similarly , for the small t ( large barrier ) limit , the flow equation depends on the length of the lattice vector @xmath182 we know that for a kagome lattice the product of the shortest reciprocal lattice vector @xmath183 and the shortest lattice vector @xmath184 is @xmath185 . then it follows that for @xmath186 , both small and large - barrier limits are _ unstable _ and there must be a stable intermediate fixed point in between . the intermediate fixed point is characterized by the mobility @xmath187 of the brownian particle under the external frictional force where @xmath188 at @xmath189 and @xmath190 at @xmath191 . thus @xmath192 for our intermediate fixed point . in general , @xmath193 depends on @xmath194 and is hard to calculate . this reminds us of a previous work by yi and kane@xcite . in it , they were able to map a quantum brownian motion model on a @xmath195 dimension honeycomb lattice to the toulouse limit of an @xmath16-channel kondo problem with @xmath5 impurity spin . like our model , for @xmath196 , the system flows into an intermediate fixed point characterized by the mobility @xmath193 . when @xmath197 , an exact description of this fixed point is possible from boundary conformal field theory since it is the same intermediate fixed point of the aforementioned three - channel @xmath5 kondo problem . in order to find an exact description of the intermediate fixed point of our problem , we should again walk down this route of mapping to the multichannel kondo problem . and @xmath2 case ( left ) and the more general case ( right ) . , width=288 ] before we start our study of the multichannel kondo problem , we should note that there is a more general situation for our resonant tunneling problem . with inversion and time - reversal symmetry , it is not guaranteed that the three tunneling processes have the same amplitude , however , although it does require @xmath198(fig . 8) . as a result , @xmath199 , rg flow equations are also modified as : @xmath200 in this subsection , we will establish the equivalence between resonant tunneling problems in a luttinger liquid and the multichannel kondo problem . this allows us to use the boundary conformal field theory technique developed for the multichannel kondo problem@xcite to obtain an exact description of our newly found intermediate fixed point . the previously mentioned work by yi and kane@xcite also utilized this method to study the intermediate fixed point of the quantum brownian motion model on the honeycomb lattice . let us recall the emery - kivelson solution of the two - channel @xmath5 kondo problem@xcite . this kondo problem can be mapped to our spinless resonant tunneling problem at @xmath19 . it was shown that with symmetric channels , at the toulouse limit , only _ half _ of the impurity spin degree of freedom is coupled to the conduction electrons resulting in the non - fermi liquid properties . if we replace the @xmath5 impurity spin by the two degenerate charge states of the dot , then the _ half _ coupling behavior of the kondo impurity spin at the toulouse limit is the same as the _ half _ occupation of the quantum dot by electrons hopping from leads(i.e . @xmath118 is a half integer when @xmath112 ) . now , for the spinful case , a suitable kondo problem would be one with two channels and three spin states for the impurity spin . naturally , this leads us to the two - channel kondo problem with @xmath3 impurity spin for which the three spin states corresponds to the three possible occupation states on the quantum dot(fig . 9 ) . the hamiltonian of a two - channel @xmath3 kondo problem reads @xmath201 where @xmath202 and @xmath203 are channel and spin indices respectively and @xmath204 is the impurity spin . @xmath3 has eight generators @xmath205 , @xmath206 and thus the electron spin operator @xmath207 . we can regroup generators of @xmath3 into three pairwise linear combinations of off - diagonal generators in analogy with the @xmath5 case as @xmath208 and @xmath209 . this allows us to write out the spin operators for electrons . following the emery - kivelson solution@xcite , we first bosonize fermions as @xmath210 where @xmath211 is a bosonic field satisfying @xmath212=-i \pi \delta_{aa ' } \delta^{ss ' } \text{sgn } ( x - x').\ ] ] then the operators at each channel are @xmath213 where @xmath214 and @xmath215 . if we further assume that our kondo problem is anisotropic meaning that the diagonal coupling constants @xmath216 and @xmath217 ( in analogy with @xmath218 in the @xmath5 case ) are not equal to the off - diagonal ones for which we call @xmath219 ( in analogy with @xmath220 in the @xmath5 case ) in general , then the hamiltonian becomes @xmath221 with @xmath222\ ] ] @xmath223\right\}\\ & + \frac{j_{\perp}}{\tau_c}\sum_{i = t , u , v}[\chi^{i_+ } s^{i_-}_a+\text{h.c . } ] \end{aligned}\ ] ] where @xmath224 is the kinetic energy of electrons and @xmath225 is the interaction between the impurity and electron spins at origin . now we introduce a unitary transformation @xmath226 to decouple @xmath227 and @xmath228 in @xmath225 in eq . ( 50 ) by setting @xmath229 . then we perform an orthogonal transformation for variables @xmath230 with @xmath231 the partition function of our anisotropic kondo problem is @xmath232 the interaction potential is given as @xmath233 @xmath234 and * * r**= @xmath235 = @xmath236 . the following mapping turns our resonant tunneling problem to the two - channel @xmath3 kondo problem : @xmath237 the essential idea is to relate the spin and charge transferred in the resonant tunneling problem to spin transferred in the kondo problem . in this way , when @xmath33 and @xmath238 , the spin @xmath5 symmetry @xmath239 @xmath240 charge symmetry can be mapped to the @xmath3 symmetry of the kondo problem . then , we can apply the same analysis and conclude that both problems flow to the same intermediate fixed point described above . so if we start off with all @xmath219s equal and @xmath241 , then the system flows into an intermediate fixed point lying on the toulouse limit line ( fig . 7 ) with the full @xmath3 symmetry ( @xmath242 ) and ( @xmath243 ) fields decouple . in this section , we utilize the previously established mapping to the multichannel kondo problem to study our tunneling problem on resonance . as a known result@xcite , at low but finite temperature , the width of the resonance line - shape vanishes as a power of temperature . the charge and spin conductance through resonance assumes a universal shape as a scaling function@xcite ( fig . 10 ) @xmath244 where @xmath245 is a non - universal constant and @xmath113 is the distance to resonance in gate voltage . using boundary conformal field theory description of the kondo problem , we calculate the on - resonance conductance @xmath246 at @xmath247 in subsection a. moreover , in subsection b , we exam all allowed operators at the fixed point of our resonant tunneling problem and identified the one relevant operator for which we could tune the system through resonance . the scaling behavior of the resonance line - shape , which depends on the critical exponent @xmath248 , is obtained from calculating the scaling dimension @xmath249 of that relevant operator . unlike the spinless case in which the fixed point is perturbatively accessible at the small - barrier limit , here , we can not extract information about conductance without an exact description of the intermediate fixed point . despite the failure of perturbative methods , since our resonant tunneling problem is equivalent to a two - channel @xmath3 kondo problem , we can study the intermediate fixed point using boundary conformal field theory . the bcft description of our kondo problem resides on the upper half - plane ( fig . 11 ) with kondo boundary condition encoded on the real axis@xcite . it is precisely this non - trivial kondo boundary condition that gives calculations of correlation functions an extra twist as reflected in appendix b. following eqs . ( 56)-(61 ) , by identifying electron operators in our tunneling problem with the two - channel @xmath3 kondo problem , we have our current - operator correspondence : @xmath250 using the kubo formula , the on - resonance kubo conductance is @xmath251 for @xmath1 and @xmath2 , the mobility @xmath252 is calculated using the boundary conformal field theory in appendix b. we have @xmath253 the physical conductance and its relation to the kubo conductance calculated above is explained in appendix c. the physical on - resonance conductance is @xmath254 for @xmath1 and @xmath2 , we have @xmath255 according to cardy@xcite , properties of the boundary operators can be obtained by conformally mapping the upper half - plane to an infinite stripe with kondo boundary conditions on both ends ( fig . 11 ) . therefore , to obtain the spectrum of hamiltonian @xmath256 in an infinite stripe with the kondo - kondo " boundary condition , we can use the double fusion " rule hypothesized by affleck and ludwig@xcite starting with the free fermion boundary condition @xmath257 " on both ends . since the conformal embedding of our kondo problem is @xmath258 , just like the case for the three - channel @xmath5 kondo problem with spin and flavor interchanged , then any boundary operators can be represented as a triplet @xmath259 where the three quantum numbers are weights in representations of lie groups @xmath260 and @xmath3 respectively@xcite . the allowed triplets are of course all possible primary fields at the intermediate fixed point . the calculation from eq . ( b12 ) shows that both the two - channel @xmath3 and the three - channel @xmath5 kondo problems@xcite flow to the same intermediate fixed point . therefore , using the latter kondo problem , if we start with the free fermion boundary condition @xmath261 , and fuse the boundary operators with the impurity spin operator @xmath262 @xmath263 twice , the resultant operators are all possible primary fields at the intermediate fixed point . their scaling dimensions are given as@xcite @xmath264 where @xmath265 is the scalar product induced by killing forms , @xmath266 is the weight of the boundary operator in the corresponding representation of its lie group , @xmath134 is the weyl vector , @xmath267 is the level and @xmath10 is the coexter number@xcite . the only relevant operators left are @xmath268 with @xmath269 and @xmath270)$ ] with @xmath271 which transform as elements of the adjoint representation of @xmath5 and @xmath3 , respectively . counting the number of available operators is the same as counting the dimension of the two adjoint representations , which gives dim(ad @xmath5)+dim(ad @xmath3)=11 possible relevant operators . however , channel symmetry and @xmath3 spin symmetry in the kondo problem all impose constraints via conservation laws . any off - diagonal elements of the adjoint representation will modify either channel number or spin numbers and thus break the conservation laws . we are left with three diagonal relevant operators . in the familiar two - channel @xmath5 kondo problem , there are two relevant diagonal operators , each from @xmath272 and @xmath273 sectors , respectively . inversion symmetry demands there should be no difference between two channels . therefore , the diagonal operator from @xmath274 @xmath5 can not be present since it will lead the system flow towards an anisotropic kondo fixed point with one channel strongly coupled and the other disconnected by breaking the @xmath274 @xmath5 symmetry@xcite . when interpreting this in the spinless resonant tunneling problem , since the fixed point is at the perfect conducting limit , then @xmath275 and @xmath276 are the two aforementioned relevant diagonal operators . inversion symmetry in this case requires the two barriers to be the same and eliminates @xmath276 . similarly , in the two - channel @xmath3 kondo problem , inversion symmetry again eliminates any relevant diagonal operator from @xmath274 @xmath5 . moreover , we know that the relevant diagonal operator in subgroup @xmath277 @xmath278 @xmath3 must vanish to make @xmath279 because when translating back to our resonant tunneling problem , time - reversal symmetry requires there be no difference between spin states so that the two @xmath280 processes are equal . note that @xmath141 might have a different amplitude . this extra degree of freedom is precisely controlled by the remaining one relevant diagonal operator from @xmath277 @xmath281 @xmath3 and can be used to tune the system to resonance . with this knowledge , at finite temperature , we are able to calculate the critical exponent @xmath282 for the resonant line - shape . the exact form of the scaling function @xmath283 can be obtained from the monte - carol simulation@xcite . since the two models flow to the same intermediate fixed point from both limits , this phenomenon manifests the level - rank duality from a quantum brownian motion model perspective.,width=288 ] bcft has granted us an exact description of our two - channel @xmath3 kondo fixed point . translating everything into the quantum brownian motion on a kagome lattice using the mapping from section slowromancap3@ , the spin - current conductance in the kondo problem characterizing the intermediate fixed point , becomes the mobility @xmath187 of fictitious particles on the lattice of a quantum brownian motion model with @xmath188 for free particles and @xmath190 for completely localized particles ( see appendix b ) . the mobility calculated in eq . ( b12 ) confirms that the quantum brownian motion on both the honeycomb lattice@xcite and the kagome lattice flows into the same strong coupling fixed point ( fig . mathematically , this can be attributed to the fact that the same conformal embedding is realized at the fixed point , namely @xmath284 . if we dig in a little further , this phenomenon is called level - rank duality relating @xmath4 conformal field theory to @xmath6 conformal field theory @xcite . however , instead of going through mind - boggling mathematical formalism , here we provide a more physical picture of this equivalency using quantum brownian motion models . this equivalency in the large - barrier limit can be assessed by comparing the mobility of the two quantum brownian motion models since both are related to kondo problems in the toulouse limit . from that calculation , a more general pattern emerges . we find that with @xmath285 fixed , any @xmath286 quantum brownian motion model flows into an intermediate fixed point with the same mobility @xmath287 $ ] . this is checked up to @xmath288 using mathematica . on the other hand , in the small - barrier limit , to the first order , the quantum brownian motion model is governed by the renormalization group eq . ( 34 ) which drives the system towards the intermediate fixed point . the general pattern stated in the previous paragraph is harder to establish since for general @xmath289 and @xmath267 , @xmath290 are complex numbers ( @xmath291 ) . therefore , it is not clear how the two systems will behave under the first order flow equation . however , for special cases with either @xmath289 or @xmath292 , there is a simple argument to show their equivalency at the small - barrier limit . first of all , it is not hard to see that the @xmath7 quantum brownian motion lives on a generalized honeycomb lattice and the @xmath8 quantum brownian motion lives on a generalized kagome lattice both in @xmath293 dimensional space . in appendix d , by choosing an appropriate origin , we have shown that all @xmath294 and @xmath295 of the honeycomb lattice have the same sign as @xmath296 of the kagome lattice . therefore , according to the flow equation , there is no difference between the two quantum brownian motion models . higher dimensional objects are always difficult to envision , so we leave detailed calculations in appendix d. here , we will stick to the simplest case with honeycomb and kagome lattices . a hand - waving explanation would be that if you blur your eyes , there is not much difference between a honeycomb lattice and a kagome lattice . of course , we can show this more rigorously . it is known that both honeycomb and kagome lattices have the triangular lattice as their bravais lattice . as a result , they also share the same reciprocal lattice . next , we can gauge " @xmath179 so that @xmath297 is the same for all shortest reciprocal lattice vectors @xmath298 . this is achieved by shifting the origin of our coordinate system to the center of hexagon ( a center of inversion ) since @xmath179 are coordinate dependent . we have @xmath299 for both honeycomb and kagome lattices . thus , the quantum brownian motion on the two lattices behaves the same whenever @xmath180 is a relevant perturbation . in this paper , we have studied the problem of resonant tunneling in a spinful luttinger liquid . we showed that along with the spinless resonant tunneling problem , they both possess symmetry protected topological phases separated by quantum critical points . for the spinless case , two insulating symmetry protected topological phases protected by inversion symmetry are found and characterized by a polarization defined as the number of charge transferred across the infinite barrier . similarly , for @xmath300 and @xmath74 , two charge and spin insulating symmetry protected topological phases protected by inversion and time - reversal symmetry are found and characterized by a polarization defined as the number of pairs of electrons with opposite spin transferred across the infinite barrier . by tuning the system through perfect resonance , we can go through quantum critical points and access all phases . the most interesting phase emerges when @xmath93 . it is a charge insulating but spin conducting phase . its transitions to other insulating phases are governed by an intermediate fixed point exhibiting non - fermi liquid properties . in order to study this intermediate fixed point , we adjust luttinger parameters to @xmath1 and @xmath2 . we solved rg flow equations for our resonant tunneling problem at this setting . at the toulouse limit , we mapped it to a quantum brownian motion model on a kagome lattice and identified the stable intermediate fixed point . with the insight from yi and kane@xcite , we found that the quantum brownian motion model is precisely the toulouse limit of a two - channel kondo problem with @xmath3 impurity spin . with the equivalency between the kondo problem and our resonant tunneling problem established , this allowed us to take advantage of the boundary conformal field theory solution of the multichannel kondo problem . we obtained an exact description of our intermediate fixed point and calculated the on - resonance conductance as well as the scaling behavior of the resonance line - shape as a function of temperature . finally , we discussed the equivalence between the two - channel @xmath3 kondo problem(@xmath4 cft ) and the three - channel @xmath5 kondo problem ( @xmath6 cft ) from a quantum brownian motion perspective . we also made a generalization establishing the equivalence between @xmath8 and @xmath7 kondo problems . we thank a. w. w. ludwig , a. stern , p. fendley , s. h. simon for helpful discussion and comments . this work was supported by a simons investigator grant to c.l.k . by the simons foundation . 56ifxundefined [ 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 10.1103/revmodphys.82.3045 [ * * , ( ) ] link:\doibase 10.1103/revmodphys.83.1057 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.95.226801 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.95.146802 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.96.106802 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.106.106802 [ * * , ( ) ] link:\doibase 10.1103/physrevb.83.035107 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.110.067205 [ * * , ( ) ] link:\doibase 10.1103/physrevb.87.155114 [ * * , ( ) ] @noop * * , ( ) @noop * * , ( ) link:\doibase 10.1103/physrevb.82.035301 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.101.176801 [ * * , ( ) ] link:\doibase 10.1103/physrevb.75.045317 [ * * , ( ) ] \doibase http://dx.doi.org/10.1016/j.aop.2009.03.005 [ * * , ( ) ] , link:\doibase 10.1103/physrevb.63.235301 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.84.1764 [ * * , ( ) ] \doibase http://dx.doi.org/10.1016/0038-1098(95)00181-6 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.68.1220 [ * * , ( ) ] link:\doibase 10.1103/physrevb.47.4631 [ * * , ( ) ] link:\doibase 10.1103/physrevb.46.15233 [ * * , ( ) ] link:\doibase 10.1103/physrevb.83.245132 [ * * , ( ) ] \doibase http://dx.doi.org/10.1016/0550-3213(90)90440-o [ * * , ( ) ] \doibase http://dx.doi.org/10.1016/0550-3213(91)90109-b [ * * , ( ) ] link:\doibase 10.1103/physrevb.57.r5579 [ * * , ( ) ] link:\doibase 10.1103/physrevb.65.195101 [ * * , ( ) ] http://stacks.iop.org/0305-4470/34/i=6/a=301 [ * * , ( ) ] \doibase http://dx.doi.org/10.1016/s0550-3213(00)00499-5 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.72.4129 [ * * , ( ) ] @noop * * , ( ) \doibase http://dx.doi.org/10.1016/0003-4916(83)90202-6 [ * * , ( ) ] link:\doibase 10.1103/physrevb.32.6190 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.51.1506 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.54.263 [ * * , ( ) ] link:\doibase 10.1103/physrevb.2.270 [ * * , ( ) ] http://stacks.iop.org/0022-3719/14/i=19/a=010 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.47.1840 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.90.246403 [ * * , ( ) ] link:\doibase 10.1103/physrevb.46.10812 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.58.602 [ * * , ( ) ] link:\doibase 10.1103/physrevb.36.3186 [ * * , ( ) ] \doibase http://dx.doi.org/10.1016/0550-3213(94)90365-4 [ * * , ( ) ] link:\doibase 10.1103/physrevb.79.235321 [ * * , ( ) ] @noop * * , ( ) \doibase http://dx.doi.org/10.1016/0550-3213(86)90596-1 [ * * , ( ) ] @noop _ _ ( , ) link:\doibase 10.1103/physrevb.45.7918 [ * * , ( ) ] link:\doibase 10.1103/physrevb.1.4464 [ * * , ( ) ] link:\doibase 10.1103/physrevb.52.r17040 [ * * , ( ) ] link:\doibase 10.1103/physrevb.52.r5539 [ * * , ( ) ] link:\doibase 10.1103/physrevb.52.r8666 [ * * , ( ) ] link:\doibase 10.1143/jpsj.65.30 [ * * , ( ) ] , link:\doibase 10.1103/physrevlett.91.206403 [ * * , ( ) ] http://stacks.iop.org/1742-5468/2006/i=02/a=p02008 [ * * , ( ) ] link:\doibase 10.1103/physrevb.56.2012 [ * * , ( ) ] here we set @xmath33 and @xmath74 and adopt the renormalization group calculation developed by anderson , yuval , and hanmann@xcite . first , we decimate all possible closely placed pairs of charges with a range between @xmath301 and @xmath302 . inserting them in between charge @xmath303 and @xmath304 , the partition function becomes @xmath305 \end{split}\ ] ] where the interaction of the dipoles with all other charges is @xmath306 \tau_c \partial_\tau \ln{\frac{\tau-\tau_j}{\tau_c}}.\ ] ] expand the exponent @xmath307 . \end{split}\ ] ] and auxiliary vectors @xmath308 , width=96 ] now , define @xmath309 ( their relation is depicted in fig . 13 ) , @xmath310 due to neutrality condition , @xmath311 gives @xmath312 . rescaling @xmath301 to @xmath302 , the partition function becomes @xmath313 \ln{\frac{\tau_i-\tau_j}{\tau_c } } ) , \end{aligned}\ ] ] which leads to the flow equations @xmath314 in this section , we calculate the universal mobility @xmath193 of quantum brownian motion model on kagome lattice at the intermediate fixed point using boundary conformal field theory results on kondo problem . the analog of @xmath315 is the spin in our kondo problem @xmath316 . therefore spin currents for our kondo problem are @xmath317 and their linear combinations are @xmath318 can be translated into velocity of brownian particles@xcite . for example , the velocity @xmath319 is mapped to the rate of spin @xmath320 injected into spin - flavor channel @xmath321 where @xmath322 the mobility @xmath187 now becomes the response of spin currents to applied potentials . @xmath323[j_{sf}^{\sigma_i}(0^+,0)- j_{sf}^{\sigma_i}(0 ^ -,0 ) ] \rangle_0 \end{aligned}\ ] ] where @xmath324 is the average with respect to the free non - interacting hamiltonian . that across the non - trivial boundary can not be translated asymptotically away from the boundary . while in b ) , correlators like @xmath325 can and thus produce trivial value which is not affected by the boundary.,width=288 ] using boundary conformal field theory@xcite , the correlation functions are calculated @xmath326 @xmath327 where @xmath202 is a universal complex number depending on kondo boundary condition and can be calculated using modular @xmath328-matrix@xcite @xmath329 ( @xmath266 is the highest weight representation of @xmath330 and @xmath203 is the highest weight of representation of impurity spin in the corresponding lie algebra ) . the general formula for calculating modular @xmath328-matrix is given as@xcite @xmath331 where @xmath332 is the number of positive roots , @xmath333(@xmath334 is the cartan matrix ) for simply - laced algebras , @xmath267 is the level , @xmath10 is the dual coexter number , @xmath335 is the weyl group , @xmath336 is the signature function and @xmath134 is the weyl vector which is half of the sum of all positive roots . for our @xmath3 case , we have @xmath330 is in the adjoint representation of @xmath3 and the impurity spin is in the fundamental representation . thus , @xmath337 $ ] and @xmath338 $ ] or @xmath339 $ ] . then @xmath340 which agrees with @xmath202 calculated in three - channel @xmath5 kondo problem@xcite . the mobility is @xmath341 it is known that the kubo conductance computed from linear response theory does not match the physical dc conductance measured in a system with fermi liquid lead@xcite . the kubo conductance describes the response of an infinite luttinger liquid , where the limit @xmath342 is taken _ before _ @xmath343 , and relates the current to the potential difference between the incident chiral modes of the luttinger liquid . however , the potential of the chiral modes is not the same as the potential of the fermi liquid leads . there is a contact resistance between the luttinger liquid and the electron reservoir where the voltage is defined . an appropriate model to account for this is to consider a 1d model for the leads in which the luttinger parameter @xmath344 in the leads@xcite . here we review that argument for the simple case of spinless electrons characterized by a single luttinger parameter @xmath10 . the generalization to include spin is straightforward . the relationship between the kubo conductance and the physical conductance can be determined by specifying the appropriate boundary condition at the interface between the luttinger liquid and fermi liquid , where @xmath345 in eq . ( 3 ) changes . charge conservation requires @xmath346 is continuous , while the condition of zero backscattering at the interface requires @xmath347 is continuous . using the equations of motion determined by eqs . 1 and 3 , we thus conclude that @xmath348 and @xmath349 are continuous . since the kubo conductance relates the current to the potential difference between the incoming _ chiral _ modes , it is useful to rewrite this boundary condition in terms of the chiral potentials @xmath350 . we thus require the continuity of @xmath351 ( charge conservation ) and @xmath352 ( no backscattering ) . applied to a single interface between @xmath353 and @xmath354 ( fig . 15 ) , we thus conclude @xmath355 elinminating @xmath356 and @xmath357 leads to @xmath358 thus , the potential of the incoming chiral mode in the fermi liquid lead is higher than potential of the chiral mode in the luttinger liquid . the contact between @xmath353 and @xmath354 effectively contributes a _ series resistance _ @xmath359 in a two terminal setup with two fermi liquid leads the series contact resistance is doubled . writing the kubo conductance as @xmath360 , where @xmath361 is the mobility , we then conclude the physical conductance is @xmath362 this reproduces the fact that for perfect transmission @xmath363 the physical conductance is @xmath364 , while the kubo conductance is @xmath365 . for the spinful case , in both the charge and the spin sectors , the contact resistance gets a factor of @xmath366 and the kubo conductance gets a factor of @xmath24 . therefore , the physical conductance gets an overall factor of @xmath24 . in this section we will discuss how our quantum brownian motion picture of level - rank duality fits to general values of @xmath289 and @xmath267 and apply the knowledge to establish proof of equivalence between quantum brownian motion models on generalized honeycom and kagome lattice . first , notice that we can represent primitive vectors of bravais lattice of the @xmath286 quantum brownian motion model as @xmath367 matrices @xmath16 with the only non - zero terms at @xmath368 and @xmath369 for @xmath370 @xmath371 basically , a primitive vector hops particle between adjacent lattice site with the same basis label . in kondo language , it refers to processes which leave intact the impurity spin . since we can choose arbitrary spin state for our impurity spin from @xmath372 , let it be @xmath124 , then each @xmath16 matrix represents the process which transfers an @xmath303 spin from channel @xmath124 to channel @xmath373 via impurity spin . clearly we can not put @xmath374 in either the first row or the first column , it leaves only @xmath375 independent spots . therefore , the matrices are actually describing a @xmath375 dimensional lattice . for the corresponding reciprocal lattice , we find primitive vectors are @xmath376 matrix @xmath377 : @xmath378 different primitive vectors can be obtained by shifting both the row and column containing entry @xmath375 around at all @xmath379 positions in the matrix . with the foundation laid , now let us proceed to the special cases with either @xmath289 or @xmath292 . we will show in the following that all @xmath294 and @xmath380 for some positive constant @xmath245 . the first trick is to choose the right origin . in general , @xmath290 will be a complex number , however , if we choose the origin of our coordinate system to be at one of the center of inversion , then we are putting @xmath180 and @xmath381 at the same footing . therefore , @xmath382 which makes it a real number . for later calculation convenience , we will choose one of the center of the bond of our generalized honeycomb lattice to be the origin for both lattices ( fig . 16 ) . next , let us again embed our @xmath293 dimensional lattices into a higher dimensional space , this time a @xmath267 dimensional space . for generalized @xmath267-honeycomb and kagome lattice , we found basis vectors are : @xmath383and @xmath384 and the shortest vectors for their reciprocal lattice are : @xmath385

baryon mass relations are compared with lattice simulations of baryon masses using different values of the light - quark masses , and hence different values of flavor - symmetry breaking . the lattice data clearly display both the and flavor - symmetry breaking hierarchies . the validity of baryon mass relations derived without assuming approximate flavor - symmetry also can be tested by lattice data at very large values of the strange quark mass . the expansion constrains the form of discretization effects ; these are suppressed by powers of by taking suitable combinations of masses . this scaling is explicitly demonstrated in the present work .

0907.0529

the @xmath0 expansion of qcd @xcite is a valuable tool for studying the nonperturbative dynamics of the strong interactions @xcite . in the limit @xmath2 , the baryon sector of qcd has an exact _ contracted _ @xmath3 spin - flavor - symmetry @xcite . for finite @xmath4 , the contracted spin - flavor - symmetry is broken by effects suppressed by powers of @xmath0 @xcite . the spin - flavor structure of the @xmath0 breaking terms is predicted at each order in the @xmath0 expansion @xcite . the spin - flavor structure of many baryon properties have been derived in a systematic expansion in @xmath0 @xcite , and the results are in excellent agreement with experiment ( for reviews , see @xcite ) . one important application of the baryon @xmath0 expansion is to baryon masses @xcite . by choosing appropriate linear combinations of the baryon masses , one can study coefficients of the baryon mass @xmath0 operator expansion with definite spin and flavor transformation properties . in the case of perturbative @xmath1 flavor - symmetry breaking , the @xmath0 analysis gives a hierarchy of baryon mass relations in powers of @xmath0 and the dimensionless @xmath1 breaking parameter @xmath5 @xcite . the analysis in ref . @xcite showed that the experimentally measured masses of the ground state octet and decuplet baryons exhibit the predicted hierarchy of the combined expansion in @xmath0 and @xmath1 flavor - symmetry breaking . the @xmath0 expansion also has been used to obtain very accurate predictions for the charm and bottom baryon masses @xcite ( to better than 10 mev accuracy ) which have been confirmed by recent experiments . in addition , ref . @xcite derived baryon mass relations which only depend on the @xmath0 expansion and which are valid even if flavor @xmath1 is not a good approximate symmetry , i.e. for large values of the strange quark mass . the predictions of the @xmath0 expansion for baryon masses are in excellent agreement with the experimental values . however , in comparing with the experimental data , one is restricted to only one value of @xmath4 and to the physical quark mass values . testing the predictions as a function of light - quark masses and @xmath4 is now possible with very accurate simulations of baryons using lattice qcd . tremendous progress in lattice qcd has been achieved recently in the simulation of baryon masses using different values of the light - quark masses . extrapolation of baryon masses on the lattice to the physical point has reproduced the experimental values at the @xmath6 level @xcite . the lattice data , however , contain important additional information about the dependence of the baryon masses on the quark masses , which can be utilized . simulations of baryon masses have been performed as a function of @xmath1 flavor - symmetry breaking ranging from small perturbative flavor - symmetry breaking to large nonperturbative flavor - symmetry breaking . there are also lattice simulations at different values of @xmath4 ( though not for baryons ) which are able to test @xmath4 scaling rules @xcite . in this paper , we show that existing lattice simulations ( at @xmath7 ) of the ground state baryon masses already are sufficiently accurate to exhibit and test interesting features of the @xmath0 and @xmath1 flavor - symmetry breaking expansions . still more accurate simulations are needed to test the most suppressed mass combinations of the @xmath0 expansion , but continued improvements in lattice simulations of baryon masses are expected in the short and long term , so it should eventually be possible to test these relations as well . we discuss in this paper how present and future lattice data can be utilized to study the spin and flavor structure of baryon masses . although we do not focus on this point here , it should eventually be possible to test the @xmath4 scaling rules in the baryon sector by lattice simulations which vary the number of colors @xmath4 . an important observation is that the @xmath0 counting rules hold at finite lattice spacing , and so are respected by the lattice results _ including _ the finite lattice spacing corrections dependent on the lattice spacing @xmath8 . thus the discretization corrections are constrained by the @xmath0 expansion . lattice computations of hadron masses are done at varying values of the light - quark masses @xmath9 , @xmath10 and @xmath11 , usually in the isospin limit @xmath12 . the lattice results are then extrapolated to the physical values of @xmath13 . it has not been possible to compute hadron masses for physical values of @xmath13 yet due to the large computational time needed , since @xmath14 . , where @xmath15 is the spatial size of the lattice . ] however , recently , with both algorithmic developments @xcite and large parallel computing machines , there are two groups simulating at or near the physical light - quark mass point @xcite . the light - quark mass dependence of the hadron masses is determined by chiral perturbation theory . there are nonanalytic in @xmath16 contributions from loop corrections which are calculable , as well as analytic terms which depend on low - energy constants ( lecs ) of the chiral lagrangian . in the baryon sector , the leading nonanalytic terms are @xmath17 and @xmath18 . the @xmath17 term , which arises from fig . [ fig : loop ] and is proportional to @xmath19 times axial couplings , radiative corrections . [ fig : loop],width=188 ] is found to be rather large , naively of order a few hundred mev . this term is absent for the pseudo - goldstone boson masses , but is present for other mesons such as the vector mesons @xcite . a surprising feature of recent lattice results is that the baryon masses as a function of @xmath20 do not show a large @xmath21 dependence . fits to the baryon masses also give much smaller values for the baryon axial couplings , and require that the @xmath22 term is almost completely cancelled by @xmath18 and @xmath23 terms . this cancellation must be accidental at the currently accessible lattice quark masses , since the terms have different @xmath16 dependence . an alternative conclusion is that lattice quark masses are too large for @xmath1 chiral perturbation theory to be valid , and that perturbative chiral behavior sets in only for much smaller quark masses than the strange quark mass . this conclusion , however , fails to explain why @xmath1 flavor - symmetry is so evident in baryon phenomenology . the @xmath0 expansion constrains the structure of baryon chiral perturbation theory @xcite . chiral corrections have to respect the spin - flavor structure of the @xmath0 expansion , and this leads to some important restrictions on the form of the chiral loop corrections . for example , the baryon mass , which is of order @xmath4 @xcite , gets nonanalytic corrections proportional to @xmath24 , which are large . this large nonanalytic contribution might lead one to expect that there should be large deviations from the gell - mann okubo mass relations , which were derived assuming that the mass operator was linear in @xmath11 . one can show , however , that the @xmath25 terms are a spin - flavor singlet , and give the same contribution to octet and decuplet baryons , whereas the @xmath24 terms are a spin - singlet flavor - octet , and the @xmath26 terms are spin - singlet flavor-@xmath27 . only the latter terms contribute to the gell - mann okubo mass combinations , so that deviations from these relations are a factor of @xmath28 smaller than naive expectation and consistent with experiment . the small size of the gell - mann okubo relation was recently confirmed for a range of light - quark masses @xcite . for qcd , the cancellation to order @xmath28 arises as a numerical cancellation between octet and decuplet intermediate states @xcite ; to see the parametric form of the cancellation in @xmath0 requires computing the chiral corrections using the @xmath1 flavor representations of baryons containing @xmath4 quarks . the mass relations of ref . @xcite project the baryon masses onto different spin - flavor channels . by studying these mass relations as a function of @xmath16 , one can investigate whether unexpected chiral behavior arises in a particular channel . the organization of this paper is as follows : in sec . ii , the baryon mass relations of the @xmath0 expansion are summarized briefly , and in sec . iii , the lattice simulation data are described . iv presents the results of a computation of @xmath0 mass combinations on the lattice for varying values of @xmath1 flavor - symmetry breaking . v discusses the lattice analysis using heavy baryon chiral perturbation theory @xcite . our conclusions are presented in sec . the @xmath0 expansion of the baryon mass operator for perturbative @xmath1 flavor - symmetry breaking is @xmath29 where the superscript denotes the flavor @xmath1 representation and the spin @xmath30 representation of each term . the @xmath0 expansion of the baryon mass operator contains eight independent operators , corresponding to the eight isospin multiplets of the ground state baryons . the arbitrary coefficients @xmath31 multiplying the @xmath0 operators are functions of @xmath0 and @xmath1 flavor - symmetry breaking . each coefficient is order @xmath32 at leading order in the @xmath0 expansion . the non - trivial content of the @xmath0 expansion is the @xmath0 suppression factors for the different terms in eq . ( [ izero ] ) . following ref . @xcite , we define the parameter @xmath33 , which is a dimensionless measure of flavor-@xmath1 symmetry breaking . @xmath1 flavor - symmetry breaking transforms as a flavor - octet , so the @xmath34 , @xmath35 , @xmath36 and @xmath37 operator expansions are zeroth , first , second , and third order in @xmath1 flavor - symmetry breaking , respectively . thus , the coefficients @xmath38 , @xmath39 and @xmath40 are naively order @xmath41 , @xmath42 and @xmath43 , respectively , in the @xmath1 flavor - symmetry breaking expansion . the @xmath0 expansion for the baryon mass operator for @xmath44 flavor - symmetry @xcite is given by @xmath45 where each operator is accompanied by an unknown coefficient which has been suppressed for simplicity . each coefficient is order unity at leading order in the @xmath0 expansion . the operator @xmath46 , where @xmath47 is the @xmath48-quark number operator . again , there are eight independent operators in the @xmath0 expansion corresponding to the masses of the eight isomultiplets of the ground state baryons . @xmath49 \hline \mreln_a & & & \left ( \sigma^ * - \sigma \right)- \left ( \xi^ * - \xi \right ) & 1/\n^2 & - \\ \mreln_b & & & { 1 \over 3 } \left ( \sigma + 2 \sigma^ * \right ) - \lambda - { 2 \over 3 } \left ( \delta - n \right ) & 1/\n^2 & - \\ \mreln_c & & & -{1 \over 4 } \left ( 2 n - 3 \lambda - \sigma + 2 \xi \right ) + { 1 \over 4 } \left ( \delta -\sigma^ * - \xi^ * + \omega \right ) & 1/\n^2 & - \\ \mreln_d & & & -{1 \over 2 } \left ( \delta - 3 \sigma^ * + 3 \xi^*- \omega \right ) & 1/\n^2 & - \\ \hline\hline \end{array}\end{aligned}\ ] ] the @xmath0 expansion continues to hold at finite lattice spacing , so that expansions of the form eqs . ( [ izero ] ) and ( [ su2u1 ] ) are valid with coefficients that depend on the lattice spacing @xmath8 . group are replaced by irreducible representations of the cubic group . ] thus , the order @xmath4 discretization correction is universal , and has the same value for all the octet and decuplet baryons . this term can be eliminated by studying baryon mass ratios or differences . the largest term that produces different discretization effects for the octet baryons is the @xmath50 term in @xmath51 , or equivalently the @xmath52 term in eq . ( [ su2u1 ] ) , which results in an @xmath53 mass splitting that is suppressed by one factor of @xmath0 relative to the leading mass term @xmath54 . the twelve @xmath0 mass relations that we study are tabulated in table [ table : relns ] . the eight mass relations @xmath55@xmath56 correspond to the first eight mass combinations in table ii of ref . these mass relations are the mass combinations which occur at definite orders in @xmath0 and perturbative @xmath1 flavor - symmetry breaking ; each relation picks out a particular coefficient in eq . ( [ izero ] ) . the mass combinations in table [ table : relns ] correspond to the coefficients listed in the table . there are two mass combinations , @xmath55 and @xmath57 , which are @xmath1 singlets and which occur at orders @xmath4 and @xmath0 , respectively . there are three flavor - octet mass combinations , @xmath58 , @xmath59 and @xmath60 , which are proportional to one factor of @xmath1 symmetry breaking @xmath41 and which occur at orders @xmath32 , @xmath0 and @xmath28 , respectively . there are two flavor-@xmath27 mass combinations , @xmath61 and @xmath62 , which occur at second order in @xmath1 breaking @xmath42 and at orders @xmath0 and @xmath28 in the @xmath0 expansion . finally , there is one mass combination @xmath56 which is suppressed by three powers of @xmath1 breaking @xmath43 and by @xmath28 in the @xmath0 expansion . the additional four mass combinations @xmath63@xmath64 correspond to the first four mass relations of ref . these baryon mass combinations are each order @xmath28 in the @xmath0 expansion . only three of the four combinations @xmath63@xmath64 are linearly independent , corresponding to the fact that there are three independent operators at order @xmath28 in eq . ( [ su2u1 ] ) . these mass relations were derived assuming only isospin flavor - symmetry , and so do not assume approximate @xmath1 flavor - symmetry and are valid even for very large @xmath1 flavor - symmetry breaking . note that the @xmath28 mass relations @xmath63@xmath64 are related to the @xmath28 relations @xmath60 , @xmath62 and @xmath56 by @xmath65 each mass relation defines a mass combination of the octet and decuplet masses @xmath66 of the form @xmath67 in this work , isospin breaking is neglected , so each @xmath66 denotes the average mass of a baryon isomultiplet . the normalization of each mass relation is arbitrary , and depends on a particular choice of normalization for coefficients in the hamiltonian , i.e. @xmath68 defines another mass relation with the same spin and flavor quantum numbers . to remove the rescaling ambiguity , ref . @xcite used the accuracy defined by @xmath69 to quantify the fractional error of a given mass relation . here we use a related quantity , the scale invariant mass combination @xmath70 dividing by @xmath71 instead of by @xmath72 avoids mixing different flavor representations via the denominator factor . the rescaled relations @xmath73@xmath74 and @xmath75@xmath76 have dimensions of mass . in our numerical analysis , we shall use the dimensionless variable @xmath77 as a measure of @xmath1 breaking , where @xmath78 gev @xcite is the scale of chiral symmetry breaking . in this work , we use the results of the recent lhpc spectrum calculation @xcite to explore the mass combinations of the @xmath0 expansion . the lhp collaboration utilized a mixed - action lattice calculation with domain - wall @xcite valence propagators computed with the asqtad improved @xcite dynamical milc gauge ensembles @xcite . the calculation was performed at one lattice spacing with @xmath79 , and a fixed spatial volume @xmath80 . the pion and kaon masses used in ref . @xcite are @xmath81 , @xmath82 , @xmath83 , @xmath84 , @xmath85 and @xmath86 , respectively , on the ` m007 ` , ` m010 ` , ` m020 ` , ` m030 ` , ` m040 ` , and ` m050 ` ensembles , where the labels denote the light - quark masses in lattice units . is defined at this order by the gell - mann okubo formula @xmath87 . ] in the dynamical ensembles and the computation of the valence propagators , the strange quark was held fixed near its physical value . ( in fact the strange quark was @xmath88 too large @xcite . ) for further details of the calculation , we refer the reader to ref . @xcite . using the _ bootstrap resampled _ lattice data , we determine the 12 mass combinations of table [ table : relns ] on each ensemble . the results are collected in table [ table : lattrelns ] . these results were determined with the absolute scale of @xmath89 on all coarse ensembles , where the scale used in ref . @xcite was determined from heavy - quark spectroscopy . we have additionally determined the mass combinations using the smoothed values of @xmath90 , where @xmath91 is determined on each different ensemble from the heavy - quark potential with @xmath92 @xcite . the values of @xmath93 determined in this way range from @xmath94 from the lightest to heaviest quark mass . these two scale - setting methods are in good agreement , as shown in the next section . @xmath95&0.2579 & 0.2380 & 0.1718 & 0.1141 & 0.0574 & 0\\ \hline \mreln_1 [ \ufont{gev } ] & 192(1 ) & 197(1 ) & 211(1 ) & 221(2 ) & 237(2 ) & 242(1 ) \\ \mreln_2 [ \ufont{gev } ] & -12.2(5 ) & -13.3(6 ) & -10.9(4 ) & -10.6(6 ) & -7.8(6 ) & -8.2(4 ) \\ \mreln_3 [ \ufont{gev } ] & -8.07(20 ) & -7.15(15 ) & -4.40(08 ) & -2.81(09 ) & -1.28(04 ) & 0\\ \mreln_4 [ \ufont{gev } ] & -0.214(10 ) & -0.181(7 ) & -0.099(5 ) & -0.056(6 ) & -0.022(3 ) & 0\\ \mreln_5 [ \ufont{gev } ] & 0.29(7 ) & 0.35(8 ) & 0.13(3 ) & 0.14(5 ) & 0.04(2 ) & 0 \\ \mreln_6 [ \ufont{gev } ] & -1.05(36 ) & -0.25(26 ) & -0.003(71 ) & 0.15(14 ) & -0.02(1 ) & 0 \\ \mreln_7 [ \ufont{gev } ] & -0.30(12 ) & -0.05(11 ) & 0.005(24 ) & 0.08(7 ) & -0.003(4 ) & 0 \\ \mreln_8 [ \ufont{gev } ] & 0.02(1 ) & 0.012(09 ) & -0.001(1 ) & 0.015(12 ) & 0.00004(6 ) & 0 \\ \hline \mreln_a [ \ufont{gev } ] & 0.0004(58 ) & 0.018(7 ) & 0.001(3 ) & 0.004(4 ) & -0.0003(19 ) & 0 \\ \mreln_b [ \ufont{gev } ] & -0.022(10 ) & -0.018(10 ) & -0.0004(30 ) & -0.003(3 ) & 0.0000(13 ) & 0 \\ \mreln_c [ \ufont{gev } ] & 0.010(4 ) & 0.002(4 ) & -0.0002(8 ) & -0.003(3 ) & 0.0001(1 ) & 0 \\ \mreln_d [ \ufont{gev } ] & -0.012(5 ) & -0.0061(45 ) & 0.0004(6 ) & -0.0076(58 ) & -0.00002(3 ) & 0 \\ \hline\hline \end{array}\end{aligned}\ ] ] a plot of the lattice baryon masses as a function of @xmath96 is given in fig . [ fig : mbt ] . . the blue squares are the pdg values and the triangles are lhp collaboration lattice data simulated with fixed @xmath11 and varying @xmath13 . the two right - most points simulate the flavor - singlet baryon masses in the exact @xmath1 flavor - symmetry limit @xmath97 . note that the value of @xmath96 at the @xmath1 flavor - symmetry point is quite large , @xmath98 mev . the ( pdg ) baryons from bottom to top are @xmath99 and @xmath100 . [ fig : mbt],width=302 ] the triangles are the lattice data , and the blue squares are the particle data group ( pdg ) values . varies slightly for the different @xmath96 values and is not exactly equal to the physical @xmath101 . ] the usual way of studying the data is to fit to the individual baryon masses . the @xmath0 analysis shows that it is useful to study the mass combinations in table [ table : relns ] instead , as these combinations have definite orders in @xmath0 and @xmath1 symmetry breaking . in this section , we present a series of plots of the 12 scale invariant baryon mass combinations @xmath102@xmath74 and @xmath103@xmath104 as a function of @xmath1 flavor - symmetry breaking . in each plot , the red triangles and green circles are the lattice data , and the errors on the mass combinations have been computed using the _ bootstrap _ data sets of the octet and decuplet masses , taking advantage of the full correlations in the lattice data . the red triangles and green circles are the lattice results using the two different scale - setting methods . the red points are determined with the absolute scale setting @xmath89 , and the green points are determined with the smoothed @xmath90 values @xcite . the blue square is obtained using the physical baryon and meson masses from the pdg @xcite , with the error bar computed using experimental uncertainties on the isospin averaged masses . for most of the plots , the error on the blue point is smaller than size of the point , and is not visible . the horizontal axis is @xmath105 in units of @xmath106 , which is a measure of @xmath1 flavor - symmetry breaking , and the vertical axis is the mass combination in mev . for mass combinations proportional to powers of @xmath1 breaking , we plot both the mass combination and the mass combination divided by the appropriate power of @xmath41 , using the definition eq . ( [ edef ] ) for @xmath41 , so that the units remain mev . the average @xmath107 mass of the octet and decuplet baryons is @xmath108 mev . naive power counting with @xmath7 implies order @xmath4 masses are @xmath109 mev , order @xmath32 masses are @xmath110 mev , order @xmath0 masses are @xmath111 mev , and order @xmath28 masses are @xmath112 mev . this @xmath0 hierarchy is evident in the pdg and lattice data for the baryon mass combinations @xmath102@xmath74 and @xmath103@xmath104 below . figure [ fig : r1 ] plots the first relation @xmath102 , which is the scale invariant mass corresponding to the @xmath113 operator in the @xmath0 expansion . this mass relation is order @xmath114 mev over the range of @xmath1 flavor - symmetry breaking simulated by the lhp collaboration and is seen to be fairly independent of @xmath1 breaking as expected . the deviation of the physical pdg point from the lattice data is presumably due to the larger than physical strange quark mass of the simulation . in principle , this relation also is susceptible to finite discretization errors , beginning at @xmath115 , although these are expected to be small for this data set @xcite . as a function of @xmath116 . the red triangles and green circles are lattice results with two different scale setting methods ( see text ) , and the blue square is using the pdg values for the hadron masses . @xmath102 is the average @xmath117 mass of the ground state baryons . [ fig : r1],width=302 ] figure [ fig : r2 ] plots the second relation @xmath118 , which corresponds to the @xmath119 operator in the @xmath0 expansion . the mass combination @xmath118 is clearly suppressed relative to the average ground state baryon mass @xmath102 . this hyperfine splitting is predicted to be order @xmath120 , and so of order @xmath121 mev in magnitude . the lattice data allow us to study this mass relation as a function of @xmath1 breaking . notice that the relation works independently of @xmath1 breaking as predicted theoretically , and that @xmath118 does not vanish at the @xmath1 symmetry point @xmath122 . the numerical results clearly show that the suppression of @xmath118 is a consequence of @xmath0 , and not due to a hidden flavor breaking suppression factor . as a function of @xmath116 . the red triangles and green circles are lattice results with two different scale setting methods ( see text ) , and the blue square is using the pdg values for the hadron masses . @xmath118 is minus the @xmath123 hyperfine mass splitting between the spin-@xmath124 decuplet and the spin-@xmath125 octet baryons . [ fig : r2],width=302 ] figure [ fig : r3 ] plots the third relation @xmath126 , which is of order @xmath41 . the relation vanishes as @xmath127 , as can be seen in the upper panel . the lower panel divides the mass relation by @xmath41 , and shows that @xmath128 is an @xmath129 mass in the @xmath0 expansion , or @xmath130 mev in magnitude . notice that @xmath128 does not vanish in the @xmath1 symmetry limit . the @xmath1 symmetry point of the lattice data is omitted in the @xmath128 plot , as it involves dividing by zero . and @xmath128 as a function of @xmath116 . @xmath126 is @xmath131 at leading order in the @xmath0 and @xmath1 flavor - symmetry breaking expansions , and @xmath128 is @xmath132 . [ fig : r3],width=302 ] figure [ fig : r4 ] plots @xmath133 and @xmath134 , which are of order @xmath135 and @xmath120 , respectively . the mass combinations @xmath126 and @xmath133 are both order @xmath41 , but @xmath133 is suppressed relative to @xmath126 by an additional factor of @xmath0 . this suppression is clearly visible in the numerical values which are @xmath136 mev in magnitude . @xmath133 is seen to vanish in the @xmath1 symmetry limit , whereas @xmath134 does not vanish at the @xmath1 symmetric point . both @xmath128 and @xmath134 display some dependence on @xmath1 symmetry breaking , which implies that there is @xmath41 dependence in the coefficients @xmath137 and @xmath138 , respectively . and @xmath134 as a function of @xmath116 . @xmath133 is @xmath139 at leading order , and @xmath134 is @xmath123.[fig : r4],width=302 ] figure [ fig : r5 ] plots @xmath140 and @xmath141 , which are of order @xmath142 and @xmath143 , respectively . the mass relation @xmath141 is consistent with being a @xmath28 mass of magnitude @xmath144 mev . the lattice data points now have relatively large errors , however . and @xmath141 as a function of @xmath116 . @xmath140 is @xmath145 at leading order , and @xmath141 is @xmath146 . [ fig : r5],width=302 ] @xmath147 , @xmath148 and @xmath74 are plotted in figs . [ fig : r6 ] , [ fig : r7 ] and [ fig : r8 ] , respectively , for completeness . in each case , the physical point ( blue square ) has the expected size , and the lattice results are compatible with @xmath0 and @xmath41 power counting expectations , but the error bars are now rather large . it would be very interesting to have more precise lattice data for the @xmath149 and @xmath148 mass relations because the leading @xmath1 flavor - symmetry breaking contribution to the flavor-@xmath36 mass splittings in chiral perturbation theory is @xmath150 rather than the naive @xmath151 of second order perturbative flavor - symmetry breaking . with very precise data , as in ref . @xcite , it would be possible to test this prediction of chiral perturbation theory . and @xmath152 as a function of @xmath116 . @xmath149 is @xmath153 , and @xmath152 is @xmath123 . [ fig : r6],width=302 ] and @xmath154 as a function of @xmath116 . @xmath148 is @xmath155 , and @xmath154 is @xmath146 . [ fig : r7],width=302 ] and @xmath156 as a function of @xmath116 . @xmath74 is @xmath157 , and @xmath156 is @xmath146 . [ fig : r8],width=302 ] the last four plots , figs . [ fig : djm1][fig : djm4 ] , as a function of @xmath116 . @xmath103 is @xmath146 . [ fig : djm1],width=302 ] as a function of @xmath116 . @xmath158 is @xmath146 . [ fig : djm2],width=302 ] as a function of @xmath116 . @xmath159 is @xmath146 . [ fig : djm3],width=302 ] as a function of @xmath116 . @xmath104 is @xmath146 . [ fig : djm4],width=302 ] are the relations @xmath103@xmath104 , which are of order @xmath28 and assume only @xmath44 flavor - symmetry . from the plots , it is evident that @xmath103@xmath104 are satisfied irrespective of the value of @xmath1 breaking . @xmath103@xmath104 are all predicted to be @xmath146 masses , or approximately @xmath160 mev . the figures show that @xmath103@xmath104 are significantly smaller than @xmath160 mev . this suppression can be understood from the perturbative @xmath1 flavor - symmetry breaking analysis , since the lattice data varies @xmath41 only over the range @xmath161 $ ] and does not contain large nonperturbative values of @xmath41 . from eq . ( [ mam5 ] ) , one sees that for small values of @xmath41 , @xmath63 and @xmath162 are @xmath145 since they are linear combinations containing @xmath60 ; @xmath163 is @xmath155 since it is a linear combination containing @xmath62 ; and @xmath64 is @xmath157 since it is proportional to @xmath56 . it would be interesting to see how well relations @xmath103@xmath104 work at even _ larger _ values of @xmath1 breaking extending to very large values for which the @xmath1 flavor - symmetry is completely broken . in the limit @xmath164 , relation @xmath103 goes over to the mass relation for heavy - quark baryons , @xmath165 , so it would be interesting to look at the crossover from light - quark mass behavior proportional to @xmath11 to heavy - quark behavior in which the hyperfine splitting is proportional to @xmath166 . similarly , in the limit @xmath164 , relation @xmath158 goes over to the heavy - quark baryon mass relation @xmath167 - { 2 \over 3 } \left ( \delta - n \right)$ ] , which is independent of the heavy - quark mass . the crossover from @xmath168 behavior to mass - independence in the @xmath164 limit also would be interesting to observe . in addition , it would be interesting to determine the value of @xmath41 at which the perturbative @xmath1 flavor - symmetry breaking analysis fails . finally , an extended range for @xmath169 would allow one to see whether relations @xmath103@xmath104 remain independent of @xmath41 even for large values of @xmath41 . as mentioned in the introduction , a recent surprise from current lattice light - quark spectrum calculations is that the baryon masses do not show a large @xmath171 behavior . in fact , the baryon masses display an unexpected ( nearly ) linear in @xmath96 scaling for a large range of pion masses @xcite . in the case of the nucleon , this finding was verified in all current @xmath172 flavor lattice calculations of the nucleon mass @xcite . these findings pose an interesting theoretical puzzle and indicate the presence of large cancellations to the baryon spectrum from the different orders in the chiral expansion . further , recent @xmath1 chiral extrapolations of the baryon masses @xcite , carried out using the next - to - leading order ( nlo ) extrapolation formulae , have found that the axial couplings , @xmath173 , @xmath174 , @xmath175 and @xmath176 , when left as free parameters , are returned from the analysis with values significantly different from those expected based upon phenomenology @xcite or the recent lattice determination @xcite . it is not clear at this point whether these findings are due to a breakdown of @xmath1 heavy baryon @xmath170pt in this mass range , or , for example , the necessity of including the next - to - next - to - leading order ( nnlo ) terms in the analysis . in this section , we perform an @xmath1 heavy baryon @xmath170pt extrapolation analysis of the various mass relations @xmath177 presented in table [ table : lattrelns ] . this allows us both to extrapolate our results to the physical @xmath178 limit and to begin exploring the combined @xmath0 and @xmath1 expansion utilizing precision lattice data . an important question to explore is whether this combined expansion displays significantly improved convergence properties to the @xmath1 extrapolations performed previously . additionally , because the different mass relations are sensitive to the different coefficients of the baryon mass expansion in eq . , we can directly extrapolate different orders in the @xmath1 chiral expansion . for example , as mentioned above , the leading contribution to @xmath61 and @xmath62 scales as @xmath179 in heavy baryon @xmath170pt . finding a definitive signal of this scaling would be a significant confirmation of the nonanalytic light - quark mass behavior predicted by chiral perturbation theory . the light and strange quark mass dependence is described by @xmath1 heavy baryon chiral perturbation theory ( hb@xmath170pt ) @xcite . the baryon chiral lagrangian is given by @xmath180 t_\mu \nonumber\\ & + b_d \tr \left ( \bar{b } \{\mc{m}_+ , b\ } \right ) + b_f \tr \left ( \bar{b } [ \mc{m}_+ , b ] \right ) \nonumber\\ & + b_0 \tr \left ( \bar{b } b \right ) \tr \left ( \mc{m}_+ \right ) \nonumber\\ & + \gamma_m\ , \bar{t}^\mu \mc{m}_+ t_\mu -\bar{\s}\ , \bar{t}^\mu t_\mu \tr \left ( \mc{m}_+ \right ) \nonumber\\ & + 2d\ , \tr \left ( \bar{b } s^\mu \{\mc{a}_\mu , b\ } \right ) + 2f\ , \tr \left ( \bar{b } s^\mu [ \mc{a}_\mu , b ] \right ) \nonumber\\ & + 2\mc{h}\ , \bar{t}^\mu s_\nu \mc{a}^\nu t_\mu + \mc{c}\ , \left ( \bar{t}^\mu \mc{a}_\mu b + \bar{b } \mc{a}_\mu t^\mu \right)\ , , \end{aligned}\ ] ] where @xmath181 and @xmath182 are the octet and decuplet fields , respectively ; @xmath183 is the four - velocity of the baryon and @xmath184 is the spin - vector . the decuplet octet mass splitting in the chiral limit is @xmath185 , and @xmath186 is the mass spurion defined by @xmath187 in terms of the light - quark mass matrix @xmath188 and @xmath189 . the covariant derivative is @xmath190\ , , \ ] ] and the vector and axial fields are @xmath191 the masses of the octet and decuplet baryons were first determined in ref . @xcite to nlo , @xmath192 . the octet baryon masses were later determined to nnlo , @xmath193 in refs . @xcite , and the decuplet baryons to nnlo in refs . @xcite . the lattice calculation of baryon masses used for this work was a mixed - action calculation @xcite . therefore , the calculation and the corresponding low - energy effective field theory for the baryons are inherently partially quenched . the partially quenched lagrangian for heavy baryons was determined by chen and savage @xcite through nlo . the extension of the partially quenched lagrangian to the mixed - action lagrangian , which includes corrections from the lattice spacing , also is known @xcite , through nnlo @xcite . at nlo , the corrections from the mixed - action are straightforward , amounting to corrections to the mixed valence - sea meson masses @xcite appearing in the one - loop self - energy corrections , fig . [ fig : loop ] , as well as the partially quenched hairpin corrections @xcite . a rigorous extrapolation to the continuum limit requires multiple lattice spacings , but the lattice data used in this work was performed at a single lattice spacing ( albeit with the size of the discretization corrections expected to be small , see ref . @xcite for details ) . however , one can make a qualitative estimate of the size of the discretization errors by comparing the extrapolations performed with the continuum extrapolation formulae and the mixed - action formulae . we now perform these chiral extrapolations . to construct the @xmath194 for the light - quark mass extrapolation , we use the _ bootstrap resampled _ lattice data to form the covariance matrix , @xmath195 - \bar{\mreln}_q \big ) \big ( \mreln_{q^\prime}[bs ] - \bar{\mreln}_{q^\prime } \big)\ , , & \end{aligned}\ ] ] where @xmath196 is a supermass index which runs over both the mass relations @xmath55@xmath62 as well as the different lattice ensembles ` m007``m050 ` . heavy baryon @xmath170pt , the relation @xmath56 is exactly zero . in the partially quenched theory , this mass relation exactly vanishes at nlo , and at nnlo , there are residual partially quenched effects which do not exactly cancel @xcite . ] we then construct the @xmath194 , @xmath197 where @xmath198 is the hb@xmath170pt extrapolation formula for the mass relation @xmath199 and depends upon the low - energy constants @xmath200 . at nlo , the baryon mass extrapolation formulae are linearly dependent upon all the @xmath201 , except for @xmath173 , @xmath174 , @xmath202 and @xmath203 , but they are linearly dependent upon @xmath204 and @xmath205 . therefore , to perform the numerical minimization , we first perform a linear least squares fit on all the linear lecs , @xmath206 , solving for the low - energy constants as functions of @xmath173 and @xmath174 , and then perform the numerical minimization . this procedure is known as the variable projection ( varpro ) method @xcite . we first perform the @xmath194 minimization using relations @xmath55@xmath62 and the nlo extrapolation formulae . in tables [ table : nlor1r7 ] and [ table : nlomar1r7 ] , we collect the results of both the continuum extrapolation and the mixed - action extrapolation . in these tables , the partially quenched / mixed - action low - energy constants @xmath207 , @xmath208 , @xmath209 , @xmath210 and @xmath211 are related to the lecs of @xmath1 hb@xmath170pt of eq . by @xmath212 then , for example , @xmath213 , where at leading order @xmath214 . in both cases , we perform the fit for four different ranges of the quark masses . the first fit includes only the lightest two mass points , while each successive fit includes an additional mass point . the results of the mixed - action and continuum fits are fairly consistent , as can be seen in tables [ table : nlor1r7 ] and [ table : nlomar1r7 ] . @xmath215 & [ \ufont{gev } ] & [ \ufont{gev } ] ^{-1 } & [ \ufont{gev } ] ^{-1 } & [ \ufont{gev}]^{-1 } & [ \ufont{gev}]^{-1 } & [ \ufont{gev } ] ^{-1 } & & \\ \hline \texttt{m007}-\texttt{m010 } & 0.04(28 ) & 0.14(6 ) & 0.999(49 ) & 0.21(16 ) & -0.58(7 ) & -0.42(16 ) & 0.68(37 ) & -0.10(6 ) & -0.36(31 ) & 0.01(12 ) & 0.32(55 ) \\ \texttt{m007}-\texttt{m020 } & 0.17(8 ) & 0.07(5 ) & 0.972(18 ) & 0.48(5 ) & -0.67(2 ) & -0.81(5 ) & 0.39(12 ) & -0.11(2 ) & 0.09(9 ) & 0.21(3 ) & 0.09(19)\\ \texttt{m007}-\texttt{m030 } & 0.21(5 ) & 0.08(3 ) & 0.972(13 ) & 0.44(5 ) & -0.67(2 ) & -0.76(3 ) & 0.49(12 ) & -0.12(1 ) & -0.00(8 ) & 0.15(2 ) & 0.18(18)\\ \texttt{m007}-\texttt{m040 } & 0.20(5 ) & 0.09(2 ) & 0.960(11 ) & 0.43(04 ) & -0.67(1 ) & -0.73(3 ) & 0.62(8 ) & -0.13(1 ) & -0.05(6 ) & 0.14(2 ) & 0.35(13 ) \\ \hline\hline \end{array}\end{aligned}\ ] ] @xmath215 & [ \ufont{gev } ] & [ \ufont{gev } ] ^{-1 } & [ \ufont{gev } ] ^{-1 } & [ \ufont{gev}]^{-1 } & [ \ufont{gev}]^{-1 } & [ \ufont{gev } ] ^{-1 } & & \\ \hline \texttt{m007}-\texttt{m010 } & 0.31(9 ) & 0.26(5 ) & 0.941(42 ) & 0.242(73 ) & -1.29(1 ) & -0.46(1 ) & 3.8(9 ) & -0.35(5 ) & -1.8(4 ) & -0.043(9 ) & 4.3(1.1)\\ \texttt{m007}-\texttt{m020 } & 0.36(3 ) & 0.19(2 ) & 0.989(15 ) & 0.365(30 ) & -1.03(1 ) & -0.70(1 ) & 0.66(22 ) & -0.24(2 ) & -0.22(11 ) & -0.009(4 ) & 0.21(26)\\ \texttt{m007}-\texttt{m030 } & 0.35(2 ) & 0.16(2 ) & 1.001(11 ) & 0.368(28 ) & -0.92(1 ) & -0.71(1 ) & 0.58(21 ) & -0.20(1 ) & -0.15(10 ) & -0.007(4 ) & 0.11(25 ) \\ \texttt{m007}-\texttt{m040 } & 0.34(2 ) & 0.15(1 ) & 0.995(8 ) & 0.376(26 ) & -0.87(1 ) & -0.70(1 ) & 0.69(18 ) & -0.20(1 ) & -0.16(8 ) & -0.001(2 ) & 0.21(21)\\ \hline\hline \end{array}\end{aligned}\ ] ] consistent with refs . @xcite , we find the values of @xmath173 and @xmath174 are significantly smaller than those determined either phenomenologically @xcite or with the recent lattice determination @xcite ( these determinations provide @xmath216 and @xmath217 ) . with the use of the varpro method , we have reduced the numerical minimization to a two - dimensional problem , thus allowing us to plot the resulting @xmath194 as a function of @xmath173 and @xmath174 . in fig . [ fig : chisq007010 ] , we provide a contour plot of the resulting @xmath194 for the continuum fit using the lightest two ensembles . in fig . [ fig : chisq007040 ] , we plot the resulting @xmath194 for the continuum extrapolation using the lightest five ensembles . figures [ fig : chisqma007010 ] and [ fig : chisqma007040 ] display the same plots constructed with the mixed - action extrapolation formulae . in all these plots , the dark ( blue ) central area satisfies @xmath218 , the @xmath219 confidence region for two parameters . each successive @xmath220 contour represents the @xmath221 confidence region . constructed using a continuum @xmath1 hb@xmath170pt @xmath194 fit to the ` m007``m010 ` data sets . the dark ( blue ) inner region represents @xmath222 , the @xmath223 confidence region for two fit parameters , @xmath173 and @xmath174 . each successive @xmath220 contour represents the @xmath224 confidence region . [ fig : chisq007010],width=302 ] constructed using a continuum @xmath1 hb@xmath170pt @xmath194 fit to the ` m007``m040 ` data sets . the dark ( blue ) inner region represents @xmath222 , the @xmath223 confidence region for two fit parameters , @xmath173 and @xmath174 . each successive @xmath220 contour represents the @xmath224 confidence region . [ fig : chisq007040],width=302 ] constructed using a mixed - action @xmath225 hb@xmath170pt @xmath194 fit to the ` m007``m010 ` data sets . the dark ( blue ) inner region represents @xmath222 , the @xmath223 confidence region for two fit parameters , @xmath173 and @xmath174 . each successive @xmath220 contour represents the @xmath224 confidence region . [ fig : chisqma007010],width=302 ] constructed using a mixed - action @xmath225 hb@xmath170pt @xmath194 fit to the ` m007``m040 ` data sets . the dark ( blue ) inner region represents @xmath222 , the @xmath223 confidence region for two fit parameters , @xmath173 and @xmath174 . each successive @xmath220 contour represents the @xmath224 confidence region . [ fig : chisqma007040],width=302 ] one observes that the mixed - action analysis returns values of @xmath173 and @xmath174 which are slightly larger than the continuum @xmath1 analysis , but the values are still smaller than expected . in table [ table : r1r7predictions ] , we use the values of the lecs from the continuum analysis to compare with the experimental values of the individual baryon masses and the @xmath0 mass combinations in table [ table : relns ] ( within a few sigma , the mixed - action extrapolations are consistent ) . despite the apparent discrepancy in the values of @xmath173 and @xmath174 , one observes the extrapolated values of both the baryon masses and the mass combinations , measured as a percent deviation , are in reasonable agreement with experiment . we caution that one can not draw strong conclusions about @xmath1 hb@xmath170pt from this analysis , in particular the small values of @xmath173 and @xmath174 . first , the values of @xmath226 used in this work are still larger than desirable for performing chiral extrapolations ( in fact the strange quark mass is known to be @xmath227% too heavy @xcite ) . second , at nlo , the baryon masses are predicted to have large @xmath228 corrections ; however , this large nonanalytic contribution is not observed in the numerical results for the baryon masses themselves , see fig . [ fig : mbt ] . consequently , there must be strong cancellations between the different orders in the chiral expansion to produce the observed results . in order for the chiral extrapolation analysis to be consistent with both the baryon masses and the expected values of @xmath173 and @xmath174 , one may need to use lighter quark masses and to include @xmath193 contributions to the masses . further , it is likely that a combined analysis of both the masses and the axial couplings will be required . .[table : r1r7predictions ] predicted baryon masses and mass relations from fit in table [ table : nlor1r7 ] . the predicted values are listed according to the heaviest light - quark mass used . the ` m030 ` fit is consistent with that of ` m020 ` and ` m040 ` . [ cols="^ , < , < , < , < " , ] performing the complete nnlo analysis of the octet and decuplet baryon masses introduces 19 new unknown lecs ( for 30 total ) . twelve of these lecs correspond to @xmath229 operators of the form @xmath230 and give rise to @xmath231 corrections to the baryon masses . the other seven lecs correspond to interactions of the baryons with the axial current , of the form @xmath232 , and give rise to baryon mass corrections of the form @xmath233 . the lattice data used in this work , ref . @xcite , are not sufficient to precisely determine all these lecs . further , the number of lecs for which the mass relations are nonlinearly dependent increases to four . in performing the full analysis , we use the varpro method to reduce the number of lecs minimized numerically to 9 . to perform the partial nnlo analysis , we add only the @xmath229 operators and are able to reduce the lecs which require numerical minimization to @xmath173 and @xmath174 , as in the nlo analysis . in figs . [ fig : chisqnnloct ] and [ fig : chisqnnloctma ] , we display @xmath194 contours as a function of @xmath173 and @xmath174 analyzed for the continuum @xmath1 and the mixed - action extrapolation formulae , respectively . in both cases , the five lightest ensembles , ` m007``m040 ` , are used . contour plot of @xmath234 constructed using a continuum @xmath1 hb@xmath170pt @xmath194 fit including partial nnlo counterterms , as discussed in the text , to the ` m007``m040 ` data sets . the dark ( blue ) inner region represents @xmath222 , the @xmath223 confidence region for two fit parameters , @xmath173 and @xmath174 . each successive @xmath220 contour represents the @xmath224 confidence region . , width=302 ] contour plot of @xmath234 constructed using a mixed - action @xmath225 hb@xmath170pt @xmath194 fit including partial nnlo counterterms , as discussed in the text , to the ` m007``m040 ` data sets . the dark ( blue ) inner region represents @xmath222 , the @xmath223 confidence region for two fit parameters , @xmath173 and @xmath174 . each successive @xmath220 contour represents the @xmath224 confidence region . , width=302 ] as can be seen , although the values of @xmath173 and @xmath174 are still consistent with zero , they are now also consistent with the phenomenological and lattice values , albeit with large uncertainties . this large allowed variation of @xmath173 and @xmath174 translates into about an order of magnitude increase in the uncertainties of the predicted baryon mass relations as compared to those in table [ table : r1r7predictions ] . the full nnlo analysis yields results in qualitative agreement with the partial nnlo minimization ; however , there are more than two lecs upon which the @xmath194 depends nonlinearly , and thus we can not display the corresponding contour plots . this analysis makes it seem plausible that a combined nnlo analysis of the baryon masses and the axial couplings could provide much more stringent constraints on the values of @xmath173 and @xmath174 , as well as the other lecs . such an analysis would allow for the first rigorous exploration of octet and decuplet properties using @xmath1 heavy baryon chiral perturbation theory . in this work , we have explored @xmath0 baryon mass relations using lattice qcd . some of these relations were derived assuming approximate @xmath1 flavor - symmetry , @xmath55@xmath56 , while other relations were derived assuming only @xmath30 flavor - symmetry , @xmath63@xmath64 . in all cases , we have found that the baryon mass relations are of the size expected by both the @xmath0 power counting as well as the power counting in the @xmath1 breaking parameter , @xmath235 . for the mass relations @xmath63@xmath64 , it would be interesting to consider strange quark values even heavier than those used in this work . two of these relations become equivalent heavy - quark baryon mass relations as @xmath236 , and this exploration would probe the transition region between @xmath237 . in the charm and bottom baryon spectra , these @xmath0 relations have led to stringent predictions for various mass combinations . it also would be interesting to study @xmath55@xmath56 , which were derived assuming approximate @xmath1 flavor - symmetry , as the strange quark mass is increased , to determine the value of @xmath11 at which the perturbative @xmath1 expansion breaks down . because of the definite spin and flavor transformation properties of the baryon mass relations , the discretization corrections to the mass relations also obey the @xmath0 and @xmath41 power counting , and are suppressed . in fact , only @xmath55 is subject to the leading @xmath115 mass corrections . the suppression of these discretization errors leads to more stringent constraints on the @xmath1 heavy baryon chiral extrapolation of the mass relations to the physical point , and a better method for chiral extrapolation . it allows one to rigorously test the predictions and convergence of @xmath1 hb@xmath170pt . in this work , we also have performed both @xmath1 and mixed - action chiral extrapolations of @xmath0 mass relations using a variable projected @xmath194 minimization . consistent with other works , including ref . @xcite ( from which our numerical data are derived ) , we find that a nlo analysis , which includes the leading nonanalytic mass corrections , returns values of the axial couplings which are significantly smaller than expected from either phenomenology or lattice qcd . however , the predicted values of the mass relations , as well as the octet and decuplet masses themselves , are in good agreement with experiment . further , we have shown that the partial inclusion of nnlo mass corrections returns values of the axial couplings @xmath173 and @xmath174 which are consistent with expectations , albeit with large error bars . unfortunately , the data set we have used in this work is insufficient to precisely constrain all the lecs in the full nnlo analysis . however , we have demonstrated that a simultaneous extrapolation of both the hyperon axial couplings as well as the @xmath0 mass relations would allow for a rigorous exploration of the predictions and convergence of @xmath1 heavy baryon @xmath170pt . it is possible , perhaps even likely , that the strange quark is too heavy for the @xmath1 theory to be convergent . this outcome , however , fails to explain the successes of flavor @xmath1 symmetry observed in nature in the baryon sector . thus , a detailed study of this phenomena is warranted . we would like to thank the members of the lhp collaboration for providing their numerical baryon mass data of ref . awl would like to thank kostas orginos for helpful discussions . the work of eej and avm was supported in part by the department of energy under grant no . de - fg03 - 97er40546 . the work of jwn was supported in part by the u. s department of energy under grant no . de - fg02 - 94er40818 . the work of awl was supported in by the u.s . doe oji grant no . de - fg02 - 07er41527 . g. t hooft , nucl . b * 72 * , 461 ( 1974 ) . e. witten , nucl . b * 160 * , 57 ( 1979 ) . coleman , @xmath238 , presented at 1979 int . school of subnuclear physics , pointlike structures inside and outside hadrons , erice , italy , 1979 . m. a. luty and j. march - russell , nucl . b * 426 * ( 1994 ) 71 c. carone , h. georgi and s. osofsky , phys . b * 322 * , 227 ( 1994 ) . r. f. dashen , e. e. jenkins and a. v. manohar , phys . d * 51 * ( 1995 ) 3697 . d. pirjol and t. m. yan , phys . d * 57 * , 1449 ( 1998 ) . c. l. schat , j. l. goity and n. n. scoccola , phys . rev . lett . * 88 * , 102002 ( 2002 ) . d. pirjol and c. schat , phys . d * 67 * , 096009 ( 2003 ) . e. e. jenkins , ann . rev . nucl . part . * 48 * ( 1998 ) 81 . a. v. manohar , _ large n qcd , _ les houches summer school in theoretical physics , session 68 : probing the standard model of particle interactions , les houches , france , 1997 , arxiv : hep - ph/9802419 . e. e. jenkins and r. f. lebed , phys . d * 52 * ( 1995 ) 282 . e. e. jenkins , phys . d * 54 * ( 1996 ) 4515 , d * 55 * ( 1997 ) r10 , d * 77 * ( 2008 ) 034012 . s. durr _ et al . _ , science * 322 * , 1224 ( 2008 ) . r. flores - mendieta et al . , d * 62 * , 034001 ( 2000 ) . m. j. teper , phys . d * 59 * , 014512 ( 1998 ) . m. j. teper , _ large n , _ arxiv:0812.0085 [ hep - lat ] . a. d. kennedy , nucl . suppl . * 140 * , 190 ( 2005 ) . s. aoki _ et al . _ [ pacs - cs collaboration ] , arxiv:0807.1661 [ hep - lat ] . e. e. jenkins , a. v. manohar and m. b. wise , phys . rev . lett . * 75 * ( 1995 ) 2272 . s. r. beane , k. orginos and m. j. savage , phys . b * 654 * , 20 ( 2007 ) [ arxiv : hep - lat/0604013 ] . e. e. jenkins and a. v. manohar , phys . b * 255 * , 558 ( 1991 ) ; * 259 * , 353 ( 1991 ) . a. manohar and h. georgi , nucl . b * 234 * ( 1984 ) 189 . a. walker - loud _ et al . _ , phys . d * 79 * , 054502 ( 2009 ) . d. b. kaplan , phys . b * 288 * , 342 ( 1992 ) . y. shamir , nucl . b * 406 * , 90 ( 1993 ) . v. furman and y. shamir , nucl . phys . b * 439 * , 54 ( 1995 ) . k. orginos and d. toussaint [ milc collaboration ] , phys . d * 59 * , 014501 ( 1998 ) . k. orginos , d. toussaint and r. l. sugar [ milc collaboration ] , phys . d * 60 * , 054503 ( 1999 ) . c. w. bernard _ et al . _ , phys . d * 64 * , 054506 ( 2001 ) . a. bazavov _ et al . _ , arxiv:0903.3598 [ hep - lat ] . s. r. beane , k. orginos and m. j. savage , int . j. mod . e * 17 * , 1157 ( 2008 ) . c. aubin _ et al . _ [ hpqcd collaboration and milc collaboration and ukqcd collaboration ] , phys . d * 70 * , 031504 ( 2004 ) . c. aubin _ et al . _ , d * 70 * , 094505 ( 2004 ) . c. amsler _ et al . _ [ particle data group ] , phys . b * 667 * ( 2008 ) 1 . s. r. beane _ et al . _ , phys . d * 79 * , 114502 ( 2009 ) [ arxiv:0903.2990 [ hep - lat ] ] . a. walker - loud , arxiv:0810.0663 [ hep - lat ] . k. i. ishikawa _ et al . _ [ pacs - cs collaboration ] , arxiv:0905.0962 [ hep - lat ] . r. flores - mendieta , e. e. jenkins and a. v. manohar , phys . d * 58 * , 094028 ( 1998 ) ; j. dai , _ et al . _ , phys . d * 53 * , 273 ( 1996 ) . h. w. lin and k. orginos , phys . d * 79 * , 034507 ( 2009 ) . e. e. jenkins , nucl . b * 368 * , 190 ( 1992 ) . r. f. lebed and m. a. luty , phys . b * 329 * , 479 ( 1994 ) . b. borasoy and u. g. meissner , annals phys . * 254 * , 192 ( 1997 ) . a. walker - loud , nucl . a * 747 * , 476 ( 2005 ) . r. f. lebed , nucl . b * 430 * , 295 ( 1994 ) . b. c. tiburzi and a. walker - loud , nucl . a * 748 * , 513 ( 2005 ) . j. w. chen and m. j. savage , phys . rev . d * 65 * , 094001 ( 2002 ) . b. c. tiburzi , phys . d * 72 * , 094501 ( 2005 ) . j. w. chen , d. oconnell and a. walker - loud , jhep * 0904 * , 090 ( 2009 ) . k. orginos and a. walker - loud , phys . d * 77 * , 094505 ( 2008 ) . g. golub and v. pereyra , inverse problems , * 19 * , r1-r26 ( 2003 ) .

inter - cell interference coordination ( icic ) and intra - cell diversity ( icd ) play important roles in improving cellular downlink coverage . modeling cellular base stations ( bss ) as a homogeneous poisson point process ( ppp ) , this paper provides explicit finite - integral expressions for the coverage probability with icic and icd , taking into account the temporal / spectral correlation of the signal and interference . in addition , we show that in the high - reliability regime , where the user outage probability goes to zero , icic and icd affect the network coverage in drastically different ways : icd can provide _ order _ gain while icic only offers _ linear _ gain . in the high - spectral efficiency regime where the sir threshold goes to infinity , the order difference in the coverage probability does not exist ; however a linear difference makes icic a better scheme than icd for realistic path loss exponents . consequently , depending on the sir requirements , different combinations of icic and icd optimize the coverage probability .

1403.0012

recently , the poisson point process ( ppp ) has been shown to be a tractable and realistic model of cellular networks @xcite . however , the baseline ppp model predicts the coverage probability of the typical user to be less than 60% if the signal - to - interference - plus - noise ratio ( sinr ) is set to 0 db even if noise is neglected . this is clearly insufficient to provide reasonable user experiences in the network . to improve the user experiences , in cellular systems , the importance of inter - cell interference coordination ( icic ) and intra - cell diversity ( icd ) have long been recognized@xcite . yet , so far , most of the ppp - based cellular analyses lack a careful treatment of these two important aspects of the network , partly due to the lack of a well - established approach to deal with the resulting temporal or spectral correlation @xcite . modeling the cellular network as a homogeneous ppp , this paper explicitly takes into account the temporal / spectral correlation and analyzes the benefits of icic and icd in cellular downlink under idealized assumptions . consider the case where a user is always served by the bs that provides the strongest signal averaged over small - scale fading but not shadowing . for icd , we consider the case where the serving bs always transmit to the user in @xmath0 resource blocks ( rbs ) simultaneously and the user always decodes from the rb with the best sir ( selection combining ) . for icic , we assume under @xmath1-bs coordination , the rbs that the user is assigned are silenced at the next @xmath2 strongest bss . note that both of the schemes create extra load ( reserved rbs ) in the network : icic at the adjacent cells and icd at the serving cell . therefore , it is important to quantify the benefits of icic and icd in order to design efficient systems . the main contribution of this paper is to provide explicit expressions for the coverage probability with @xmath1-bs coordination and @xmath0-rb selection combining . notably , we show that , in the high - reliability regime , where the outage probability goes to zero , the coverage gains due to icic and icd are qualitatively different : icd provides _ order _ gain while icic only offers _ linear _ gain . in contrast , in the high - spectral efficiency regime , where the sir threshold goes to infinity , such order difference does not exist and icic usually offers larger ( linear ) gain than icd in terms of coverage probability . the techniques presented in this paper have the potential to lead to a better understanding of the performance of more complex cooperation schemes in wireless networks , which inevitably involve temporal or spectral correlation . generally speaking , inter - cell interference coordination ( icic ) assigns different time / frequency / spatial dimensions to users from different cells and thus reduces the inter - cell interference . conventional icic schemes are mostly based on the idea of frequency reuse . the resource allocation under cell - centric icic is designed offline and does not depend on the user deployment . while such schemes are advantageous due to their simplicity and small signaling overhead , they are clearly suboptimal since the pre - designed frequency reuse pattern can not cope well with the dynamics of user distribution and channel variation . therefore , there have been significant efforts in facilitating icic schemes , where the interference coordination ( channel assignment ) is based on real user locations and channel conditions and enabled by multi - cell coordination . different user - centric ( coordination - based ) icic schemes in ofdma - based networks are well summarized in the recent survey papers @xcite . conventionally , most of the performance analyses of icic are based on network - level simulation , and the hexagonal - grid model is frequently used @xcite . since real cellular deployments are subject to many practical constraints , recently more and more analyses are based on randomly distributed bss , mostly using the ppp as the model . these stochastic geometry - based models not only provide alternatives to the classic grid models but also come with extra mathematical tractability @xcite . in terms of the treatment of icic , the most relevant papers to this one are @xcite , where the authors analyzed partial frequency reuse schemes using independent thinning . the authors in @xcite considered bs coordination based on clusters grouped by tessellations . different from these papers , this paper focuses on user - centric icic schemes where the spatial correlation of the coordinated cells is explicitly accounted for . it is worth noting that icic is closely related to multi - cell processing ( mcp ) and coordinated multipoint ( comp ) transmission , see @xcite and the references therein . mcp / comp emphasizes the multi - antenna aspects of the cell coordination , while the form of icic considered in this paper does not take into account the use of mimo ( joint transmission ) techniques and thus is not subject to the considerable signaling and processing overheads of typical mcp / comp schemes , which include symbol - level synchronization and joint precoder design @xcite . thus it can be considered as a simple form of mcp / comp that is light on overhead . intra - cell diversity ( icd ) describes the diversity gain achieved by having the serving bs opportunistically assigns users with their best channels . in cellular systems , diversity exists in space , time , frequency and among users@xcite . it is well acknowledged that diversity can significantly boost the network coverage . however , conventional analyses of diversity usually do not include the treatment of interference , _ e.g.,_@xcite . in order to analytically characterize diversity in wireless networks with interference , a careful treatment of interference correlation is necessary , otherwise the results may be misleading . therefore , there have been a few recent efforts in understanding this correlation@xcite . notably , @xcite shows that in an _ ad hoc _ type network , simple retransmission schemes do not result in diversity gain if interference correlation is considered . analyzing the intra - cell diversity ( icd ) under interference correlation , this paper shows that a diversity gain _ can _ be obtained in a cellular setting where the receiver is always connected to the strongest bs , in sharp contrast with the conclusion drawn from _ ad hoc _ type networks in @xcite . the rest of the paper is organized as follows : section [ sec : sysplpspf ] presents the system model and discusses the comparability of icic and icd . sections [ sec : icic ] and [ sec : icd ] derive the coverage probability for the case with icic or icd only , respectively , and provide results on the asymptotic behavior of the coverage probability in the high - reliability as well as high - spectral efficiency regimes . the case with both icic and icd is analyzed in section [ sec : icic&icd ] . we validate our model and discuss fundamental trade - offs between icic and icd in section [ sec : numer ] . the paper is concluded in section [ sec : conclu ] . considering the typical user at the origin @xmath3 , we use a homogeneous poisson point process ( ppp ) @xmath4 with intensity @xmath5 to model the locations of bss on the plane . to each element of the ground process @xmath6 , we add independent marks @xmath7 and @xmath8 , where @xmath9 $ ] and @xmath10,$ ] , to denote the set @xmath11 . ] to denote the ( large - scale ) shadowing and ( power ) fading effect on the link from @xmath12 to @xmath3 at the @xmath13-th resource block ( rb ) , and the combined ( marked ) ppp is denoted as @xmath14 . in particular , under power law path loss , the received power at the typical user @xmath3 at the @xmath13-th rb from a bs at @xmath15 is @xmath16 where @xmath17 is the path loss exponent . in this paper , we focus on rayleigh fading , _ i.e.,_@xmath18 is exponentially distributed with unit mean but allow the shadowing distribution to be ( almost ) arbitrary . [ fig : icicill ] shows a realization of a ppp - modeled cellular network under @xmath1-bs coordination with lognormal shadowing . due to the shadowing effect , the @xmath1 strongest bss under coordination are not necessarily the @xmath1 nearest bss . . the network is under @xmath1-bs ( @xmath19 ) coordination with lognormal shadowing . the typical user is denoted by @xmath20 , the bss by @xmath21 , the serving bs by @xmath22 and the coordinated non - serving bs by @xmath23 . , width=264 ] the base station locations ( ground process @xmath24 ) and the shadowing random variables @xmath25 are static over time and frequency ( i.e. , over all rbs ) , which is the main reason of the spectral / temporal correlation of signal and interference . in comparison , the ( small - scale ) fading @xmath26 is iid over rbs . both @xmath25 and @xmath26 are iid over space ( over @xmath12 ) . the user is assumed to be associated with the strongest ( without fading ) bs and is called _ covered _ ( without icic ) at the @xmath13-th rb iff @xmath27 where @xmath28 is the serving bs . the path loss process with shadowing ( plps ) @xmath29 is the point process on @xmath30 mapped from @xmath31 , where @xmath32 and the indices @xmath33 are introduced such that @xmath34 for all @xmath35 . note that the plps is an _ ordered _ process . it captures the effect of shadowing and spatial node distribution of the network at the same time , and consequently , determines the bs association . the plps @xmath29 is a one - dimensional ppp with intensity measure @xmath36)=\lambda \pi r^\delta { \mathbb{e}}[s^\delta]$ ] , where @xmath37 , @xmath38 and @xmath39 means equality in distribution . [ lem : plpsisppp ] the proof of lemma [ lem : plpsisppp ] is analogous to that of ( * ? ? ? * lemma 1 ) and is omitted from the paper . the intensity measure of the plps demonstrates the necessity of the @xmath40-th moment constraint on the shadowing random variable @xmath25 . without this constraint , the aggregate received power ( with or without fading ) is unbounded almost surely . similar to the construction of @xmath31 , we construct a marked plps @xmath41 , where we put two marks on each element of the plps @xmath29 : @xmath42,\ ; x\in\phi$ ] , are the iid fading random variables directly mapped from @xmath31 ; @xmath43 indicates whether a bs represented by @xmath44 is transmitting at the rb(s ) assigned to the typical user , _ i.e.,_each bs either transmits at all the @xmath0 rbs or does not transmit at any of these rbs . ] . in the case where no ambiguity is introduced , we will use @xmath45 as an abbreviation for @xmath46 and @xmath47 as a short of @xmath48 . for example , if no icic is considered , we have @xmath49 , and the coverage condition in can be written in terms of the marked plps as @xmath50 with icic , the value of @xmath47 is determined by the scheduling policy . given @xmath47 , the coverage condition at the @xmath13-th rb under @xmath1-bs coordination can be expressed in terms of the marked plps as @xmath51 by @xmath1-bs coordination ( icic ) , we assume the @xmath2 strongest non - serving bss of the typical user do not transmit at the rbs to which the user is assigned strongest bss and then reserve the rbs at all of them . ] . thus , we have @xmath52\setminus \{1\}$ ] .. ] for @xmath53 , the exact value of @xmath47 is hard to model since the bss can either transmit to its own users in the rb(s ) assigned to the typical user or reserve these rb(s ) for users in nearby cells , and the muted bss can effectively coordinate " with multiple serving bss at the same time . therefore , the resulting density of the active bss outside the @xmath1 coordinating bss is a complex function of the user distribution , ( joint ) scheduling algorithms and shadowing distribution . in order to maintain tractability , we assume @xmath54 are iid bernoulli random variables with ( transmitting ) probability @xmath55 . such modeling is justified by the random distribution of the users and the shadowing effect @xcite . here , @xmath56 $ ] is called the _ effective load _ of icic . @xmath57 implies all the coordinating bs clusters do not overlap while @xmath58 represents the scenario where all the users assigned to the same rb(s ) in the network share the same @xmath2 muted bss . the actual value of @xmath59 lies between these two extremes while @xmath60 . but this does not have a large influence on the accuracy of the analyses as is shown in detail in section [ sec : numer ] . ] and is determined by the scheduling procedure which this paper does not explicitly study . however , we assume that @xmath59 is known . the accuracy of this model will be validated in section [ sec : numer ] . let @xmath61 be the event of coverage at the @xmath13-th rb . we consider the coverage probability with inter - cell interference coordination ( icic ) and intra - cell diversity ( icd ) formally defined as follows . the coverage probability with @xmath1-bs coordination and @xmath0-rb selection combining is @xmath62 in other words , the typical user is covered iff the received sir at any of the @xmath0 rbs is greater than @xmath63 . the superscript c denotes _ coverage _ and @xmath64 stresses that @xmath65 is the probability of being covered in _ at least one of the m rbs_. ( if there is no possibility of confusion , we will use @xmath65 and @xmath66 interchangeably . ) in the baseline case without icic and icd , each user occupies a single rb at the serving bs . with ( only ) @xmath0-rb selection combining , each user occupies @xmath0 rbs at the serving bs . thus , the system load is increased by a factor of @xmath0 . the load effect of icic can be described by the effective load @xmath59 since , as discussed above , in a network with @xmath1-bs coordination there are @xmath55 of the bss actively serving the users in a single rb whereas each bs serves one user in every rb in the baseline case , _ i.e.,_the load is increased by a factor of @xmath59 due to icic . the fundamental comparability of icic and icd comes from the similarity in introducing extra load in the system and will be explored in more detail in section [ sec : numer ] . this section focuses on the effect of icic on the coverage probability . since no icd is considered , we will omit the superscript @xmath13 on the fading random variable @xmath67 , for simplicity . our analysis will be relying on a statistical property of the marked plps @xmath68 stated in the following lemma . for @xmath69 , let @xmath70 and @xmath71 , where @xmath72 . for all @xmath73 , the two random variables @xmath74 and @xmath75 are independent . [ lem:2worlds ] if @xmath76 , the lemma is trivially true , since @xmath77 while @xmath78 has some non - degenerate distribution . for @xmath79 , @xmath80 $ ] and @xmath81 , the joint ccdf of @xmath82 and @xmath83 can be expressed as @xmath84 \\ & \peq{a } { \mathbb{e}}_{\xi_k}\left[{\mathbb{p}}\left(\frac{\xi_1}{\xi_k } > x_1\right){\mathbb{p}}\left(\frac{\xi_k^{-1}}{i_k } > x_2\right ) \mid \xi_k\right ] \\ & \peq{b } { \mathbb{p}}\left(\frac{\xi_1}{\xi_k } > x_1\right ) { \mathbb{e}}_{\xi_k}\left[{\mathbb{p}}\left(\frac{\xi_k^{-1}}{i_k } > x_2\right ) \mid \xi_k\right ] \\ & = { \mathbb{p}}\left(\frac{\xi_1}{\xi_k } > x_1\right){\mathbb{p}}\left(\frac{\xi_k^{-1}}{i_k } > x_2\right ) , \end{aligned}\ ] ] where ( a ) is due to the fact that @xmath85 and @xmath86 are conditionally independent given @xmath87 by the poisson property and @xmath88 , @xmath89 are iid and independent from @xmath29 . ( b ) holds since , conditioning on @xmath87 implies that there are @xmath90 points on @xmath91 . thus , thanks to the poisson property , it can be shown that given @xmath87 , @xmath82 follows the same distribution as that of the minimum of @xmath90 iid random variables with cdf @xmath92 . of intensity measure @xmath93 , given there are @xmath94 points on @xmath95 the joint distribution of the locations of the @xmath94 points is the same as that of @xmath94 iid random variables with cdf @xmath96 ( * ? ? ? * theorem 2.25 ) . ] since the resulting conditional distribution of @xmath82 does not depend on @xmath87 , this distribution is also the marginal distribution of @xmath82 as is stated in the lemma . furthermore , due to lemma [ lem : plpsisppp ] , it is straightforward to obtain the ccdf of @xmath97 , which is formalized in the following lemma . for all @xmath98 , the ccdf of @xmath82 is @xmath99.\ ] ] [ lem : xi1xik ] as discussed in the proof of lemma [ lem:2worlds ] , by the poisson property and the intensity measure of @xmath29 given in lemma [ lem : plpsisppp ] , conditioned on @xmath100 , @xmath101 , where @xmath102 denotes the @xmath103-th order statistics of @xmath90 iid random variables with cdf @xmath104 . thus , @xmath105 is the probability that all the @xmath90 iid random variables are larger than @xmath12 . for @xmath106 , let @xmath107 for @xmath108 . the laplace transform of @xmath109 is @xmath110 c(s ) \rho^\delta \right),\ ] ] where @xmath111 and @xmath112 is the gauss hypergeometric function . [ lem : ltildeirho ] first , we can calculate the laplace transform of @xmath113 using the probability generating functional ( pgfl ) of ppp @xcite , _ i.e.,_@xmath114 \\ & = { \mathbb{e}}_{\xi}\prod_{\xi\in \xi \cap(\rho,\infty ) } { \mathbb{e}}_{h,\chi}[\exp(-s\chi h\xi^{-1 } ) ] \\ & = \exp\left ( - { \mathbb{e}}_{h,\chi } \left[\int_\rho^\infty ( 1-e^{-s\chi h / x } ) \lambda({\textnormal{d}}x)\right ] \right),\end{aligned}\ ] ] where @xmath115 is a bernoulli random variable with mean @xmath116 , @xmath93 is the intensity measure of @xmath29 and by lemma [ lem : plpsisppp ] , @xmath117\delta x^{\delta-1 } { \textnormal{d}}x$ ] . then , straightforward algebraic manipulation yields @xmath118{\mathbb{e}}_h\left [ ( sh)^\delta \gamma(1-\delta,\frac{sh}{\rho } ) - \rho^\delta(1-e^{-\frac{sh}{\rho } } ) \right ] \right).\end{gathered}\ ] ] since , for an arbitrary random variable @xmath119 and constant @xmath120 , @xmath121 , we have @xmath122 c(s ) \rho^\delta \right),\ ] ] where @xmath123 $ ] . the proof is completed by considering the exponential distribution of @xmath124 . here , @xmath113 can be understood as the interference from bss having a ( non - fading ) received power weaker than @xmath125 . in the case without shadowing , _ i.e.,_@xmath126 , it can also be understood as the interference coming from outside a disk centered at the typical user with radius @xmath127 . the laplace transform of @xmath128 is @xmath129 where @xmath130 . [ lem : xiktildiklaplace ] first , the pdf of @xmath87 can be derived analogously to the derivation of ( * ? ? ? * lemma 3 ) as @xmath131)^k\frac{\delta x^{k\delta-1}}{\gamma(k)}\exp\left(-\lambda\pi{\mathbb{e}}[s^\delta ] x^\delta\right).\ ] ] then , thanks to lemma [ lem : ltildeirho ] , the laplace transform of @xmath132 can be obtained by deconditioning @xmath133 ( given @xmath134 ) over the distribution of @xmath87 . this leads to @xmath135 where @xmath136 . note that although the path loss exponent @xmath17 is not explicitly taken as a parameter of @xmath137 , @xmath137 depends on @xmath17 by definition . thus , the value of @xmath17 affects all the results . since we consider rayleigh fading , the coverage probability without icic is just the laplace transform of @xmath138 . the special case of @xmath139 of lemma [ lem : xiktildiklaplace ] corresponds to the well - known coverage probability in cellular networks ( under the ppp model ) without icic or icd @xcite , @xmath140 note that since we consider the full load case and @xmath56 $ ] , @xmath60 implies @xmath141 . in the more general case @xmath142 , @xmath59 depends on the user distribution and the scheduling policy and thus is hard to determine . however , treating @xmath59 as a parameter we obtain the following theorem addressing the case with non - trivial coordination . the coverage probability for the typical user under @xmath1-cell coordination ( @xmath142 ) is @xmath143 where @xmath130.[thm : kcoord ] the coverage probability can be written in terms of the plps as @xmath144 where @xmath145 is exponentially distributed with mean 1 , and thus @xmath146 . since @xmath147 and @xmath148 are statistically independent ( lemma [ lem:2worlds ] ) , we can calculate the coverage probability by @xmath149 where @xmath150 is the cdf of @xmath151 given by lemma [ lem : xi1xik ] . the theorem is thus proved by change of variables . the finite integral in ( [ equ : kcoor - integral ] ) can be straightforwardly evaluated numerically . the gauss hypergeometric function can be cumbersome to evaluate numerically , especially when embedded in an integral , as in . alternatively , @xmath152 can be expressed as @xmath153 and @xmath154 . here , @xmath155 , @xmath156 $ ] , is the _ upper _ incomplete beta function , which can be calculated much more efficiently in many cases . in matlab , the speed - up compared with the hypergeometric function is at least a factor of 30 . a useful trick in the numerical evaluation of the integral in is to exploit the alternative representation of @xmath153 where @xmath157 $ ] is the _ upper _ incomplete beta function ( note @xmath154 ) . although widely used in the literature and ( in this case ) being a more concise representation of @xmath158 , the gauss hypergeometric function is typically quite cumbersome to evaluate numerically especially when embedded in a numerical integral . the fact that @xmath158 can be expressed in terms of the incomplete beta function allows potentially much faster numerical evaluation . in particular , when evaluated in matlab , the beta function - based expression runs at least 30 times faster than the equivalent hypergeometric function - based expression . [ fig : kcoordcoverage ] demonstrates the effect of icic on the coverage probability for @xmath159 and @xmath57 . the former case may be interpreted as a lower bound and the latter case an upper bound . as expected , the larger @xmath1 , the higher the coverage probability for all @xmath63 . on the other hand , the marginal gain of cell coordination decreases with increasing @xmath1 since the interference , if any , from far away bss is attenuated by the long link distance and affects the sir less . [ fig : kcoordcoverage ] also shows that larger @xmath59 results in larger coordination gain in terms of sir . this is due to the fact that coordination not only mutes the strongest @xmath2 interferers but also thins the interfering bss outside the coordinating cluster . however , this does not mean that the system will be better - off by implementing a larger @xmath59 . instead , from the load perspective , the ( sir ) gain is accompanied with the loss in bandwidth ( increased load ) since fewer bss are actively serving users . the sir - load trade - off will be further discussed along with the model validation in section [ sec : numer ] . [ c][t]@xmath160 [ c][c]@xmath161 [ c][c]@xmath159 kcoor1to5_kappa_eq_1k.eps ( 53,38)(4,3)30 ( 25,33)@xmath162 while the finite integral expression given in theorem [ thm : kcoord ] is easy to evaluate numerically , it is also desirable to find a simpler estimate that lends itself to a more direct interpretation of the benefit of icic . this subsection investigates the asymptotic behavior of icic when @xmath163 . note that @xmath164 refers to the high - reliability regime since in this limit the typical user is covered almost surely . in practice , the high - reliability regime ( @xmath164 ) is usually where the control channels operate . in the lte system ( narrowband ) , the lowest mcs mode for downlink transmission supports an sinr about -7 db and thus may also be suitable for the high - reliability analysis . in wide - band systems ( _ e.g.,_cdma , uwb ) , the system is more robust against interference and noise . thus , @xmath63 is much smaller and the high - reliability analysis is more applicable . let @xmath165 be the outage probability of the typical user for @xmath166 . then , @xmath167 where @xmath168 and @xmath169 is the ( pochhammer ) rising factorial . [ prop : pok1theta->0 ] see appendix [ app : pok1theta->0 ] . proposition [ prop : pok1theta->0 ] shows that for pure icic schemes , the number of coordinating bss only linearly affects the outage probability in the high - reliability regime . however , depending on the value of @xmath63 , even the linear effect may be significant . in fig . [ fig : akalpha ] , we plot the coefficient @xmath170 for @xmath171 as a function of the path loss exponent @xmath17 , assuming @xmath159 . the difference ( in ratio ) between @xmath170 for different @xmath1 indicates the usefulness of icic , and this figure shows that icic is more useful when the path loss exponent @xmath17 is large . this is consistent with intuition , since the smaller the path loss exponent , the more the interference depends on the far - away interferers and thus the less useful the local interference coordination is . for other values of @xmath59 , the same trend can be observed . [ c][t]@xmath170 akalpha.eps ( 30,25)@xmath19 ( 50,49)@xmath60 ( 53,47)(-3,-4)12 the other asymptotic regime is when @xmath172 . in this regime , the coverage probability goes to zero while the spectral efficiency goes to infinity . thus , it is of interest to study how the coverage probability decays with @xmath63 . the coverage probability of the typical user for @xmath1-bs ( @xmath142 ) coordination satisfies @xmath173 where @xmath174 . [ prop : pck1theta->inf ] we prove the proposition by studying the asymptotic behavior of @xmath175 . using theorem [ thm : kcoord ] and a change of variable , we have @xmath176 considering the sequence of functions ( indexed by @xmath63 ) @xmath177 we have @xmath178 , and @xmath179 converges to @xmath180 as @xmath181 . therefore , @xmath182 where ( b ) is due to the monotone convergence theorem . further , since @xmath183 and @xmath184 , we have @xmath185 . this allows replacing the inequality ( a ) with equality and completes the proof . proposition [ prop : pck1theta->inf ] shows , just like in the high - reliability regime , that @xmath186 is not affected by the particular choice of @xmath1 and @xmath59 . since @xmath187 , the coverage probability decays faster when @xmath17 is smaller in the high spectral efficiency regime , consistent with intuition . icic creates additional load to the neighboring cells by reserving the rbs at the coordinated bss . the extra load improves the coverage probability as it reduces the inter - cell interference . in contrast , with selection combining ( sc ) , the serving bs transmits to the typical user at @xmath0 rbs simultaneously , and the user is covered if the maximum sir ( over the @xmath0 rbs ) exceeds @xmath63 . like icic , sc can also improve the network coverage at the cost of introducing extra load to the bss . different from icic , sc takes advantage of the intra - cell diversity ( icd ) by reserving rbs at the serving cell . this section provides a baseline analysis on the coverage with icd ( but without icic ) . the joint success probability of transmission over @xmath0 rbs ( without icic ) is @xmath188 [ thm : jointsuccess1 ] since the proof of theorem [ thm : jointsuccess1 ] is a degenerate version of that of a more general result stated in theorem [ thm : jointsuccess ] , we defer the discussion of the proof to section [ sec : icic&icd ] . a similar result was obtained in @xcite where a slightly different framework was used and the shadowing effect not explicitly modeled . due to the inclusion - exclusion principle , we have the coverage probability with selection combining over @xmath0 rbs : the coverage probability over @xmath0 rbs without bs - coordination is @xmath189 where @xmath190 is given by theorem [ thm : jointsuccess1 ] . [ cor : puc1 m ] fig . [ fig : icd_m1to5 ] compares the coverage probability under @xmath0-rb selection combining , @xmath191 for @xmath192 . as expected , the more rbs assigned to the users , the higher the coverage probability . also , similar to the icic case , the marginal gain in coverage probability due to icd diminishes with @xmath0 . however , comparing figs . [ fig : icd_m1to5 ] and [ fig : kcoordcoverage ] , we can already observe dramatic difference : with the same overhead , the coverage gain of icd looks more evident than that of icic in the high - reliability regime , _ i.e.,_when @xmath163 . this observation will be formalized in the following subsection . [ c][t]@xmath191 icd_m1to5.eps ( 43,48)(4,3)20 ( 30,45)@xmath193 ( 65,65)@xmath194 let @xmath195 be the outage probability of the typical user under @xmath0-rb selection combining . we have @xmath196 where @xmath197 and @xmath198 is the confluent hypergeometric function of the first kind . [ prop : puo1mtheta->0 ] the proof of proposition [ prop : puo1mtheta->0 ] can be found in appendix [ app : puo1mtheta->0 ] . although proposition [ prop : puo1mtheta->0 ] provides a neat expression for the constant in front of @xmath199 in the expansion of the outage probability , numerically evaluating the @xmath0-th derivative of the reciprocal of confluent hypergeometric function may not be straightforward . a relatively simple approach is to resort to fa di bruno s formula . alternatively , one can directly consider and simplify it by introducing the bell polynomial @xcite : @xmath200 where @xmath201 and @xmath202 which can be efficiently evaluated numerically was mistaken as @xmath203 . the typo this corrected in this manuscript . ] . to better understand proposition [ prop : puo1mtheta->0 ] we introduce the following definition of the diversity gain in interference - limited networks , which is consistent with the diversity gain defined in @xcite and is analogous to the conventional diversity defined ( only ) for interference - less cases , see _ e.g.,_@xcite . the diversity ( order ) gain , or simply diversity , of interference - limited networks is @xmath204 clearly , proposition [ prop : puo1mtheta->0 ] shows that a diversity gain can be obtained by selection combining in sharp contrast with the results presented in @xcite , where the authors show that there is no such gain in retransmission . the reason of the difference lies in the different association assumptions . @xcite considers the case where the desired transmitter is at a fixed distance to the receiver which is independent from the interferer distribution . however , this paper assumes that the user is associated with the strongest bs ( on average ) . in other words , the signal strength from the desired transmitter and the interference are correlated . proposition [ prop : puo1mtheta->0 ] together with @xcite demonstrates that this correlation is critical in terms of the time / spectral diversity . [ c][t]@xmath205 icd_m1to5asymp.eps ( 43,64)(3,-4)22 ( 35,66)@xmath193 ( 65,30)@xmath194 propositions [ prop : pok1theta->0 ] and [ prop : puo1mtheta->0 ] quantitatively explain the visual contrast between figs . [ fig : kcoordcoverage ] and [ fig : icd_m1to5 ] in the high - reliability regime ( @xmath164 ) . while icic reduces the interference by muting nearby interferers , the number of coordinated bss only affects the outage probability by the coefficient and does not change the fact that @xmath206 as @xmath164 . in contrast , icd affects the outage probability by both the coefficient and the exponent . [ fig : icd_m1to5asymp ] compares the asymptotic approximation , _ i.e.,_@xmath207 , with the exact expression provided in corollary [ cor : puc1 m ] . a reasonably accurate match can be found for small @xmath63 , and the range where the approximation is accurate is larger when @xmath0 is smaller . thus , despite the fact that the main purpose of proposition [ prop : puo1mtheta->0 ] was to indicate the qualitative behavior of icd , the analytical tractability of @xmath208 also provides useful approximations in applications with small coding rate , _ e.g.,_spread spectrum / uwb communication , node discovery , _ etc_. for completeness , we also consider the high spectral efficiency regime where @xmath172 . the coverage probability of the typical user under @xmath0-rb selection combining satisfies @xmath209 where @xmath210 . [ prop : pc1mtheta->inf ] we proceed by ( first ) considering @xmath211 . by theorem [ thm : jointsuccess1 ] , we have @xmath212 where ( a ) comes from ( * ? ? ? since @xmath213 , we have @xmath214 , which leads to the proposition thanks to corollary [ cor : puc1 m ] . comparing propositions [ prop : pck1theta->inf ] and [ prop : pc1mtheta->inf ] , we see that unlike the high - reliability regime , the coverage probabilities of icic and icd do _ not _ have order difference in the high spectral efficiency regime . however , the difference in coefficients ( @xmath215 and @xmath216 ) can also incur significant difference in the coverage probability . [ fig : bkbm_kappa1 ] compares the coefficients for different path loss exponent @xmath17 assuming @xmath58 . note that @xmath58 corresponds to the smallest possible @xmath215 . yet , even so , @xmath215 still dominates @xmath216 for most realistic @xmath17 . this implies that icic is often more effective than icd in the high spectral efficiency regime . [ c][c]@xmath215 and @xmath216 [ c][c]@xmath215 [ c][c]@xmath216 [ c][c]@xmath17 bkbm_kappa1.eps ( 60,15)@xmath217 ( 60,20)(-1,2)15 sections [ sec : icic ] and [ sec : icd ] provided the coverage analysis in cellular networks with icic and icd separately . this section considers the scenario where the network takes advantage of icic and icd at the same time . in particular , we will evaluate the coverage probability @xmath65 when the typical user is assigned with @xmath0 rbs with independent fading at the serving bs and all the @xmath0 rbs are also reserved at the @xmath2 strongest non - serving bss . in order to derive the coverage probability , we first generalize lemma [ lem : xiktildiklaplace ] beyond rayleigh fading . in particular , for a generic fading random variable @xmath218 , we introduce the following definition . for a plps @xmath219 , let @xmath220 be the interference from the bss weaker ( without fading ) than the @xmath221-th strongest bs , _ i.e.,_@xmath222 , where @xmath223 are iid . similarly , we define @xmath224 to be the interference from bss with average ( over fading ) received power less than @xmath125 . then , we obtain a more general version of lemma [ lem : xiktildiklaplace ] as follows . for general fading random variables @xmath225 and @xmath226<\infty$ ] , the laplace transform of @xmath227 is @xmath228\right)^k}. \label{equ : xiktildiklaplacegeneral}\ ] ] [ lem : xiktildiklaplacegeneral ] the proof of the lemma follows exactly that of lemmas [ lem : ltildeirho ] and [ lem : xiktildiklaplace ] . the only difference is that we do not factor in the distribution of the fading random variable @xmath218 . more precisely , we can first show @xmath229\rho^\delta{\mathbb{e}}_h[e^{-sh}+s^\delta h^\delta \gamma(1-\delta , sh)-1]\right).\end{gathered}\ ] ] then , integrating @xmath134 over the distribution of @xmath87 gives the desired result . note that the condition @xmath226<\infty$ ] in lemma [ lem : xiktildiklaplacegeneral ] is sufficient ( but not necessary ) to guarantee the existence of the laplace transform . as will become clear shortly , for the purpose of this section , the most important case of @xmath218 is when @xmath218 is a gamma random variable with pdf @xmath230 , where @xmath231 . for this case , we have the following lemma . for @xmath232 , if @xmath218 is a gamma random variable with pdf @xmath230 , @xmath233 [ lem : lxikikm ] almost trivially based on lemma [ lem : xiktildiklaplacegeneral ] , lemma [ lem : lxikikm ] helps to show the following theorem . for all @xmath10 and @xmath142 , the joint coverage probability over @xmath0-rbs under @xmath1-cell coordination is @xmath234 [ thm : jointsuccess ] let @xmath235 be the fading coefficient from the @xmath103-th strongest ( on average ) bs at rb @xmath13 for @xmath236 $ ] . by definition , we have @xmath237\right ) \label{equ : ptckdef}\ ] ] due to the conditional independence ( given @xmath29 ) across @xmath13 , can be further simplified as @xmath238 where the inner expectation in is taken over @xmath235 for @xmath236 $ ] and @xmath33 , and due to the independence ( across @xmath13 and @xmath103 ) and ( exponential ) distribution of @xmath235 , @xmath239 are iid gamma distributed with pdf @xmath240 . further , writing @xmath241 as @xmath242 and letting @xmath243 , we obtain the following expression by taking advantage of the statistical independence shown in lemma [ lem : plpsisppp ] : @xmath244 where @xmath245 is given in lemma [ lem : lxikikm ] . the proof is completed by plugging in lemma [ lem : xi1xik ] . note that although theorem [ thm : jointsuccess ] does not explicitly address the case @xmath60 , the same proof technique applies to this ( easier ) case , where the treatment of the random variable @xmath151 is unnecessary since it has a degenerate distribution ( @xmath246 ) . thus , the proof of theorem [ thm : jointsuccess1 ] is evident and omitted from the paper . due to the inclusion - exclusion principle , we immediately obtain the following corollary . the coverage probability over @xmath0 rbs with @xmath1 bs - coordination is @xmath247 where @xmath248 is given by theorem [ thm : jointsuccess ] . [ c][t]@xmath249 m1m2.eps ( 28,53)(1,-2)7 ( 13,56)@xmath162 ( 40,40)(2,-3)10 ( 45,18)@xmath162 let @xmath250 be the outage probability of the typical user under @xmath0-rb selection combining and @xmath1-bs coordination . we have @xmath251 where @xmath252 . [ prop : puokmtheta->0 ] proposition [ prop : puokmtheta->0 ] combines proposition [ prop : pok1theta->0 ] and [ prop : puo1mtheta->0 ] . its proof is analogous to that of proposition [ prop : puo1mtheta->0 ] ( but more tedious ) and is thus omitted from the paper . proposition [ prop : puokmtheta->0 ] gives quantitative evidence on why a pure icd scheme maximizes the coverage probability in the high - reliability regime . in fig . [ fig : t1t2 ] , we plot the outage probability for difference number of coordinated cell @xmath162 and in - cell diversity @xmath253 assuming @xmath141 and observe the consistency with proposition [ prop : puokmtheta->0 ] . in section [ sec : sysplpspf ] , we introduced the effective load @xmath59 and modeled the impact of the out - of - cluster coordination on the interference by independent thinning of the interferer field with retaining probability @xmath55 . although , remarkably , @xmath59 disappears when considering the diversity order of the network , it is still of interest to evaluate the accuracy of such modeling in the non - asymptotic regime . to this end , we set up the following icic simulation to validate the effective load model . we consider the users are distributed as a homogeneous ppp @xmath254 with density @xmath255 independent from the bs process @xmath24 . we assume a single channel and a random scheduling policy where we pick every user exactly once at a random order . a picked user is scheduled iff its strongest bss are not ( already ) serving another user or coordinating ( _ i.e.,_being muted ) with user(s ) in other cell(s ) and its second to @xmath1-th strongest bss are not transmitting ( serving other users ) . thus , after the scheduling phase , there are at most @xmath256 users scheduled where @xmath257 is the simulation region since there are at most @xmath256 serving bss . in reality , the number of scheduled users is often much less than @xmath256 since 1 ) there is always a positive probability that there are empty cells due to the randomness in bs and user locations ; 2 ) when @xmath142 some bss are muted due to coordination . the ratio between the number of bss @xmath256 and the number of scheduled users ( which equals the number of serving bss ) is consistent with the definition of the effective load @xmath59 and thus is a natural estimate . under lognormal shadowing with standard deviation @xmath258 , we empirically measured @xmath59 as in table [ tab : kappa ] . it is observed that our simulation results in the estimates @xmath259 that can be well approximated by an affine function of @xmath1 and the function depends on the shadowing variance . the fact that more severe shadowing results in smaller @xmath59 can be explained in the case @xmath60 . in this case , the only reason that @xmath260 is the existence of empty cells and the larger @xmath59 is the more empty cells there are . independent shadowing reduces the spatial correlation of the sizes of nearby poisson voronoi cells and thus naturally reduces the variance of the number of users in each cell , resulting in a smaller number of empty cells . .estimated @xmath59 [ cols="^,^,^,^,^,^",options="header " , ] [ tab : kappa ] fig . [ fig : kcoord_sim ] compares the coverage probability under @xmath1-bs coordination predicted by theorem [ thm : kcoord ] using the estimated @xmath59 from table [ tab : kappa ] with the simulation results . we picked the case where @xmath261 db since this is the _ worst case _ in terms of matching analytical results with the simulation due to the size correlation of poisson voronoi cells . to see this more clearly , consider the case @xmath60 , where the simulation and analysis match almost completely in the figure . the match is expected but not entirely trivial since the process of transmitting bss is no longer a ppp . more specifically , a bs is transmitting iff there is at least one user in its voronoi cell , _ i.e.,_the ground process @xmath24 is thinned by the user process @xmath254 . however , the thinning events are spatially dependent due to the dependence in the sizes of the voronoi cells . as a result , clustered bss are less likely to be serving users at the same time and thus the resulting transmitting bs process is more _ regular _ than a ppp . yet , fig . [ fig : kcoord_sim ] shows that the deviation from a ppp is small when @xmath262 . is still quite good for smaller user densities , say @xmath263 . ] shadowing breaks the spatial dependence of the interfering field ( transmitting bss ) and consequently improves the accuracy of the analysis . when @xmath264 db , the difference between the simulated coverage probability and the one predicted in theorem [ thm : kcoord ] are almost visually indistinguishable ( fig . [ fig : kcoord_sim_sigmav10 ] ) . these results validate the effective load model for analyzing icic in the non - asymptotic regime . [ c][t]@xmath160 kcoord_sim_sigmav0.eps ( 15,45)@xmath162 ( 50,47)(4,1)30 [ fig : kcoord_sim ] [ c][t]@xmath160 kcoord_sim_sigmav10.eps ( 15,45)@xmath162 ( 50,47)(4,1)30 [ fig : kcoord_sim_sigmav10 ] the analyses in sections [ sec : icic ] , [ sec : icd ] and [ sec : icic&icd ] shows the significantly different behavior of icic and icd schemes despite the fact that both the schemes improves the coverage probability through generating extra load in the system . in particular , propositions [ prop : pok1theta->0 ] , [ prop : puo1mtheta->0 ] and [ prop : puokmtheta->0 ] show that when @xmath164 , icd will have a larger impact on the coverage probability due to the diversity gain . in contrast , propositions [ prop : pck1theta->inf ] and [ prop : pc1mtheta->inf ] suggest that when @xmath181 , icic will be more effective since the linear gain is typically larger ( fig . [ fig : bkbm_kappa1 ] ) . intuitively , a icic - icd combined scheme should present a trade - off between the performance in these two regimes . to make a fair comparison between different icic - icd combined schemes , we need to control their load on the system in terms of rbs used . by the construction of the model , we observe that the load introduced by icic is the effective load @xmath59 times the load without icic since @xmath55 is the fraction of active transmitting bss , which , in the single - channel case , is proportional to the number ( or , density ) of users being served . similarly , the load introduced by icd is @xmath0 times the load without icd since @xmath0 rbs are grouped to serve a single user ( while without icd they could be used to serve @xmath0 users ) . thus , under both @xmath1-bs coordination and @xmath0-rb selection combining , the system load is proportional to @xmath265 which we term _ icic - icd load factor_. fig . [ fig : icicicdsimana_dbl ] plots the coverage probability of three icic - icd combined schemes with different but similar icic - icd load factor @xmath265 using both the analytical result and simulation . as is shown in the figure , a hybrid icic - icd scheme ( _ i.e.,_with @xmath266 ) provides a trade - off between the good performance of icic and icd in the two asymptotic regimes . in general , a hybrid scheme could provide the highest coverage probability for intermediate @xmath63 , and the crossing point depends on all the system parameters . [ c][t]@xmath65 -bs coordination and @xmath0-rb selection combining @xmath65 with different combinations @xmath267 . here , @xmath268 , @xmath269 db . the left figure shows the part for @xmath270 $ ] and the right figure for @xmath271 $ ] . , title="fig : " ] another more fundamental trade - off is between the load and the icic - icd combined schemes . in other words , how to find the optimal combination @xmath267 that takes the load into account . while the complexity of this problem prohibits a detailed exploration in this paper , we give a simple example to explain the trade - off . assume all the users in the network are transmitting at the same rate @xmath272 and the network employs the random scheduling procedure as described in section [ subsec : effectiveload ] . then the ( average ) throughput of the typical scheduled user is @xmath273 in the interference - limited network . under the icic and icd schemes , the number of user being served per rb is ( on average ) @xmath274 times those who can be served in the baseline case without icic and icd . therefore , for fixed @xmath63 , the spatially averaged ( per user ) throughput is proportional to @xmath275 . intuitively , it is the product of the probability of a random chosen user being scheduled ( @xmath276 ) and the probability of successful transmission ( @xmath65 ) . then we can find optimal combination @xmath277 using exhaustive search . alternatively , we can enforce an outage constraint and find the optimal @xmath267 combination such that @xmath278}(\mathsf{p}^{\cup \textnormal{c}}_{k , m } ) \label{equ : kstarmstaroutage}\ ] ] fig . [ fig : kstarmstar ] plots the exhaustive search result for @xmath279 defined in and . in the simulation , we limit our search space for both @xmath1 and @xmath0 to @xmath280 and we use the affine function @xmath281 to approximate @xmath59 , which turns out to be an accurate fit in our simulation ( see the data in table [ tab : kappa ] ) . fig . [ fig : kstarmstar ] shows that as @xmath63 increases , it is beneficial to increase @xmath1 . this is consistent with the result derived in propositions [ prop : pck1theta->inf ] and [ prop : pc1mtheta->inf ] and fig . [ fig : bkbm_kappa1 ] , which show that icic is more effective in improving coverage probability for large @xmath63 . if there is no outage constraint , it is more desirable to keep both @xmath1 and @xmath0 ( and thus the load factor ) small . this is true especially for small @xmath63 since the impact of @xmath267 on @xmath65 is small ( @xmath282 ) but both @xmath59 ( a function of @xmath1 ) and @xmath0 linearly affect the load factor and thus the average throughput . the incentive to increase @xmath267 is higher if an outage constraint is imposed . although it is still more desirable to increase @xmath1 ( both due to its usefulness in the high - spectral efficiency regime and its smaller impact on the load factor ) , the increase in @xmath0 also has a non - trivial impact : a slight increase in @xmath0 could significantly reduce the optimal value of @xmath1 . this is an observation of practical importance , since the cost of increasing @xmath1 is usually much higher than that of increasing @xmath0 due to the signaling overhead that icic requires . [ c][t]@xmath65 as a function of @xmath63 . the top subfigure is optimized for average throughput , see . the bottom subfigure is optimized for average throughput under an outage constraint , see with @xmath283 . @xmath284 db shadowing is considered . , title="fig : " ] this paper provides explicit expressions for the coverage probability of inter - cell interference coordination ( icic ) and intra - cell diversity ( icd ) in cellular networks modeled by a homogeneous poisson point process ( ppp ) . examining the high - reliability regime , we demonstrate a drastically different behavior of icic and icd despite their similarity in creating extra load in the network . in particular , icd , under the form of selection combining ( sc ) , provides diversity gain while icic can only linearly affect the outage probability in the high - reliability regime . in contrast , in the high - spectral efficiency regime , icic provides higher coverage probability for realistic path loss exponents . all the analytical results derived in the paper are invariant to the network density and the shadowing distribution . the fact that icd under selection combining provides diversity gain in cellular networks even with temporal / spectral interference correlation contrasts with the corresponding results in _ ad hoc _ networks , where @xcite shows no such gain exists . this shows that the spatial dependence between the desired transmitter and the interferers is critical in harnessing the diversity gain . in the non - asymptotic regime , we propose an effective load model to analyze the effect of icic . the model is validated with simulations and proven to be very accurate . using these analytical results , we explore the fundamental trade - off between icic - icd and system load in cellular systems . we prove the theorem by calculating @xmath285 . first , consider the case where @xmath60 . since @xmath165 , we have @xmath286 where @xmath287 , and the last equality is due to lhospital s rule and the fact that @xmath288 . moreover , we have @xmath289 due to the series expansion of the gauss hypergeometric function @xmath290 . thus , we proved is true for @xmath60 . for @xmath291 , by theorem [ thm : kcoord ] , we have @xmath292 where the integral , by change of variable @xmath293 , can be written as @xmath294 where @xmath295 . therefore , we have @xmath296 where the rhs can be simplified by ( repetitively ) applying lhospital s rule as follows : @xmath297 where @xmath298 note that @xmath299 and thus @xmath300 , which by lhospital s rule can be further simplified as @xmath301 . therefore , thanks to , we have @xmath302 combining and completes the proof . in order the prove proposition [ prop : puo1mtheta->0 ] , we first introduce two useful lemmas . letting @xmath303 , the following lemma states a simple algebraic fact which will turn out to be useful in the asymptotic analysis . we prove the lemma by induction . first , consider the case where @xmath315 . then for all @xmath316 , we have @xmath317 where the @xmath318-th derivative can be expanded by the leibniz rule , _ i.e.,_@xmath319 which is @xmath320 when @xmath321 if @xmath322 . when @xmath323 , the only non - zero term in the sum is the one with @xmath324 and thus is @xmath325 . therefore , the lemma is true for @xmath315 . second , we prove it for the case @xmath326 with general @xmath311 by induction . assume @xmath327 } } = 0 , \end{gathered}\ ] ] for all @xmath328 nonnegative integers @xmath329 with @xmath330 . then , we consider the case for @xmath311 , move all the @xmath331 to the front and , analogous to the @xmath315 case , obtain @xmath332}}. \label{equ : nderi}\end{gathered}\ ] ] expanding only the @xmath333-th order derivative using lemma [ lem : dkuc1um ] , we can express as @xmath334 where @xmath335 is independent from @xmath336 and @xmath337 for all @xmath338 $ ] . we then can move the derivative operators inside the summation . further , since @xmath339 , we have @xmath340 for all @xmath341 in the summation , which leads to the observation that @xmath342 } } = 0,\ ; \forall j\in[k_n]\ ] ] by our assumption on the @xmath328 case . thus the lemma is proved for the case @xmath326 for all @xmath343 . for the case @xmath344 , we see , by the first part of the proof , that @xmath342}}\ ] ] can be non - zero only if @xmath345 . thus can be simplified to @xmath346}},\ ] ] which is @xmath347 if we assume the lemma is true for @xmath328 . since @xmath311 is arbitrarily chosen , the lemma is proved for all @xmath343 . by corollary [ cor : puc1 m ] , we have @xmath348 we then proceed the proof by considering the taylor expansion of @xmath349 at @xmath350 . to find the @xmath311-th derivative of @xmath349 we treat @xmath351 as a composite of @xmath352 and @xmath353 , where the derivatives of @xmath353 is available by the series expansion of hypergeometric function mentioned before . then , by fa di bruno s formula @xcite , we have @xmath354 where @xmath355 is the set of @xmath311-tuples of non - negative integers @xmath356 with @xmath357 , and @xmath358 . directly leads to the taylor expansion of @xmath359 , which combined with leads to a series expansion of @xmath205 as function of @xmath63 , @xmath360 where @xmath361 . rearranging the sums and products in expression above yields @xmath362 , where @xmath363 recall that @xmath364 indicates @xmath357 . by lemma [ lem : summmmki ] , we have @xmath365 for all @xmath366 , _ i.e.,_@xmath367 as @xmath163 . further , lemma [ lem : summmmki ] helps us to obtain the coefficient in front of @xmath199 , _ i.e.,_@xmath368 which leads to the concise expression in the proposition by reusing fa di bruno s formula . m. haenggi , `` second - order properties of wireless networks : correlation effects in space and time , '' keynote lecture at the 2013 international workshop on spatial stochastic models for wireless networks ( spaswin13 ) , tsukuba science city , japan . slides available at http://www3.nd.edu/~mhaenggi/talks/spaswin13.pdf . g. fodor , c. koutsimanis , a. rcz , n. reider , a. simonsson , and w. mller , `` intercell interference coordination in ofdma networks and in the 3gpp long term evolution system . '' _ journal of communications _ , vol . 4 , no . 7 , 445453 , aug . a. hamza , s. khalifa , h. hamza , and k. elsayed , `` a survey on inter - cell interference coordination techniques in ofdma - based cellular networks , '' _ ieee communications surveys & tutorials _ , vol . 15 , no . 4 , pp . 16421670 , mar . 2013 . c. kosta , b. hunt , a. quddus , and r. tafazolli , `` on interference avoidance through inter - cell interference coordination ( icic ) based on ofdma mobile systems , '' _ ieee communications surveys & tutorials _ , vol . 15 , no . 3 , pp . 973995 , mar . h. elsawy , e. hossain , and m. haenggi , `` stochastic geometry for modeling , analysis , and design of multi - tier and cognitive cellular wireless networks : a survey , '' _ ieee communications surveys & tutorials _ , vol . 15 , no . 3 , pp . 9961019 , jul . 2013 . t. d. novlan , r. k. ganti , a. ghosh , and j. g. andrews , `` analytical evaluation of fractional frequency reuse for heterogeneous cellular networks , '' _ ieee transactions on communications _ , 60 , no . 7 , pp . 20292039 , jul . 2012 . k. huang and j. andrews , `` an analytical framework for multicell cooperation via stochastic geometry and large deviations , '' _ ieee transactions on information theory _ , vol . 59 , no . 4 , pp . 25012516 , apr . 2013 . s. akoum and r. heath , `` multi - cell coordination : a stochastic geometry approach , '' in _ 2012 ieee 13th international workshop on signal processing advances in wireless communications ( spawc ) _ , 2012 , pp . 1620 . d. gesbert , s. hanly , h. huang , s. shamai shitz , o. simeone , and w. yu , `` multi - cell mimo cooperative networks : a new look at interference , '' _ ieee journal on selected areas in communications _ , vol . 28 , no . 9 , pp . 13801408 , sep . 2010 . m. haenggi and r. smarandache , `` diversity polynomials for the analysis of temporal correlations in wireless networks , '' _ ieee transactions on wireless communications _ , vol . 12 , no . 11 , pp . 59405951 , nov . 2013 . u. schilcher , c. bettstetter , and g. brandner , `` temporal correlation of interference in wireless networks with rayleigh block fading , '' _ ieee transactions on mobile computing _ , vol . 11 , no . 12 , pp . 21092120 , dec . r. tanbourgi , h. dhillon , j. andrews , and f. jondral , `` effect of spatial interference correlation on the performance of maximum ratio combining , '' _ ieee transactions on wireless communications _ , 2014 , to appear . z. gong and m. haenggi , `` interference and outage in mobile random networks : expectation , distribution , and correlation , '' _ ieee transactions on mobile computing _ , vol . 13 , no . 2 , 337349 , feb . 2014 .

we present the x - ray grating spectra of the recurrent nova rs ophiuchi during its 2006 outburst , obtained with the xmm - newton and chandra observatories . two weeks after optical maximum , the x - ray spectrum was hard and dominated by emission lines of h - like and he - like ions . the x - ray luminosity was 2.4 10 erg in the 0.33 - 10 kev range . the spectra indicate a collisionally dominated plasma with a broad range of temperatures . all the lines are blue - shifted , with the velocity shift increasing with lower ionization state and longer wavelength . two weeks later , the spectrum was still dominated by emission lines , although the line ratios present indicate cooling . during this observation , a soft x - ray flare occurred in which a new system of higher velocity emission lines appeared in the soft end of the spectrum . towards the end of the second month the emission spectrum became composite , with both absorption and emission features . the dominant component during the third month was the supersoft continuum with the broad absorption features of a hot white dwarf atmosphere . the absorption lines did not appear to be significantly blue - shifted as previously observed in another classical nova . a preliminary fit of an atmospheric model available in the literature indicates a white dwarf temperature slightly exceeding 800,000 k. the x - ray luminosity reached at least 9 10 erg in the 0.2 - 1 kev range , while the intrinsic nebular absorption decreased by a factor of five since the first observation . the spectral fits indicate a massive white dwarf , with a mass of at least 1.2 m . therefore , rs oph may be an important type ia supernova progenitor . we show that the data are consistent with mass loss ending before day 54 after the outburst , and nuclear burning ending around day 69 . a rapid decay in x - ray luminosity followed after week 10 . the x - ray luminosity 5 , 7 and 8 months after optical maximum dropped by more than two orders of magnitude , and the spectra are dominated by soft x - ray emission lines typical of a collisionally excited and ionized plasma . they do not appear to consistent with emission from an accretion disk .

0709.4512

rs ophiuchi ( rs oph ) is a member of a rare subclass of cataclysmic variable known as recurrent novae ( rn ) . as the name suggests , rn are systems which are observed to go into outburst more than once . nova outbursts are believed to be the result of thermonuclear runaway ( tnr ) on the surface of a white dwarf accreting material from a binary companion ( e.g. schatzman 1950 , starrfield et al . 1974 , prialnik et al . the tnr occurs once the pressure at the base of the shell of accreted material reaches a high enough value . following fujimoto ( 1982a ) the critical pressure required for the tnr ignition is given by @xmath5 where m@xmath6 and r are the mass and radius of the white dwarf , and @xmath7m@xmath8 is the mass of the accreted shell at the onset of the tnr . as long as accretion continues , all novae are expected to recur with timescales ranging from 10@xmath9 years to only a few decades . for systems hosting high mass white dwarfs , and those with higher accretion rates , critical envelope masses will be reached in a shorter time . therefore , the fact that we see the recurrent novae repeat at all during a human lifetime is an indication that we are dealing with higher mass white dwarfs . if the white dwarfs in recurrent novae can grow in mass with each subsequent accretion outburst cycle , they may eventually approach and even exceed the chandrasekhar limit . therefore , rn are interesting and important supernova ia progenitor candidates . rs oph went into its 6th recorded outburst on feb 12.83 , 2006 ( narumi 2006 ) , having previously been detected in outburst in 1898 , 1933 , 1958 , 1967 and 1985 . it is possible that other outbursts have occurred since the discovery of the system , but were missed by observers . long term lightcurves of rs oph obtained by the aavso show that the system undergoes a characteristic dimming just before the outburst , and an overshoot at the end of decay from maximum beyond the mean inter - outburst magnitude ( oppenheimer 1993 ) . schaefer ( 2004 ) reported that a reduction in magnitude was detected on plates obtained in 1907 , just after the end of the seasonal observing gap . he attributed these low values to the post - outburst dip observed in other events , and suggested that rs oph experienced an outburst in 1907 while it was hidden by the sun . similarly , dips in magnitude were detected both before and after the 1945 observing gap , suggesting another outburst was missed at that time ( oppenheimer & mattei 1996 ) . rs oph is believed to host a massive white dwarf accreting material from a red giant companion of spectral type k7 ( mrset & schmid 1999 ) . the most recent determination of the orbital period of the binary is 456 days ( fekel et al . 2000 ) . the secondary is believed to underfill its roche lobe , and so accretion proceeds through a wind and not through roche lobe overflow ( rlo ) , as in cataclysmic variables and classical novae ( e.g. dobrzycka et al . 1996 , hachisu et al . the distance to the system has been discussed extensively in the literature , and disagreements remain . bode ( 1987 ) gives 1.6 @xmath10 0.3 kpc . results from the current outburst give a distance ranging from 540 pc ( monnier et al . 2006 ) to 1.31.7 kpc ( hachisu et al . this large spread in distance makes the exact parameters of the system difficult to determine . we will follow bode in this work , and assume a distance of 1.6 @xmath10 0.3 kpc . the system was observed in x - rays by exosat on several occasions during the 1985 outburst ( mason et al . , 1987 ) . however , only low resolution spectra were obtained as x - ray grating spectroscopy was not available in 1985 . since then the situation has vastly improved , with no less than 4 active x - ray observatories equipped with advanced instrumentation capable of high resolution grating spectroscopy or microsecond timing accuracy . data have been obtained at frequent intervals by the rossi x - ray timing experiment ( rxte ) , the x - ray telescope ( xrt ) on board swift , and of course chandra and xmm - newton . in this paper , we present the chandra and xmm - newton data , which include the first grating spectra obtained of any of the recurrent novae . the temporal coverage offered by such an extensive set of observations allows us for the first time to test models of the evolution of an outburst in the rs oph system . we present a detailed timing analysis of the data set in an accompanying paper ( leibowitz et al . thanks to a generous allocation of director s discretionary time ( ddt ) , we were able to obtain a total of 9 grating observations of rs oph ; 5 xmm - newton and 4 chandra pointings . table 1 outlines the details of each observations . the quoted exposure times are the net exposures after being corrected for periods of bad telemetry . these observations were initially carried out every two weeks from days 14 to 67 . after that , one pointing each in june , september and october was obtained , following the evolution of the system back to its inter - outburst state . the chandra observations were carried out using the hetg / acis - s in february , and then the letg / hrc - s in all subsequent pointings . the data were reduced using ciao v. 3.3.1 with caldb v. 3.2.2 . for each observation , the _ mkgrmf _ task was used to create response matrices for each order , followed by _ mkgarf _ to produce the ancillary response files . we worked with the + 1 and -1 orders for each grating , which contained most of the flux . the xmm - newton observatory has a suite of 6 instruments ; two reflection grating spectrographs ( rgs ) , 3 european photon imaging cameras ( epic ) , and the optical monitor ( om ) , an optical and near uv telescope . the epic utlilize two slightly different technologies . the epic - mos cameras are arrays of front - side illuminated ccds . in contrast , the epic - pn camera consists of an array of 12 back - side illuminated ccds , sensitive to energies from 0.15 to 10 kev . it has the highest soft energy response of any of the instruments onboard xmm - newton , and when operated in timing mode can resolve events down to 0.03 ms . all of the epic - pn data we present here were obtained in timing mode , necessary because of the high count rate encountered in rs oph . this minimized the problem of pile - up " in the data ( see @xmath113.4 below ) . all of the xmm - newton instruments gather data during each observation , producing simultaneous broadband imaging spectroscopy and high resolution grating spectroscopy data . the xmm - newton data were reduced using sas v. 6.5 . before we present and analyze each observation in more detail , an overview of the early evolution can be obtained by looking at the epic - pn spectra obtained in timing mode ( fig . the spectrum became much softer and brighter over the course of the first 40 days . the total count rate increased by a factor of @xmath121000 between day 14 and day 54 , with most of this increase occuring below 1 kev . at the same time , the count rate above 2 kev decreased by a factor of 100 , demonstrating the dramatic change in hardness ratio of the system . in light of the x - ray evolution of other classical novae ( e.g. v382 vel , orio et al . 2001a , and 2002 ) we interpret this behavior as follows : the early x - ray emission originated in the ejecta , and as the plasma surrounding the white dwarf became optically thinner to supersoft x - rays , the hot hydrogen burning white dwarf became visible . the most striking difference with classical novae is the rapidity of this evolution . in the following sections , we show how the grating spectra support this interpretation , and give unprecedented details of the physics of both the ejecta and the inner source . rs oph was observed with x - ray instruments multiple times from day 3 onwards . early observations were carried out with the xrt onboard swift , and the proportional counter array ( pca ) on rxte . both detected bright , hard emission from the system . for the swift observation on day 3 bode et al . ( 2006 ) obtained a best fit thermal bremsstrahlung temperature of 70 kev ( 8.1 @xmath0 10@xmath13 k ) . the data obtained with rxte in the following three weeks showed the spectrum shifting from hard to soft energies , which was interpreted as emission from cooling gas initially shock heated to extremely high temperatures by the blast wave from the nova eruption ( sokoloski et al . 2006 ) . both chandra and xmm - newton first observed the system 14 days from the start of the outburst , on feb 26th . as the source was so hard at this time , the chandra observation was carried out with the high energy transmission grating ( hetg ) and acis - s . the observation suffered from pile - up " , a common problem for high count rate observations . pile - up is the recording of two or more photons as a single event of higher energy . although the use of gratings reduces this problem to some extent , pile - up can still occur for the very brightest objects , and necessitates some additional reduction steps . during the initial stages of the chandra reduction pipeline , events with energies greater than 15 kev are rejected and flagged as bad . with grating observations , this leads to a `` hole '' in the zeroth order data i.e. a region with zero counts . this can result in bad centroiding in the source detection phase , creating extraction regions centered on an inaccurate zeroth order position . the consequence is a reduction in the accuracy of the wavelength calibration of the spectrum . to avoid this , we re - extracted the spectrum , using new coordinates for the zeroth order image . we used the coordinates of the intersection between the frame - shift streak and the trace of the meg arm in the level 1 event file . this is the best method available for determining the zeroth order position . this re - extraction ensures the most accurate wavelength responses possible are obtained . of the suite of xmm - newton instruments , the epic - pn camera and the rgs gratings produced useful data . the epic - mos instruments were operated in imaging mode in case there was sudden obscuration of the source like that observed in the classical nova v4743 sgr ( ness et al . , 2003 ) . however , the count rate remained high for the duration of the observation , and the resultant pile - up on the mos cameras rendered the data unusable . the xmm - newton and chandra observations overlap in time substantially , giving combined high spectral resolution from 1.2 to 35 . the grating data taken on day 14 are shown in figures 2 and 3 . the total unabsorbed flux in the range covered by each grating , obtained using the isis spectral analysis package ( houck & denicola , 2000 ) , is given in table 1 . the values range from 1.6 to 5 @xmath0 10@xmath14 erg s@xmath15 @xmath2 depending on the energy range covered by each instrument . in the 514 region , which is covered by all the grating instruments , the absorbed flux values agree to within 5% . emission lines from h and he - like ions of metals dominate these spectra , and very little continuum emission is present . we first identified the brightest lines using the guide package , which makes use of the astrophysical plasma emission database ( aped ) through atomdb v.1.3 . lines of species from fe xxv to n vii are present in the wavelength range covered by the combination of the hetg and the rgs . where the signal to noise in a line was high enough , we obtained estimates of the centroid velocity and full width at half maximum ( fwhm ) by fitting a gaussian ( for the h - like ions ) or sum of gaussians ( for the he - like ions ) to the data . the results are presented in table 2 . for the he - like ions , we constrained the separation of the components and fit the resonance line central wavelength , since this was the strongest line present in each triplet . an example of this is given in fig . fitting the line profiles yielded a very interesting result - all lines are blueshifted , and the magnitude of the velocity shift increases for lower ionization states and longer wavelengths ( see table 2 ) . additionally , as the wavelength increases , so does the broadening of the lines . next , we determined the dominant source of ionization in the plasma using the he - like triplets observed in the spectrum . we used the isis package to convert the count rate spectrum to flux , and then used the best gaussian fit to calculate the flux in each component of the triplet . our best fit line fluxes are given in table 3 , along with the value of the so - called g ratio , _ ( f+i)/r_. this ratio is a useful plasma diagnostic as its value is sensitive to the dominant source of ionization in the plasma . porquet et al . ( 2001 ) showed that _ ( f+i)/r _ is @xmath16 for collisionally dominated plasmas . the three lines we obtain ratios for all have _ ( f+i)/r _ = 1 within errors , indicating that at this early stage the plasma is collisionally dominated . we must bear in mind that the spectra shown in figures 2 and 3 are unique and unprecedented - no similar spectra have been obtained for any other nova , because in general novae are less x - ray luminous than rs oph at this stage of the outburst . since the he - like line ratios indicate a collisionally dominated plasma , we attempted to fit the data with the apec models available in xspec , which model line emission from these plasmas . following the work done by both bode et al . ( 2006 ) and sokoloski et al . ( 2006 ) for the prompt x - ray emission , we took the simplest possible approach and first tried to fit a single temperature , solar abundance model to both the chandra and xmm - newton spectra simultaneously . we velocity broadened the model by 1000 kms@xmath15 to match the widths observed in the data . the fits to both datasets are not good , and we find that a single temperature model can not fit the spectrum . this is not surprising given that fe xxv and n vii do not exist at the same temperature fe xxv does not exist in plasmas below 10@xmath17 k , while very little n vii exists above 6.3 @xmath0 10@xmath18 k ( arnaud & rothenflug , 1985 ) . it is clear that at this stage of the outburst , the x - ray emission is much more complex than a single , thermal brehmsstrahlung model . therefore , we decided to add additional temperature components to account for the range of lines seen in the data and improve our model . this is where one of the advantages of high resolution over broadband data becomes clear ; we can make use of the simultaneous detection of both h - like and he - like ions of the same element to determine the range of temperatures required to produce the observed spectrum . the ratio of the hydrogen - like ly @xmath19 line flux to the sum of the fluxes of the three components of the helium - like triplet is highly sensitive to temperature , and allows us to input real temperature values from the data into our models . we calculated the ratios for the s , si , mg lines in the chandra spectra , and the o lines in the rgs spectrum , and present these results in table 4 , along with the temperature derived for each ion . using these temperature values , we then fit a multiple temperature component , solar abundance apec model to the data . we froze the components with values determined from the line ratios , and included an additional component of higher temperature , ( which we allowed to vary ) to fit the fe xxv line observed at 1.8 . we also allowed the normalization factor , which depends on the x - ray emission measure of the plasma , to vary for each component . our final , best fit model gave an absorbing column of 1.2 @xmath0 10@xmath20 @xmath2 , and four apec components with temperatures 16.84 , 2.31 , 0.92 and 0.64 kev . introducing additional temperature components did not improve the fit . the normalization factors suggest x - ray emission measures for each plasma component of a few 10@xmath21 @xmath22 . this is approximately a factor of 10 larger than the emission measure determined by sokoloski et al . ( 2006 ) from the rxte data on the same day in the 0.510 kev range . the fit to the chandra data is reasonably good , although not perfect . the continuum is reproduced very well , and the flux in the majority of the lines matches the data to within a factor of 2 . however , the model gives a much poorer fit to the rgs data ( see fig . , the model drastically underestimates the strengths of the o and n lines , by at least a factor of 10 in the case of oxygen , and a factor of 100 for nitrogen . introducing additional lower temperature components does not produce the required flux in the o viii or n vii ly @xmath19 lines , and is also inconsistent with the weak o vii and n vi lines observed in the data . increasing the normalization factor , which depends on emission measure , produces too much flux at shorter wavelengths . an explanation for such strong o viii and n vii lines is enhancement of o and n in the x - ray emitting material . livio and truran ( 1994 ) showed that available data suggests that all novae are enhanced in heavy metals , with significant enhancements in cno cycle nuclei for most fast " novae . therefore , a likely explanation for the observed line strengths , particularly in the soft part of the spectrum , is that the emission arises in the nova ejecta . the model derived fluxes are in good agreement with the isis values , even for the rgs data . despite the poor fit the o viii and n vii ly @xmath19 lines , these features actually contribute very little to the total x - ray flux from the system at this early stage . unabsorbed fluxes were then obtained by correcting for the best fit column density of 1.2 @xmath0 10@xmath20 @xmath2 . assuming a distance of 1.6 kpc , we used these results to calculate a luminosity in the band covered by each instrument . for the february data , we find a luminosity in the range 535 of 2 @xmath0 10@xmath1 erg s@xmath15 for the rgs data , 2.78 @xmath0 10@xmath23 erg s@xmath15 in the range 1.514 for the heg , and 3.81 @xmath0 10@xmath23 erg s@xmath15 in the range 2.525 for the meg . in order to have a complete understanding of the observed spectra , we also need to account for the dependence of the velocity shifts and broadening on wavelength and ionization stage . this dependence seems to suggest that the lines are formed in different regions of the nova wind , at different distances from the star . the same dependence has been clearly observed in another object , the o star @xmath24 puppis . chandra observations of this star showed a similar wavelength dependent behavior of centroid velocity blueshift and broadening to rs oph ( cassinelli et al . , 2001 ) . there , the authors state that this behavior indicates a connection between radial velocity , wind absorption , and doppler broadening of the line emission " . recent work by waldron and cassinelli ( 2007 ) and cassinelli et al . ( 2007 ) showed that a possible origin for x - ray emission in o - star atmospheres is the formation of bow shocks around clumps that form in the winds . bow shocks produce a range of post shock temperatures consistent with that seen in o stars , and interestingly , in rs oph . therefore , one possible origin for the line emission seen in rs oph is the formation of bow shocks in the nova wind , either around clumps in the nova ejecta , or clumps from the red giant wind which are still present in the system ( see crowley , 2006 for a discussion of clumps in the stellar winds of red giants in symbiotic systems ) . more complex theoretical models will be required to explore this possibility further . xmm - newton observed the system again on day 26 ( march 10th ) . this observation lasted approximately 10000s for the epic - pn . however , the rgs were turned on 2000s after the start of the observation , and were marred by many intervals of high background . the epic - pn count rate in the region 0.151 kev has increased by a factor of 2 , from 14 to 35 cts / s . while the count rate in the region 110 kev has decreased by a factor of 2 . the total absorbed flux in the region 535 is 1.6 @xmath0 10@xmath14 erg s@xmath15 @xmath2 , approximately the same as on day 14 . the spectrum is still dominated by emission lines , although the strengths of the lines has changed , reflecting the shift seen in the epic - pn data . the strong lines of si xiv , si xiii , mg xi and ne x observed in the february spectrum decrease in strength by a factor of @xmath122 , mirrored by a similar drop in total flux in the region 514 from 2 @xmath0 10 @xmath14 to 10 @xmath14 erg s@xmath15 @xmath2 . in contrast , o viii and n vii lines at the soft end of the spectrum increase by a factor of @xmath123 , and there is a significant increase in flux between 20 and 30 . the he - like triplet of n vi is also detected , and is at least 10 times brighter than in february , also demonstrating cooling of the emitting plasma . the temperature indicated by the mg h / he ratios is 8 @xmath0 10@xmath18 k , a decrease of 10@xmath18 k over the course of the 2 weeks since the previous observation . the total absorbed flux is slightly higher , although very similar to the value obtained in february ( see table 1 ) . we extracted lightcurves for the observation from both the epic - pn and rgs instruments , and found that the count rate was variable , with changes in brightness on times scales of a few hours . in particular , about 5000s into the observation ( 3000s for the rgs ) , a rapid brightening occurred where the count rate increased from @xmath1270 to @xmath2590 cts / s over the course of just a few minutes . while the flare was detected by both the epic - pn and rgs instruments , we show in the upper panel of figure 5 the epic - pn lightcurve extracted in the energy range 0.15 0.4 kev ( where the flare was most noticeable ) since the epic - pn was turned on @xmath122000 s before the rgs . this bright period lasted for @xmath125000s , after which the count rate dropped again , although it remained higher than its initial value for the duration of the observation . dramatic changes in x - ray brightness appear to be ubiquitous in novae , and often exhibit detectable periodic oscillations ( see e.g. v382 vel ( orio et al . 2003 ) , v1494 aql ( drake et al . 2003 ) and v4743 sgr ( ness et al . , 2005 ) ) . we performed timing analysis on the epic - pn and rgs lightcurves . while we give a detailed discussion in a companion paper ( leibowitz et al . , 2007 ) we summarize the results here as follows . during the first part of the observation , prior to the flare , no period is found . then , for @xmath125000s after the rapid brightening , a period of 35.6 s is found in the power spectrum of the observation . this period then turns off " and is not present for the remainder of the observation , even though the count rate remains high . similar periods are detected by swift in the subsequent two weeks ( osborne et al . , 2006b ) . this xmm - newton observation is the first detection of this period in the current outburst of rs oph . in order to examine the possible origin of these temporal phenomena in more detail , we split the rgs data into three time periods - before the flare ( time interval 1 ) , during the peak of the flare while the @xmath1235 s period is detected ( interval 2 ) , and the end of the observation ( time interval 3 ) after the @xmath1235 s period disappears . we then re - extracted the spectrum for each time interval , applying a less conservative bad interval flagging . we show a comparison of the first and second time periods in the middle and lower panels of fig . the spectrum extracted in interval 1(shown in black in the figure ) is very similar to february , albeit with different line strengths . the onset of the flare , however , introduces some dramatic differences into the spectrum . the second spectrum , is is essentially identical at wavelengths below 20 , with the same line shifts , broadening and fluxes . however , longward of 20 a dramatic increase in count rate takes place . the n vi intercombination and forbidden lines of the n vi triplet increase in strength by a factor of 2 , although the strength of the resonance line remains the same . in the region longward of the n vii ly @xmath19 line , a large number of new lines with count rates similar to both the n vi and n vii lines suddenly appear . the lines are still present in the last time interval , once the 35s period disappears , although their count rates are lower reflecting the drop in overall count rate also seen in the lightcurves of the observation . no significant continuum is detected , and we conclude that these new lines alone are the source of the dramatic increase in count rate at low energies seen in the epic - pn data . the ly series lines of c vi lie in this region of the spectrum , and we can identify most of the lines as lines of c vi or n vi if we allow for large blueshifts of 800010000 km s@xmath15 . however some features , most notably the broad line at 26.7 , remain difficult to identify even allowing for a large velocity shift . in the lower panel of fig . 5 , we show ids of the new lines taking the @xmath128000 km s@xmath15 shift into account . such velocity shifts are much larger than those observed on day 14 , or those of the harder lines detected in the same spectrum . we note that 8000 km s@xmath15 is consistent with the escape velocity of a massive white dwarf . we find evidence , therefore , that the additional soft flux is due to emission lines appearing in material ejected suddenly at very high velocities . we also note that the appearance of the new emission lines coincides with the detection of the 35s period in the lightcurve , suggesting that the two phenomena are related . after day 26 , swift monitoring revealed considerable progress in the trend toward softer energies ( osborne et al . 2006b ) , so a new observation was obtained with the chandra letg / hrc - s on day 40 . the exposure lasted for a little under 10000s ( see table 1)and is shown in black in figure 6 , where we plot the fluxed spectrum for each of the sss phase observations . since the fluxed spectra are corrected for the instrumental response , they allow a comparison between datasets obtained with different instruments . the day 40 spectrum is markedly different from that obtained just two weeks earlier with xmm - newton . the total absorbed flux in the 583 region is now 1.7 @xmath0 10@xmath26 erg s@xmath15 @xmath2 , a factor of @xmath12100 larger than the day 26 observation when calculated in the same wavelength range . longward of 14 , a dramatic increase in the level of the continuum has occurred . several absorption features are present which are predicted by white dwarf model atmospheres , most notably lines of fe xvii , o viii and n vii . the spectrum is similar to that observed in v4743 sgr during its supersoft phase , although the rs oph spectrum is significantly harder . also , as far as we can determine , there is no significant blueshift of the absorption lines as was the case for v4743 sgr . we also detect lines due to o i at 23.51 and o ii at 23.35 that are known to be associated with absorption by cold gas ( e.g. yao & wang , 2006 ) , which we attribute to absorption by interstellar or circumstellar material . in addition to the absorption features , the emission line features observed on days 14 and 26 are still present across the spectrum in figure 6 . in the harder portion of the spectrum ( between 5 and 14 ) , we still detect several of the emission lines that were prominent in the previous spectra , most notably the lines of mg xii , xi and ne x at 8.4 , 9.1 and 12.1 respectively . the lines have continued to fade , and the total flux in the 514 is now 6.8 @xmath0 10@xmath27 erg s@xmath15 @xmath2 30% lower than in the xmm - newton observation of day 26.1 . in the softer part of the spectrum , emission lines of fe xvii , o viii , o vii , n vii and n vi that arise in the nebula are present , superimposed on the continuum . we extracted the lightcurve for the observation using the ciao tool _ ( see fig . 7 , also leibowitz et al . , 2007 ) , and found that rs oph was still highly variable . unlike the xmm - newton observation 2 weeks earlier , the chandra data show large scale fluctuations in brightness throughout the entire observation , with no obvious extended `` bright '' period . we did extract a separate spectrum for the period marked by the dotted vertical lines in fig 8 , since during this period the lightcurve is relatively flat . however , we found no evidence of a dramatic softening of the spectrum as we did for the epic - pn data earlier in the month . the flat region , compared to the rest of the spectrum , shows a higher count rate but no evidence of a dramatic shift in energy distribution . the change in count rate is confined to the wavelength range 1430 . however , since the shape of the spectrum and the absorption features do not seem to change , the variability may be due to changing absorption in the clumpy ejecta . two more observations were carried out in the subsequent 4 weeks . the first was with xmm - newton on day 54 . again , only the rgs and epic - pn obtained useful information , and even then both instruments suffered from pile - up problems due to very high count rate . this effect has been encountered in another observation of a bright nova , v4743 sgr in 2003 ( leibowitz et al . , 2006 ) . in the rgs , pile - up results in first order events being incorrectly identified as higher order . this results in a loss of signal in the first order spectrum . this can be corrected to some extent by examining the pulse height amplitudes ( pha ) of the higher order events . piled up events increase in pha by an integer multiple of the intrinsic energy . therefore , events with unreasonably large pha in higher orders can be identified as lower order pile - up artifacts . a correction can then be applied , where the higher order counts are multiplied by the order and added to the first . this gives the corrected first order count distribution . this works very well for extremely soft sources , as no intrinsic higher order counts are present . therefore , all higher order events can be confidently identified as artifacts of pile - up . however , this correction proved to be more difficult for rs oph than for v4743 sgr , as the source is harder and therefore has real higher order events . the softest end of the second order event distribution is intrinsic to the source , while the hard second order events are due to pile - up . therefore , the correction to first order is only valid within a certain wavelength range , which has to be estimated . we therefore lose the softest part of the first order rgs spectrum , although we have regained the counts at the hard end , and we end up extracting the spectrum in the region 528 . although this correction sacrifices some spectral coverage , none of the rgs orders would be suitable for model fitting if this procedure had not been carried out . as fig . 1 demonstrated , by the xmm - newton observation on day 54 , rs oph had become an incredibly bright , supersoft source . the rgs spectrum ( shown in green in fig . 6 ) is similar in shape to the letg spectrum of two weeks earlier , although it has become even brighter , reflecting the huge increase in count rate seen in the broadband data . the total absorbed flux can only be evaluated in the range 528 , and because of the method used to correct for pile - up , we could not use isis to estimate the flux . instead , we rely on the fit to the model described below , and estimate the flux in the region 528 to be at least 4 @xmath0 10@xmath26 erg s@xmath15 @xmath2 . this is a factor of 2.5 higher than the letg observation of day 40 when evaluated in the same wavelength range . the absorption features are deeper , and the emission features appear to be less prominent . the ly series of n vii in absorption is now easily identifiable between 19 and 25 . the neutral oxygen absorption features are also more pronounced . the harder lines between 5 and 14 are still present , and once again are fainter than the previous observation . we extracted the lightcurve for the epic - pn and both rgs instruments , and found that the large scale brightness fluctuations observed two weeks earlier with chandra are no longer present . instead , for the duration of the observation the count rate remains constant . another observation was carried out using chandra on day 67 , again with the letg / hrc - s , with a net exposure time of @xmath126500s . the spectrum ( shown in red in fig . 6 ) is remarkably similar to that observed with xmm - newton on day 54 . the total absorbed flux measured with isis is 3.8 @xmath0 10@xmath26 erg s@xmath15 @xmath2 in the range 583 . in the same range as the corrected rgs spectrum ( 528 ) we find that the absorbed flux is 3.4 @xmath0 10@xmath26 erg s@xmath15 @xmath2 about 15% lower than the day 54 observation . again , it appears that the absorption features are less contaminated by superimposed emission , although both the harder lines between 5 and 14 and the n vi he - like triplet are clearly detected , showing that some nebular emission is still present at this time . the hard nebular component has faded further , and the total 514 flux is now just 3.5 @xmath0 10@xmath27 erg s@xmath15 @xmath2 . extracting the lightcurve , we again found no large scale fluctuations in brightness , with the lightcurve looking similar to one obtained with xmm - newton two weeks earlier . the @xmath1235 s period is missing on day 40 , but a period of 17.86 s is clearly detected in the data . in a companion paper by leibowitz et al . we present a discussion of the possible origin of this period , that seems to be due emission originating near the poles of the white dwarf which are at different temperatures . if this interpretation is correct , the poles must have reached a uniform temperature by day 54 , as a period of 34.72s was detected in the epic - pn and rgs data . later , no clear period is identified in the letg data obtained on day 67 . since on days 54.0 and 67 no large scale fluctuations were detected in the observation lightcurves , the appearance of the @xmath1235 period is at this point no longer associated with increases or decreases in the lightcurve or with spectral variability . the dramatic increase in the continuum level between 14 and 35 in these spectra allowed us to attempt to fit white dwarf atmospheric models to the data and investigate the temperature of the star . many of the non local thermonuclear equilibrium ( nlte ) models commonly used for extremely hot white dwarfs and central stars of planetary nebulae , for example , those developed by hartmann & heise , ( 1997 ) are not suitable because they do not reach the extremely high temperatures of post - nova white dwarfs . we used a series of plane parallel , static models calculated by rauch ( 2003 ) . these models extend up to 10@xmath18 k and include an extensive network of elements from h to fe for line blanketing calculations to produce state of the art atmospheric spectra . we use models with two sets of abundances , solar or halo ( which were developed for modeling systems in the magellanic clouds ) . we also used a series of models developed specifically for the classical nova v4743 sgr ( rauch et al . , 2005a , b ) , with abundnaces which follow the composition of the white dwarf through multiple cycles of accretion and nova outbursts ( rauch , private communication ) . nuclear processing and dredge - up of the underlying white dwarf lead to surfaces that are enhanced in he and cno cycle elements . nuclear processing of the accreted material transforms c into n via the cno bi - cycle , leading to a c / n ratio that is significantly lower than solar . this type of model fitting allows us to explore the surface composition and determine whether the white dwarf shows signs of having retained its cno processed accreted material . if this is the case , the white dwarf can grow in mass and possibly explode as a type 1a supernova . a coarse grid of abundances is available , allowing the c / n ratio to be varied from the solar value , to 0.001 times solar . we note that these abundances have not been fine - tuned for the rs oph system , but at the present time they are the only suitable models available to use in the literature . by showing that even with these models useful constraints are obtained , we hope to encourage further development of white dwarf atmospheric models specifically for use with rs oph . we constrained our fits to the region longward of 14 where the white dwarf continuum is strongest , and to avoid the harder lines of the nebular emission . for the chandra observations we modeled both the + 1 and -1 orders simultaneously , and for the xmm - newton observation both rgs1 and 2 . first , we fit all available models to each observation in order to explore the dependence of best fit photospheric temperature on abundance . since the atmospheric abundances are uncertain , and our nova models are not finely tuned to the current dataset , this was necessary to help estimate the uncertainty in our best fit temperature , and absorbing column . with the exception of the halo model , all the temperatures found agree to within @xmath1210,000 k. the halo model was developed for novae in the lmc and smc , and since optical and infrared studies of the rs oph system seem to indicate that the red giant has near solar abundances , we believe this is an inappropriate model to consider for rs oph . therefore , we believe the temperature that model fitting returns is relatively independent of the choice of abundances . next , we determined the best fit model . we did this primarily by eye , since the presence of emission features arising in the nova ejecta , and absorption lines due to the ism / csm , make determining a formal statistical fit difficult , especially without a fine tuned model . the fit to the day 40 data obtained with chandra is the worst of the three , due to the superposition of many emission features on the white dwarf continuum , and we can not determine a reliable temperature for this observation ( the fit is however useful for obtaining lower limits to the flux , as we show below ) . however , as the continuum continues to increase in brightness , and the superimposed emission features become weaker , the spectral fits improve . on day 54 , the best fit is obtained with the v4743 sgr model which has a c / n ratio of 0.001 times the solar value . this model is necessary to explain the lack of a c vi k - edge at 25.37 and weak c vi ly series lines between 26 and 30 in the observed spectrum . our best fit gives a temperature of @xmath12820,000 @xmath10 10,000 k , and an absorbing column of 2.3 @xmath0 10@xmath28 @xmath2 . while this value is slightly lower than the interstellar value of 2.4@xmath010@xmath28 @xmath2 ( hjellming et al . 1986 ) , we believe it to be a lower limit , since the lack of data longward of 28 inhibits our ability to fit this parameter with a large degree of confidence . the chandra data of day 67 provides the most convincing evidence of a high temperature for the white dwarf photosphere . the continuum level is so high that the contribution from the emission component is almost insignificant over most of the spectrum . the observation is also free of the instrumental issues that affected the xmm - newton observation two weeks earlier . we include two figures showing the fit to the day 67 data . the first , figure 8 , shows our best fit model with a blackbody for comparison . this figure clearly demonstrates the importance of nlte effects in modeling the spectra of these hot white dwarfs . figure 9 demonstrates the difference between our nova model with c / n = 0.001 solar and a model of solar abundance . again , no c vi k - edge or ly series lines are detected in the tail of the continuum , showing clearly that our nova abundance model is more appropriate than the solar abundance model . we note that the o k - shell edge at 23.4 is partially filled , which seems to be due to superimposed emission . for day 67 , we find a best fit temperature of 809,000 @xmath10 10,000 k and an intervening absorbing column of 3.16 @xmath0 10@xmath28 @xmath2 . this is slightly lower than the value obtained for day 54 , showing that the white dwarf has started to cool . the day 67 best fit value of n(h ) is higher than the interstellar value . although a large uncertainty is associated with these values because of the poor fit at shorter wavelengths , constraining n(h ) to the interstellar value produces model fits with an excess of emission at long wavelengths . therefore we conclude that there is still some intrinsic absorbing material present in the system . using the best fit model to the day 40 data , and correcting for the intervening absorbing column , we derive a value for the luminosity in this range , l@xmath29 = 8.1 @xmath0 10@xmath1 erg s@xmath15 . given the difference between the isis derived flux ( which is essentially the directly measured flux and includes the contribution from the nebula ) and the flux from the model ( which is only for the white dwarf atmosphere ) , we consider this luminosity a lower limit - the true luminosity could be as much as 20% higher , since here we assume all additional flux arises in the nebula . on day 54 , we derive a value for l@xmath30 = 5.8 @xmath0 10@xmath1 erg s@xmath15 . here , we can also use the fit to the epic - pn data to give us additional information . using the same model , we find a lower limit to the absorbed flux in the energy range 0.1510 kev of @xmath123 @xmath0 10@xmath26 erg s@xmath15 @xmath2 , and l@xmath31 = 9.2 @xmath0 10@xmath3 erg s@xmath32 . the luminosity value determined on day 67 is a little lower ; l@xmath33 = 3.31 @xmath0 10@xmath3 erg s@xmath15 . we estimate the errors on our luminosity to be between 10 and 20% because of the nebular contribution which the atmosphere model does not account for . all the models we fit to the data had log(g ) = 9 , as these were the only models that could produce a high enough temperature to match the data . although the grid of effective gravity value of rauch s models is coarse , models with g=7 , 8 and 9 are available . obviously the error in the log(g ) determination is large , but fitting models with different abundances that the effective temperature is determined in a relatively narrow range and even fitting a blackbody we obtain a temperature close to 800,000 k. we can express the surface gravity in terms of temperature assuming that the white dwarf luminosity does not exceed ( at least not by much and not for a long time ) the eddington luminosity given by @xmath34then , substituting r@xmath35 = gm / g into the expression l = 4@xmath36r@xmath35@xmath37t@xmath38 , and equating to the expression for the eddington luminosity above , we can write @xmath39 therefore , the main constrain on the effective gravity is determined by the high effective temperature , and it appears that in our case g must be at least few times 10@xmath40 cm s@xmath41 . a less compact wd would result in a bolometric luminosity exceeding the eddington limit by a large amount . by early june , rs oph was no longer detected as a luminous supersoft source . swift observations obtained during april and may showed a linear decline in count rate in the 0.310 kev region , from a maximum of @xmath12400 c / s on april 12 , to 0.3 c / s by may 31st . between may 31 and june 11 , there were no significant changes in this count rate ( osborne et al , 2006c , d ) . chandra observed the system again on june 4th ( day 111 ) , 6 weeks after the previous grating observation . the data confirm the huge drop in flux seen in the swift observations . while the day 67 spectrum had a high count rate and strong continuum emission , the day 111 spectrum ( shown in the upper panel of figure 10 ) is dominated by emission lines with only a low continuum level . the count rate has also decreased by a factor of @xmath12170 , to only 0.22 c / s . we obtained the absorbed flux in the region 535 using isis and obtained a value of 2 x 10@xmath27 erg s@xmath15 @xmath2 . it seems clear from the dramatic change in the system from the end of april that surface hydrogen burning has stopped . where the signal to noise was high enough , we estimated spectral line velocity shifts by fitting a gaussian model . the mg xii line at 8.42 the fe xvii line at 15.01 and the ly @xmath19 lines all show evidence of a blue shift of @xmath12 500 kms@xmath15 . none of the other lines showed a detectable shift , and within the uncertainties of the model fit all the observed lines could be at their rest wavelengths . this is in stark contrast with the huge blueshifts observed in february and march . the spectra obtained in september ( day 205 ) and october ( day 239 ) with the rgs are very similar to the june letg data . figure 11 shows the two spectra . again , the dramatic blue shifts of the emission lines observed in march and april are no longer present , and instead lines are detected at their rest wavelengths . there is little change between the two , which were obtained 4 weeks apart . the count rate measured by epic - pn in the region 0.1510 kev on these dates falls from 0.76 c / s to 0.59 c / s . we attempted to fit a variety of models to these data sets , and once again did not achieve a good fit . the absorbed fluxes for both observations are given in table 1 , and show that the luminosity has decreased by a factor of @xmath1210 since june . despite the lack of spectral resolution , it is still clear from the epic - pn data that much of the emission is from lines . interestingly , emission from fe xxv at 6.7 kev , and the fe k@xmath19 complex at @xmath126.4 kev is still detected in september ( see fig . 12 ) , although the count rate is a factor of @xmath121000 lower than in february , when the fe xxv he - like triplet was detected with chandra and the hetg , and also with epic - pn on xmm - newton . by october , the count rate in this region has become too low for a confident identification of the fe features . we attempted to fit the data with various thermal plasma models available in xspec , and did not achieve a good fit with any of them . we also tried to fit the data with models suitable for accreting cvs , including a cooling flow and models developed for photoionized plasmas , and once again could not obtain a good fit . the photoionized plasma models , typical for most magnetic cvs ( mukai et al . , 2003 ) , predict a stronger continuum especially at hard energies , than we see in the data . the cooling flow models for accretion disks ( mukai et al . , 2003 ) fail to reproduce the line ratios observed in the data , and we note that the o viii ly @xmath19 , n vii ly @xmath19 and fe xvii lines between 15 and 24 were particularly hard to reproduce . for the observations in september and october , conversion of the epic - pn count rate to an equivalent rosat pspc count rate shows that the x - ray flux is still at least an order of magnitude higher than in quiescence ( orio 1993 , orio et al . the only similar spectrum observed in a nova that of the recent nova , v382 velorum ( nova velorum 1999 ) . v382 vel provides a very interesting comparison system to rs oph . ness et al . , ( 2005 ) , present a chandra observation of the system also obtained around 8 months after the outburst . the authors observed a similar emission line dominated spectrum which they attributed to emission in a collisionally ionized and excited shell . the v382 vel letg spectrum is included for comparison in the lower panel of fig . the spectrum of rs oph has stronger lines in the region 1030 , particularly those of o viii ly @xmath19 and n vii ly @xmath42 . the fe xvii lines between 15 and 17 are present and bright , unlike in v382 vel where their absence was suggested to imply fe under - abundance by ness et al . the lines at longer wavelengths seen in v382 vel are not present in rs oph in june despite a similar absorbing column , suggesting the plasma is hotter at this stage . however , these lines emerge in the rgs spectra obtained in september and october . a similar spectrum was also seen during the low " state of the v4743 sgr letg observations discussed in 3 . in that observation , the white dwarf appears to have been suddenly obscured , and in the place of the strong supersosft continuum , only a very soft emission line spectrum was observed showing lines of n vii and n vi ( ness et al , 2003 ) . the similarity of these spectra make it likely that they share a common emission mechanism , most likely originating in the material ejected by the novae . the appearance of additional plasma at low energies on day 27 , associated with very high velocity material , is extremely interesting . its appearance may be related to the jet - like structure detected in radio data ( see obrien et al . , 2006 , lane et al . , 2007 ) . if these new lines are associated with the jet - like outflow , then the absence of a system of red - shifted lines means we are only seeing one side of the jet at this time . the high effective temperature of the wd determined at the end of march and in april implies an effective gravity of at least a few times 10@xmath13 cm s@xmath41 . as discussed previously , only models with log(g ) = 9 can produce any flux shortward of @xmath1216 . therefore , in order for the observed luminosity not to be extremely super - eddington , the white dwarf must be extremely compact , and therefore massive . early work by paczyski ( 1971 ) on the evolution of planetary nebulae nuclei showed that only the most massive cores , those above 1.2 m@xmath43 , ever reached temperatures of log(t@xmath44 ) @xmath45 5.7 during their shell burning phase . sequences of white dwarf configurations for hot , massive white dwarfs were calculated by althaus et al . ( 2005 ) for various effective temperatures and chemical compositions . although their grids only reach 150,000 k , their results clearly indicate that only the most massive white dwarfs with m@xmath6 @xmath46 1.18 m@xmath4 can reach log(g ) @xmath46 9 for any temperature . therefore , both our high photospheric temperature , and large surface gravity , only appear to be consistent with a massive white dwarf , of at least 1.2 m@xmath43 , and values larger than this seem likely . swift monitoring of rs oph in april and may ( osborne et al . , 2006d ) showed that after april 28.7 ( day 75 of the outburst ) the soft x - ray flux began to fade , with an almost linear decline in count rate throughout may . our data obtained on day 67 with the letg ( which is more sensitive to the soft part of the spectrum than swift ) show that the white dwarf was already less luminous and most likely cooler on day 67 than on day 54 . we attribute this drop in luminosity to the end of nuclear burning . hachisu and kato ( 2007 ) assumed that hydrogen burning ended on day 75 , and from model fitting determined that the mass of the white dwarf was 1.35 m@xmath4 . however , given that burning duration is inversely proportional to white dwarf mass , we conclude from our data that the white dwarf mass may be even higher than this estimate . this places rs oph extremely close to the chandrasekhar limit . hachisu and kato also predicted a continuum driven wind phase lasting 42 days for a model with hydrogen fraction x = 0.17 . we note that the lightcurve of rs oph was extremely variable in march , both in the xmm - newton and chandra observations . we examined archival observations obtained with swift during this period , and although many of the exposures are only a few thousands seconds in length , we also find large variability in those data . however , in april the lightcurve becomes almost stable , with no large changes in brightness . if all variability phenomena observed in march are associated with the evolution of the nova ejecta , as we see in our observation on day 27 , we can conclude that mass loss turns off at the end of march , which is consistent with hachisu and katos prediction that mass loss ends on day 42 - firmly between our day 39 and day 54 observations . rs oph was detected in outburst on february 12th , 2006 . the evolution of the outburst during the first 9 months , as seen with high resolution grating spectra , can be summarized as follows . the spectra are initially dominated by h - like and he - like lines of fe xxv through n vii , produced as the nova ejecta interact with the red giant wind filling the rs oph system . the lines are blue - shifted by velocities between 500 and 1300 km s@xmath15 , and broadened by @xmath122000 km s@xmath41 . the strengths of the o and n lines are consistent with the ejecta being enhanced in these elements , as seen in many other novae . on day 27 , additional plasma emitting lines in the region 2430 appears @xmath125000 s into the observation . many of these features are consistent with lines of c vi and n vi blue - shifted by @xmath128000 km s@xmath15 . if this is the case , they could be associated with the outflow observed in the vlbi radio data obtained just a few days later . temperatures obtained from h - like to he - like line ratios indicate multiple temperatures between 10@xmath47 and 10@xmath48 k , with substantial cooling between the two observations . the system then evolves rapidly toward a softer state , as the nebular component continues to cool and the underlying white dwarf atmosphere begins to emerge . at the end of march and in april , rs oph appears as a bright supersoft source . we fitted nlte atmospheric models in order to determine some physical parameters of the white dwarf . we find log(g ) @xmath129 , and a photospheric temperature of @xmath12800,000 k. the best fits are obtained with models which have c / n abundance ratios of at most 0.1 times the solar value , indicative of cno cycle processed material on the wd surface , which may indicate that some mass is retained after the accretion - outburst cycle . the spectral changes observed from days 39 to 69 are consistent with mass loss ending at the end of march , and nuclear burning turning off at the end of april . just 8 weeks later , the bright supersoft emission from april has been replaced in june by an emission line dominated spectrum with very little continuum . many of the same lines observed in february and march are still detected , although the large velocity shifts have disappeared . the spectrum is similar to that seen in other post novae , including v382 vel and v4743 sgr , although the range of lines and inferred temperatures is different . in those systems , the emission is believed to arise in the nova ejecta , and is most likely the case for rs oph . we compared the late spectra to those of several accreting cvs , and conclude that an accretion disk is not the source of the x - ray emission . the system has still not returned to its quiescence level in x - rays , exhibiting a flux at least 10 times higher than the maximum quiescence flux measured with rosat . continued monitoring in both the x - ray and optical regimes will be crucial in understanding the behavior of this system as it begins to accrete again and return to its quiescent state . we are extremely grateful to harvey tananbaum , the director of chandra , and to norbert schartel , the esa xmm - newton project scientist , for their generous allocation of discretionary observing time . we thank the anonymous referee for their useful and insightful comments . this research was supported by nasa grant for xmm - newton research , by chandra - smithsonian award , and by an italian space agency ( asi ) grant 2006 for x - ray astronomy data analysis . the latter provided partial salary support to tn . this project was completed while tn and mo visited the kavli institute for theoretical physics ( kitp ) in santa barbara , and mo gratefully acknowledges the financial support of kitp . ccccccc 02/26/2006 15:18:49 & chandra & hetg / acis - s meg & 9912 & 13.81 & 3.34 @xmath10 2.75 @xmath0 10@xmath41 & 3.3 @xmath0 10@xmath14 ( 2.525 ) + & & heg & 9912 & & 1.63 @xmath10 2.36 @xmath0 10@xmath41 & 4.1 @xmath0 10@xmath14 ( 1.514 ) + 02/26/2006 16:51:47 & xmm - newton & epic - pn & 21411 & 13.87 & 91.64 @xmath10 6.77 @xmath0 10@xmath41 & 5.0 @xmath0 10@xmath14 ( 0.1510 kev ) + & & rgs1 & 23838 & & 1.85 @xmath10 9.35 @xmath0 10 @xmath49 & 1.6 @xmath0 10@xmath14 ( 535 ) + & & rgs2 & 23833 & & 2.81 @xmath10 1.14 @xmath0 10 @xmath41 & 1.6 @xmath0 10@xmath14 ( 535 ) + 03/10/2006 23:03:39 & xmm - newton & epic - pn & 10358 & 26.13 & 73.01 @xmath50 8.86 @xmath0 10 @xmath41 & 2.3 @xmath0 10@xmath14 ( 0.1510 kev ) + & & rgs1 & 1142 & & 1.36 @xmath10 3.54 @xmath0 10 @xmath49 & 1.6 @xmath0 10@xmath14 ( 535 ) + & & rgs2 & 1119 & & 2.63 @xmath10 4.95 @xmath0 10 @xmath41 & 1.6 @xmath0 10@xmath14 ( 535 ) + 03/24/2006 12:24:18 & chandra & letg / hrc - s & 9969 & 39.69 & 17.24 @xmath10 5.41 @xmath0 10@xmath41 & 1.7 @xmath0 10@xmath26 ( 583 ) + 04/07/2006 20:47:51 & xmm - newton & epic - pn & 3850 & 54.04 & 1947 @xmath10 0.71 & 3.0 @xmath0 10@xmath26 ( 0.1510 kev ) + & & rgs1 & 9802 & & 165 @xmath10 0.14 & 4.0 @xmath0 10@xmath26 ( 528 ) + & & rgs2 & 18640 & & 131 @xmath10 9.33 @xmath0 10 @xmath41 & 4.0 @xmath0 10@xmath26 ( 528 ) + 04/20/2006 17:23:48 & chandra & letg / hrc - s & 6523 & 66.89 & 38.95 @xmath10 9.43 @xmath51 10@xmath41 & 3.8 @xmath0 10@xmath26 ( 583 ) + 06/04/2006 12:04:54 & chandra & letg / hrc - s & 19966 & 111.67 & 0.22 @xmath10 1.57 @xmath0 10@xmath41 & 2.0 @xmath0 10@xmath27 ( 583 ) + 09/06/2006 01:41:10 & xmm - newton & epic - pn & 25552 & 205.24 & 0.756 @xmath52 5.94 @xmath0 10 @xmath49 & 1.2 @xmath0 10@xmath53 ( 0.1510 kev ) + & & rgs1 & 40257 & & 3.139 @xmath0 10@xmath41 @xmath10 1.52 @xmath0 10 @xmath49 & 9.1 @xmath0 10@xmath54 ( 535 ) + & & rgs2 & 40247 & & 4.306 @xmath0 10@xmath41 @xmath10 1.62 @xmath0 10 @xmath49 & 9.1 @xmath0 10@xmath54 ( 535 ) + 10/09/2006 23:20:16 & xmm - newton & epic - pn & 43339 & 239.14 & 0.585 @xmath52 4.00 @xmath0 10 @xmath49 & 9.1 @xmath0 10@xmath54 ( 0.1510 kev ) + & & rgs1 & 48753 & & 1.938 @xmath0 10@xmath41 @xmath10 1.28 @xmath0 10 @xmath49 & 5.38 @xmath0 10 @xmath54 ( 535 ) + & & rgs2 & 48747 & & 3.045 @xmath0 10@xmath41 @xmath10 1.39 @xmath0 10 @xmath49 & 5.38 @xmath0 10 @xmath54 ( 535 ) + s xv & meg & 7.2 & -595 @xmath10 178 & 1741 @xmath10 119 + si xiv & meg & 7.2 & -496 @xmath10 56 & 1776 @xmath10 49 + si xiii & meg & 7.0 & -632 @xmath10 45 & 1403 @xmath10 45 + mg xii & meg & 7.0 & -926 @xmath10 73 & 2022 @xmath10 36 + mg xi & meg & 6.8 & -1145 @xmath10 74 & 1918 @xmath10 99 + o viii & rgs2 & 6.5 & -1202 @xmath10 47 & 2029 @xmath10 48 + n vii & rgs2 & 6.3 & -1307 @xmath10 72 & 2108 @xmath10 73 + ccccc s xv & 2.01@xmath10 0.27 & 0.43 @xmath10 0.24 & 1.18 @xmath10 0.18 & 0.8 @xmath10 0.23 + si xiii & 1.79 @xmath10 0.09 & 0.58 @xmath10 0.07 & 1.35 @xmath10 0.07 & 1.08 @xmath10 0.07 + mg xi & 1.22@xmath10 0.21 & 0.37 @xmath10 0.19 & 0.72 @xmath10 0.12 & 0.9 @xmath10 0.26 + rrrrrr & & & & & + s & [email protected] & 1.6 @xmath0 10@xmath17 & & & + si & [email protected] & 2.0 @xmath0 10@xmath17 & & & + mg & [email protected] & 7.9 @xmath0 10@xmath18 & & & + mg & [email protected] & 8.9 @xmath0 10@xmath18 & & [email protected] & 6.3 @xmath0 10@xmath18 + o & [email protected] & 5.6 @xmath0 10@xmath18 & & & +

located in the perseus cluster , ngc 1271 is an early - type galaxy with a small effective radius of 2.2 kpc and a large bulge stellar velocity dispersion of 276 km s for its-band luminosity of . we present a mass measurement for the black hole in this compact , high - dispersion galaxy using observations from the integral field spectrograph nifs on the gemini north telescope assisted by laser guide star adaptive optics , large - scale integral field unit observations with ppak at the calar alto observatory , and _ hubble space telescope _ wfc3 imaging observations . we are able to map out the stellar kinematics both on small spatial scales , within the black hole sphere of influence , and on large scales that extend out to four times the galaxy s effective radius . we find that the galaxy is rapidly rotating and exhibits a sharp rise in the velocity dispersion . through the use of orbit - based stellar dynamical models , we determine that the black hole has a mass of and the-band stellar mass - to - light ratio is ( uncertainties ) . ngc 1271 occupies the sparsely - populated upper end of the black hole mass distribution , but is very different from the brightest cluster galaxies ( bcgs ) and giant elliptical galaxies that are expected to host the most massive black holes . interestingly , the black hole mass is an order of magnitude larger than expectations based on the galaxy s bulge luminosity , but is consistent with the mass predicted using the galaxy s bulge stellar velocity dispersion . more compact , high - dispersion galaxies need to be studied using high spatial resolution observations to securely determine black hole masses , as there could be systematic differences in the black hole scaling relations between these types of galaxies and the bcgs / giant ellipticals , thereby implying different pathways for black hole and galaxy growth .

1506.05129

over the past 15 years , it has become increasingly clear that supermassive black holes are an essential component of a galaxy , as demonstrated by the correlations connecting black hole masses and galaxy bulge properties ( e.g. , @xcite ) . supermassive black holes are thought to regulate galaxy properties and influence star formation via feedback processes @xcite , however the black hole bulge relations can also arise because of the inherent averaging associated with random galaxy mergers , without the need for the black hole to actively influence its host galaxy @xcite . roughly 80 dynamical black hole mass ( @xmath7 ) measurements have been made to date @xcite , almost exclusively through the use of high angular resolution facilities such as the _ hubble space telescope _ ( _ hst _ ) and 8@xmath810 m ground - based telescopes with the aid of adaptive optics ( ao ) . despite the growing number of black hole mass measurements , the local black hole mass census is highly incomplete . gaining a more complete picture of black hole demographics and a deeper understanding the mechanisms that drive black hole / galaxy evolution requires the secure measurement of many more black holes , particularly those at the extremes of the black hole mass scale and in a wider range of galaxy types . with the goal of finding more objects suitable for future dynamical black hole mass measurements , the het massive galaxy survey obtained long - slit spectra of @xmath91000 galaxies using the hobby - eberly telescope ( het ) at mcdonald observatory @xcite . the survey uncovered a sample of interesting early - type galaxies with small sizes and high stellar velocity dispersions for their luminosities . more quantitatively , these galaxies have an effective radius , @xmath10 , below 3 kpc , a central stellar velocity dispersion larger than @xmath11 km s@xmath0 , and @xmath1-band luminosities @xmath9@xmath12 . the het spectra hint that these compact , high - dispersion galaxies could host some of the largest black holes known ( @xmath13 ) , and that the black holes could weigh a high fraction of its host galaxy s mass . six example objects were highlighted in @xcite , and orbit - based stellar dynamical models were calculated for one galaxy , ngc 1277 . in the case of ngc 1277 , @xcite found a @xmath14 black hole that is surprisingly 59% of the galaxy s bulge mass , or 14% of the galaxy s total mass . while @xcite infer a similar @xmath7 from seeing - limited , large - scale integral field unit ( ifu ) data , @xcite show a smaller black hole of a few billion solar masses can also reasonably reproduce the observed het long - slit kinematics presented in @xcite . obtaining secure black hole mass measurements for the other compact , high - dispersion galaxies found through the het massive galaxy survey is important for addressing questions concerning the upper end of the black hole host galaxy relationships . with the present sample of black hole mass measurements , the slope , intrinsic scatter , and even the shape of the correlations for high - mass black holes are not well established ( e.g. , @xcite ) . also , the black hole mass stellar velocity dispersion relation ( @xmath7 @xmath15 ) and the black hole mass bulge luminosity relation ( @xmath7 @xmath16 ) are in direct conflict at the upper end , and make drastically different predictions for the inferred number density of the most massive black holes @xcite . not only are the compact , high - dispersion galaxies from the het survey useful for filling in the poorly sampled high - mass end of the black hole relations , but also the mass estimates from @xmath7 @xmath15 and @xmath7 @xmath16 differ by a factor of at least three . consequently , the galaxies are also useful for testing which of the correlations is more fundamental and a better predictor of @xmath7 at the high - mass end of the scaling relations . while recent progress has been made in searching for , and revising measurements for , black holes with masses larger than @xmath17 @xcite , many of these galaxies are giant ellipticals or brightest cluster galaxies ( bcgs ) , which are often large ( with @xmath18 kpc ; e.g. , @xcite ) , have cored surface brightness profiles , and are dispersion - supported showing little to no rotation . in contrast , the compact , high - dispersion galaxies found through the het survey are small , rapidly rotating , and generally exhibit cuspy surface brightness profiles . such host galaxy environments havent been extensively explored on the black hole host galaxy relationships . besides ngc 1277 , only the compact galaxies ngc 1332 @xcite , ngc 4342 @xcite , ngc 4486b @xcite , and m60-ucd1 @xcite have dynamical black hole mass measurements , with ngc 1332 and ngc 4342 being most like ngc 1277 . for these galaxies , the black hole mass measurements are in agreement with @xmath7 @xmath15 given the intrinsic scatter of the relation , but are positive outliers from the @xmath7 @xmath16 relation @xcite . we note that there are uncertainties associated with the bulge luminosity for ngc 1332 ( see @xcite for details ) and the black hole mass measurement for ngc 4486b ( see @xcite for details ) . also , tidal stripping is believed to be the cause of the over - massive black hole in the ultracompact dwarf galaxy m60-ucd1 @xcite , and there is some debate as to whether ngc 4486b and ngc 4342 have been stripped as well ( e.g. , @xcite ) . nevertheless , additional similar galaxies need to be studied because there could be systematic differences in the scaling relations between the compact , high - dispersion galaxies and the giant ellipticals / bcgs . if so , that would imply that the black holes in the two types of galaxies grew in different ways . while the compact , high - dispersion galaxies are unusual in the present - day universe , they are qualitatively similar to the typical @xmath19 quiescent galaxies , which are also small , have disk - like features , and could have high velocity dispersions @xcite . the @xmath19 red nuggets are believed to be the progenitors of the massive early - type galaxies seen today , evolved in size and mass ( e.g. , @xcite ) . thus , the compact , high - dispersion galaxies found through the het survey could provide clues to the link between local galaxies and the @xmath19 red nuggets , the ultra compact sub - mm galaxies at @xmath20 @xcite , and the early massive black holes found in @xmath21 quasars @xcite . we have begun to obtain the imaging and spectroscopic observations necessary for a detailed examination of the compact , high - dispersion galaxies from the het massive galaxy survey . this includes _ hst _ and ao - assisted ifu observations to probe the region over which the black hole dominates the galaxy s potential ( the black hole sphere of influence ; @xmath22 ) , and ifu observations that sample the large - scale stellar kinematics out to several effective radii . in this paper , we focus on measuring the mass of the black hole for the first compact , high - dispersion galaxy for which we have completed ao ifu observations . ngc 1271 has not been widely investigated in the literature and is given an uncertain sb0 classification according to the nasa / ipac extragalactic database ( ned ) . the galaxy is located within the perseus cluster , at @xmath23 , and we adopt a distance to ngc 1271 of 80 mpc , which is the hubble flow distance derived from the @xcite virgo + great attractor + shapley supercluster infall velocity field model assuming a hubble constant of @xmath24 km s@xmath0 mpc@xmath0 , a matter density of @xmath25 and a cosmological constant of @xmath26 . the sloan digital sky survey @xmath27 color is 1.6 and absolute @xmath28-band magnitude is -20.8 for the galaxy . long - slit spectra of ngc 1271 were obtained through the het massive galaxy survey , and the reported [ ] /h@xmath29 and [ ] /h@xmath30 emission - line ratios measured within a 35 aperture @xcite place ngc 1271 just within the composite galaxies section near the active galactic nuclei side of @xcite . also , @xcite do not find the presence of any broad emission lines . in section [ sec : obs ] , we describe the imaging and spectroscopic observations , including the data reduction procedures . we present the luminous mass model for the galaxy in section [ sec : stellarmassmodel ] and the stellar kinematics in section [ sec : stellarkin ] . our determination of the point - spread function ( psf ) for the spectroscopic observations is discussed in section [ sec : psf ] , and an overview of the orbit - based stellar dynamical models is given in section [ sec : models ] . in section [ sec : results ] , we present the results of the models and examine possible sources of systematic uncertainty . finally , in sections [ sec : discussion ] and [ sec : conclusion ] , we study the galaxy s orbital structure , discuss ngc 1271 within the context of the @xmath7 host galaxy relationships , and summarize our findings . for ngc 1271 , we obtained imaging observations with the _ hst _ wide - field camera 3 ( wfc3 ) in order to measure the galaxy s surface brightness distribution . we also acquired spectra with the near - infrared integral field spectrometer ( nifs ; @xcite ) on the 8.1 m gemini north telescope assisted by the alttitude conjugate adaptive optics for the infrared @xcite system . the nifs data is important for constraining @xmath7 , as it resolves the black hole sphere of influence . we also acquired large - scale spectra with the postdam multi aperture spectrograph ( pmas ; @xcite ) in the pmas fiber pack ( ppak ; @xcite ) mode at the 3.5 m telescope at calar alto observatory . although long - slit spectroscopic observations along the galaxy major axis have been previously made using the het , measuring a large - scale , two - dimensional ( 2d ) velocity field is preferable over a single slit observation for constraining the stellar mass - to - light ratio and the stellar orbital distribution . hence , we use the ppak ifu observations in place of the major - axis het measurements . below we describe the wfc3 , nifs , and ppak observations and data reduction methods . we observed ngc 1271 with _ wfc3 and the ir / f160w filter under program go-13050 . the observation was composed of three dithered full array exposures of 450 s , and four dithered subarray exposures of 1.7 s , leading to a total integration time of 1354 s. the short subarray exposures were chosen to ensure the nucleus would not become saturated . the flattened , calibrated images were corrected for geometric distortions , cleaned , and combined using astrodrizzle @xcite . since the exposures were dominated by galaxy light , we found that the standard astrodrizzle sky subtraction overestimated the background flux . we therefore manually measured the background level in each of the images . for the full array images , we measured the flux from the corners of each image , while for the subarray exposures we measured the flux difference between the sky - subtracted full frames and the subarray frames . with the background level determined , the exposures were combined to produce a super - sampled image with a spatial resolution of 006 pixel@xmath0 . not only is the _ hst _ image suitable for determining the luminous distribution on near the black hole , but due to the small size of ngc 1271 , we were also able to measure the luminosity out to larger galaxy scales , extending to @xmath95 @xmath10 ( adopting an effective radius of 56 , or 2.2 kpc , measured from a single component srsic fit to the _ hst _ f160w image ; see section [ subsec : bhrels ] ) . the nifs laser guide star ( lgs ) ao observations were acquired over three nights , on 2012 dec 27 , 2012 dec 29 , and 2013 jan 8 , in queue mode under program gn-2012b - q-51 . we used the @xmath31 filter and the @xmath1 grating with a central wavelength of 2.2 @xmath32 m to obtain spectra over a 3@xmath333 field - of - view and a spectral resolution of r@xmath95290 . we recorded 900 s exposures of the galaxy nucleus , following an object - sky - object observing sequence , totaling 3 hours of on - source integration . an @xmath34 mag star located 17 away from the galaxy was used as the tip - tilt reference . in addition , we observed the tip - tilt star to monitor the psf during each of the three nights and the a0 v stars hip 10559 and hip 22842 for telluric correction . the data were reduced using iraf tasks within the gemini / nifs package version 1.11 , utilizing the example nifs processing scripts . the reduction included sky subtraction , flat fielding , interpolation over bad pixels , cosmic - ray cleaning , and spatial rectification and wavelength calibration using ronchi mask and arc lamp exposures . the spectra were then corrected for telluric features , using an a0 v star spectrum , after interpolating over the br@xmath35 absorption line and dividing by a black body with a temperature of 9480 k. next , a data cube was produced , having @xmath36 and @xmath37 spatial dimensions , each with a scale of 005 pixel@xmath0 , and one spectral dimension , @xmath38 , using a common wavelength range and sampling for the individual science exposures . the relative spatial positions between the data cubes were determined by summing along the wavelength axis and cross - correlating the resulting flux maps . the offsets were used to align and combine the 12 individual exposures , generating the final data cube of the galaxy . data reduction of the psf star observations followed a similar procedure . the ppak observations of ngc 1271 were acquired as part of a campaign to obtain large - scale spectroscopy of the compact , high - dispersion galaxies . the observations , data reduction , and kinematic measurements for ngc 1271 will be presented in @xcite , but follow closely the ppak observations for two other compact , high - dispersion galaxies described in @xcite . for completeness , we briefly review the pertinent information below and in section [ subsec : ppakkin ] . the wide - field ifu observations were taken over three nights , from 2013 jan 4 - 6 , using the v500 grating to provide coverage of @xmath39 with a spectral resolution of r@xmath9850 at 5000 . we used three dithers to fully sample the 331 27-diameter fibers , to increase the spatial resolution of the data , and to address effects of vignetting and bad pixels . an additional 36 fibers , located 72 away from the center of the instrument field - of - view , were used to measure the sky . during each of the three nights , two 1200 s science exposures were taken at each of the three dither positions , leading to a total of 6 hours of on - source integration . the data reduction followed the procedure adopted for the calar alto legacy integral field spectroscopy area survey . the main steps included bias subtraction , flat - fielding , cosmic ray cleaning , extraction of spectra , wavelength calibration , sky subtraction , and flux calibration using spectrophotometric standard stars . spectra from the three pointings were then combined and resampled into a data cube , followed by a correction for differential atmospheric refraction . we note that the line spread function is measured as a function of wavelength and fiber position from arc lamps during the wavelength calibration step , and then homogenized to a common value prior to the extraction of the ppak kinematics . the details of these steps are discussed at length by @xcite and @xcite and we refer the reader to those publications for additional information . we generated a luminous mass model for ngc 1271 by parameterizing the _ hst _ wfc3 f160w image as the sum of 2d gaussians using the multi - gaussian expansion ( mge ) formalism @xcite . the mge method is able to reproduce a wide range of galaxy surface brightness profiles and allows for an analytic deprojection to determine the intrinsic luminosity density . here , we use the image decomposition package galfit @xcite because it takes into consideration an error map during the fit and allows for the detailed examination of model residuals , but we utilize the implementation of @xcite to determine suitable starting parameter values for the initial run with galfit . when constructing the mge model , we account for the wfc3 psf , which we adopt from @xcite . this psf was generated with tiny tim @xcite for the f160w filter at the center of the wfc3 detector assuming a g2 v spectral type , and drizzled to produce a psf with the same spatial scale as our final science image of ngc 1271 . in addition , we identified foreground objects using the program sextractor @xcite and masked these regions during the mge fit . the final mge model contained 11 components , where each gaussian was set to have the same position angle and center and the projected axis ratio ( @xmath40 ) was required to be larger than 0.25 . due to the degeneracy associated with fitting a large number of gaussians , we chose to restrict @xmath41 in order to avoid highly flattened components that place very stringent constraints on the viewing angles for which a model can be deprojected . the final mge model is an good representation of the galaxy , as can be seen in figure [ fig : mge ] . we present the final mge parameters , after correction for galactic extinction using the @xcite value given by ned , in table [ tab : mge ] . + lcccccccc 1 & & 6.074 & & 0.073 & & & & 0.25 + 2 & & 4.778 & & 0.175 & & & & 0.83 + 3 & & 4.786 & & 0.399 & & & & 0.56 + 4 & & 4.183 & & 0.765 & & & & 0.77 + 5 & & 3.172 & & 1.838 & & & & 0.66 + 6 & & 3.756 & & 2.018 & & & & 0.26 + 7 & & 3.391 & & 4.306 & & & & 0.25 + 8 & & 3.054 & & 5.857 & & & & 0.38 + 9 & & 2.474 & & 8.876 & & & & 0.49 + 10 & & 1.879 & & 12.946 & & & & 0.72 + 11 & & 0.889 & & 24.772 & & & & 0.99 from the nifs and ppak data cubes , we measured the line - of - sight velocity distribution ( losvd ) as a function of spatial location . the losvd was described using the first four gauss - hermite ( gh ) moments : the radial velocity ( @xmath42 ) , the velocity dispersion ( @xmath43 ) , and @xmath44 and @xmath45 , which describe the losvd s asymmetric and symmetric deviations from a gaussian . high signal - to - noise ( s / n ) spectra , typically @xmath46 , are required in order to reliably extract the higher order gh moments ( e.g. , * ? ? ? * ; * ? ? ? * ) , and thus we use the voronoi binning algorithm @xcite in order to construct spatial bins that optimize the balance between spatial resolution and s / n . we then used the penalized pixel fitting ( ppxf ) method of @xcite to measure the stellar kinematics in each bin . this procedure determines the best - fitting losvd by convolving with a stellar template to match the observed galaxy spectra . errors on the kinematics were determined using monte carlo simulations , in which random gaussian noise was added to the spectrum based upon the ppxf model residuals . we performed 100 realizations and from the distributions measured the standard deviation to determine 1@xmath43 uncertainties . during the monte carlo runs , the penalization term was set to zero to produce realistic errors . we measured the stellar kinematics from the three primary @xmath1-band co bandheads [ @xmath47co , @xmath48co , and @xmath49co ] in 127 spatial bins by using ppxf to fit the wavelength region between @xmath50 m . we made use of the nifs spectral template library v2.0 @xcite , which contains 28 stars observed using the nifs ifu with the @xmath1 grating and @xmath51 filter . the library includes spectral types ranging from g8 - m5 giant stars , k3 - m3 supergiants , and a g8 ii star . we first created an optimal stellar template by fitting a high s / n spectrum , constructed by adding together all spectra in the galaxy data cube . this optimal template was composed of six stars , and was dominated by m5 iii , m3 iii , and k3 iab stars that make up 37% , 26% , and 21% of the total flux , respectively . next , we measured the gh moments in each spatial bin with ppxf by keeping the relative weights of the stars that make up the template fixed , but allowing the coefficients of a second degree additive legendre polynomial and a second degree multiplicative legendre polynomial to vary . such polynomials are needed to account for differences in the optimal stellar template and the galaxy spectra shape , as the continuum of the stars in the nifs spectral template library has been previously removed . the kinematics were in good agreement with those measured when fitting a new optimal stellar template to each spatial bin and fitting only two gh moments . finally , we bi - symmetrize the kinematics using the machinery presented @xcite . since the dynamical models are only able to produce symmetric kinematics , this step is commonly performed ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) in order to reduce the noise in the observations . with the symmetrization routine , the systematic offsets in the odd gh moments , such as the galaxy s recession velocity , are removed as well . from the nifs data , we find that the galaxy is rotating quickly , with stars reaching velocities of @xmath52 km s@xmath0 . there is also a sharp peak in the velocity dispersion , which rises from 205 km s@xmath0 at a radius of @xmath91 to 396 km s@xmath0 at the center . the map of @xmath44 is anti - correlated with the velocity map , while @xmath45 shows a slight increase at the center . the s / n in each spatial bin ( measured as the ratio between the median value of the spectrum and standard deviation of the ppxf model residuals ) ranged between 33 and 96 with a median value of 66 . therefore , we were able to place excellent constraints on the kinematics , with median errors over all spatial bins of @xmath53 km s@xmath0 , @xmath54 km s@xmath0 , @xmath55 , and @xmath55 for @xmath42 , @xmath43 , @xmath44 , and @xmath45 , respectively . we present example spectra and fits with ppxf at three different locations within the nifs data cube in figure [ fig : spec_nifs ] and the bi - symmetrized nifs kinematics in table [ tab : nifskin ] . cccccccccc -0.027 & 0.022 & -8.879 & 7.817 & 396.368 & 10.793 & 0.017 & 0.016 & 0.029 & 0.017 + -0.027 & -0.028 & -38.385 & 6.554 & 394.762 & 8.844 & 0.017 & 0.014 & 0.033 & 0.018 + 0.055 & 0.022 & 37.125 & 5.608 & 395.394 & 7.818 & -0.015 & 0.012 & 0.031 & 0.015 + 0.055 & -0.028 & 11.045 & 6.073 & 392.972 & 8.368 & -0.001 & 0.012 & 0.029 & 0.014 + -0.077 & -0.003 & -45.778 & 6.763 & 392.601 & 9.207 & 0.026 & 0.014 & 0.028 & 0.017 + we measured the stellar kinematics in 268 spatial bins over a wavelength range of 4200@xmath87000 , which includes a number of absorption features such as the h@xmath30 , mg @xmath56 , and fe 5015 lines . during the fit with ppxf , we masked sky features and emission lines and included a 15th degree additive polynomial . the kinematics were extracted using the indo - u.s . library of coud feed stellar spectra @xcite , and the optimal stellar template was dominated by g9 v , g9 iii , k0 iii , and a0p stars . as a final step , we bi - symmetrize the kinematics and subtract off the systematic offsets in the odd gh moments using the procedure in @xcite . the kinematics from the ppak data were measured out to @xmath924 , or @xmath94 @xmath10 . the large - scale kinematics exhibit features that are similar to the measurements made from the high spatial resolution nifs data . in particular , the stars show rotation with velocities of @xmath57231 , a peak in the velocity dispersion to values of 297 km s@xmath0 from 102 km s@xmath0 at radius of @xmath924 , and there is a @xmath58 anti - correlation . the difference in the peak velocity dispersions measured from the ppak and nifs data can be attributed to the very different spatial resolutions of the two data sets . typical errors on the ppak kinematics for @xmath42 , @xmath43 , @xmath44 , and @xmath45 are 8 km s@xmath0 , 12 km s@xmath0 , 0.04 , and 0.05 . we present example fits to the galaxy spectra at several spatial locations in figure [ fig : specfit_ppak ] and provide the bi - symmetrized ppak kinematics in table [ tab : ppakkin ] . ngc 1271 has an uncertain sb0 classification according to ned , but we do not see an obvious bar feature in the _ hst _ f160w image or in the ppak / nifs kinematics . n - body simulations have shown that common kinematic signatures associated with bars include a `` double - hump '' feature in the rotation curve and an @xmath58 correlation over the projected length of the bar @xcite . instead , even the unsymmetrized kinematics clearly show a smooth increase in the radial velocity from the southeast side of the galaxy to the northwest side , and that @xmath44 is anti - correlated with @xmath42 , as is expected for axisymmetric systems . cccccccccc -0.063 & -0.208 & -28.540 & 4.759 & 296.999 & 6.723 & 0.010 & 0.013 & 0.027 & 0.016 + -0.063 & 0.792 & 18.003 & 4.564 & 294.131 & 6.221 & -0.013 & 0.013 & 0.025 & 0.016 + 0.937 & -0.208 & 60.969 & 4.491 & 282.870 & 5.858 & -0.030 & 0.013 & 0.035 & 0.017 + -1.063 & -0.208 & -85.825 & 4.277 & 278.884 & 5.984 & 0.043 & 0.012 & 0.036 & 0.017 + -0.063 & -1.208 & -60.969 & 4.491 & 282.870 & 5.858 & 0.030 & 0.013 & 0.035 & 0.017 + the psf of the nifs and ppak observations are important inputs into the stellar dynamical models . in order to estimate these quantities , we convolve the mge model presented in section [ sec : stellarmassmodel ] with the sum of two concentric , circular 2d gaussians in order to match the collapsed nifs and ppak data cubes . the psf is parameterized by the dispersion and relative weight of each gaussian component . we find dispersions of 016 and 043 with relative weights of 0.61 and 0.39 , respectively , for the nifs psf , while the ppak psf can be described with dispersions of 152 and 545 with relative weights of 0.82 and 0.18 . in addition , the comparison between the _ hst _ image and the collapsed data cubes allows for the center of the nifs and ppak apertures to be defined . the core of the nifs psf is larger than expected for ao observations ( e.g. , @xcite ) , which is likely the result of using a fairly faint , off - axis tip - tilt star . in section [ subsec : errorbudget ] , we test the effect of our assumed nifs psf on the inferred black hole mass by instead estimating the psf from nifs observations we acquired of the tip - tilt star itself . black hole masses are often measured from stellar kinematics by constructing dynamical models based upon the schwarzschild superposition method @xcite , and here we use the three - integral , triaxial schwarzschild code of @xcite . this technique finds a self - consistent distribution function from the observables without any assumptions about the orbital anisotropy . in the model , the black hole , the stars , and dark matter all contribute to the galaxy s gravitational potential . the stellar potential is determined by deprojecting the observed surface brightness assuming a viewing orientation and a stellar mass - to - light ratio ( @xmath59 ) that is constant with radius . we then generate a representative orbit library in the potential , and the orbits are numerically integrated while keeping track of their intrinsic and projected properties . during the modeling , the effects of the psf and aperture binning are taken into account . finally , we assign weights to each orbit such that the superposition matches the observed kinematics and total light distribution . we calculate many models varying the parameters of interest ( @xmath7 , @xmath59 , the viewing orientation parameters , and the dark matter halo parameters ) , and the best - fit model is the one with the lowest @xmath60 . for ngc 1271 , we adopt a ( nearly ) oblate axisymmetric shape , with an intermediate to long axis ratio of 0.99 . we assume axisymmetry given that ngc 1271 looks highly flattened and exhibits rapid rotation , and we do not find any evidence for kinematic twists . with this assumption , the inclination angle ( @xmath61 ) is the only viewing orientation parameter needed to describe the galaxy s intrinsic shape . in our final model , we adopt @xmath62 ( where @xmath63 corresponds to an edge - on view ) . the mge description of ngc 1271 presented in section [ sec : stellarmassmodel ] includes a couple of flat gaussian components , which can only be deprojected for angles @xmath64 . therefore , we chose to run models at a fixed inclination angle that lies midway between the extremes . our choice of inclination angle is further supported by a nuclear dust disk that appears highly inclined in an _ hst _ wfc3 f814w image . while the f814w image clearly shows a regular dust lane , the f160w image and the @xmath1-band nifs data cube used in our analysis do not appear to be significantly affected by dust . in addition , the ngc 1271 models include a spherically symmetric dark matter halo following a navarro - frenk - white ( nfw ) form @xcite . the parameters describing the nfw halo are the concentration index ( @xmath65 ) and the fraction of dark matter ( @xmath66 ) , where @xmath67 and @xmath68 is the halo virial mass and @xmath69 is the stellar mass . hence , our models have four free parameters : @xmath7 , @xmath70 , @xmath65 , and @xmath66 . we began by calculating models on a coarse grid that spanned a wide range of values before generating models on a smaller , more finely sampled grid focused around the minimum @xmath60 . ultimately , our final model grid contained 21 @xmath7 values , 31 @xmath70 values , 8 @xmath65 values , and 20 @xmath66 values with @xmath71 $ ] , @xmath72 $ ] , @xmath73 $ ] , and @xmath74 $ ] . the orbit library samples 29 equipotential shells with radii between 0003 to 100 , and 8 angular and 8 radial values at each energy . we note that triaxial orbit families ( e.g. , box orbits ) continue to be included in our orbital libraries because we are running a triaxial schwarzschild code in the axisymmetric limit . moreover , we employ a dithering method , in which 125 orbits with adjacent initial conditions are bundled together , to establish a smooth distribution function when constructing schwarzschild models . thus , the galaxy models are made with 696,000 orbits . the models were fit to the observed nifs and ppak kinematics , where 4 gh moments were measured in a total of 395 bins , resulting in 1580 observables . the results of the final model grid are summarized in figure [ fig : chi2plots ] , which shows the @xmath60 as a function of @xmath7 , @xmath70 , @xmath65 , and @xmath66 after marginalizing over the other three parameters . we find best - fit values of @xmath75 , @xmath76 , @xmath77 , and @xmath78 ( corresponding to @xmath79 ) . the comparison between the observed nifs kinematics and model predictions is shown in figure [ fig : nifs_datamodel ] and the comparison between the ppak data and model predictions is displayed in figure [ fig : ppak_datamodel ] . the model is an excellent match to the observed kinematics , and is able to reproduce the sharp increase in the velocity dispersion and the slight peak in @xmath45 at the nucleus . the @xmath60 per degree of freedom ( @xmath80 ) is 0.3 . if the unsymmetrized kinematics are used instead , the best - fit model has @xmath81 and an @xmath7 and @xmath82 that are consistent within the final uncertainties given below in section [ subsec : errorbudget ] . as can be seen in figure [ fig : chi2plots ] , we are able to place strong constraints on @xmath7 and @xmath70 . by searching for the range of @xmath7 , or @xmath70 , values that caused the minimum @xmath60 to increase by 1 and 9 , we estimate the 1@xmath43 and 3@xmath43 model fitting uncertainties . we find @xmath83 ( 1@xmath43 ) and @xmath84 ( 3@xmath43 ) , as well as @xmath85 ( 1@xmath43 ) and @xmath86 ( 3@xmath43 ) . in contrast , the dark halo parameters are not well constrained , and the @xmath65 and @xmath66 values are highly uncertain . specifically , the marginalized @xmath60 curve for @xmath65 is unconstrained at the upper end , with @xmath87 at the 3@xmath43 level . although the best - fit value for @xmath65 is at the maximum value considered in our final model grid , previous runs using coarsely sampled grids showed no sign of convergence for large values of @xmath65 . similarly , the marginalized @xmath60 curve for @xmath66 is essentially flat , and all @xmath66 values sampled by our grid ( @xmath88 ) are allowed within 3@xmath43 . the best - fit value found for @xmath66 is the minimum value considered in our final grid , but we also ran models without a dark halo and found a significantly worse fit , such that the @xmath60 increased by 123 relative to the best - fit model with a dark halo . thus , a dark halo is required to match the observed kinematics , but its properties can not be pinned down with the current datasets and we do not present any further details associated with the dark halo parameters . the formal model fitting uncertainties quoted in the previous section are the statistical errors associated with the dynamical models , however systematic effects can have a significant impact on the inferred black hole mass and mass - to - light ratio . in this section , we evaluate some common sources of uncertainty that are not already incorporated into the statistical errors , such as those associated with the assumed form of the dark matter halo , the adopted inclination angle , the number of orbits used in the models , the nifs psf model , details associated with the extraction of the nifs kinematics , and the symmetrization of the input kinematics . _ dark matter halo _ : previous work has clearly shown that the dark halo could be an important component in the stellar dynamical models due to the degeneracy between the dark halo and stellar mass - to - light ratio , which in turn is also degenerate with the black hole ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? if @xmath89 is spatially well resolved , then the degeneracy between the black hole and mass - to - light ratio can be mitigated , and the exclusion of a dark halo will have minimal impact on the inferred @xmath7 . in contrast , if @xmath89 is not very well resolved and a dark halo is not included in the stellar dynamical models , then @xmath59 will be artificially elevated to account for the missing mass at large radii . since @xmath59 is taken to be constant with radius , a smaller @xmath7 is then required to fit the observed central kinematics . when generating a grid of models without a dark halo for ngc 1271 , and assuming @xmath62 to match our fiducial model presented in section [ sec : results ] , we measure @xmath90 and @xmath91 . in other words , the inferred @xmath7 is underestimated by a factor of @xmath92 when a dark halo is excluded from the modeling . we note that when fitting dynamical models without a dark halo to only the small - scale nifs kinematics , we recover very similar parameter values as those found from our fiducial model , with @xmath92 and @xmath93 . moreover , we tested how the assumed shape of the dark halo affects @xmath7 . our fiducial model was calculated assuming a spherical nfw halo , but another common form is a halo with a cored logarithmic potential @xcite , given by @xmath94 the parameters @xmath95 and @xmath96 are the asymptotic circular velocity and radius within which the dark matter density is constant . thus , the halo from a logarithmic potential yields smaller densities at small radii compared the the nfw halo . we constructed models with a dark halo from a cored logarithmic potential , sampling @xmath97 , @xmath98 @xmath99 , @xmath100 km s@xmath0 , and @xmath101 kpc . we recovered similar results to the models with an nfw halo , namely that @xmath102 and @xmath103 , corresponding to a 7% increase in @xmath7 and a 4% decrease in @xmath82 compared to the fiducial model . although the black hole mass is sensitive to the inclusion of a dark halo in the stellar dynamical models , the form of the halo has a small impact on @xmath7 . for this reason , being unable to constrain the dark halo parameters is not a concern for the purposes of this paper , as long as reasonable halos are sampled over when constructing the orbit - based models . _ inclination angle _ : all of the models presented in this paper assume an axisymmetric shape with an inclination angle of @xmath62 . however , we also ran a grid of models that sampled 13 inclination angles from @xmath104 to @xmath105 . this corresponds to the range of angles for which the mge model in section [ sec : stellarmassmodel ] can be deprojected . it is computationally expensive to calculate a model grid that samples over @xmath7 , @xmath82 , @xmath61 , @xmath65 , and @xmath66 simultaneously , so we instead varied the first three parameters while sampling over five nfw halos . the five dark halos were those with the lowest @xmath60 from the model grid at the beginning of section [ sec : results ] , and the halos span a range of @xmath65 and @xmath66 values ( @xmath106 and @xmath107 ) . from this test , we determine that @xmath108 , which is within 10% of the best - fit value in section [ sec : results ] , and @xmath109 , which is the same as the best - fit value in section [ sec : results ] . moreover , the best - fit inclination angle was @xmath110 , however the angle was not well constrained . all angles between @xmath111 and @xmath105 were allowed within the 3@xmath43 statistical uncertainties . such behavior is not surprising , and other stellar dynamical work have also found it difficult to infer the inclination angle from 2d line - of - sight kinematics @xcite . _ number of orbits _ : the fiducial model presented in section [ sec : results ] was calculated using orbits that covered 29 equipotential shells with 8 angular and 8 radial values at each energy . when accounting for the orbital dithering , this translates into a total of 696,000 orbits . we also tested the effect on @xmath7 and @xmath82 when the number of orbits is about doubled , such that 37 equipotential shells with 10 angular and 10 radial values at each energy are used . again 125 orbits with adjacent starting positions were bundled together , resulting in 1,387,500 orbits . due to a large increase in computational time for a single model , we constructed a model grid that samples over @xmath7 , @xmath82 , and the top five nfw halos from section [ sec : results ] . we found no change in the best - fit @xmath7 or @xmath59 values compared to the fiducial model . _ nifs psf _ : the nifs psf was measured by comparing the mge model of the _ hst _ image to the collapsed nifs data cube . while this approach is commonly used in black hole mass measurement work ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) , estimation of the psf from ao observations is notoriously difficult due to the constantly changing quality of the ao correction and the combination of data cubes from multiple nights . therefore , we also estimated the psf in a different manner in order to assess how strongly the adopted nifs psf affects @xmath7 . utilizing the nifs observations of the tip - tilt star , we fit the sum of four concentric , circular 2d gaussians to the collapsed nifs data cube using galfit . we found that the psf is best described by gaussians with dispersions of 004 , 008 , 021 , and 043 with relative weights of 0.08 , 0.36 , 0.26 , and 0.30 . a four - component psf model provided a significantly better fit to the collapsed data cube than a simpler two - component model , and the residuals between the four gaussian model and the data have a standard deviation of just 9% out to a radius of 1 . the psf is typical of what one would expect from the gemini ao system ( e.g. , * ? ? ? * ; * ? ? ? * ) , and the quality of the ao correction is better than that implied by the psf adopted in the fiducial model . this is not completely surprising as the nifs observations of the star used the star itself for tip - tilt corrections , whereas the observations of the galaxy were made off - axis , using the star for guiding . nonetheless , calculating stellar dynamical models using this better psf and comparing to the results using the poorer psf should cover the range of possible black hole masses due to the uncertainty in the nifs psf . we calculated models using the new four - gaussian psf and further assumed that the center of the nifs spaxel with the largest flux coincides with the galaxy nucleus . we varied @xmath7 and @xmath82 , while sampling over the top five nfw halos from section [ sec : results ] , and found @xmath112 and @xmath113 . therefore , the black hole mass and mass - to - light ratio change by 10% and 4% compared to the fiducial values in section [ sec : results ] . _ measuring the nifs kinematics : _ we measured the nifs kinematics with ppxf using a second degree additive legendre polynomial and a second degree multiplicative polynomial to correct for shape differences between the losvd - broadened optimal stellar template and the observed galaxy spectrum . we selected this continuum correction because it was one of the simplest models that still provided a good fit to the data , and produced kinematics that were in good agreement with those measured using combinations of degree @xmath114 additive / multiplicative polynomials , with the exception of the lowest order polynomials . since the continuum has been previously divided out of the the spectral templates in the nifs library , but not removed from the observed galaxy spectra , using the lowest order polynomials ( combinations of degree @xmath115 additive / multiplicative polynomials ) produced visibly poor spectral fits . we also constructed a new library containing stellar templates that have not been continuum - corrected and thus have a very similar shape as the galaxy spectra . we retrieved @xmath1-band nifs observations from the gemini archive of stars that are part of the nifs spectral template library and two stars observed under program gn-2010a - q-112 . these twelve stars are k0 - m5 giants , a k5 supergiant , and an m0 supergiant . we reduced the observations following the main procedure outlined in section [ subsec : obsnifs ] , with the additional steps of extracting a one - dimensional spectrum , rebinning to a common wavelength range and sampling , and shifting the stars to rest . with this new template library , we are able to obtain good fits to the galaxy spectra using low - order polynomials with ppxf . in order to examine possible effects on @xmath7 and @xmath82 due to uncertainties associated with the choice of the ppxf polynomial degree , we fit dynamical models to the nifs kinematics extracted using the new stellar template library and an additive constant , along with the ppak kinematics presented in section [ subsec : ppakkin ] . we sampled over @xmath7 , @xmath82 , and the top five nfw halos from section [ sec : results ] , finding @xmath116 and @xmath113 . this corresponds to a change of 20% and 4% from the best - fit black hole mass and mass - to - ratio from the fiducial model . these results are likely representative of a maximum change in best - fit parameter values , as using this particular continuum correction with the new stellar template library resulted in the largest number of bins with kinematics inconsistent at the 1@xmath43 level ( all bins were consistent at the 2@xmath43 level ) compared to the nifs kinematics from section [ subsec : nifskin ] . _ symmetrizing the kinematics : _ the nifs and ppak kinematics were bi - symmetrized prior to using them as inputs into the stellar dynamical modeling code . we used the method outlined in @xcite , which averages the measurements of a gh moment in a four - fold symmetric manner around the minor and major axes . during the averaging of the measurements for a single gh moment , the kinematic error of the bin and the fraction of flux a spaxel contributes to that bin are taken into account . such modifications to the input kinematics and their errors are a common way in which to reduce observational noise , and often kinematics are bi - symmetrized in order to obtain reasonable results from an axisymmetric modeling code and are point - symmterized for use with a triaxial modeling code . however , symmterization routines are never perfect , as discussed in @xcite . we therefore tested how symmetrization affects @xmath7 and @xmath82 by running additional models with kinematics that were point - symmetrized . we find that the black hole mass increases to @xmath117 , or by 20% of the fiducial value , and the @xmath4-band mass - to - light ratio decreases to @xmath118 , or 4% of the fiducial value from section [ sec : results ] . _ summary : _ we derive the final range of black hole masses and mass - to - light ratios for ngc 1271 by adding in quadrature the formal model fitting 1@xmath43 uncertainty from section [ sec : results ] and the additional sources of systematic uncertainty above . ultimately , we determine that @xmath119 and @xmath120 . the dominant source of systematics for ngc 1271 are those associated with the continuum - correction model used to extract the nifs kinematics and the symmetrization of the kinematics . both affect @xmath7 and @xmath82 at the 20% and 4% level , respectively . in addition to examining possible sources of systematic uncertainty and incorporating the effects into the final error budget , we ran other tests to assess the robustness of the ngc 1271 @xmath7 measurement . we describe these tests below . _ nuclear dust disk : _ a dust disk is present at the center of ngc 1271 , and is visible in the f814w wfc3 image . however , dust does nt appear to be significant in the near - infrared imaging and in the nifs data cube . we tested constructing a new mge of the f160w image after excluding the dust disk by using the f814w image as guide for generating the mask . we ran dynamical models using the modified mge , following the procedure in section [ subsec : app_to_n1271 ] and sampling over the top five nfw halos from section [ sec : results ] . we found best - fit values of @xmath121 and @xmath103 , which is well within the final uncertainties adopted for ngc 1271 discussed in section [ subsec : errorbudget ] . _ stellar mass - to - light ratio variation : _ our dynamical models assume that @xmath82 remains constant with radius . in order to determine whether there is an obvious change in stellar population , we generated an mge of the f814w image following the methods described in section [ sec : stellarmassmodel ] . during the fit , we masked out the nuclear dust disk and foreground objects in the f814w image , and we accounted for the psf using a bright , isolated star in the image . from the mge models of the f814w and f160w images , we do nt see evidence for color gradients , finding that the color changes by at most 0.16 mag from 02 to 24 . although ngc 1271 exhibits a fairly uniform color , we further examined dynamical models that are fit to only the nifs kinematics . given the limited radial extent of the nifs kinematics , which extend out to a radius of @xmath91 , or @xmath9390 pc , systematics associated with mass - to - light ratio gradients ( and dark matter halos ) are mitigated . when fitting to only the nifs kinematics , we recover consistent results to those presented in [ subsec : errorbudget ] , where @xmath122 and @xmath123 ( 1@xmath43 uncertainties ) . we show contours of @xmath60 as a function of black hole mass and mass - to - light ratio for this nifs - only model grid in figure [ fig : nifsonlygrid ] . _ ppak kinematics : _ when measuring the stellar kinematics from the ppak data , we masked out spectral regions containing possible emission lines and sky residuals , as can be seen in figure [ fig : specfit_ppak ] . in order to verify that our choice of a spectral mask does not bias the kinematic measurements and inferred black hole mass , we decreased the number and width of the excluded wavelength regions . we ran dynamical models using the modified ppak kinematics while sampling over the top five dark matter halos in section [ sec : results ] . we measured @xmath124 and @xmath103 , which is within the final uncertainties given for ngc 1271 in section [ subsec : errorbudget ] . also , the observed velocity dispersion of the ppak kinematics presented in section [ subsec : ppakkin ] drops below the instrumental resolution in many of the spatial bins located @xmath12510away from the nucleus . while care was taken to homogenize the line spread function to a common value such that there was no variation with wavelength or fiber position , measuring dispersions well below the instrumental resolution is a difficult task . we therefore also tested the effect on @xmath7 and @xmath126 when excluding the spatial bins in which the dispersion was below @xmath127 km s@xmath0 . when calculating dynamical models that fit to the adjusted ppak kinematics and that sample over the top five dark matter halos in section [ sec : results ] , we find no change from the best - fit values presented in section [ sec : results ] . ngc 1271 harbors a black hole with @xmath128 and has a stellar mass - to - light ratio of @xmath129 . we note that the final uncertainty on the black hole mass we use is comparable to the formal 3@xmath43 statistical uncertainty . some @xcite have suggested 3@xmath43 statistical errors should be used in place of 1@xmath43 errors as a conservative way in which to account for the effect of unknown systematics on @xmath7 . with a black hole mass of @xmath130 and adopting 276 km s@xmath0 for the bulge stellar velocity dispersion ( see section [ subsec : bhrels ] ) , @xmath131 044 . thus , the nifs observations have resolved the black hole sphere of influence . below we discuss the galaxy s orbital structure and place the galaxy on the @xmath7@xmath8host galaxy relations . in addition to determining the mass of the black hole in ngc 1271 , the schwarzschild models provide information about the galaxy s orbital structure . using our best - fit model in section [ sec : results ] , we show the ratio @xmath132 as a function of radius in figure [ fig : orbstructure ] . the tangential velocity dispersion is defined as @xmath133 , and @xmath134 are the usual spherical coordinates . we find that ngc 1271 is roughly isotropic at all radii covered by our kinematic measurements , deviating by at most 30% from @xmath135 , but we observe a trend in which @xmath132 declines at radii outside the black hole sphere of influence . as expected , short - axis tube orbits dominate in this oblate system , making up more than 85% of the orbits at all radii . long - axis tube orbits , which are important for triaxial and prolate systems , are negligible , while the fraction of box orbits increases at small radii but still make up only 15% of the orbits near the nucleus . furthermore , we use our best - fit stellar dynamical model to examine the mass distribution as a function of average radius , @xmath136 , and spin , @xmath137 , of the orbits , where @xmath138 . here , @xmath139 is the average angular momentum along the @xmath140-direction and @xmath141 is the average second moment of the orbit . ngc 1271 shows several dynamical components , as can be seen in figure [ fig : massdens ] , including a clear non - rotating bulge ( with @xmath142 ) , a highly co - rotating component ( with @xmath1430.5 ) , and a maximally co - rotating component ( with @xmath144 ) . the bulge component accounts for 12% of the mass , whereas the rotating components total 75% of the mass within the radial extent of the kinematic measurements . qualitatively these components agree with the classification of ngc 1271 as a fast - rotating s0 galaxy with a classical bulge . although the dynamical decomposition from the best - fit stellar dynamical model presented above can be used to place ngc 1271 on the @xmath7 bulge relationships , we follow the more common approach of carrying out a photometric decomposition to determine the galaxy s bulge luminosity and bulge effective radius . using galfit , we find a single srsic component fit is an insufficient description of the galaxy , with the percent difference between the model and data reaching as high as 60% . the fit is significantly improved with the addition of one or two other srsic components , and in the later case the percent difference between the model and data is under 15% . in table [ tab : galfitresults ] , we present the best - fit parameters of galfit models with one , two , and three srsic components , as well as the f160w luminosity for each component . the three - component model provides the best match to the _ hst _ image , but it is difficult to unambiguously identify a `` bulge '' component because the components all have rather low srsic indices . therefore , we conservatively assume that the innermost component of the three - component model provides a lower limit on the bulge luminosity and effective radius , while the innermost component of the two - component model gives an upper limit . this yields a @xmath1-band bulge luminosity of @xmath145 when correcting for galactic extinction using the @xcite wfc3 f160w value of @xmath146 , assuming an @xmath147 color of @xmath148 @xcite , and a @xmath1-band solar absolute magnitude of @xmath149 . alternatively , the bulge has a mass of @xmath150 when applying the best - fit mass - to - light ratio from our dynamical models to the luminosities in table [ tab : galfitresults ] . the corresponding bulge effective radius ranges between 06 and 52 ( or @xmath151 kpc ) . ccccccccccc 1 & & 10.72 & & 7.7@xmath152 & & 5.56 & & 4.78 & & 0.41 + + 1 & & 10.95 & & 6.2@xmath152 & & 5.23 & & 6.54 & & 0.54 + 2 & & 12.47 & & 1.5@xmath152 & & 6.63 & & 0.93 & & 0.20 + + 1 & & 12.39 & & 1.6@xmath152 & & 0.61 & & 2.12 & & 0.68 + 2 & & 11.80 & & 2.8@xmath152 & & 5.24 & & 1.17 & & 0.27 + 3 & & 11.90 & & 2.6@xmath152 & & 10.46 & & 1.30 & & 0.62 in order to determine the bulge stellar velocity dispersion for ngc 1271 , we use the best - fit stellar dynamical model from section [ sec : results ] and predict the luminosity - weighted second moment within a circular aperture whose radius equals the galaxy s bulge effective radius , following the approach used by @xcite . due to the uncertainty in the bulge effective radius for ngc 1271 , we measure the effective stellar velocity dispersion ( @xmath153 ) for three different bulge effective radii corresponding to the largest @xmath10 estimate from the galfit decomposition , the smallest estimate of @xmath10 , and the average of the two . additionally , some previous black hole studies have chosen to exclude data within @xmath89 when determining @xmath153 because the stellar kinematics are under the direct influence of the black hole in this region ( e.g. , @xcite ) . when excluding the region within @xmath89 , we find effective stellar velocity dispersions of @xmath154 km s@xmath0 , @xmath155 km s@xmath0 , and @xmath156 km s@xmath0 , whereas when the region within @xmath89 is included we measure @xmath157 km s@xmath0 , @xmath158 km s@xmath0 , and @xmath159 km s@xmath0 , for bulge effective radii of 52 , 29 , and 06 , respectively . as a comparison , the het massive galaxy survey reports a central velocity dispersion of 317 km s@xmath0 @xcite , which is the observed stellar velocity dispersion within a 35 aperture from the major axis long - slit data . figure [ fig : bhrels ] shows the location of ngc 1271 on the most recent versions of the @xmath160 and @xmath161 relations by @xcite . for the purposes of placing ngc 1271 on the @xmath7 @xmath15 correlation , we use @xmath162 km s@xmath0 with uncertainties that include the @xmath163 measurements made for bulge effective radii of 52 and 06 when excluding data within @xmath89 . when placing ngc 1271 on the @xmath7 @xmath16 relation , we set the faint end of the bulge luminosity error bar assuming the luminosity of the innermost component of the three - component srsic fit to the _ hst _ image and the high end of the error bar assuming the luminosity of the innermost component of the two - component srsic fit . we adopt a @xmath1-band bulge luminosity of @xmath164 , which is the midpoint of the range of possible bulge luminosities . we find that ngc 1271 consistent with the @xmath7 @xmath15 relation , but is an order of magnitude above the black hole mass prediction from the @xmath7 @xmath16 correlation . in order to demonstrate that the black hole in ngc 1271 must be larger than that expected from @xmath165 , in figure [ fig : comparemodels ] we present the nifs observations of the velocity dispersion and @xmath45 along with the predictions from the best - fitting model with a black hole mass of @xmath166 and a @xmath167 black hole . the @xmath167 black hole is expected from @xmath7 @xmath16 when conservatively using the galaxy s total @xmath1-band luminosity of @xmath168 , which is derived from the single srsic fit to the _ hst _ image , after correcting for galactic extinction and assuming a @xmath169 . clear differences between the kinematic predictions and the observations can be seen by eye . the best - fit model with @xmath170 is able to nicely reproduce the sharp rise in the velocity dispersion and the slight peak in @xmath45 at the nucleus , while the less massive black hole predicted from @xmath7 @xmath16 fails to do so . ngc 1271 has an apparent ellipticity of @xmath171 and a specific stellar angular momentum of @xmath172 within one effective radius based on the ppak data . here , @xmath173 , @xmath42 , and @xmath43 are the radius , velocity , and velocity dispersion and the brackets denote a luminosity weighted average @xcite . using the dividing line between slow and fast rotators from the atlas@xmath174 survey , such that fast rotators have @xmath175 @xcite , ngc 1271 falls well within in this fast rotator regime . ngc 1271 appears similar to the other compact galaxies ngc 1277 , ngc 1332 , ngc 4342 , ngc 4486b , and m60-ucd1 . all six of these early - type galaxies have small sizes , are rotating , show large stellar velocity dispersions for their luminosities , and have black holes that are too massive for their host galaxy s bulge luminosity . the black holes , however , are consistent with @xmath7 @xmath15 given the intrinsic scatter of the relation . in the case of m60-ucd1 , @xcite suggest that the ultracompact dwarf galaxy ( ucd ) was once the nucleus of a larger galaxy that has since been tidally stripped by the giant elliptical m60 , whose center lies at a projected distance of just 6.6 kpc away from the ucd . while tidal stripping is a natural explanation for the presence of an over - massive black hole , in the case of ngc 1271 , we do not see signs of active stripping in the _ hst _ image . the isophotes appear extremely regular , and no massive galaxies immediately neighbor ngc 1271 like in the case of m60-ucd1 . ngc 1271 is @xmath9270 kpc in projection from the bcg of perseus . further evidence could come from counting the number of globular clusters , as the galaxy would be stripped of its globular clusters first . while ngc 1271 appears not to have been stripped with our current data , we can not rule out an event in the distant past . interestingly , the behavior of the compact , high - dispersion galaxies being consistent with @xmath7 @xmath15 but being large positive outliers on @xmath7 @xmath16 could be in conflict with recent observations of bcgs , which instead may hint that black hole mass becomes independent of @xmath15 at high black hole mass while the @xmath7 @xmath16 correlation remains unchanged at large luminosities @xcite . clearly , more compact , high - dispersion galaxies and bcgs / giant ellipticals need to be examined . there could be systematic differences in the scaling relations between the two types of galaxies , thereby imply different mechanisms for black hole growth . since the compact , high - dispersion galaxies like ngc 1271 look similar to the quiescent @xmath19 red nuggets , they could be relics that somehow avoided the same fate that ultimately produced the giant ellipticals observed today . perhaps the compact , high - dispersion galaxies are left over from an era when the local black hole scaling relations did not apply and galaxies instead contained over - massive black holes . to summarize , we obtained ao - assisted gemini nifs observations of ngc 1271 to map out the stellar kinematics on scales comparable to the black hole sphere of influence , and large - scale ifu data with ppak , which are useful for constraining the galaxy s stellar mass - to - light ratio and orbital distribution . using an _ hst _ wfc3 @xmath4-band image along with the spectral information , we constructed orbit - based stellar dynamical models . we measure @xmath176 and @xmath177 . the quoted errors combine the 1@xmath43 model fitting uncertainties with some common sources of systematic uncertainty that affect stellar dynamical models . the black hole in ngc 1271 is at the upper end of the black hole mass distribution ( @xmath178 ) . yet , this compact , rapidly rotating galaxy , with a high stellar velocity dispersion for its luminosity is very different from the giant elliptical galaxies and bcgs that are expected to harbor the most massive black holes in the universe . such host galaxy environments have yet to be widely explored on the @xmath7 host galaxy relations . with our mass measurement , we find that the black hole is too large for the galaxy s @xmath1-band bulge luminosity of @xmath179 , falling an order of magnitude above the expectation from the @xmath7 @xmath16 correlation , but the black hole mass is consistent with expectations from the @xmath7 @xmath15 relationship assuming @xmath180 km s@xmath0 . this behavior has also been observed in the few other compact galaxies that have dynamical black hole mass measurements to date . carrying out more black hole mass measurements in similar galaxies using high spatial resolution observations from _ hst _ and ao is necessary in order to determine if there are systematic differences in the black hole scaling relations between the large ellipticals / bcgs and these compact , high - dispersion galaxies . the compact , high - dispersion galaxies could be remnants of the @xmath19 red nugets that for some reason did not evolve into the largest ellipticals observed today , and instead reflect a time when black holes were too large for their bulges . more broadly , additional black hole mass measurements are needed in order to enlarge and better fill in undersampled regions of galaxy parameter space . targeting such a large and carefully selected sample with high spatial resolution facilities is a natural step toward gaining a more complete census of local black holes and a better understanding of the role that black holes play in galaxy evolution . j. l. w. has been supported by an nsf astronomy and astrophysics postdoctoral fellowship under award no . 1102845 . based on observations obtained at the gemini observatory , which 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 national research council ( canada ) , conicyt ( chile ) , the australian research council ( australia ) , ministrio da cincia , tecnologia e inovao ( brazil ) and ministerio de ciencia , tecnologa e innovacin productiva ( argentina ) , under program gn-2012b - q-51 . also based on observations made with the nasa / esa hubble space telescope , obtained at the space telescope science institute , which is operated by the association of universities for research in astronomy , inc . , under nasa contract nas 5 - 26555 . these observations are associated with program # 13050 . this work is further based on observations collected at the centro astronmico hispano alemn ( caha ) at calar alto , operated jointly by the max - planck institut fr astronomie and the instituto de astrofsica de andaluca ( csic ) . the authors acknowledge the texas advanced computing center ( tacc ; http://www.tacc.utexas.edu ) at the university of texas at austin for providing hpc resources that have contributed to the research results reported within this paper . the authors also made use of the grant - funded cyberinfrastructure at indiana university . this material is based upon work supported by the national science foundation under grant no . cns-0723054 , and in part by lilly endowment , inc . , through its support for the indiana university pervasive technology institute , and in part by the indiana metacyt initiative . the indiana metacyt initiative at indiana university is also supported in part by lilly endowment , inc . this research has made use of the nasa / ipac extragalactic database which is operated by the jet propulsion laboratory , california institute of technology , under contract with nasa .

in our model , spinless fermions ( or hardcore bosons ) on a square lattice hop to nearest neighbor sites , and also experience a hard - core repulsion at the nearest neighbor separation . this is the simplest model of correlated electrons and is more tractable for exact diagonalization than the hubbard model . we study systematically the dilute limit of this model by a combination of analytical and several numerical approaches : the two - particle problem using lattice green functions and the t - matrix , the few - fermion problem using a modified t - matrix ( demonstrating that the interaction energy is well captured by pairwise terms ) , and for bosons the fitting of the energy as a function of density to schick s analytical result for dilute hard disks . we present the first systematic study for a strongly - interacting lattice model of the t - matrix , which appears as the central object in older theories of the existence of a two - dimensional fermi liquid for dilute fermions with strong interactions . for our model , we can ( lanczos ) diagonalize the system at all fillings and the system with four particles , thus going far beyond previous diagonalization works on the hubbard model .

cond-mat0207571

since the discovery of the high - temperature superconductors in 1986 , there has been intense study of a number of two - dimensional models that are believed to model the electronic properties of the cuo@xmath2 plane of the cuprate superconductors , for example , the hubbard model , the @xmath3 model , and the heisenberg model.@xcite two - dimensional quantum models with short - range kinetic and interaction terms are difficult to study . in one dimension , there are exact solutions using the bethe ansatz and a host of related analytical techniques,@xcite and there is a very accurate numerical method , the density - matrix renormalization group ( dmrg),@xcite that can be applied to large systems relatively easily . in two dimensions , on the other hand , there are few exact solutions ( one famous nontrivial case is the hubbard model with one hole in a half - filled background , the nagaoka state@xcite ) , and current numerical methods are not satisfactory ( quantum monte carlo is plagued by the negative sign problem@xcite at low temperatures and at many fillings of interest and the dmrg in two dimensions@xcite is still in early development stage ) . the most reliable method for studying complicated quantum systems is exact diagonalization , which means enumerating all basis states and diagonalizing the resulting hamiltonian matrix . of course , this method is computationally limited by the growth of the hilbert space which is in general exponential in the number of particles and the lattice size . the @xmath4 hubbard model with 16 electrons , 8 spin - up and 8 spin - down , after reduction by particle conservation , translation , and the symmetries of the square , has 1,310,242 states in the largest matrix block,@xcite and can be diagonalized using the well - known lanczos method . the hubbard model has been diagonalized for the @xmath4 lattice ( see e.g. , ref . ) , and at low filling ( four electrons ) for @xmath5@xcite with extensive employment of symmetries . we have asked the question : _ is there a model that contains the basic ingredient of short - range hopping and interaction but is simpler , in the exact diagonalization sense , than the hubbard model ? _ the answer is yes : we can neglect the spin . we obtain the following hamiltonian for _ spinless _ fermions , @xmath6 where @xmath7 and @xmath8 are spinless fermion creation and annihilation operators at site @xmath9 , @xmath10 the number operator , @xmath11 the nearest - neighbor hopping amplitude , and @xmath12 the nearest - neighbor interaction . note that with spinless fermions , there can be at the most one particle per site ; no on - site interaction ( as that in the hubbard model ) is possible , and we have included in our hamiltonian nearest - neighbor repulsion . the spinless fermion model , eq . ( [ eq - ham ] ) , is a two - state model , and the number of basis states for a @xmath13-site system is @xmath14 , which is a significant reduction from the @xmath15 of the hubbard model . we can further reduce the number of basis states by taking the nearest - neighbor interaction @xmath16 , i.e. , infinite repulsion , which excludes nearest neighbors , giving roughly @xmath17 states . the spinless fermion model with infinite repulsion eq . ( [ eq - ham ] ) contains a significant reduction of the hilbert space . after using particle conservation and translation symmetry ( but not point group symmetry ) , the largest matrix for the @xmath0 system has @xmath18 states ( for 11 particles ) , and we can therefore compute for all fillings the @xmath0 system whereas for the hubbard model @xmath4 is basically the limit . this of course means that for certain limits we can also go much further than the hubbard model , for example , we can handle four particles on a @xmath1 lattice where the number of basis states is @xmath19 . this extended capability with our model has enabled us to obtain a number of results that are difficult to obtain with the hubbard model . an added feature of our model is that the basis set for the spinless fermion problem is identical to that for the hardcore boson problem , because with hardcore repulsion , there can be at the most one boson at one site also . therefore , without computational difficulty , we can study numerically both the spinless fermion and hardcore boson problem . spinless fermions can also be realized in experiments , for example , the spin polarized @xmath20he due to a strong magnetic field , or ferro or ferri - magnetic electronic systems where one spin - band is filled . the one - dimensional spinless fermion model with finite repulsion is solved exactly using bethe ansatz.@xcite the infinite - dimensional problem is studied in ref . . a very different approach using the renormalization group for fermions is done in ref . . a monte carlo study of the two - dimensional model at half - filling only and low temperatures is in ref . , which , dating back to 1985 , is one of the earliest quantum monte carlo simulations for fermions . ( it is no coincidence that they chose the model with the smallest hilbert space . ) considering the tremendous effort that has been devoted to the hubbard model and the close resemblance of our model , eq . ( [ eq - ham ] ) , to the hubbard model , it is surprising that works on this spinless model have been rather sparse , though it has been commented that the spinless model offers considerable simplifications.@xcite this paper is one of the two that we are publishing to study systematically the two - dimensional spinless fermion and hardcore boson model with infinite nearest - neighbor repulsion . the present paper focuses on the dilute limit , treating the problem of a few particles , and the other paper@xcite will focus on the dense limit , near half - filled,@xcite where stripes ( that are holes lining up across the lattice ) are natural objects ( see ref . for a condensed study of stripes in this model ) . we will use lanczos exact diagonalization , exploiting the much - reduced hilbert space of our model , and a number of analytical techniques , for example , in this paper , lattice green functions and the t - matrix . one of the goals of these two papers is to advertise this model of spinless fermions to the strongly - correlated electron community , as we believe that it is the simplest model of correlated fermions and deserves more research effort and better understanding . the prior work most comparable to ours may be the studies of four spinless electrons in a @xmath5 lattice , with coulomb repulsion , by pichard _ et al _ ; @xcite their motivation was the wigner crystal melting and the competition of coulomb interactions with anderson localization when a disorder potential is turned on . at the dilute limit of our model , the scattering t - matrix is of fundamental importance . for two particles , we expect that , at least when the potential @xmath12 is small , we can write a perturbative equation for energy , @xmath21 which is to say that the exact interacting energy of two particles is the noninteracting energy @xmath22 , for a pair of momenta @xmath23 and @xmath24 , plus a correction term @xmath25 due to the interaction @xmath12 . and with more than two particles , at least when the particle density is low , we expect to have @xmath26 eq . ( [ eq - perturbmany ] ) is central in fermi liquid theory , where it is justified by the so - called `` adiabatic continuation '' idea , which says that interacting fermion states correspond one - to - one to noninteracting ones as we slowly switch on a potential . in the boson case , because many bosons can occupy one quantum mechanical state and form a condensate , eq . ( [ eq - perturbmany ] ) should be modified , but with only two bosons , we expect eq . ( [ eq - perturb ] ) should be valid ( in that the correction vanishes in the dilute limit ) . ( [ eq - perturb ] ) and ( [ eq - perturbmany ] ) are used when we look at a list of noninteracting energies and draw correspondences with the interacting energies , the energy shift being packaged in the term @xmath25 . one possible objection to the above formulas ( eqs . ( [ eq - perturb ] ) and ( [ eq - perturbmany ] ) ) is that they appear to be perturbative , yet the interaction potential in our problem is infinitely strong , so the first - order ( first born approximation ) scattering amplitude , being proportional to the potential , is infinite too . however , this singular potential scattering problem ( e.g. , hard - sphere interaction in 3d ) has been solved ( see ref . ) by replacing the potential with the so - called scattering length , which is finite even when the potential is infinite . as we review in appendix [ sec - physical ] , a perturbation series ( born series ) can be written down ( that corresponds to a series of the so - called ladder diagrams ) and even though each term is proportional to the potential , the sum of all terms ( the t - matrix , @xmath25 in eqs . ( [ eq - perturb ] ) and ( [ eq - perturbmany ] ) ) is finite . because the t - matrix captures two - body interaction effects , it is the centerpiece of dilute fermion and boson calculations with strong interactions . field - theoretical calculations in both three and two dimensions are based on the ladder diagrams and the t - matrix . see fetter and walecka@xcite for the 3d problem , schick@xcite for the 2d boson problem and bloom@xcite the 2d fermion problem . for lattice fermion problems , kanamori@xcite derived the t - matrix for a tight - binding model that is essentially a hubbard model ( this work is also described in yosida@xcite ) . and in ref . , the t - matrix is worked out explicitly for the hubbard model , and kanamori s result is obtained . ref . also evaluated the t - matrix for the dilute limit in three dimensions and obtained a functional dependence on particle density . rudin and mattis@xcite used the t - matrix expression derived in refs . and and found upper and lower bounds of the fermion t - matrix in two dimensions in terms of particle density . rudin and mattis s result for the low - density limit of the two - dimensional hubbard model is of the same functional form as bloom s diagrammatical calculation for the two - dimensional fermion hard disks.@xcite since the discovery of high - temperature superconductors , bloom s calculation has received a lot of attention because of its relevance to the validity of the fermi liquid description of dilute fermions in two dimensions . there have been a number of works on the 2d dilute fermi gas@xcite and on the dilute limit of 2d hubbard model,@xcite all using the t - matrix , but these results have not been checked by numerical calculations . in fact , we are not aware of a systematic study of the t - matrix for a lattice model . in this paper , we present the first such study for the two - particle problem in sec . [ sec - tmat ] ( for bosons and fermions ) and the few - fermion problem in sec . [ sec - tmat2 ] . we check the t - matrix results with exact diagonalization data and show that our t - matrix on a lattice is the sum of the two - body scattering terms to infinite order . in this paper , we will study systematically the dilute limit of our model eq . ( [ eq - ham ] ) , focusing on the problem of a few particles . our paper is divided into four parts . in sec . [ sec - green ] , the two - particle ( boson and fermion ) problem is studied . we formulated the two - particle schrodinger equation using lattice green functions , employ some of its recursion relations to simplify the problem , and obtain the two - boson ground state energy in the large - lattice limit . using the two - particle result , we then study the problem of a few particles and obtain an expression for ground state energy on a large lattice . in sec . [ sec - tmat ] , the two - particle problem is then cast into a different form , emphasizing the scatterings between the two particles . the result is the t - matrix , that is exact for the two - particle problem and contains all two - body scattering terms . we will study the two - particle t - matrix in great detail , showing the differences between the boson and fermion cases , and demonstrating that the first t - matrix iteration is often a good approximation for fermion energy . in appendix [ sec - physical ] , we show explicitly that the t - matrix we obtain is the sum total of all two - body scattering terms . the problem of a few fermions is taken up in sec . [ sec - tmat2 ] . first , the fermion shell effect is discussed and demonstrated from diagonalization , and we show the difference for bosons and fermions . we show the modifications to the two - fermion t - matrix that enable us to calculate energies for three , four , and five particles . using this t - matrix , we can compute the interaction corrections to the noninteracting energy and trace the change in the energy spectrum from the nointeracting one to the interacting one . finally , in sec . [ sec - dilute ] , we discuss the energy per particle curve for dilute bosons and fermions . we have studied the two - dimensional results derived by schick@xcite for bosons and bloom@xcite for fermions by fitting the data from diagonalization for a number of lattices . schick s result for dilute bosons is checked nicely , and we explain that for spinless fermions in our model we will need the p - wave scattering term , which is not included in bloom s calculation . in appendix [ sec - diag ] , we discuss briefly our exact diagonalization computer program , which can handle arbitary periodic boundaries specified by two vectors on the square lattice and uses translation symmetry to reduce the matrix size . the two - particle problem has appeared in many different contexts . the most familiar one is the hydrogen atom problem in introductory quantum mechanics textbooks . the two - magnon problem is closely related mathematically to our two - particle problem , and it has been solved in arbitary dimensions for ferromagnets ( see e.g. , ref . ) . another important two - particle problem is the cooper problem , with two electrons in the presence of a fermi sea ( see e.g. , ref . ) . and motivated by the possibility of cooper pair formation in high - temperature superconductors , there have also been a number of studies on bound states on a two - dimensional lattice.@xcite the two - electron problem in the plain two - dimensional repulsive hubbard model is studied in ref . , and ground state energy in the large - lattice limit is obtained analytically . in this section , we present a rather complete calculation for the two - particle problem in our model , treating both bosons and fermions . with infinite repulsive interaction in our model , we are not interested in finding bound states . we calculate eigenenergies for all states for a finite - size lattice , and our calculation is more complicated than the hubbard model@xcite case because of nearest - neighbor ( in place of on - site ) interaction . where the green function in the hubbard case was a scalar object , in our case it is replaced by a @xmath4 matrix , corresponding to the four nearest neighbor sites where the potential acts . this green function study of the two - particle problem is closely related to the treatment of the two - electron problem in the hubbard model@xcite and that in an extended hubbard model.@xcite we will show the use of lattice symmetry and recursion relations to simplify the problem with nearest - neighbor interactions . in this two - particle calculation , we will work in momentum space , and we will start with a hamiltonian more general than eq . ( [ eq - ham]),@xcite @xmath27 here we have allowed hopping and interaction between any two lattice sites , but we require that both depend only on the separation between the two vectors and both have inversion symmetry . that is @xmath28 , @xmath29 , @xmath30 , and @xmath31 . in momentum space , ( [ eq - tg ] ) and ( [ eq - ug ] ) become , @xmath32 where @xmath33 with @xmath34 and @xmath35 . eqs . ( [ eq - tg ] ) and ( [ eq - ug ] ) reduce to our nearest - neighbor hamiltonian eq . ( [ eq - ham ] ) if we take , @xmath36 where we have taken the lattice constant to be unity,@xcite and the nearest - neighbor vectors will be called @xmath37 then we have , @xmath38 note that the structure of later equations depends sensitively on having four sites in eq . ( [ eq - v ] ) where @xmath39 m , but does not depend much on the form of eq . ( [ eq - t ] ) and the resulting dispersion eq . ( [ eq - ek ] ) . using momentum conservation of eq . ( [ eq - hamg ] ) , the two - particle wave function that we will use is , @xmath40 where the sum is over the whole brillouin zone , and the coefficient @xmath41 satisfies , @xmath42 where @xmath43 for bosons and @xmath44 for fermions . applying the more general form of the hamiltonian operator eqs ( [ eq - tq1 ] ) and ( [ eq - uq ] ) to the state eq . ( [ eq - psig ] ) , the schrodinger equation @xmath45 becomes @xmath46 eq . ( [ eq - sch ] ) is a matrix equation @xmath47 where @xmath48 . if @xmath12 is not infinity , this @xmath49 matrix @xmath50 can be diagonalized , and @xmath51 and @xmath41 are respectively the eigenvalue and eigenvector . to deal with @xmath16 , we need some further manipulations . we consider the case when @xmath52 , for any @xmath53 , which is to say , the energy @xmath51 is not the energy of a noninteracting pair . the ( lattice ) fourier transform of the coefficients @xmath41 is @xmath54 this is just the real - space wavefunction in terms of the relative coordinate @xmath55 . define the lattice green function , @xmath56 then after dividing the first factor from both sides of our schrodinger equation , eq . ( [ eq - sch ] ) , and fourier transforming , we obtain @xmath57 in the following we return to the nearest - neighbor potential @xmath58 in eq . ( [ eq - v ] ) . the green function sum in eq . ( [ eq - greeneqg ] ) then has only four terms , @xmath59 summed over the separations in eq . ( [ eq - rlist ] ) . if we also restrict @xmath55 to the four nearest - neighbor vectors,@xcite then eq . ( [ eq - greeneq ] ) becomes , @xmath60 if we define the @xmath4 matrix , @xmath61 and a @xmath62 vector @xmath63 , then we obtain a simple matrix equation , @xmath64 we can also rewrite this equation as an equation for energy using the determinant , @xmath65 with @xmath16 , we have even simpler equations @xmath66 and @xmath67 notice we write @xmath68 to denote the limit as @xmath69 ; it would not do to write simply @xmath70 in eqs . ( [ eq - iminusgnull ] ) and ( [ eq - null ] ) , since @xmath71 as @xmath72 ( being the amplitude of the relative wavefunction at the forbidden separations @xmath73 . ) . for the hubbard model , there is only on - site interaction , so @xmath58 is nonzero only when @xmath74 , and the sum in eq . ( [ eq - greeneqg ] ) has only one term . ( [ eq - gtr ] ) is simply a scalar equation , which , after @xmath75 cancels from both sides of the equation and using eq . ( [ eq - green ] ) , gives , @xmath76 which is exactly the result in ref . . we specialize to the case of total momentum @xmath77 and rectangular - boundary lattices . we have from eqs . ( [ eq - green ] ) and ( [ eq - grdef ] ) , @xmath78 where the potential is nonzero on the sites @xmath73 given by eq . ( [ eq - rlist ] ) and in the last step we have used the symmetry properties of the dispersion relation @xmath79 . obviously eq . ( [ eq - grij ] ) is a function of displacements @xmath80 , which ( in view of eq . ( [ eq - rlist ] ) can be ( 0,0 ) , ( 1,1 ) , ( 2,0 ) , or any vector related by square symmetry . it is convenient for this and later sections to define a new notation for the green function @xmath81 , emphasizing its dependence on @xmath82 , @xmath83 where the sum is over the @xmath13 wavevectors @xmath84 with @xmath85 and @xmath86 ( for one brillouin zone ) , and we have used the expression for @xmath87 from eq . ( [ eq - ek ] ) ( and taken @xmath88 ) . this green function for rectangular - boundary lattices satisfies the following reflection properties , @xmath89 and if we have a square lattice ( @xmath90 ) we also have @xmath91 eq . ( [ eq - grij ] ) can be written as , @xmath92 using the reflection properties of @xmath93 , eq . ( [ eq - rectsym ] ) , and the definition eq . ( [ eq - grgamma ] ) , our green function matrix becomes , @xmath94 where @xmath95 , @xmath96 , @xmath97 , and @xmath98 . the eigenvalues and eigenvectors of this matrix are , @xmath99 where @xmath100 and @xmath101 are complicated functions of @xmath102 , @xmath103 , @xmath104 , and @xmath105 . the exact energy @xmath51 makes the matrix @xmath106 singular , which means that one of the eigenvalues has to be zero . from eq . ( [ eq - null ] ) , the null eigenvector of @xmath107 is @xmath108 , in terms of @xmath73 as in eq . ( [ eq - rlist ] ) . the relative wavefunction should be odd or even under inversion , depending on statistics , i.e. @xmath109 which follows immediately from eqs . ( [ eq - ft ] ) and ( [ eq - g ] ) . inversion , acting on nearest - neighbor vectors eq . ( [ eq - rlist ] ) , induces @xmath110 and @xmath111 ; thus with @xmath77 , we should have @xmath112 and @xmath113 for fermions , and @xmath114 and @xmath115 for bosons . inspecting the eigenvectors we obtained in eq . ( [ eq - evals ] ) , we see that those corresponding to @xmath116 are antisymmetric under inversion corresponding to a `` p - wave - like '' ( relative angular momentum 1 ) state for fermions . setting @xmath117 , we get @xmath118 , or setting @xmath119 @xmath120 , which respectively mean @xmath121 associated with the even eigenvectors are @xmath122 which are identified as boson eigenvalues . the boson eigenvalues , eq . ( [ eq - evals ] ) , are rather complicated for general rectangular - boundary lattices . for a square - boundary lattice , using eq . ( [ eq - squaresym ] ) , we get @xmath123 in the matrix eq . ( [ eq - gmat ] ) , which makes the fermion eigenvalues @xmath124 degenerate . the boson eigenvalues in eq . ( [ eq - evals ] ) simplify greatly to @xmath125 and @xmath126 , which means that the boson energy equations are , @xmath127 the corresponding eigenvectors simplify too , to @xmath128 and @xmath129 respectively , which may be described as `` s - wave - like '' and `` d - wave - like '' , i.e. relative angular momentum 0 and 2 . ( [ eq - fermioneq0 ] ) , ( [ eq - fermioneq ] ) , ( [ eq - bosoneq1 ] ) and ( [ eq - bosoneq2 ] ) are much better starting points for analytical calculations than the original determinant equation eq . ( [ eq - det ] ) . in the center of the problem is the lattice green function @xmath93 defined in eq . ( [ eq - greenmn ] ) . many of the lattice calculations come down to evaluating these lattice green functions.@xcite in this section , we derive the large - lattice two - boson energy using the recursion and symmetry relations of the green function @xmath93 . the green function @xmath93 for general @xmath131 and @xmath132 and finite lattice are difficult to evaluate . the good thing is that there are a number of recursion relations connecting the green functions at different @xmath131 and @xmath132.@xcite these are trivial to derive after noting that eq . ( [ eq - green ] ) ( for @xmath77 ) can be written @xmath133 g(e,0;{{\bf r}},{{\bf r}}')= \delta_{{{\bf r}}=0}\delta_{{{\bf r}}'=0}\ ] ] where @xmath134 is the discrete laplacian , @xmath135 for any function @xmath136 , where the sum is over neighbor vectors eq . ( [ eq - rlist ] ) . the two recursion relations that we will use are @xmath137 using eqs . ( [ eq - squaresym ] ) , ( [ eq - recur1 ] ) , and ( [ eq - recur2 ] ) , the boson equation eq . ( [ eq - bosoneq1 ] ) for square - boundary lattices simplifies to @xmath138 with eigenvector @xmath128 . next we compute the leading form of @xmath139 for large @xmath130 of a square - boundary lattice . the calculation is close to that in ref . for the hubbard model . we define @xmath140 . because the lowest energy of an independent particle is @xmath141 , @xmath25 is the energy correction to two independent particle energy at zero momentum . then we have , from eq . ( [ eq - greenmn ] ) , @xmath142 we should discuss the number of approximations we have made to extract this leading dependence in @xmath130 . first except in the @xmath143 term we have ignored the @xmath25 term , assuming it is small as compared to @xmath144 ( with @xmath145 ) . this is justified as we only want the leading term in the large-@xmath130 limit . using an integral for a lattice sum is another approximation . we choose the lower limit of integration to be @xmath146 corresponding to the first wavevectors after @xmath147 is taken out of the sum . we also used the quadratic approximation for the energy dispersion @xmath87 appearing in the denominator . using the boson energy equation eq . ( [ eq - bosoneq3 ] ) and the large-@xmath130 limit of the green function eq . ( [ eq - g00l ] ) , we get , @xmath148 in the large-@xmath130 limit , @xmath149 ( as it is the interaction correction to the noninteracting energy ) , so we get , to the leading order of @xmath130 , @xmath150 we will check eq . ( [ eq - del ] ) in sec . [ sec - afewparticles ] . the procedure used in sec . [ sec - bosonlargel ] for two bosons can also be applied to problems with a few particles . for a few particles on a large lattice with short - range ( here nearest - neighbor ) interaction , two - particle interaction is the main contribution to energy . we write for two particles , @xmath151 here in this section we use the notation @xmath152 and @xmath153 to denote the @xmath154-particle exact and noninteracting ground state energies respectively and emphasize the dependence of @xmath25 on @xmath130 by using @xmath155 . it is reasonable to expect that the energy for @xmath154 particles is the noninteracting energy plus interaction corrections from the @xmath156 pairs of particles . we then have , @xmath157 for bosons , @xmath158 , because in the ground state , all bosons occupy the zero - momentum state . on the other hand , for fermions , because of pauli exclusion , no two fermions can occupy the same state , the noninteracting ground state is obtained from filling the @xmath154 fermions from the lowest state ( @xmath159 ) up . ( [ eq - eml ] ) implies that plotting @xmath160 versus @xmath130 for different @xmath154 should all asymptotically at large @xmath130 approach @xmath155 . in fig . [ fig - bflargel ] , we do such plots , for bosons and fermions with @xmath161 . the fermion results , from p - wave scattering ( as our spinless fermion wave function has to be antisymmetric ) , are much smaller than the boson results ( bold curves ) from s - wave scattering . versus @xmath130 for @xmath161 . all curves appear to converge at large @xmath130 . the fermion ( p - wave ) result is much less than the boson result ( s - wave ) . the @xmath162 plot goes to @xmath163 and the @xmath164 plot to @xmath165 . the boson @xmath164 curve is too high to be included in this plot . ] note that in our calculation for @xmath139 eq . ( [ eq - g00l ] ) , we have neglected the contribution of @xmath25 in the denominator except for the first term ( @xmath143 ) . now with the leading form of @xmath25 eq . ( [ eq - del ] ) , we can obviously plug @xmath166 into eq . ( [ eq - g00l ] ) to get the form of the next term , @xmath167 using eqs . ( [ eq - eml ] ) and ( [ eq - deltae ] ) , we get , for a few bosons ( @xmath158 ) , @xmath168 in fig . [ fig - bosonlargel2 ] , we plot @xmath169 versus @xmath170 for @xmath161 , using the boson data in fig . [ fig - bflargel ] . quadratic polynomial fitting is done for @xmath171 , where we have more data than @xmath172 . the coefficient @xmath173 for both fits , implying , from eq . ( [ eq - deltae ] ) , the leading - order term in @xmath155 is indeed @xmath174 . @xmath175 and @xmath176 from two fits are also comparable . versus @xmath170 for @xmath161 . quadratic polynomial fitting is done for @xmath177 and @xmath178 . the fitted constant coefficients are approximately one , and the other coefficients from @xmath177 and @xmath178 are comparable . ] to summarize , from eqs . ( [ eq - eml ] ) and ( [ eq - deltae ] ) and fitting in fig . [ fig - bosonlargel2 ] , we find that in our model the energy of a small number @xmath154 of bosons on a large @xmath179 lattice is to the leading order of @xmath130 , @xmath180 for two _ fermions _ on a large @xmath179 lattice , the noninteracting energy the lead term in eq . ( [ eq - fewbosons ] ) is obviously lower for @xmath181 than for @xmath182 . we have not worked out the asymptotic behavior for @xmath183 . in sec . [ sec - bosonlargel ] and sec . [ sec - afewparticles ] , we studied the ground state energy of a few particles on a large lattice , and we showed that the energy of @xmath154 particles can be approximated by summing the energy of the @xmath156 pairs . in this section , we reformulate the equations for two particles and derive a scattering matrix , the t - matrix . the t - matrix gives us equations of the form eqs . ( [ eq - perturb ] ) and ( [ eq - perturbmany ] ) which are more precise statements of the ideas presented in sec . [ sec - bosonlargel ] and sec . [ sec - afewparticles ] . they apply to small lattices and to excited states . to have an equation in the form of eq . ( [ eq - perturb ] ) , we start with any pair of momentum vectors @xmath23 and @xmath24 and write noninteracting energy of the pair @xmath184 and total momentum @xmath185 . because our hamiltonian , eqs . ( [ eq - tq1 ] ) and ( [ eq - uq ] ) , conserves total momentum , we can restrict our basis states to @xmath186 . it is tempting to take @xmath187 and @xmath188 as our nonperturbed states , but there can be other two - particle states with the same total momentum @xmath189 and energy @xmath190 . in fact , using our energy dispersion function eq . ( [ eq - ek ] ) , if we write @xmath191 and @xmath192 , and define @xmath193 and @xmath194 then we have , @xmath195 and @xmath196 . we call this fact , that component exchanges in the pair @xmath23 and @xmath24 gives a pair @xmath197 and @xmath198 that have the same total momentum and energy , the _ pair component exchange symmetry _ of our energy dispersion function @xmath87 . this symmetry is is due to the fact that our @xmath87 is separable into a @xmath199 part and a @xmath200 part ( i.e. , @xmath201 where @xmath202 in our model).@xcite the pair component exchange symmetry says that if @xmath203 and @xmath204 , then the state @xmath205 , with @xmath197 and @xmath198 defined above using component exchange , has the same total momentum and energy as @xmath206 . the degenerate perturbation theory requires @xmath205 should be included in the set of nonperturbed states with @xmath206 . with a noninteracting two - particle energy @xmath190 and total momentum @xmath189 , we divide the @xmath13 wavevectors into two disjoint sets , @xmath207 note that if @xmath208 then @xmath209 . denote @xmath210 the number of elements in @xmath211 and @xmath212 the number of elements in @xmath213 . with this separation of @xmath53 , our eigenstate eq . ( [ eq - psig ] ) becomes @xmath214 where the first sum contains all states whose noninteracting energy is degenerate . using the idea of degenerate perturbation theory , we expect to be able to find a secular matrix @xmath215 , @xmath216 , for the degenerate states in @xmath211 only , and @xmath215 will eventually be our momentum space t - matrix , which we will derive now . note that using eq . ( [ eq - g ] ) , the number of independent states in the first sum of eq . ( [ eq - psi2 ] ) is less than @xmath210 . we include both @xmath186 and @xmath217 in our calculation because we are considering boson and fermion problems at the same time : the symmetric solution @xmath218 is a boson solution and the antisymmetric solution @xmath219 is a fermion solution ( see eq . ( [ eq - g ] ) ) . our purpose is to derive a set of closed equations for @xmath41 , the coefficent in our two - particle state eq . ( [ eq - psi2 ] ) , with @xmath208 . the schrodinger equation for the two - particle state @xmath220 , eq . ( [ eq - sch ] ) , can now be written as , @xmath221 where @xmath222 is the fourier transform of @xmath41 as defined in eq . ( [ eq - ft ] ) . for @xmath223 , if we assume that @xmath52 , eq . ( [ eq - schr ] ) becomes @xmath224 where we have defined a green function for the set @xmath213 , @xmath225 and a fourier transform with vectors in @xmath211 , @xmath226 by restricting to the nearest - neighbor repulsion potential eq . ( [ eq - v ] ) , eq . ( [ eq - g2 ] ) becomes , @xmath227 summed over neighbor vector eq . ( [ eq - rlist ] ) . now restricting @xmath228 in eq . ( [ eq - g2r ] ) , we get a set of four equations , @xmath229 where we have written @xmath230 and @xmath231 and @xmath232 . ( [ eq-4eqs ] ) is a matrix equation , @xmath233 where @xmath234 is @xmath4 , @xmath235 and @xmath236 are @xmath62 , and @xmath12 is a scalar ( strength of potential ) . and we can invert the matrix to get , @xmath237 this is a key result , as we have expressed the desired function @xmath75 , a fourier transform of @xmath41 including all @xmath53 , in terms of @xmath238 which includes only @xmath208 ; the information about other @xmath223 was packaged into the green function @xmath239 . now we go back to eq . ( [ eq - schr ] ) , restrict the summation to @xmath240 , and substitute in @xmath241 from eq . ( [ eq - invert ] ) , and we get , @xmath242 where in the last step we have used the fourier transform of @xmath243 eq . ( [ eq - ft1 ] ) and defined , @xmath244 if we restrict @xmath208 in eq . ( [ eq - gqeq ] ) , then we have , @xmath245 which means , @xmath246 where we have written @xmath247 and left out the dependence on @xmath189 . @xmath248 is the t - matrix in momentum space . both @xmath53 and @xmath249 in eq . ( [ eq - tqv ] ) are in @xmath211 , which means that if there are @xmath210 elements in @xmath211 then the matrix @xmath250 is @xmath216 . ( [ eq - tq ] ) is our desired equation that shows explicitly the interaction correction to the noninteracting energy @xmath190 . in appendix [ sec - physical ] , we show the physical meaning of @xmath251 in the language of diagrammatic perturbation theory , namely it is the sum total of all the terms with repeated scattering of the same two particles . this t - matrix formalism for the two - particle problem is therefore exact , and it is exactly equivalent to the schrodinger equation and the green function formalism in sec . [ sec - green ] . the resulting equation is an implicit equation on @xmath51 , of the form @xmath252 of eq . ( [ eq - perturb ] ) , and we will show in a later section that for fermions the approximation @xmath253 is often very good . note also that for our case @xmath16 , the t - matrix expression eq . ( [ eq - gq ] ) becomes @xmath254 where the potential @xmath12 cancels out , giving a finite value . this is one of the advantages of the t - matrix formalism that it can deal with infinite ( singular ) potential , for which straightforward perturbation theory would diverge . the definition of @xmath251 in eq . ( [ eq - gq ] ) is a fourier transform of the real space quantity @xmath255 . here @xmath234 is @xmath4 because we have nearest - neighbor interaction . when there is only on - site interaction , as is in the usual hubbard model case , @xmath256 , eq . ( [ eq - green2 ] ) , is a scalar . then , we can simply use the scalar quantity @xmath257 , which is the t - matrix that has appeared in kanamori,@xcite mattis,@xcite rudin and mattis,@xcite and yosida.@xcite our expression , eq . ( [ eq - gq ] ) , is more complicated because we have nearest - neighbor interaction ( and thus the relevance of @xmath258 ) . in sec . [ sec - eqforrec ] , after deriving the general green function equation using @xmath259 , we specialized to rectangular - boundary lattices and used lattice reflection symmetries to diagonalize the @xmath4 matrix @xmath259 and obtained scalar equations . here our t - matrix equation eq . ( [ eq - tq ] ) requires us to find the eigenvalues of the t - matrix @xmath215 . in this section , we use particle permutation symmetry and pair component exchange symmetry to diagonalize the @xmath216 t - matrix @xmath250 for a few special cases . there is only one momentum vector in @xmath211 . let us write @xmath261 ( this implies that @xmath262 ) . then there is only one unperturbed two - particle basis state @xmath263 ( see eq . ( [ eq - psi2 ] ) ) . this must be a boson state , and @xmath250 is a number . we write the resulting scalar equation as , @xmath264 here @xmath266 with @xmath267 . the basis states are @xmath206 and @xmath268 . the symmetric ( boson ) combination is @xmath269 , and the antisymmetric ( fermion ) combination is @xmath270 . these have to be the eigenvectors of @xmath250 . and that is to say that if we define @xmath271 then we have @xmath272 , @xmath273 , and @xmath274 here @xmath275 and @xmath276 are scalars that correspond to boson and fermion symmetries respectively . and our t - matrix equation eq . ( [ eq - tq ] ) is reduced to two scalar equations , @xmath277 for bosons and fermions respectively . our notation for the eigenvalues of @xmath250 is always to write @xmath278 with subscripts that are the coefficients ( in order ) of the @xmath279 two - particle basis vectors . here @xmath281 with @xmath282 . the basis states are @xmath206 , @xmath268 , @xmath205 , and @xmath283 . using particle permutation symmetry , we get two states with even symmetries appropriate for bosons , which generically would be @xmath284 and two odd ( fermion - type ) states , @xmath285 where @xmath102 and @xmath103 are arbitrary coefficients to be determined . recall @xmath280 means the pair @xmath286 has the same total momentum and energy as @xmath287 , which may happen for various reasons . when the reason is the pair component exchange symmetry ( of sec . [ sec - setup ] ) , _ i.e. _ @xmath193 and @xmath194 , then @xmath288 , due to a hidden symmetry under the permutation @xmath289 . the only effect this permutation has on the momentum transfers @xmath290 is to change the sign of one or both components ; but the potential @xmath58 is is symmetric under reflection through either coordinate axis , hence @xmath291 is invariant under the permutation . since the t - matrix depends only on @xmath291 , it inherits this symmetry . next , if we define @xmath292 then we have @xmath293 , @xmath294 , and @xmath295 becomes diagonal with four eigenvalues of @xmath250 : @xmath296 , @xmath297 , @xmath298 , and @xmath299 . and our t - matrix equation eq . ( [ eq - tq ] ) is reduced to @xmath300 for bosons and @xmath301 for fermions . the three cases @xmath302 , and @xmath280 with pair component exchange symmetry are three special cases in which we know the eigenvectors of @xmath215 and can therefore diagonalize @xmath215 from symmetry considerations easily . different or larger values of @xmath210 are possible when @xmath189 has a special symmetry , e.g. when @xmath303 , @xmath304 generically since @xmath211 includes pairs such as @xmath305 . for these general cases , we return to eq . ( [ eq - tq ] ) and diagonalize @xmath215 numerically . for example , on a @xmath179 lattice , the pairs ( 0,1)@xmath306 and ( 1,0)@xmath307 have the same total energy and momentum , but this is not due to the pair component exchange symmetry . in this case , we numerically diagonalize the @xmath4 matrix @xmath250 , and we find that in the fermion eigenvectors , eqs . ( [ eq - fstate1 ] ) and ( [ eq - fstate2 ] ) , @xmath308 . the example system that we will study here is @xmath309 with @xmath182 . the noninteracting and interacting energies of the system are in table [ t-10by11 ] . it can be seen that all of the energies listed in table [ t-10by11 ] are of the three cases discussed in sec . [ sec - symmetrycon ] : @xmath260 , @xmath265 , and @xmath280 due to pair component exchange symmetry . . the 12 low - lying noninteracting and exact two - particle energies of the @xmath309 lattice with total momentum @xmath182 . @xmath23 and @xmath310 are the momentum vectors . [ cols="^,^,^,^,^",options="header " , ] [ t - slopes ] in fig . [ fig - benergycurve ] , the solid line is the function @xmath311 using @xmath312 , @xmath313 , and @xmath314 , and we obtain a good fit up to @xmath315 . for bosons , quantum monte carlo can be used to obtain zero temperature energies for reasonably large systems . for a dilute boson gas on a square lattice with on - site hardcore but not nearest - neighbor interaction , ref . has fitted the first term of schick s formula eq . ( [ eq - schick ] ) , and ref . has used higher - order terms and included the fitting of the chemical potential also . the agreement is good in both studies . for fermions , it customary to write the energy per particle expansion in terms of the fermi wavevector @xmath316 . for two - dimensional dilute hard disk fermions with a general spin @xmath317 , the energy per particle from diagrammatic calculations , is obtained by bloom@xcite @xmath318 ( see ref . for the spin-1/2 calculation and ref . for general @xmath317 ) . ( [ eq - bloom2 ] ) means that for our spinless fermions ( @xmath319 ) , the leading order correction to the noninteracting energy in eq . ( [ eq - bloom2 ] ) is zero , which is due to the fact that eq . ( [ eq - bloom2 ] ) is derived for s - wave scattering . in our model , without spin , only antisymmetric spatial wavefunctions are allowed for fermions , and therefore the leading - order correction to the noninteracting energy should be from p - wave scattering . ref . contains a formula for p - wave scattering in three dimensions where the leading - order correction to @xmath320 is proportional to @xmath321 while the s - wave correction is proportional to @xmath322 . we are not aware of a two - dimensional p - wave calculation in the literature,@xcite and we have not worked out this p - wave problem in two dimensions . we expect that the p - wave contribution to energy should be considerably smaller than that from the s - wave term . in sec . [ sec - afewparticles ] , we have considered the case of a few fermions on a large @xmath179 lattice , and in fig . [ fig - bflargel ] we have studied the interaction correction to the noninteracting energy @xmath25 . it was shown there that @xmath25 for our spinless fermions is much smaller than that for bosons . using @xmath323 , we can rewrite eq . ( [ eq - bloom2 ] ) as @xmath324 in this form , it is revealed that the second term of eq . ( [ eq - bloom2n ] ) is identical to the first term of the boson expression eq . ( [ eq - schick ] ) , apart from the replacement @xmath325 . in other words , the dominant interaction term for spinfull fermions is _ identical _ to the ferm for bosons , provided we replace @xmath132 by the density of all spin species but one , i.e. of the spin species which can s - wave scatter off a given test particle . we have studied a two - dimensional model of strongly - interacting fermions and bosons . this model is the simplest model of correlated electrons . it is very difficult to study two - dimensional quantum models with short - range kinetic and potential terms and strong interaction . there are very few reliable analytical methods , and many numerical methods are not satisfactory . with our simplified model of spinless fermions and infinite nearest - neighbor repulsion , we can use exact diagonalization to study systems much larger ( in lattice size ) than that can be done with the hubbard model . one of our goals is to publicize this model in the strongly - correlated electron community . in this paper , we made a systematic study of the dilute limit of our model , using a number of analytical techniques that so far have been scattered in the literature . we studied the two - particle problem using lattice green functions , and we demonstrated the use the lattice symmetry and green function recursion relations to simplify the complications brought by nearest - neighbor interactions . we derived in detail the two - particle t - matrix for both bosons and fermions , and we showed the difference between the boson and fermion cases and that for fermions the first t - matrix iteration is often a good approximation . we applied the two - fermion t - matrix to the problem of a few fermions , with modifications due to pauli exclusion , and showed that the t - matrix approximation is good for even small lattices . it is somewhat puzzling that with the essential role the t - matrix plays in almost every calculation in the dilute limit with strong interactions , no systematic study of the t - matrix for a lattice model has been made , as far as we know . we believe that our work on the two - particle t - matrix and the few - fermion t - matrix is first such study . some approximations that are routinely made in t - matrix calculations are graphically presented , especially the use of first t - matrix iteration in calculating fermion energy . and we demonstrate the qualitative difference between the boson and fermion t - matrices . we believe that this study is a solid step in understanding dilute fermions in two dimensions , and is of close relevance to the 2d fermi liquid question . the dilute boson and fermion energy per particle curves were studied in sec . [ sec - dilute ] . the boson curve was fitted nicely with a previous diagrammatic calculation , and our work on dilute bosons complements quantum monte carlo results.@xcite for the fermion problem in our model , the leading order contribution to energy is from p - wave scattering ; therefore , the series of results based on s - wave calculations by bloom,@xcite bruch,@xcite and engelbrecht , _ et al._@xcite are not directly available . hopefully , the work in progress on p - wave scattering will be completed , and our diagonalization data can shed light to the interesting problem of two dimensional dilute fermions . our model of spinless fermions and hardcore bosons with infinite nearest - neighbor repulsion involves a significant reduction of the size of the hilbert space as compared to the hubbard model . this enables us to obtain exact diagonalization results for much larger lattices than that can be done with the hubbard model , and this also enables us to check the various analytical results ( green function , t - matrix , diagrammatics ) in the dilute limit with diagonalization for much larger systems than that has been done in previous works . this paper and a companion paper@xcite on the dense limit are the first systematic study of the spinless fermion model in two dimensions . we hope that the comprehensiveness of this paper can not only draw more attention to this so far basically overlooked model but also serve as a guide for diagonalization and analytical studies in the dilute limit . this work was supported by the national science foundation under grant dmr-9981744 . we thank g. s. atwal for helpful discussions . this section describes briefly our exact diagonalization program . it is indebted to refs . and , which are guides for coding exact diagonalization in one dimension . here we only describe the necessary considerations in more then one dimension , and focuses on the use translation symmetry to reduce the problem . our underlying lattice is the square lattice , and we take the lattice constant to be unity . the periodic boundary conditions are specified by two lattice vectors @xmath326 and @xmath327 , such that for any lattice vector @xmath328 we have @xmath329 , where @xmath330 and @xmath331 are two integers . in fig . [ fig-2lattice ] , we show two systems . the first one has @xmath332 and @xmath333 so the number of lattice sites is @xmath334 . the second one has @xmath335 and @xmath336 so @xmath337 . from this example we see immediately the advantage of having skewed boundary conditions : we can have reasonably shaped systems with number of sites ( here 19 ) not possible for an usual rectangular system . on the left and @xmath338 on the right . site numbers are shown , following the numbering convention , upward and rightward . ] a site order is needed to keep track of the order of the fermion sign . the convention that we use is starting the zeroth site from the lower left corner and move progressive upward and rightward following the square lattice structure until we encounter boundaries of the lattice defined by the periodic boundary condition vectors @xmath326 and @xmath327 ( see fig . [ fig-2lattice ] ) . a basis state with @xmath154 particles is then represented by an array of the @xmath154 occupied site numbers , with nearest neighbors excluded ( because @xmath16 in our hamiltonian eq . ( [ eq - ham ] ) ) . denote such a basis state @xmath339 and we have @xmath340 where @xmath341 denotes the set of states created by hopping one particle in @xmath339 to an allowed nearest - neighbor site and for bosons @xmath342 always and for fermions @xmath343.@xcite and the matrix element is @xmath344 if @xmath345 and 0 , otherwise . in order to calculate for large systems , it is necessary to use symmetry to block diagonalize the hamiltonian matrix . in our code , we use lattice translation symmetry because it works for arbitrary periodic boundaries . the eigenstate that we use is the bloch state@xcite @xmath346 in this expression @xmath347 is a wavevector ( one of @xmath13 , where @xmath13 is the number of sites ) , @xmath348 is a lattice vector @xmath349 is a short hand notation for translation by @xmath350 , and @xmath351 is a normalization factor . the original basis states are divided by translation into classes and any two states in the same class give the same bloch state with an overall phase factor . what we need to do is to choose a representative from each class , and use this state consistently to build bloch states . for a state @xmath339 we denote its representative @xmath352 . to compute the hamiltonian matrix elements using the bloch states eq . ( [ eq - bloch ] ) , let us start from a representative state @xmath352 . we have , as in eq . ( [ eq - hn ] ) , @xmath353 , then @xmath354 , where we have used the fact that @xmath349 commutes with the hamiltonian . we have , @xmath355 next because we are interested in matrix elements between representative states , we want to connect @xmath356 in the preceding equation to @xmath357 . if @xmath358 , then @xmath359 . so we have @xmath360 we should note that for all @xmath361 there can be more than one element having the same representative @xmath362 . that is to say in the sum in eq . ( [ eq - hnk ] ) , there can be more than one term with @xmath357 . we write a new set @xmath363 . then we can write our matrix element equation as follows , @xmath364 eq . ( [ eq - hmn ] ) is the centerpiece of the bloch state calculation . it includes many of the complications that come with the bloch basis set . ( see ref . for the corresponding equation in one dimension . ) let us use @xmath365 to denote the number of bloch basis states for one @xmath366 . @xmath365 is the dimensionality of the matrix that we need to diagonalize . for @xmath365 in the order of thousands , full diagonalization ( with storage of the matrix ) is done using lapack,@xcite and a @xmath367 matrix ( @xmath0 with @xmath368 ) takes about 27 minutes.@xcite for larger @xmath365 , the lanczos method is implemented following the instructions in ref . , we have two options . first , we store information about the matrix ( i.e. , for each column , a set of ( @xmath369 , @xmath370 , @xmath371 ) described above that contains information about the nonzero entries of the hamiltonian matrix in this column ) . the @xmath372 case on @xmath0 lattice with @xmath373 and tolerance @xmath374 takes about 45 minutes ( 32 lanczos iterations ) and uses about 1.5 gb of memory . this basically reaches our memory limit . on the other hand , we can also do lanczos without storing matrix information . the same @xmath372 case on @xmath0 uses only 200 mb of memory but takes more than four hours ( 263 minutes ) , for a larger tolerance @xmath375 ( therefore fewer lanczos iterations , 14 ) . without storing matrix information , we can calculate for larger matrices : the @xmath376 case on @xmath0 , with @xmath377 ( the largest for the @xmath0 system ) and tolerance @xmath375 , is done in 10 hours , using less than 400 mb of memory . the largest matrix we computed for this work is @xmath378 , i.e. , about 2.5 million bloch states , for @xmath162 on @xmath1 . this takes 10 hours and uses about 550 mb of memory , for a tolerance of @xmath375 . in addition , we have also installed arpack@xcite that uses the closely related so - called arnoldi methods and can obtain excited state eigenvalues and eigenvectors as well . if we only need information about the ground state , our lanczos program is considerably faster than arpack . the @xmath0 lattice , with maximum @xmath365 around 2 million , is basically the largest lattice for which we can calculate eigenenergies at all fillings . the exponential growth is very rapid after this . the @xmath379 lattice with 8 particles has @xmath380 bloch states , and with one more particle , @xmath372 , there are @xmath381 , i.e. , more than 30 million states . in this section we give yet another derivation of the t - matrix which makes more explicit the physical meaning of @xmath251 eq . ( [ eq - gq ] ) . before we get into a lot of algebra , let us describe the physical idea . in scattering theory we know that the born series is a perturbation series of the scattering amplitude in terms of the potential . in fig . [ fig - graph ] we show the first three terms graphically , where the first term , the first born approximation , is particularly simple it is the fourier transform of the potential . we also know that when the potential is weak the first few terms are an good approximation to the scattering amplitude , but when the potential is strong , we need all terms . in this section , we will show that our t - matrix @xmath382 is the sum of all such two - body scattering terms . we start with eq . ( [ eq - sch ] ) which we copy here for convenience , @xmath385 for @xmath208 we break up the sum over @xmath249 into two terms and get , @xmath386 for @xmath223 we can rewrite eq . ( [ eq-1 ] ) to get , @xmath387 plug eq . ( [ eq-3 ] ) into the second sum in eq . ( [ eq-2 ] ) and rearrange terms , we get , @xmath388 where we have defined , @xmath389 now break the sum over @xmath390 in eq . ( [ eq-4 ] ) into two parts , and we get @xmath391 plug eq . ( [ eq-3 ] ) into the last term of eq . ( [ eq-6 ] ) and we get @xmath392 where we have defined what we have done here is the traditional perturbation theory using iteration . ( [ eq-9 ] ) is the born series for scattering amplitude . the first term @xmath395 , the fourier transform of the potential @xmath58 , is the first born approximation . the content of higher order terms @xmath396 , @xmath397 , ... can be obtained from their definition . ( [ eq-5 ] ) says that @xmath396 involves two scatterings under @xmath12 , and eq . ( [ eq-8 ] ) says that @xmath397 involves three scatterings . thus the born series eq . ( [ eq-9 ] ) can be graphically depicted at the ladders in fig . [ fig - graph],@xcite and it involves multiple scatterings to all orders . note that each term in the born series is infinite for infinite potential @xmath12 . next we will show that summing all the terms in the series gives the t - matrix and the potential @xmath12 cancels out , giving a finite value . n. g. zhang and c. l. henley , unpublished ( cond - mat/0206421 ) . note that with spinless fermions and hardcore bosons and infinite nearest - neighbor repulsion , the filling ( particle per lattice site ) in our model goes from 0 to 1/2 only . pichard , g. benenti , g. katomeris , f. selva , and x. waintal , to appear in _ exotic states in quantum nanostructures _ ( kluwer , dordrecht ) , ed . s. sarkar ( cond - mat/0107380 ) . one could speculate that , if their model included our infinitely strong nearest - neighbor repulsion , the hilbert space ( identical to ours ) would be substantially reduced , without very much error in the energies . we are slightly abusing the notation @xmath12 . @xmath58 is the potential function for all space . @xmath12 is the value of the potential at four points ( @xmath406 for example ) . and @xmath407 the fourier transform of @xmath58 . we will differentiate these meanings by including explicitly the argument , @xmath55 or @xmath366 . without argument , we then mean @xmath406 . incidentally the interesting problem of computing the resistance between two points on a lattice resistor network turns out to be an application of the lattice green functions . see j. cserti , am . * 68 * , 896 ( 2000 ) . k. huang and c. n. yang , phys . rev . * 105 * , 767 ( 1957 ) ; k. huang , c. n. yang , and j. m. luttinger , phys * 105 * , 776 ( 1957 ) ; t. d. lee and c. n. yang , phys . rev . * 105 * , 1119 ( 1957 ) ; t. d. lee , k. huang , and c. n. yang , phys . rev . * 106 * , 1135 ( 1957 ) . @xmath411 is in fact a function of @xmath131 and @xmath132 , but in our discussion in this paper , we always consider a fixed column @xmath132 , so we omitted the @xmath132 dependence in the notation @xmath411 .

we propose to simulate a dirac field near an event horizon using ultracold atoms in an optical lattice . such a quantum simulator allows for the observation of the celebrated unruh effect . our proposal involves three stages : ( 1 ) preparation of the ground state of a massless 2d dirac field in minkowski spacetime ; ( 2 ) quench of the optical lattice setup to simulate how an accelerated observer would view that state ; ( 3 ) measurement of the local quantum fluctuation spectra by one - particle excitation spectroscopy in order to simulate a de witt detector . according to unruh s prediction , fluctuations measured in such a way must be thermal . moreover , following takagi s inversion theorem , they will obey the _ bose - einstein _ distribution , which will smoothly transform into the _ fermi - dirac _ as one of the dimensions of the lattice is reduced .

1606.09505

the path towards quantum gravity opened a territory full of surprises : quantum field theory in curved spacetime @xcite . bekenstein s phenomenological thermodynamics of black holes @xcite received a strong support from hawking , when he found that a black hole must emit thermal radiation @xcite . the discovery hinted that thermal effects might appear without any underlying stochasticity . fulling , davies , and unruh proposed that a similar effect existed in an essentially flat spacetime , i.e. , rindler spacetime : an accelerated observer through an empty minkowski spacetime will perceive a thermal bath of particles , at a temperature proportional to its acceleration @xcite . both phenomena are intimately related : in both cases , an event horizon , which prevents communication between different regions of spacetime , is developed . furthermore , in order to observe hawking s radiation one must stay at rest near a black hole , and therefore feel an acceleration . a further surprise was revealed when takagi studied the relation between dimensionality and the unruh thermal spectrum @xcite . in 3 + 1d , an accelerated detector of bosonic particles in minkowski spacetime will record a bose - einstein distribution , and a detector of fermionic particles will find a fermi - dirac distribution . but this is only true if the dimension of space is odd . otherwise , an apparent _ statistics inversion _ phenomenon takes place : bosons are detected with a fermi - dirac distribution , while fermions are detected with a bose - einstein distribution . the unruh effect is not just an exotic curiosity : it bears a deep relation to entanglement @xcite and black hole thermodynamics , and it plays a central role in jacobson s derivation of einstein equations as equations of state for spacetimes in thermal equilibrium @xcite . these results point to a fundamental nature of the unruh effect as a quantum counterpart of the principle of equivalence , which it corrects @xcite . moreover , the unruh effect can be regarded as a particular case of _ parametric amplification _ of the vacuum fluctuations @xcite , which puts it in the same class of phenomena as the dynamical casimir effect @xcite . the latter can be seen as a flat spacetime analog of the hawking effect and connected with the unruh thermal bath close to the black hole horizon . the intriguing relation between the unruh and dynamical casimir effects has been also explored in the context of brane physics @xcite . the fundamental relevance of the unruh effect provides a strong motivation to measure it and related phenomena in the laboratory @xcite ( see also @xcite for some more recent proposals ) . given the difficulty of the task , a different approach has been to develop analogue gravity systems where hawking radiation might show up @xcite . one of the first ideas @xcite was to build a _ sonic analogue _ of a black hole in a moving medium , whose speed of sound replaces the speed of light . if the relative velocity between parts of the propagating medium is larger than the speed of sound , an effective horizon appears . the medium can be either water @xcite or a bose - einstein condensate ( bec ) @xcite , which can be employed also to probe the dynamical casimir effect @xcite . a specific proposal for measuring the unruh effect in this setting , using an accelerated impurity as de witt detector , was proposed in @xcite ( impurities can be used also as detectors of casimir forces and quantum friction @xcite ) . other very interesting approaches are to use a non - linear optical medium in which a refractive index perturbation moves at high speed @xcite , or to exploit the geometric properties of graphene sheets @xcite . the use of engineered lattices of superconducting qubits @xcite has been already used to probe the dynamical casimir effect @xcite and proposed for unruh physics . in this work , we take a different strategy and propose a new framework for simulating the unruh effect , which is based on the _ quantum simulation _ of dirac fermions using ultracold fermionic atoms in a 2d optical lattice @xcite . the possibility of simulating the dirac hamiltonian in certain spacetime metrics was recently put forward by some of us , where the information about the metric is encoded in the tunneling terms shaped by the lasers @xcite . building upon that framework , we propose to start the experiment by setting up an optical lattice whose dynamics simulates the massless dirac hamiltonian in 2 + 1d in minkowski spacetime , where the fermi velocity , analogue to the speed of sound in a bec , plays the role of the speed of light . by achieving the ground state , we can assume that our quantum state is the dirac vacuum in minkowski spacetime . now , we can _ quench _ the system by suddenly changing the tunneling terms in the lattice to the values corresponding to the dirac hamiltonian in a rindler metric . in other terms , the same dirac physics minkowski vacuum of dirac fermions but viewed by an accelerated observer . canonical observation of the unruh effect should be performed now by a local _ de witt detector _ @xcite , a device whose purpose is to couple minimally to the quantum fluctuations of the field and interchange energy @xcite . the full spectrum of local fluctuations obtained is predicted to follow both unruh and takagi s predictions . what is the novelty of our approach ? our setup is a quantum simulator , i.e. , a _ quantum computer of special purpose _ @xcite that allows for a systematic study of gravitating quantum matter . for instance , within our quantum simulator it is possible to change the fermi velocity or the shape of the metric . more importantly , it provides a framework for systematically studying quantum many - body systems @xcite . beyond the free fields studied in this work , let us emphasize that the setup we propose for simulating the unruh effect can be used also for studying interacting fermions in curved spacetime , and that it allows for the subtle manipulations needed to simulate experiments in relativistic quantum information @xcite . another parameter which can be easily tuned is the dimensionality of the artificial spacetime , thus allowing us to probe the aforementioned inversion theorem of takagi @xcite . the investigation of the unruh effect bears a strong relation to the study of boundary effects . indeed , the horizon can be considered as a boundary for fields which are accessible to the accelerated observer . ensuring that the boundary conditions do not spoil the unitarity of the theory imposes certain conditions on the hamiltonian @xcite which , as we will show , are fulfilled naturally for the dirac hamiltonian in rindler spacetime , and provides a procedure to perform the right discretization . surprisingly , our hamiltonian has the same form as one of the candidates to solve the riemann conjecture via the hilbert - polya approach , @xmath0 @xcite . in a different line , our model bears relation to the hyperbolically deformed hamiltonians @xcite and to the techniques of off - diagonal confinement in optical lattices @xcite . as our work is meant to be directed to a wide audience , we try as much as possible to keep it self - contained . in section [ sec : rindler ] we provide a pedagogical overview of the relativistic physics for an accelerated observer , both classical and quantum . section [ sec : dirac ] introduces the dirac hamiltonian in rindler spacetime and discusses its discretization . readers mainly interested in the proposed quantum simulation of the unruh effect could go directly to section [ sec : proposal ] , where we detail our quench strategy , provide numerical simulations of the expected results and suggest a possible experimental implementation . we finish in section [ sec : conclusions ] with conclusions and proposals for further work . this section is a review of the physics of an accelerated observer . we will discuss in a pedagogical fashion the basics of the rindler metric , the thermalization theorem and the unruh effect . let us briefly review rindler physics , i.e. , minkowski spacetime viewed by an accelerated observer @xcite . let us consider an observer moving with constant acceleration @xmath1 ( for convenience in the following we take the speed of light to be @xmath2 ) in the positive @xmath3-axis , at rest at @xmath4 and @xmath5 . physics seen by this observer is more properly described in a co - moving reference frame , obtained by the _ fermi - walker _ transport procedure . let @xmath6 be the co - moving time coordinate for this observer , and @xmath7 the co - moving space coordinate . they are called _ rindler coordinates _ , and can be found using this transformation ( see fig . [ fig : rindler ] ) @xmath8 in particular , the considered trajectory corresponds to @xmath9 for all @xmath6 . notice the similarity with polar coordinates , where @xmath7 plays the role of a radius and @xmath6 is an angle in hyperbolic geometry . the principle of equivalence states that physics seen by a non - inertial observer can be absorbed by a change in her metric . indeed , in these coordinates , the minkowski metric becomes @xmath10 which is known as the rindler metric . notice that the rindler time direction corresponds to a symmetry of the metric , i.e. , it constitutes a killing vector which is inequivalent to the usual minkowski time direction . in fact , it corresponds to a boost transformation . in the polar coordinates view , it is the generator of hyperbolic rotations . furthermore , the pole @xmath11 corresponds to a singularity in the coordinate system , because the coefficient of @xmath12 vanishes . this is the hallmark of an _ event horizon_. in fact , one can consider the rindler metric as a particular example of _ optical metric _ where the only non - trivial entry of the metric is @xmath13 , which becomes position - dependent . these are called optical metrics because propagation of light along the geodesics is equivalent to `` galilean '' propagation in a refractive medium with a position - dependent refractive index @xmath14 , which implies the `` local '' speed of light @xmath15 . the corresponding optical metric is of the form @xmath16 for rindler spacetime , @xmath17 . as @xmath18 , the local speed of light vanishes , which implies that signals can not propagate beyond that point . thus , spacetime is separated into two parts which do not communicate : the two rindler wedges , @xmath19 and @xmath20 . it is remarkable that an event horizon can appear even in a flat spacetime . ( rindler time ) and @xmath7 ( rindler space ) . the rindler wedge , delimited by dashed lines , is the domain of validity of the coordinate patch . constant @xmath6 lines ( green ) are spacelike , and constant @xmath7 lines ( red ) are timelike . for simplicity , we plot the trajectories only in the right wedge , @xmath21 , as the ones for the wedge @xmath22 can be obtained by reflection around the @xmath23-axis . note that as detailed in sect . [ sec : proposal ] both wedges are realized and are at the same footing in our proposal . , width=340 ] let us return to the proposed accelerated observer , which in rindler spacetime just sits at rest at @xmath9 . from her point of view , light moves at her left more slowly than usually , and faster at her right . near the horizon , @xmath11 , light moves more and more slowly , coming to stop at @xmath11 , i.e. , its local speed of light is zero ( while the actual speed of light stays obviously constant to 1 ) . let us now consider objects which are static with respect to the accelerated observer , i.e. , objects at rest in rindler spacetime at different values of @xmath7 . tracing back their trajectories to minkowski spacetime , it can be checked that they correspond also to accelerated trajectories , with acceleration @xmath24 . this implies that , in order to keep pace with an observer with acceleration @xmath25 in front of you , you must accelerate faster than that @xcite . this result is known as _ bell s spaceship paradox_. the interplay between thermodynamics and general relativity gives raise to surprising properties . let us restrict ourselves to spacetimes which contain a time - like killing vector , i.e. , spacetimes whose metric can be said to be time invariant , where we have a well defined concept of energy . in that case , _ tolman - eherenfest s theorem _ @xcite states that for any field in thermal equilibrium within a stationary curved spacetime , the product of the local temperature and the modulus of the local time - like killing vector is constant , @xmath26 const . there is a simple way to visualize this result . photons emitted at one point @xmath27 in space with frequency @xmath28 will reach another point in @xmath29 with a redshifted frequency , @xmath30 . thus , the same factor should be applied to energies and to temperatures . thus , in rindler spacetime , temperature at any point is inversely proportional to the distance to the horizon . or , in other words , it is proportional to the acceleration of an observer stationary at that point . notice that this does not entail a non - zero temperature . the theorem still holds if the temperature is zero everywhere . but the biggest surprises show up when we introduce quantum mechanics @xcite . let us consider a free fermionic field in minkowski spacetime , with hamiltonian @xmath31 , described in terms of local creation operators @xmath32 . the physical vacuum is the ground state of its hamiltonian , @xmath33 , and it does not correspond to the fock vacuum , @xmath34 , which is defined by @xmath35 for all @xmath3 . in the physical vacuum , all the negative energy single - particle modes will be occupied @xmath36 where @xmath37 creates the @xmath38-th mode , and @xmath39 is its energy . according to the usual convention , @xmath37 creates a particle if @xmath40 and an anti - particle if @xmath41 . therefore , the physical vacuum is built by occupying all the anti - particle states , and none of the particle ones . let us now consider an accelerated observer moving through this vacuum . she will see physics displayed not on minkowski spacetime , but on a rindler metric . let @xmath42 be the appropriate hamiltonian operator , which is also a free fermionic hamiltonian . its single - particle modes are known as the rindler modes , @xmath43 . they have energies @xmath44 and are solutions to the wave equation in the rindler metric ( here the index @xmath45 labels the eigenstates , but does not correspond to momentum in the acceleration direction , since translational invariance is broken ) . the ground state of @xmath42 is @xmath46 again , the rindler modes will qualify either as particles , if @xmath47 , or anti - particles if @xmath48 . the pure state @xmath33 does not need to be an eigenstate of @xmath42 , much less its ground state . from the point of view of the accelerated observer , who measures energies with @xmath42 , @xmath33 is not the true vacuum any more . how does this state look like to her ? it is crucial to realize that the rindler metric has a horizon , which separates space into two parts which can not communicate . thus , she will not detect @xmath33 , but the _ reduced density matrix _ which results of tracing out the hidden part @xmath49 where @xmath50 means a trace over the left - out degrees of freedom . a reduced density matrix can always be formally written as a thermal state @xmath51 where @xmath52 is called the entanglement hamiltonian @xcite . since the minkowski vacuum is a slater determinant , we can use wick s theorem in reverse to prove that the entanglement hamiltonian must be a free fermionic hamiltonian @xcite . in other terms , the accelerated observer will see a thermal state of free particles . the minkowski vacuum is invariant under lorentz boosts , which correspond to time translations in rindler spacetime . for @xmath53 , this property implies @xmath54 . \label{commutator}\ ] ] thus , @xmath55=0 $ ] , i.e. , the entanglement hamiltonian and the rindler hamiltonian must commute . in fact , they can be non - trivially proved to be proportional , and the constant of proportionality can be read as an inverse temperature @xmath56 where @xmath57 here @xmath58 is known as the _ unruh temperature _ and this result , which is far more general than the particular case studied here , is the _ thermalization theorem _ @xcite . thus , the unruh temperature does not appear because of any underlying stochasticity . the loss of information which gives rise to the thermal effect is related to the presence of the horizon . an important consequence of thermalization theorem is that minkowski vacuum _ appears to be stationary to an accelerated observer_. indeed , while minkowski vacuum is clearly not an eigenstate of dirac hamiltonian in rindler spacetime and , thus , evolves non - trivially in rindler time , _ what an accelerated observer detects is invariant under such time evolution _ as the thermal state is a diagonal density matrix in the rindler eigenbases . this observation is crucial in our proposal for simulating the unruh effect with ultracold atoms , see sect . [ sec : proposal ] . let us consider the canonical transformation between minkowski ( @xmath37 ) and rindler modes ( @xmath43 ) @xmath59 this is a bogoliubov transformation in disguise , because a positive energy rindler mode ( particle ) requires both positive and negative minkowski modes for its expansion ( particle and anti - particle ) . in fact , as the acceleration is position dependent , the relevant bogoliubov transformation has to be defined locally . in the continuous limit , the global bogoliubov transformation is even ill - defined as the eigenstates are not normalizable . in order to properly define it we have to consider normalized states @xcite , for instance wave packets centered around a generic point @xmath60 . on physical terms , this means that we can associate a well - defined acceleration to the rindler wave - packet . for practical purposes , the wave - packet normalization is equivalent to restricting the scalar product of the unnormalized modes to a small region @xmath61 such that @xmath62 of space around @xmath60 . with this definition , the occupation of each rindler mode on the minkowski ground state is given by @xmath63 and the thermalization theorem ensures that @xmath64 with @xmath65 according to eq . , where @xmath3 is the spatial distance of the point @xmath66 from the horizon . but the _ unruh effect _ goes beyond the thermalization theorem , because it is defined _ operationally _ , in terms of what a local observer can measure . the so - called _ de witt _ detector @xcite is a device carried along with the observer , which couples minimally to the fermionic field at a spatial point @xmath60 , and can emit and absorb particles . under a large variety of circumstances it can be proved that the probability amplitude of absorption / emission is given solely by the _ wightman function _ @xmath67 here , @xmath68 is the trajectory for the observer for simplicity we consider trajectories parallel to the @xmath3-axis and we omit remaining constant spatial coordinates and @xmath69 is the creation operator for a fermion at event @xmath70 . the fourier transform of @xmath71 , @xmath72 , is the _ detector response function _ , which should be experimentally accessible , as we will discuss later . the formula makes equal sense in minkowski or in rindler spacetimes , if we are allowed to abuse notation and let @xmath3 and @xmath73 denote the coordinates in both . in rindler spacetime , the trajectory of an accelerated observer will be just a constant @xmath74 . let us define two different basis changes , from rindler space - localized states to rindler and minkowski modes , respectively . at time @xmath4 , if @xmath32 creates a particle at point @xmath3 , we have @xmath75 where the unitary matrices @xmath76 and @xmath77 are the single - particle wave functions of minkowski and rindler modes , respectively , and determine the unitary transformation @xmath78 in , @xmath79 . here and in the following by the bar we denote the complex conjugate of the matrix elements . so we get @xmath80 thus , the detector response function is strongly dependent on the form of the rindler and minkowski modes , through @xmath81 and @xmath82 . going beyond the thermalization theorem , unruh predicted that the distribution function @xmath72 will be _ thermal_. but a surprise is hiding behind eq . due to the spatial dependence of the rindler and minkowski modes . if the dimension of space is _ odd _ , then the response function of a fermionic field will follow the fermi - dirac distribution function , as one would expect . but if the dimension of space is _ even _ , @xmath83 will follow a bose - einstein distribution . the opposite is true for a free bosonic field . this fact , known as takagi s inversion theorem @xcite stems from dimensional effects in wave propagation , analogous to those observed for light propagating radially . in odd dimension , the huygens principle holds , and a pointlike perturbation after a time @xmath73 is concentrated in a spherical shell of radius @xmath84 , where @xmath85 is the propagation velocity . in even dimension , however , not all the scattered waves propagate at the same @xmath85 , the huygens principle does not hold and the perturbation becomes radially spread with time . alternative physical meanings of the detector response function are worth mentioning . the first is a measure of quantum fluctuations : @xmath83 is the power spectrum of the quantum noise @xcite . the second is related to the dynamical casimir effect . let us consider a physical plane in space , whose interaction with our fermionic field can be expressed as a dirichlet boundary condition . now let us move this plane with constant acceleration @xmath25 . then , the stress - energy tensor at any point will depend on its current distance to the plane . in fact , it can be proved @xcite that the limit @xmath86 can be made meaningful , thus providing a well defined stress - energy tensor for the rindler vacuum , which induces quantum fluctuations that are probed by expression . in this section we describe the behavior of dirac fermions in rindler spacetime for one and two spatial dimensions . in particular , we explicitly construct the corresponding hamiltonian in a square lattice . indeed , since the rindler metric has a time - like killing vector field , we can use a hamiltonian formalism and discretize it to get a simple tunneling model . the resulting model bears a surprising resemblance to the @xmath87 hamiltonian used in the hilbert - polya approach to proving the riemann conjecture . this point is further detailed in appendix [ sec:1drindler ] . let us consider a relativistic massless fermionic field in two dimensions , governed by the dirac equation in minkowski spacetime @xmath88 where the @xmath89 are a representation of the clifford algebra , @xmath90 , where @xmath91 is the minkowski metric and @xmath92 . in , as well as in the rest of the section , sums over repeated indices are left implicit according to einstein s convention . as it stands , the equation is manifestly lorentz covariant . let us shift to a hamiltonian view , which is more convenient for simulation . in other words , we single out the time - derivative @xmath93 let us make the following choice for the @xmath94 matrices in two dimensions , @xmath95 , @xmath96 , @xmath97 . we obtain @xmath98 equation can be formulated on a general ( curved ) background metric @xmath99 as well . for spinor systems it is very convenient to introduce the _ vielbein _ , which is a set of vectors defined on the tangent manifold , @xmath100 , such that @xmath101 . the parallel transport for the vielbein vectors defines the _ spin - connection _ , @xmath102 , and allows a compact expression for the covariant derivative of a spinorial field @xcite @xmath103 where @xmath104 $ ] . by making use of it , the dirac equation reads @xmath105 where the _ curved _ gamma matrices @xmath106 are defined by @xmath107 , and the curved indices @xmath108 are lowered and raised by contracting with the metric @xmath99 and its inverse @xmath109 . when we single out the time - derivative , we obtain again a schrdinger equation of the form @xmath110 where @xmath111 . let us now consider the specific case of the 2d rindler metric @xmath112 . the only non - vanishing element of the spin - connection is @xmath113 . with the aforementioned choice for the @xmath94 matrices , we get @xmath114 thus , the hamiltonian density becomes @xmath115 which is the single - particle rindler hamiltonian . its second - quantized form is simply @xmath116 the same expression can obviously be derived by taking the legendre transformation of the dirac lagrangian in rindler spacetime . in intuitive terms , the @xmath117 term is related to the volume form , @xmath118 . the @xmath119 term comes for the covariant derivative and it is essential to ensure the hermiticity of @xmath42 . indeed , this factor cancels the so - called _ deficiency indices _ @xcite , i.e. , allows us to treat the horizon at @xmath120 as a boundary , ensuring that any boundary condition can be imposed while respecting self - adjointness of the hamiltonian . this property is more evident once is cast in symmetric fashion , i.e. , the spatial derivatives act symmetrically both on @xmath121 and @xmath122 , @xmath123 in this form , the propagation in rindler metric is sensitive only to the overall scale factor which determines a fermi velocity that changes linearly along the @xmath3 direction . it is worth noticing that the equivalent symmetric formulation of single - particle hamiltonian is @xmath124 which is also manifestly hermitian . it will be further discussed in the appendix [ sec:1drindler ] , in relation with the riemann conjecture . the minkowski and rindler dirac hamiltonians in one and two spatial dimensions can be suitably discretized on the lattice @xcite . as shown in detail in the appendix [ subsec : discr ] , a convenient way of doing this is to consider a 1d - chain or a 2d - square lattice with non - interacting spinless fermions @xmath125 where the sum runs over all pairs of nearest neighbors sites @xmath126 . since the 1d - dirac models can be realized as a slice along @xmath3 ( defined as the direction perpendicular to the rindler horizon ) of the 2d - dirac ones , we focus on the latter case . hamiltonian can represent the dynamics of each of the chiral components of the minkowski dirac hamiltonian if the tunneling terms @xmath127 have all the same modulus and the sum of their phases around each plaquette is @xmath128 . this corresponds to the well known @xmath128-flux hamiltonian @xcite . all possible choices of phases respecting the @xmath128-flux condition are equivalent , as they are related by gauge transformations . we will focus on the one corresponding to the symmetry gauge for the synthetic gauge field associated to the phases . precisely , @xmath129 where we adopt cartesian coordinates to parametrize the lattice , @xmath130 and denote with @xmath131 the lattice spacing . the discretized version of the rindler dirac hamiltonian can also be chosen to be of the form , but with spatially modulated tunnelings , @xmath127 . each tunneling rate has to be proportional to the average @xmath3 coordinate of each link , which represents the distance from the horizon . we place the horizon at @xmath120 accordingly to the coordinates chosen in . the tunneling phases have to satisfy the same @xmath128-flux condition as for the dirac hamiltonian in minkowski space . for the symmetric gauge choice of we have @xmath132 the numerical simulation and the experimental implementation in optical lattices of the hamiltonians and will be discussed in the next section . we would like to remark here that in principle any other lattice realization of the dirac hamiltonian like the ones in bichromatic @xcite , hexagonal @xcite and brick - wall lattices @xcite , which do not involve artificial gauge fields , can be considered and deformed by shaping the tunneling term to reproduce the dirac hamiltonian in rindler spacetime . other artificial lattice dirac systems such as nano - patterned 2d electron gases , photonic crystals , micro - wave lattices @xcite or polaritons @xcite could also be used . since the unruh effect is a single - particle and purely kinematic effect , it could be studied using both bosonic and fermionic systems . the latter offers a simple route to explore the relativistic ( linear dispersion relation ) regime , as detailed in the next section . in this section we present our proposal to study the unruh effect for dirac fermions in an optical lattice , in one and two spatial dimensions . the crucial idea behind our proposal is that all measurements made by an accelerated observer on the minkowski vacuum of the dirac field can be simulated by _ quenching _ the dirac hamiltonian from the one in minkowski spacetime to the one in rindler spacetime , which amounts to quench the tunneling amplitudes from constant to properly position - dependent values . as a by far non - trivial consequence of thermalization theorem ( _ cf . _ [ sect : thermalization ] ) , the minkowski vacuum will now be seen as a thermal state in rindler , which we will subsequently probe with a suitable analogue of de witt detectors , yielding the local fluctuation spectrum predicted by the unruh effect . as a thermal state corresponds only to populations of rindler modes , the minkowski vacuum is stationary , that is to say that is invariant under time translations in rindler spacetime ( this property is not so surprising because rindler time translations correspond to lorentz boosts in the original minkowski coordinates ) . we start by providing an overview of the experimental procedure . our scheme relies crucially on one - particle excitation spectroscopy , which we discuss in detail . the robustness of our scheme is then validated by performing a numerical simulation of the response function in realistic experimental conditions . we conclude by proposing an experimental implementation of the protocol which is accessible using state - of - the - art techniques . as explained in sect . [ sec : unruh_effect ] , the observation of the unruh effect requires a measurement of the wightman two - point correlation function in the frequency domain for an accelerated observer in the minkowski vacuum . in other terms , we have to measure the fourier transform of two - point correlations in time . for a dirac system as the one we consider , the minkowski vacuum is the fermi sea and what needs to be measured is the overlap between the state corresponding to a one - hole excitation at different times . furthermore , this one - hole excitation must follow an accelerated trajectory . traditionally , an accelerated observer is considered in order to observe the unruh effect , with the minkowski vacuum at rest and the one - hole excitation moving . for instance , this is the approach considered in @xcite , where the one - hole excitation is created by the coupling to an impurity . however , due to the equivalence principle , the measurement can actually be done in any reference frame . we choose to perform it in the rest frame of the observer and the one - hole excitation . there , the time evolution is governed by the dirac hamiltonian in rindler spacetime ( eq . ( [ rindler.ham ] ) ) , and the response function is simply the overlap between the one - hole excitation at rest at different times . the measurement of the wightman spectral function can then be interpreted as the creation of a one - hole excitation at a fixed location @xmath133 in the fermi sea , the evolution of this state with the dirac hamiltonian in rindler spacetime for a time @xmath73 , and the creation of a particle at @xmath133 . this is exactly what one - particle excitation spectroscopy , a standard technique in cold atom experiments , determines @xcite . thus , our protocol to observe the unruh effect consists of three steps : 1 . _ preparation of the minkowski vacuum _ by achieving the ground state of the dirac hamiltonian with a uniform fermi velocity ( dirac hamiltonian in minkowski spacetime ) ; 2 . _ quench to an accelerated frame _ governed by the dirac hamiltonian with a spatially dependent fermi velocity ( dirac hamiltonian in rindler spacetime ) . the quench introduces an event horizon in the middle of the gas , effectively disconnecting it in two halves ; 3 . _ measurement of the wightman function _ in the accelerated ( rindler ) frame using local one - particle excitation spectroscopy at point @xmath133 . the first two steps provide a convenient method for preparing the minkowski vacuum as the ground state of a hamiltonian which can be easily implemented experimentally , and for making it evolve into an accelerated ( rindler ) frame . the third step , local one - particle excitation spectroscopy , is the crucial ingredient of our proposal . it creates a one - hole excitation in the gas , whose dynamics in the accelerated ( rindler ) frame produces the bogoliubov transformation . and it is this transformation which is responsible for the thermalization theorem and the unruh effect . given its importance , we describe it in detail in the next section . our proposal for observing the unruh effect relies on the use of one - particle excitation spectroscopy for measuring the wightman function . this technique consists in transferring a fraction of atoms of the gas to an auxiliary energy band which is initially unoccupied and has a considerably smaller bandwidth , so that it can be neglected . the process requires a field coupling both bands , and can be implemented in a variety of fashions ( radio - frequency , one - photon or two - photon laser transitions ) depending on the atomic species chosen . in our case , we require the process to be local , since the wightman function is defined locally ( at point @xmath133 ) and the unruh temperature varies as a function of the distance to the horizon . if we consider the ensemble of the two bands as an effective two - level system , the effect of the coupling can be modeled in the interaction picture as @xmath134 where @xmath135 represents the detuning between the frequency of the field and the energy of the auxiliary band where the operator @xmath136 ( @xmath137 ) creates ( destroys ) an atom , and we assume an integration over all momenta . since the measurement is performed after the quench , the operator @xmath138 ( @xmath139 ) evolves with the rindler hamiltonian . now , let us compute the occupation of the auxiliary band at a later time . as it is highly excited , we can assume it to be initially empty . the initial state is thus @xmath140 taking a sufficiently small coupling @xmath141 allows us to treat at first order in perturbation theory . we find @xmath142 then , the occupation @xmath143 of the auxiliary state for @xmath144 is @xmath145 [ eq : nb ] where we have used translation invariance in time . the calculation above not only demonstrates that local one - particle excitation spectroscopy measures the wightman spectral function . it also clearly shows that it is the time evolution under the rindler hamiltonian which is responsible for the observed thermal response . we have calculated the occupation @xmath143 assuming that the pulse started at @xmath4 , i.e. , immediately after the quench we have used the original minkowski vacuum to start with. however , the integral depends only on the duration of the pulse , and not on the actual moment at which the pulse starts . this is an experimental manifestation of the stationary of minkowski vacuum in rindler spacetime and of the time invariance of the associated populations as rindler particles . let us now discuss in detail the range of validity of our approach and review some possible limitations . in our scheme we do not implement the dirac hamiltonian ( both in minkowski and rindler spacetimes ) in the continuum , but only a lattice version of it . this introduces a characteristic length scale in the system , the lattice spacing , and an associated uv energy cut - off . measurements of the wightman function below this length scale are not meaningful . however , the finite spatial resolution that one - particle excitation spectroscopy will have in the experiment naturally smears out these discretization artifacts . we will show below that a measurement of the response function convolved over two lattice sites is sufficient to suppress most of them . another consequence of implementing the dirac hamiltonian in an optical lattice is that the relativistic dispersion relation only holds in a certain range of energies , in the vicinity of the dirac points . thus , the measurements must be restricted to this energy range , which is given by the local tunneling rate . this limitation is common to other proposals for simulating relativistic effects with cold atoms . for instance , using the bogoliubov excitations of a bose - einstein condensate as relativistic particles is only valid in the phonon - like regime of the bogoliubov dispersion relation , and breaks down away from it . our protocol relies on a change of reference frame , from a rest frame to an accelerated one . this step is done by quenching the hamiltonian from minkowski to rindler spacetime . the change of reference frame should be instantaneous , an approximation which is valid if the quench time is much shorter than the smallest characteristic timescale of the system given by the inverse of the largest tunneling rate . we will see in sect . [ sect : implementation ] that experimentally this is a reasonable assumption . the main effect of the quench is to introduce an artificial horizon in the middle of the lattice that effectively disconnects the left and right halves . placing the horizon exactly in the middle of the system is important to minimize the distortions induced by the finite system size . finally , let us remark that quenching the hamiltonian of a quantum system normally triggers a temporal evolution of its initial state . in our case , however , as we have observed at the end of sect . [ sect : thermalization ] and further argued in this section , the minkowski vacuum looks stationary to the accelerated observer and this dynamics is absent . indeed , the rindler hamiltonian is proportional to the entanglement hamiltonian of both halves of the system . thus , the density matrices of both halves are time - invariant as they are diagonal and depend only on the populations . the unruh effect implies that measurements of the wightman function at different distances from the horizon , and thus different accelerations , will yield different values of the unruh temperature . in order to compare these measurements , the rates should be measured with respect to the proper time @xmath146 . for the rindler metric , eq . , @xmath147 with latexmath:[$\xi \propto same time , the fourier transformation of the wightman function @xmath71 has to be performed with respect to the proper time . in the frequency domain it is then given by @xmath149 . finally , up to now we have been assuming that the minkowski vacuum can be exactly realized in the experiment . or , in other terms , that it is possible to prepare perfectly the ground state of the dirac hamiltonian in the homogeneous tunneling lattice at half filling ( exactly up to the dirac points ) . in real experiments , however , the actual temperature of the fermionic gas will not be zero but rather on the order of the tunneling rate . we will show that the signatures of the unruh effect can still be appreciated when starting with a finite - temperature sample . in order to validate our scheme and address the effects presented above we have performed numerical simulations , which we present in this section . the calculations have been done using the @xmath128-flux realizations of the dirac hamiltonian ( eqs . ( [ pi.flux.hamiltonian ] ) and ( [ rindler.pi.flux.hamiltonian ] ) ) in one and two spatial dimensions . in both cases we have simulated numerically the complete scheme , starting with an initial state in minkowski spacetime , assuming an instantaneous quench , and computing then the wightman response function as will measured by one - particle excitation spectroscopy ( see eq . ) . in the calculations we fix the system sizes @xmath150 , @xmath151 . the natural energy scale of the system is the bandwidth of the dirac hamiltonian in minkowski spacetime ( proportional to the tunneling strength @xmath152 in ) . therefore , all energies ( i.e. frequencies and temperatures ) are measured in units of @xmath152 . the amplitude @xmath153 that characterizes the tunneling strength of the dirac hamiltonian in rindler spacetime is in principle arbitrary due to the overall scale invariance of the rindler space we choose it so that the maximal tunneling rate is equal to the minkowski value , @xmath154 . the lattice spacing is fixed as @xmath155 and we attach @xmath156 coordinates to each site in a symmetric way with respect to the horizon , i.e. the @xmath120 line . thus , while @xmath157 always runs over integers , @xmath158 , @xmath159 runs over integers for @xmath160 odd , @xmath161 , and over half - integers for @xmath162 even , @xmath163 . note that fixing the horizon exactly in the middle of the gas is important to minimize the distortions introduced by the finite size of the system . as we mentioned in the previous section , in a discrete realization of the dirac hamiltonian only measurements performed at length scales above the lattice spacing @xmath131 are meaningful . the finite spatial resolution of the measurements will automatically perform the required coarse - graining . we simulate it numerically by considering a convolution of the wightman function with a gaussian of standard deviation corresponding to @xmath164 lattice sites along the @xmath3 direction . the raw data obtained before the convolution , and further details concerning it are included in appendix [ sect : a3 ] . the frequency dependence of the response is evaluated at five different positions , at linearly increasing distances from the horizon . the top panel of fig . [ fig : unruh_1d ] ( a ) shows the convoluted results obtained for a 1d system of size @xmath165 . the tallest ( red ) curve is the closest to the horizon , and the lowest ( blue ) one is the most distant . for frequencies close to zero ( i.e. the fermi energy ) the response functions all have a behavior resembling a fermi - dirac distribution , with strong lattice artifacts at large negative frequencies . in order to compare the different results , we rescale the curves with respect to the proper time @xmath146 . the proper frequency is then @xmath166 , and the proper rate of detection is @xmath149 . for our choice of units @xmath167 , @xmath168 . thus , @xmath11 corresponds to the horizon and @xmath9 to the edge of the system . [ fig : unruh_1d ] ( b ) presents the same curves as fig . [ fig : unruh_1d ] ( a ) , but in rescaled units . for frequencies close to @xmath169 , they reproduce fermi - dirac distributions whose temperature increases as we approach the horizon . the distributions are _ not _ normalized , since they are defined up to a global constant . notice that in fig . [ fig : unruh_1d ] we have restricted the displayed frequency range to the regime where the energies are lower than the local tunneling range @xmath170 , since it is only there that the dispersion relation remains linear and the description of the particles in terms of dirac fermions is valid . in rescaled units this condition becomes @xmath171 . in the following we will restrict ourselves to this frequency range . after the quench . the colors denote different distances to the horizon , @xmath159 , expressed in lattice sites : blue ( lower ) is far away and red ( taller ) is closest to it . ( b ) wightman response function of the same system in the frequency domain , measured with respect to the proper time @xmath172 . the proper frequency is @xmath166 , while taking the fourier transform with respect to @xmath146 requires rescaling @xmath173 . as explained in the main text , @xmath149 represents what a static de witt detector in rindler spacetime would observe . notice that the curves collapse to fermi - dirac distributions of increasing temperatures as we approach the horizon . for @xmath174 lattice artifacts ( deviations from the relativistic dispersion relation ) distort the response.,title="fig:",width=321 ] after the quench . the colors denote different distances to the horizon , @xmath159 , expressed in lattice sites : blue ( lower ) is far away and red ( taller ) is closest to it . ( b ) wightman response function of the same system in the frequency domain , measured with respect to the proper time @xmath172 . the proper frequency is @xmath166 , while taking the fourier transform with respect to @xmath146 requires rescaling @xmath173 . as explained in the main text , @xmath149 represents what a static de witt detector in rindler spacetime would observe . notice that the curves collapse to fermi - dirac distributions of increasing temperatures as we approach the horizon . for @xmath174 lattice artifacts ( deviations from the relativistic dispersion relation ) distort the response.,title="fig:",width=321 ] fig . [ fig : unruh_2d ] shows the corresponding results for a 2d system , which differs strongly from its 1d counterpart as predicted by takagi s inversion theorem . [ fig : unruh_2d ] ( a ) shows the rescaled response function for a @xmath175 lattice , measuring at linearly increasing positions from the horizon as in the 1d case . the displayed curves include the spatial gaussian convolution , along with an energy coarse - graining @xmath176 . the latter simulates the finite energy resolution of the measurement , limited by the finite system size . the raw data , prior to convolution and coarse - graining is presented in appendix [ sect : a3 ] . as predicted by takagi , the results are now similar to a bose - einstein distribution . finally , in fig . [ fig : unruh_2d ] ( b ) we study the transition between 1d and 2d , by showing the rescaled response functions for a set of lattices with dimensions @xmath177 ( red ) , @xmath178 , @xmath179 and @xmath180 ( black ) , always measured at a point @xmath181 lattices sites away from the horizon . notice that the fermi - dirac distribution disappears very fast when we increase the transverse dimension @xmath151 . after the quench . as previously , the colors denote different distances from the horizon , @xmath159 , expressed in lattice units : blue ( + signs ) is far away , red ( full squares ) is closest to it . ( b ) same for strips of different widths , @xmath177 to @xmath180.,title="fig:",width=302 ] after the quench . as previously , the colors denote different distances from the horizon , @xmath159 , expressed in lattice units : blue ( + signs ) is far away , red ( full squares ) is closest to it . ( b ) same for strips of different widths , @xmath177 to @xmath180.,title="fig:",width=302 ] as a last step , we study the robustness of our protocol under an increase in the physical temperature of the gas . this results in an imperfect preparation of the minkowski vacuum , which is the starting point of the protocol . fig . [ fig : unruh_thermal ] ( a ) compares the rescaled response functions for a 1d system at physical temperature @xmath182 and @xmath183 , measured at two different points , one close to the horizon and one far from it . at @xmath183 the distributions are rounded near @xmath169 , but we can still see that the one closest to the horizon is more curved and presents a larger probability for positive energy excitations . the 2d case is more robust , as shown in fig . [ fig : unruh_thermal ] ( b ) . there , we can see that the rescaled distributions at @xmath184 measured near and far from the horizon are clearly distinguished , and keep the same global features than at @xmath182 . the explicit expression used to calculate the response functions for a thermal gas is given in the appendix [ sect : a3 ] . after the quench , for two different physical temperatures , @xmath182 and @xmath183 , and at two different locations , near and far from the horizon . the two response functions at @xmath183 resemble fermi - dirac distributions at finite temperature , but the one measured closer to the horizon corresponds to a higher temperature than the one measured far away . ( b ) same comparison for a 2d system of size @xmath175 . the response curves keep their global features when increasing the system temperature from @xmath182 to @xmath184 . the response function far from the horizon , @xmath185 , shows a local maximum at positive frequencies before decaying . this behavior is analogous to the one expected for a thermal gas of dirac fermions in the homogeneous tunnelling lattice ( see the appendix [ sect : a3 ] , fig . [ fig:2d_mink_gas ] ) . indeed , in rindler spacetime the limit @xmath186 corresponds to zero acceleration and converges to the results of minkowski spacetime.,title="fig:",width=302 ] after the quench , for two different physical temperatures , @xmath182 and @xmath183 , and at two different locations , near and far from the horizon . the two response functions at @xmath183 resemble fermi - dirac distributions at finite temperature , but the one measured closer to the horizon corresponds to a higher temperature than the one measured far away . ( b ) same comparison for a 2d system of size @xmath175 . the response curves keep their global features when increasing the system temperature from @xmath182 to @xmath184 . the response function far from the horizon , @xmath185 , shows a local maximum at positive frequencies before decaying . this behavior is analogous to the one expected for a thermal gas of dirac fermions in the homogeneous tunnelling lattice ( see the appendix [ sect : a3 ] , fig . [ fig:2d_mink_gas ] ) . indeed , in rindler spacetime the limit @xmath186 corresponds to zero acceleration and converges to the results of minkowski spacetime.,title="fig:",width=302 ] our proposal to implement experimentally the dirac hamiltonian in minkowski and rindler spacetimes is based on the recent experimental realizations of the hofstadter model with ultracold atoms @xcite , but in the symmetric gauge and using fermionic atoms instead . as sketched in fig . [ fig.implementationh ] , a two - dimensional square lattice with bare tunneling matrix elements @xmath187 along the @xmath3 and @xmath23 directions , and lattice spacing @xmath188 ( where @xmath189 is the wavelength of the lattice beams ) , is subjected to a potential gradient oriented along the diagonal direction of the lattice . this leads to an energy offset between neighboring sites @xmath190 which inhibits tunneling . the offset value could depend on the state of the atom , but should be identical along the @xmath3 and @xmath23 directions . a pair of raman laser beams collinear with the lattice beams , of wave vectors @xmath191 and frequencies @xmath192 result in an additional optical potential @xmath193 with @xmath194 and @xmath195 . the potential amplitude @xmath196 is assumed to be a slowly varying function of @xmath66 . the effect of the raman beams is to restore tunneling along the two directions when the condition @xmath197 is fulfilled , but with a spatial dependence of the phase . in the high frequency limit @xmath198 , the system is then described by the effective hamiltonian @xmath199 here the phase factor is @xmath200 . the dirac hamiltonians and are special cases of . for the raman laser propagation directions displayed in fig . [ fig.implementationh ] and a raman laser wavelength @xmath201 , the phases are @xmath202 , which corresponds to the @xmath128-flux hamiltonian in the symmetric gauge . the laser assisted tunneling amplitudes are then given by @xmath203 where @xmath204 is the bessel function of the first kind . this expression is valid in the limit @xmath205 and for slowly varying @xmath206 , which allows to use as average amplitude of the potential its value at the center of the link . this scheme allows for the simulation of the whole family of optical metrics @xcite considered by some of us in @xcite , and also of extensions of this family to include a mild time dependence in the metric @xcite . the realization of the dirac hamiltonian in minkowski spacetime requires laser - assisted tunneling amplitudes @xmath207 homogeneous across the cloud , which corresponds to a constant value of the raman optical potential amplitude . this could be realized using gaussian raman beams of waist @xmath208 much larger than the cloud size ( see fig . [ fig.implementationh]a ) . in order to implement the dirac hamiltonian in rindler spacetime , we need instead tunneling amplitudes which increase linearly along the @xmath3 direction , @xmath209 or , equivalently , a raman optical potential amplitude proportional to @xmath3 . using a tem@xmath210 hermite - gauss mode @xcite for the @xmath23 raman beam results in the large beam limit in a raman optical potential @xmath211 which , as follows from , leads to the required spatial dependence of @xmath73 ( see fig . [ fig.implementationh]b ) . the quench between the two situations ( minkowski and rindler ) could be performed by a sudden change of the mode of the @xmath23 raman beam , on a timescale of @xmath212s . this is well below the shortest timescale of the system , given by the inverse of the highest tunneling rate , which will typically be on the order of @xmath213 ms . we thus consider the quench as instantaneous . finally , this scheme can be easily modified , adding for example a superlattice potential along the @xmath23 direction , in order to interpolate between the 1d and 2d situations and observe the inversion of statistics . for measuring the wightman function @xmath72 we propose to perform local spectroscopy of the energy bands and determine their occupation as a function of energy . this information is contained in the transfer rate from an atomic state experiencing the dirac hamiltonian in rindler spacetime after the quench , to an auxiliary atomic state with a different dispersion relation . it could thus be measured using one - particle excitation spectroscopy , as demonstrated in @xcite and more recently used to characterize spin - orbit coupled fermi gases @xcite . in order to perform local measurements and determine the dependence of the detector response function with the distance to the horizon , the transfer could be performed using a spectroscopy beam @xcite focused at different @xmath3 positions ( see fig . [ fig.implementationh]c ) . note that the finite waist of this measurement beam , larger than the lattice spacing , would remove from the measurement some of the discretization artifacts discussed previously , and is equivalent to the convolution procedure used in the numerics ( see appendix [ sect : a3 ] ) . experimentally , the most challenging requirement for this local spectroscopy scheme is to realize @xmath128-flux model using an atomic species where one atomic state is subjected to the dirac hamiltonian in minkowski and rindler spacetimes , whereas the second ( auxiliary ) state experiences a different dispersion relation . this situation could be achieved exploiting the ground ( @xmath214 ) and long - lived metastable excited state ( @xmath215 ) of the fermionic isotopes of yb or sr , since the different polarizability of the two states leads to different optical potentials for a broad range of lattice and raman beam wavelengths @xmath216 and @xmath217 . @xmath216 would then be chosen such that the auxiliary band has a negligible bandwidth compared to the initial one . in this scheme the potential gradient leading to the site offset @xmath218 should be realized optically as well . the spectroscopy would be performed using a single laser tuned to the clock transition . this ensures an excellent energy resolution , below the tunneling energy scale , as recently demonstrated experimentally in refs . we have developed a proposal for a quantum simulator of the unruh effect in 1d and 2d massless fermionic fields using ultracold atoms in an optical lattice . the addition of interacting fields and disorder is possible in our approach , which therefore constitutes a full framework for the study of the theoretical implications of quantum field theory in curved spacetime . moreover , our simulator provides a setting for the study of relativistic quantum information theory in an experimentally accessible system . the implementation of this quantum simulator is within experimental reach using state of the art experimental techniques . the detection methods proposed here are potentially relevant also for detecting topological properties in simulators of topological insulators and to assess the properties of quantum systems out of equilibrium . in this work we have restricted ourselves to the study of the rindler metric , i.e. minkowski spacetime viewed by an accelerated observer . nonetheless , the formalism and experimental tools described here may be extended to the study of more complex spacetimes , for instance , non static or even non - stationary ones . our work can also be considered as the first mandatory step prior to the inclusion of matter back - reaction in the artificial metric , and to the simulation of dynamical gravity fields . thus , the present work paves the way to experiments that are not only fascinating _ per se _ , but are also able to access phenomena that are not fully understood theoretically , such as gravitating quantum matter in interaction . finally , the recent conceptual developments towards a combination of quantum mechanics and general relativity , such as quantum graphity @xcite or the maldacena - susskind notion of relating entanglement and spacetime in order to avoid the firewall problem @xcite , might also be amenable to quantum simulation using a similar approach . this work has been supported by spanish mineco ( severo ochoa grant sev-2015 - 0522 , foqus fis2013 - 46768 , fis2014 - 59546-p and fis2012 - 33642 ) , the generalitat de catalunya ( sgr 874 ) , fundaci privada cellex , dfg ( for2414 ) and eu grants equam ( fp7/2007 - 2013 grant no . 323714 ) , osyris ( erc-2013-adg grant no . 339106 ) , siqs ( fp7-ict-2011 - 9 no . 600645 ) , quic ( h2020-fetproact-2014 no . 641122 ) and pcig13-ga-2013 - 631633 . the authors want to acknowledge a. enciso , i. fuentes , j. korbicz , j. len , d. peralta , c. sabn , and g. sierra for very fruitful discussions . let us restrict ourselves to 1d . since the @xmath22 and @xmath21 regions are effectively separated , we may restrict ourselves to the right half - line . consider the spinless 1d version of eq . , @xmath219 . notice that @xmath220 . let us define @xmath221 , taking the horizon to @xmath222 . so , @xmath223 . the eigenvalue equation is let us discuss how to discretize @xmath232 , the 1d rindler hamiltonian , appropriately . consider an open 1d lattice with spacing @xmath131 , and lattice points @xmath233 , with @xmath234 and even @xmath235 . thus , the wavefunctions only take components @xmath236 . let us use a central differences discretization for @xmath237 , i.e. , @xmath238 . let us call @xmath82 the discrete version of the @xmath239 , for later convenience : this means that the tunneling between sites @xmath159 and @xmath244 must be @xmath245 , independently of @xmath131 . this is not surprising , since both @xmath82 and rindler spacetime are _ scale invariant_. a good approximation is obtained by replacing the _ geometric mean _ by the _ arithmetic mean _ : @xmath246 . the discrete hamiltonian can be analytically diagonalized @xcite . its spectrum becomes continuous with constant energy level density as @xmath247 , but the convergence rate is very slow : the distance between levels scales as @xmath248 . let us now focus on the 2d case . in order to formulate the rindler hamiltonian on a square lattice it is convenient to start with the symmetric continuous formulation . explicitly , by writing the spinor in terms of its chiral components , @xmath249 , we have @xmath250 we can now exploit the bipartition of the lattice for discretizing separately the two chiralities in the two checkerboard sublattices and write the kinetic term in terms of the tunneling between the two @xmath251 at this point we notice that the above hamiltonian can be rewritten as the @xmath128-flux hamiltonian once we do not distinguish fermions in the different sublattices . denoting the annihilation ( creation ) operators by @xmath252 ( @xmath253 ) , @xmath254 by applying the gauge transformation @xmath255 we can recover the @xmath128-flux hamiltonian in the symmetric gauge . , and the color ( and point - type ) denoting the distance to the horizon are the same as in fig . [ fig : unruh_1d ] a. ( b ) raw response function in the frequency domain for a 2d system prior to gaussian convolution as in and the frequency rescaling described in sect . [ sect : signature ] . the system size , @xmath165 , and the color code ( and point - type ) denoting the distance to the horizon are the same as in fig . [ fig : unruh_2d ] a.,title="fig:",width=321 ] , and the color ( and point - type ) denoting the distance to the horizon are the same as in fig . [ fig : unruh_1d ] a. ( b ) raw response function in the frequency domain for a 2d system prior to gaussian convolution as in and the frequency rescaling described in sect . [ sect : signature ] . the system size , @xmath165 , and the color code ( and point - type ) denoting the distance to the horizon are the same as in fig . [ fig : unruh_2d ] a.,title="fig:",width=321 ] the dirac hamiltonian is of interest in very different areas , not only of physics , but also of mathematics . indeed , fermionic models are regularly used as mathematical tools in differential geometry and analytic number theory . for instance , by studying the number of non - trivial solutions of a dirac operator in a given manifold , it is possible to determine the topological properties of the manifold itself as proved by the celebrated atiyah - patodi - singer index theorem @xcite . the dirac hamiltonian in rindler spacetime considered in this paper and its non - spinorial 1d equivalent @xmath256 provides a handle for proving the riemann conjecture , which is one of the most famous and relevant open problems in mathematics . riemann conjectured that the non - trivial zeroes of the riemann zeta function @xmath257 in the complex plane all have real part @xmath119 @xcite . one of the established routes towards proving this conjecture the hilbert - polya route is specially interesting for physicists as it attempts the construction of a hermitian operator whose eigenvalues are the imaginary parts of the non - trivial riemann zeroes . in physics , natural occurrences of hermitian operators are , of course , quantum hamiltonians @xcite . in 1999 , berry and keating proposed the @xmath0 hamiltonian and showed how the statistical behavior of its eigenvalues corresponded to the statistical average behavior of the imaginary parts of the non - trivial riemann zeroes @xcite . in fact , the classical hamiltonian @xmath0 must be supplemented with a quantization prescription . the most natural one is i.e. , the 1d version of the dirac hamiltonian in rindler spacetime . the discovery of this berry - keating hamiltonian led to a series of attempts to extend the model in several directions @xcite , including a recent spinorial extension , which is eq . @xcite . flux lattice , the response has a maximum at positive frequency @xmath135 . the response above coincides with the one obtained in rindler spacetime in fig . [ fig : unruh_thermal ] at very large distance from the horizon.,width=321 ] in sect . [ sec : unruh_effect ] , we have derived the expression for the wightman response function in the frequency domain , @xmath72 , for the ideal case of fermionic atoms at zero temperature , . for a realistic gas at finite temperature @xmath259 considered at the end of sect . [ sect : signature ] , the minkowski vacuum , @xmath33 , appearing in , has to be replaced by the thermal mixed state @xmath260 note that for @xmath264 , @xmath265 , and one recovers the zero temperature response function . as defined in the main text , eqs . and , the unitary matrices @xmath78 and @xmath77 are determined from the single - particle modes of dirac hamiltonian in minkowski and rindler spacetime . in sect . [ sec : unruh_effect ] , in order to smear out lattice artifacts , we have considered a convolution of the response function with a gaussian . in fact , as explained in sect . [ sect : implementation ] , such convolution is what is really detected by one - particle excitation spectroscopy . the gaussian convolution consists of the following . by defining

we consider , a gap version of the shortest vector in a lattice problem . this problem is known to be in but is not known to be in or in . we prove that it lies inside , the quantum analogue of . this is the first non - trivial upper bound on the quantum complexity of a lattice problem . the proof relies on two novel ideas . first , we give a new characterization of , called . working with the formulation allows us to circumvent a problem which arises commonly in the context of : the prover might use entanglement between different copies of the same state in order to cheat . the second idea involves using estimations of autocorrelation functions for verification . we make the important observation that autocorrelation functions are positive definite functions and using properties of such functions we severely restrict the prover s possibility to cheat . we hope that these ideas will lead to further developments in the field .

quant-ph0307220

the field of quantum algorithms has witnessed several important results ( e.g. , @xcite ) in the last decade , since the breakthrough discovery of shor s quantum algorithm for factoring and discrete logarithm in 1994 @xcite . despite these important developments , two problems in particular had little progress in terms of quantum algorithms : graph isomorphism ( gi ) , and gap versions of lattice problems such as the shortest vector in the lattice problem ( gapsvp ) and the closest vector in the lattice problem ( gapcvp ) . to understand why these problems are interesting in the context of quantum computation , let us first recall their definitions and what is known about them classically . graph isomorphism is the problem of deciding whether two given graphs can be permuted one to the other . it is known to be in @xmath6 @xcite and therefore , it is not np complete unless the polynomial hierarchy collapses . @xmath7 is the problem of deciding whether the shortest vector in a given @xmath8-dimensional lattice @xmath9 is shorter than @xmath10 or longer than @xmath11 . @xmath12 is the following problem : given a lattice and a vector @xmath13 , decide whether @xmath14 or @xmath15 where @xmath16 is the minimal distance between @xmath13 and any point in @xmath9 . both problems have important cryptographic applications @xcite . regarding their complexity , it is easy to see that they both lie in @xmath2 for any @xmath17 . the results of lagarias et al . @xcite imply that when @xmath18 , both problems are in @xmath19 . for @xmath20 these lattice problems are not known to be in @xmath19 but as shown in @xcite , they are in @xmath21 ( and in fact in the class statistical zero knowledge ) . this implies that for @xmath20 the problems are not np complete unless the polynomial hierarchy collapses . the fact that the graph isomorphism problem and the two lattice problem with the above parameters are very unlikely to be np complete , and that they possess a lot of structure , raised the hope that quantum computers might be able to solve them more efficiently than classical computers . despite many attempts , so far all that is known in terms of the quantum complexity of these problems are reductions to problems for which quantum algorithms are also not known @xcite , and negative results regarding possible approaches @xcite . progress in designing an algorithm for one of these problems is the holy grail of quantum algorithmic theory . in light of the difficulty of finding efficient algorithms for these problems , a weaker question attracted attention : can any quantum upper bound be given on these problems , which does not follow trivially from the classical upper bounds ? regarding graph isomorphism , which is known to be in coam , the natural question to ask is whether it is in coqma , the quantum analog of conp . it is more natural to speak in this context , and in the rest of the paper , about the _ complements _ of the problems we described , and so the question is whether the graph _ non_-isomorphism ( gni ) problem lies inside qma . qma can be viewed as the quantum analog of np , and was recently studied in various papers @xcite . strictly speaking , qma is actually the analog of merlin arthur , the probabilistic version of np , since in the quantum world it is more natural to consider probabilistic classes . attempts to prove that gni is in qma have so far failed . as for lattice problems , since @xmath22 , it follows from the classical result @xcite that if @xmath18 the complements of the problems we described , namely @xmath23 and @xmath24 , lie in @xmath4 . the interesting question , however , is whether these problems are still in @xmath4 for lower gaps , such as @xmath25 . notice that this does not follow from the classical results . in this paper we solve the question of containment in qma for one of the aforementioned problems . this is the first non trivial quantum upper bound for a lattice problem . [ thm : main ] the problem @xmath26 is in @xmath4 for some constant @xmath27 . one of the new ideas in the proof of theorem [ thm : main ] is the important connection between quantum estimations of inner products , or autocorrelation estimates , and properties of positive definite functions . the technique of using positive definite functions to analyze quantum protocols is likely to prove useful in other contexts , due to its generality : the property of positive definiteness applies to autocorrelation functions over any group , and not only over @xmath28 as in our case . another important issue in the proof theorem [ thm : main ] is a problem that arises commonly in the analysis of qma protocols . namely , in certain situations , we would like to repeat a test on several copies of the witness but the prover might use entanglement between the copies in order to cheat . we circumvent this problem by giving a new characterization of @xmath4 , named @xmath5 . we start by proving that indeed @xmath29 and then , using this new characterization , we prove the soundness of our protocol . hopefully , both the new characterization of qma and the new technique of verification using positive definite functions will help in proving that other important problems such as @xmath30 and @xmath31 lie in @xmath4 . in more generality , in this work we gain a better understanding of the class qma and the techniques used to analyze it . we hope that this work will lead to an even better understanding of this important class . understanding classical np led to a few of the most important results in theoretical computer science , including pcp and hardness of approximation . a few indications that qma is fundamental for quantum computation have already been given in @xcite . our results might also lead to progress in terms of quantum algorithms for lattice problems . in this context , it is interesting to consider theorem [ thm : main ] in light of a recent paper by aharonov and ta - shma @xcite . @xcite showed that if the state we use as the quantum witness in the qma protocol can be _ generated _ efficiently , it can be used to provide a @xmath32 algorithm for the lattice problem . the result we present here shows that certain properties of the state of @xcite can be _ verified _ efficiently , which might be a stepping stone towards understanding how to generate the state efficiently , thus providing an efficient algorithm for the lattice problem . finally , we mention that similar techniques to the one used in the proof of @xmath29 , might also prove useful in other contexts , for example for proving security of quantum cryptographic protocols . the paper starts with an overview of the proof . we continue with preliminaries in section [ sec : prelim ] . the proof of theorem [ thm : main ] is obtained by combining three theorems . the proof of each of the theorems is independent and is presented in a separate section . first , in section [ sec : qma+ ] we define the class qma+ and show that it is equal to qma . then , in section [ sec : cvp ] , we show that @xmath33 , a version of @xmath23 , is in qma+ . finally , in section [ sec : svptocvp ] we show that if @xmath33 is in qma then so is @xmath24 . assume we are given a witness which we would like to verify . usually , we apply a certain unitary transformation and measure the output qubit . if the witness is correct , the outcome should be @xmath10 . hence , we reject if the outcome is @xmath34 . consider , however , a situation where our unitary transformation is such that for the correct witness the outcome is @xmath10 with probability @xmath35 , for some @xmath36 . thus , it is natural to consider the following stronger test : we apply a unitary transformation and accept if the _ probability _ of measuring @xmath10 is close to some number @xmath35 . we call a verifier that performs such tests a super - verifier and denote the corresponding class by @xmath5 . our first theorem is [ thm : qma+ ] @xmath29 showing that @xmath4 is contained in @xmath5 is easy ; essentially , the super - verifier can say that the probability of measuring @xmath10 should be close to @xmath37 . the other direction is more interesting . given a super - verifier we can construct a verifier that accepts a witness which is composed of many copies of the original witness . the verifier can then apply the unitary transformation to each one of the copies and measure the results . finally , it can compute the fraction of times @xmath10 was measured and check if it is close to @xmath35 . indeed , if the prover does not cheat and sends many copies of the original witness we should measure @xmath10 in around a @xmath35 fraction of the measurements . however , it seems that the prover might be able to cheat by using entanglement between the different copies . using the markov inequality , we show that this is impossible . next , we show [ thm : gapcvp ] the problem @xmath38 is in @xmath5 for some constant @xmath27 . @xmath39 is a variant of @xmath40 where we are given the additional promise that the shortest vector in @xmath9 is longer than @xmath11 . the proof of this theorem is very involved , but the idea is as follows . the correct quantum witness @xmath41 for @xmath33 , i.e. , the witness in case @xmath13 is far from the lattice , is defined as follows ( a similar state appears in @xcite which can be seen as the quantum analogue of the probability distribution of @xcite ) . consider the ` probability distribution ' obtained by choosing a random lattice point and adding to it a gaussian of radius @xmath42 . we define @xmath41 as the superposition corresponding to this probability distribution . see figure [ fig : witness ] . actually , the state @xmath41 can not be defined as above , since we can not represent a point in @xmath28 with infinite precision , so we need to work over a very fine grid . moreover , the number of grid points in @xmath28 is infinite . hence , we restrict the state to grid points inside the basic parallelepiped of the lattice . we will define this formally later ; it is best to keep in mind the continuous picture . given this superposition , for some constant @xmath43 , solving @xmath38 ( and in fact also @xmath44 ) is easy : it is done by estimating the inner product of the above state with the same state shifted by @xmath13 . if @xmath45 then the inner product is almost zero since the gaussians and their shifted version do not intersect . if @xmath46 , the inner product is large since the two states are almost the same . to show containment in @xmath5 , we will use this state as the correct witness . hence , it remains to show how a super - verifier can verify that the prover is not cheating . cheating in this context means that @xmath46 but the prover claims that @xmath47 and sends some witness which is not necessarily the correct witness . we now define the verification process . define @xmath48 to be the real part of the inner product of the given witness state with itself shifted by @xmath49 . we call @xmath50 the autocorrelation function of the witness . it is a function from @xmath28 to @xmath51 such that @xmath52 . we define @xmath53 to be the same , for the correct witness @xmath41 . an important property of @xmath50 is that for any @xmath49 , there exists a quantum circuit whose probability of outputting @xmath10 is directly related to @xmath48 . hence , since a super - verifier can check the probability of outputting @xmath10 , it can effectively check that @xmath48 is close to some value . since we expect to see the correct witness , we construct a super - verifier that checks that @xmath48 is close to @xmath54 for some vectors @xmath49 . more precisely , with probability half the super - verifier chooses the vector @xmath13 and otherwise it randomly chooses a polynomially short vector . in order to complete the description of the super - verifier , we have to show that it can compute @xmath54 for the points chosen above . later in the paper we analyze the function @xmath53 and it turns out to have a familiar form : it is very close to a periodic gaussian , like the one shown in figure [ fig : witness ] . therefore , @xmath55 is approximately zero since @xmath13 is far from the lattice and @xmath54 for short vectors @xmath49 has the form @xmath56 . in both cases , the super - verifier knows the value of @xmath53 and can therefore perform the verification procedure described above . we remark that analyzing @xmath53 involves some technical calculations ; it is here that we need the assumption that the shortest vector in the lattice is large , so that the gaussians are well separated and do not interfere with each other . the proof of soundness of this test uses the observation that autocorrelation functions are necessarily positive definite . a function @xmath57 is positive definite ( pd ) if for any @xmath58 and any @xmath59 points @xmath60 , the @xmath61 matrix @xmath62 defined by @xmath63 is positive semidefinite . notice that no matter what witness the prover gives , the function @xmath50 must be pd since it is an autocorrelation function . we will complete the proof by showing that no pd @xmath50 exists which passes the above test if @xmath64 , i.e. , no pd function exists which is both close to @xmath34 at a vector @xmath13 whose distance to @xmath9 is at most @xmath65 , and also close to a gaussian at many randomly chosen points polynomially close to the origin . why does nt such a pd function exist ? intuitively , our proof relies on certain non - local behaviors of positive definite functions . namely , we will show that changing the value of a pd function at even one point affects the function at many other points . we assume that @xmath66 is close to @xmath34 and @xmath67 . let @xmath68 be a point which is equal to @xmath13 modulo the lattice ( i.e. , @xmath69 ) such that @xmath70 . such a point exists since @xmath67 . as we will see later , we can guarantee that @xmath50 is periodic on the lattice and hence @xmath71 is close to @xmath34 . we start with a simple property of positive definite functions which can be obtained from using @xmath72 matrices in the definition : if @xmath73 is close to @xmath34 then @xmath74 is at most @xmath75 and similarly , @xmath76 is at most @xmath77 . by repeating the argument we derive an upper bound on @xmath78 where @xmath79 for some @xmath80 . the point @xmath81 is polynomially close to the origin and the upper bound is much smaller than the correct gaussian value , @xmath82 . this shows that the super - verifier can detect a cheating prover by choosing the point @xmath81 . however , the super - verifier does not know where @xmath13 is relative to the lattice and therefore he can not compute @xmath68 or @xmath81 . the probability that our randomly chosen point happens to be @xmath81 is negligible . thus , we will have to derive stronger properties of the function @xmath50 . these will be obtained by considering the positive definite condition with @xmath83 matrices . essentially , we will show that for any point @xmath49 which is almost orthogonal to @xmath81 , it can not be that @xmath48 , @xmath84 and @xmath85 are all close to their correct values @xmath86 . this means that one of the points in the triple @xmath87 is such that the verifier detects a cheating prover by choosing it . using the fact that @xmath81 was chosen to be polynomially short , we will argue that all three points in a triple have roughly the same probability to be chosen by the verifier . hence , a cheating prover is caught with non - negligible probability , and the soundness of the protocol follows . curiously , it seems essential in our proof to use gaussians and not spheres . this is unlike the classical proof of @xcite that seems to work both with spheres and with gaussians . essentially , the difference between the two distributions is in the behavior of their autocorrelation functions . for gaussians , the autocorrelation with a short vector @xmath49 behaves like @xmath88 while for spheres it behaves like @xmath89 where @xmath90 are some constants . in the proof , using properties of positive definite functions obtained from @xmath83 matrices , we obtain an upper bound of the form @xmath91 for some constant @xmath92 . this yields a contradiction since @xmath93 . however , if we used spheres , we would not obtain any contradiction since @xmath94 for short vectors @xmath49 . to complete the proof of theorem [ thm : main ] , we need the final theorem : [ thm : cvptosvp ] for any @xmath95 , if @xmath96 is in @xmath4 then so is @xmath97 . the proof of this theorem uses an idea similar to @xcite . essentially , an instance of @xmath97 can be translated into @xmath8 instances of @xmath96 . if there is no short vector in the original lattice then in all the @xmath98 instances the target vector is far from the lattice . otherwise , if there exists a short vector then in at least one of the @xmath98 instances , the target vector is close to the lattice . based on this idea , we construct a quantum verifier for @xmath97 . the witness it expects to see is a concatenation of the @xmath8 witnesses of the corresponding @xmath96 problems . it applies a @xmath96 verifier to each one of the copies and accepts if and only if they all accept . for @xmath99 , define @xmath100 as @xmath101 . for any @xmath102 , we will often denote @xmath103 by @xmath104 . let @xmath105 denote the @xmath8-dimensional unit ball and let @xmath106 denote its volume . for a vector @xmath102 let @xmath107 denote the @xmath108 dimensional subspace orthogonal to @xmath49 . for a vector @xmath102 and a subspace @xmath109 let @xmath110 denote the projection of @xmath49 on the subspace @xmath109 . we will slightly abuse notation by denoting the projection of @xmath49 on the subspace spanned by a vector @xmath13 as @xmath111 . for an introduction to lattices , see @xcite . a lattice in @xmath28 is defined as the set of all integer combinations of @xmath8 linearly independent vectors . this set of vectors is known as a basis of the lattice and is not unique . given a basis @xmath112 of a lattice @xmath9 , the fundamental parallelepiped is defined as @xmath113 when the basis is clear from the context we will use the notation @xmath114 instead of @xmath115 . note that a lattice has a different fundamental parallelepiped for each possible basis . for a point @xmath102 we define @xmath116 as the minimum of @xmath117 over all @xmath118 . for a lattice @xmath119 and a point @xmath102 we define @xmath120 as the unique point @xmath121 such that @xmath122 is an integer combination of @xmath123 ( see , e.g. , @xcite ) . notice that a function @xmath124 can be naturally extended to a function @xmath125 by defining @xmath126 . we will often refer to values of functions outside of @xmath114 , in which case we mean the periodicity above . we will also use , for technical proofs , the notion of a voronoi cell of l , denoted @xmath127 , which is the set of all points in @xmath28 which are closer to the origin than to any other lattice point . in addition , @xmath128 denotes the unique point @xmath129 such that @xmath130 . notice that @xmath131 . the shortest ( non - zero ) vector of @xmath9 is the vector @xmath132 , such that @xmath133 and is minimal . the following is the gap version of the shortest vector problem : for any gap parameter @xmath134 the promise problem @xmath97 is defined as follows . the input is a basis for a lattice @xmath9 . it is a @xmath135 instance if the length of the shortest vector is more than @xmath136 . it is a @xmath137 instance if the length of the shortest vector is at most @xmath10 . we also define the gap version of the closest vector problem and a non - standard variant of it which will be used in this paper : for any gap parameter @xmath134 the promise problem @xmath138 is defined as follows . the input is a basis for a lattice @xmath9 and a vector @xmath13 . it is a @xmath135 instance if @xmath139 . it is a @xmath137 instance if @xmath140 . for any gap parameter @xmath134 the promise problem @xmath96 is defined as follows . the input is a basis for a lattice @xmath9 and a vector @xmath13 . it is a @xmath135 instance if @xmath139 and the shortest vector in @xmath9 is of length at least @xmath136 . it is a @xmath137 instance if @xmath14 . each vector in the input basis @xmath123 is given with polynomially many bits . without loss of generality , we assume that the target vector @xmath13 is given to us in the form @xmath141 where each @xmath142 is represented by at most @xmath143 bits . we are interested in the quantum analog of the class np . for an introduction to this class , the reader is referred to a recent survey by aharonov and naveh @xcite and to a book by kitaev , shen and vyalyi @xcite . strictly speaking , this class is the quantum analogue of ma , the probabilistic version of np , and so it is denoted qma . it is also sometimes denoted bqnp @xcite . [ def : qma ] a language @xmath144 if there exists a quantum polynomial time verifier @xmath145 , polynomials @xmath146 , and efficiently computable functions @xmath147 , such that : * @xmath148 * @xmath149 , * @xmath150 , and the @xmath151 s are density matrices of @xmath152 qubits . a @xmath61 matrix @xmath62 is positive semidefinite ( psd ) if it is hermitian and for any vector @xmath153 , @xmath154 is real and non - negative . the requirement that @xmath62 is hermitian is redundant since this is already implied by the requirement that @xmath154 is real for all @xmath153 . the next two claims list some simple properties of positive semidefinite matrices . [ claim : psd_simple ] let @xmath155 denote two positive semidefinite matrices . then the following matrices are also positive semidefinite : @xmath156 , @xmath157 , @xmath158 and @xmath159 where @xmath27 is real and @xmath159 is the matrix obtained by taking the real part of every entry of @xmath62 . clearly , all four matrices are hermitian . let @xmath68 be any vector in @xmath160 . then , @xmath161 and @xmath162 . also , @xmath163 . finally , @xmath164 which is positive semidefinite according to the previous cases . [ claim : psd_det ] the determinant of a positive semidefinite matrix @xmath62 is non - negative . since @xmath62 is hermitian , it can be diagonalized with orthogonal eigenvectors and real eigenvalues . moreover , since it is positive semidefinite , its eigenvalues are non - negative . hence , the determinant of @xmath62 , which is the product of its eigenvalues , is non - negative . next , we define a positive definite function over an arbitrary group @xmath165 . in this paper , @xmath165 will always be a grid in @xmath28 , i.e. , a discrete additive subgroup of @xmath28 . [ def : positive_definite ] let @xmath165 be a group . a function @xmath166 is positive definite ( pd ) if for any integer @xmath167 and any set of group elements @xmath168 , the @xmath59 by @xmath59 matrix @xmath62 defined by @xmath169 is positive semidefinite . the following two corollaries follow directly from definition [ def : positive_definite ] and claims [ claim : psd_simple ] , [ claim : psd_det ] : [ cor : pd_simple ] let @xmath170 be two positive definite functions . then the following functions are also positive definite : @xmath171 , @xmath172 , @xmath173 where @xmath27 is real . [ cor : pd_determinant ] let @xmath166 be a positive definite function for some group @xmath165 . then , for any integer @xmath167 and any set of group elements @xmath168 , the @xmath59 by @xmath59 matrix @xmath62 defined by @xmath169 has a non - negative determinant . using corollary [ cor : pd_determinant ] we derive the following two useful lemmas . these lemmas describe known properties of positive definite functions ( see , e.g. , @xcite ) . [ pdf2x2 ] let @xmath174 be a real positive definite function such that @xmath175 . then for any @xmath176 , @xmath177 and @xmath178 . choose @xmath179 in definition [ def : positive_definite ] and choose @xmath34 and @xmath49 as the two group elements . then , @xmath180 is positive semidefinite . hence , @xmath62 is hermitian and @xmath181 . moreover , @xmath182 therefore , @xmath183 [ pdf3x3 ] let @xmath174 be a real positive definite function such that @xmath175 . then , for any @xmath176 such that @xmath184 exists , @xmath185 . choose @xmath186 in definition [ def : positive_definite ] and choose @xmath34 , @xmath49 and @xmath187 as the three group elements . let @xmath188 denote @xmath54 and @xmath189 denote @xmath190 . then , @xmath191 according to lemma [ pdf2x2 ] , @xmath192 . hence we have @xmath193 which implies @xmath194 the following claim shows the important fact that autocorrelation functions are always positive definite . [ cl : pd ] let @xmath57 be a function from a group @xmath165 to the complex numbers , and let @xmath50 be its autocorrelation function defined by @xmath195 then @xmath50 is a positive definite function . let @xmath58 and @xmath168 be arbitrary and consider the @xmath61 matrix @xmath62 defined by @xmath196 . according to definition [ def : positive_definite ] , it is enough to show that @xmath62 is psd . for any vector @xmath153 , a `` super - verifier '' is given by a classical polynomial - time randomized algorithm that given an input @xmath49 outputs a description of a quantum circuit @xmath145 and two numbers @xmath198 $ ] . this can be thought of as follows . assume that we are given a witness described by a density matrix @xmath151 . then , consider @xmath199 where @xmath200 is the projection on the space where the output qubit of @xmath145 is @xmath201 ( this is equal to the probability of measuring an output qubit of @xmath201 ) . then , @xmath202 represents an estimate of this value and @xmath203 is the accuracy of the estimate . [ def : qma+ ] a language @xmath204 if there exists a super - verifier and polynomials @xmath205 such that : * @xmath206 + ( i.e. , there exists a witness such that with probability @xmath10 the super - verifier outputs @xmath145 which accepts the witness with probability which is close to @xmath202 ) * @xmath207 + ( i.e. , for any witness , with some non - negligible probability , the super - verifier outputs a circuit @xmath145 that accepts the witness with probability which is not close to @xmath202 ) where probabilities are taken over the outputs @xmath208 of the super - verifier and @xmath151 is a density matrix over @xmath209 qubits . in the rest of this section we prove theorem [ thm : qma+ ] . we note that for simplicity we defined @xmath5 with perfect completeness in the yes case ; the same theorem holds also with non - perfect completeness . the following lemma proves the easy direction of the theorem . it will not be used in this paper and is presented here mainly for the sake of completeness . @xmath210 note that using amplification @xcite , any language in @xmath4 has a verifier with completeness @xmath211 and soundness @xmath212 . given such a verifier @xmath145 , construct a super - verifier that simply outputs @xmath213 . this satisfies the definition of @xmath5 , using @xmath214 , for example . we now prove the more interesting direction : [ thm : inqma ] @xmath215 given a super - verifier for a language @xmath216 with polynomials @xmath205 , we construct a qma verifier @xmath217 for @xmath9 . let @xmath218 be a large enough parameter to be determined later . the witness given to @xmath217 consists of @xmath219 qubits which can be thought of as @xmath59 registers of @xmath209 qubits each . given an input @xmath49 , the verifier @xmath217 starts by calling the super - verifier with the input @xmath49 . the result is a description of a circuit @xmath145 and numbers @xmath220 $ ] . next , @xmath217 applies @xmath145 to each of the @xmath59 registers and measures the results . let @xmath221 denote the number of 1s measured divided by @xmath59 . @xmath217 accepts if @xmath222 and rejects otherwise . let @xmath132 and let @xmath151 be as in definition [ def : qma+ ] . the witness for @xmath217 will be @xmath223 . note that the probability to measure @xmath10 in each register is @xmath224 . let us denote this probability by @xmath225 and let us choose @xmath226 . then , according to the chernoff bound , the probability that @xmath227 is at most @xmath228 . by definition [ def : qma+ ] , the triples @xmath229 given by the super - verifier are such that @xmath230 and @xmath231 which implies that @xmath217 accepts with probability exponentially close to @xmath10 . it suffices to show that if @xmath232 then @xmath217 rejects with probability at least @xmath233 ( which is polynomially bounded from @xmath34 ) . essentially , the reasoning is based on a markov argument , as we will see shortly . let @xmath234 be any witness for @xmath217 . we first define a witness @xmath151 for the circuits @xmath145 that the super - verifier outputs . let @xmath235 be the reduced density matrix of @xmath236 to the @xmath237th register , and let @xmath151 to be the average of the reduced density matrices : @xmath238 . for an output of the super - verifier @xmath229 we again let @xmath225 denote the probability to measure @xmath10 given @xmath151 , namely @xmath239 . we observe that [ claim : exp ] for a fixed witness @xmath234 and a fixed circuit @xmath145 , the expectation of the random variable @xmath221 is @xmath225 . the random variable @xmath221 is the average of @xmath59 indicator variables . the expected value of the @xmath237th indicator variable is @xmath240 . therefore , using linearity of expectation , the expected value of @xmath221 is @xmath241 . according to definition [ def : qma+ ] , with probability at least @xmath242 , @xmath229 is such that @xmath243 . then , it is enough to show that for such triples @xmath229 , @xmath217 rejects with probability at least @xmath244 . so , in the following fix one such triple @xmath229 . using claim [ claim : exp ] , we obtain that the expected value of @xmath221 is either less than @xmath245 or more than @xmath246 . we now use a markov argument ; in the first case , since @xmath221 is a non - negative random variable , the probability that it is more than @xmath247 ( so that @xmath217 may accept ) is at most @xmath248 similarly , for the second case , consider the non - negative random variable @xmath249 . the probability that it is greater than @xmath250 is at most @xmath251 in this section we prove theorem [ thm : gapcvp ] . recall that an input to @xmath252 is a pair @xmath253 . by choosing a large enough constant @xmath43 and scaling we can assume that in @xmath135 instances , @xmath254 and the shortest vector in @xmath9 is of length at least @xmath255 and that in @xmath137 instances @xmath64 . in the case of a @xmath135 instance , the prover provides a quantum state that represents a gaussian distribution around the lattice points . we will use the periodicity of the lattice and present our state as a superposition over points inside the parallelepiped @xmath114 . we would have liked to consider the superposition over all points in the parallelepiped @xmath114 with weights that depend on the distance to the lattice : @xmath256 however , this state is ill defined since the register contains points in @xmath28 , which we need infinite precision in order to represent . we will therefore discretize space , and consider points on a very fine lattice @xmath257 . in order to prevent confusion , we will refer to @xmath257 as a ` grid ' and not a lattice . we discuss this in the following . [ sec : disc ] * discretization issues : * the grid @xmath257 is obtained by scaling down the lattice @xmath119 by a factor of @xmath258 for some @xmath259 . formally , @xmath257 is the set of all integer combinations of the vectors @xmath260 where @xmath261 is chosen such that the following requirements are satisfied : @xmath262 the diameter of one parallelepiped of @xmath257 , @xmath263 , is at most @xmath264 , and @xmath262 @xmath265 where @xmath266 was defined as the precision in which @xmath13 is given . note that we can choose @xmath267 to be polynomial in @xmath8 because @xmath268 . to store a vector in @xmath269 in the quantum register , we store its coefficients in terms of the basis vectors @xmath270 . each coefficient is a number of the form @xmath271 for @xmath272 and so we need @xmath267 bits to store @xmath273 . since we need @xmath8 coefficients , the register consists of @xmath274 qubits . the formal definition of the witness is : @xmath275 where @xmath276 and @xmath277 is a normalization factor chosen so that @xmath278 our verification process is based on autocorrelation tests which we define in the following . for @xmath279 , @xmath280 is defined to be the bijection @xmath281 from @xmath269 into itself . the function @xmath282 is defined as @xmath283 . note that @xmath54 is equal to @xmath284 [ def:1bit ] for any @xmath279 define the circuit @xmath285 as follows . given an input register , add one qubit ( called the control qubit ) in the state @xmath286 . then apply @xmath280 to the register conditioned that the control qubit is @xmath10 , and otherwise do nothing . finally , apply the hadamard matrix @xmath287 on the control qubit . the control qubit is the output qubit . [ claim : purestate_correlation ] given a pure state @xmath288 , the probability of measuring 1 after applying @xmath285 is @xmath289 . after adding the control qubit to @xmath288 , the state is @xmath290 . after performing a conditioned @xmath280 , the state is @xmath291 . finally , after the hadamard transform , the state is @xmath292 the probability of measuring @xmath10 is therefore @xmath293 the next lemma provides a good approximation to @xmath54 : [ le : integrals ] let @xmath9 be a lattice whose shortest vector is of length at least @xmath255 . then , for any @xmath279 , @xmath294 the proof is fairly complicated technically , and we delay it to the appendix . the super - verifier randomly chooses one of the following two cases : * * autocorrelation with respect to @xmath13 * + output the circuit @xmath295 , together with @xmath296 and @xmath297 . * * autocorrelation with respect to short vectors * + let @xmath298 denote the ball of radius @xmath299 around the origin . choose a vector @xmath300 from the uniform distribution over @xmath301 . let @xmath302 be either @xmath49 or @xmath303 with equal probability . output the circuit @xmath304 , together with @xmath305 and @xmath297 . the verifier works on points in @xmath269 . note that the map @xmath306 for @xmath307 is well defined and is a bijection on @xmath308 , and so is its inverse . this means that these maps can be applied efficiently by a quantum computer . this follows from a basic result in quantum computation , which states that if @xmath309 and its inverse can be applied efficiently classically , then they can be applied efficiently and without garbage bits by a quantum computer @xcite . next , we describe a procedure that picks a point uniformly at random from @xmath301 . first , pick a point @xmath310 uniformly from the ball @xmath311 . represent it as a combination of the basis vectors @xmath312 . then , let @xmath279 be the point obtained by rounding the coefficients of @xmath313 down to multiples of @xmath314 . if @xmath315 then output @xmath49 . otherwise , repeat the procedure again . the probability of outputting each @xmath316 is proportional to the probability that @xmath313 is in @xmath317 . since @xmath318 , @xmath319 and therefore the above probability is proportional to the volume of @xmath320 . this volume is the same for all @xmath49 and hence the output is indeed uniform over @xmath301 . the procedure has to be repeated when @xmath321 . this can only happen if @xmath322 . but the probability of this is at most @xmath323 and therefore the procedure stops after a polynomial number of steps with probability exponentially close to 1 . finally , we note that we can not really choose a uniform point @xmath313 in the ball since its representation is not finite ; this can be easily fixed by choosing an approximation to @xmath313 and then arguing that the distance of the output distribution from the uniform distribution on @xmath301 is exponentially small . let @xmath9 be a lattice whose shortest vector is of length at least @xmath255 and @xmath13 a vector such that @xmath324 . then , given the witness @xmath325 described in section [ sec : quantum_witness ] , the super - verifier outputs triples @xmath229 such that @xmath326 . first assume that the super - verifier outputs @xmath295 . by lemma [ le : integrals ] , @xmath55 is exponentially small and therefore , using claim [ claim : purestate_correlation ] , @xmath327 is in the range @xmath328 $ ] . otherwise , the super - verifier outputs a circuit @xmath304 for some short vector @xmath302 . notice that @xmath329 , since the lattice has no short vectors . by lemma [ le : integrals ] , @xmath330 is exponentially close to @xmath331 and hence @xmath332 is exponentially close to @xmath333 . let @xmath9 be a lattice and @xmath13 be a vector such that @xmath64 . then , given any witness @xmath151 , with probability at least @xmath334 , the super - verifier outputs triples @xmath229 such that @xmath335 . we will need the following definitions : we say @xmath49 is `` good '' for a real function @xmath50 if @xmath336 and @xmath337 . otherwise , we say @xmath49 is `` bad '' for @xmath50 . we say that @xmath50 is @xmath338-gaussian approximating on the set @xmath339 if all except at most @xmath338 fraction of the vectors in @xmath339 are good for @xmath50 . the idea of the proof is as follows . let @xmath151 be any witness and assume by contradiction that @xmath64 and that with probability at least @xmath340 the super - verifier outputs @xmath229 such that @xmath341 . we use @xmath151 to define a pd function @xmath50 , and show that by the conditions of the theorem @xmath342 and that @xmath50 is @xmath343-gaussian approximating on @xmath344 . we then show that such a pd function does nt exist , if @xmath64 , which derives a contradiction . [ [ definition - of - h ] ] definition of @xmath50 : + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + we can write @xmath151 as @xmath345 for some weights @xmath346 and pure states @xmath347 . also , write @xmath348 for some @xmath349 . this form of @xmath347 is without loss of generality , because by our choice of the number of qubits in the register , and by the definition of @xmath257 , each possible basis state represents a point in @xmath308 . define the functions @xmath350 , @xmath351 we let @xmath352 be the function @xmath353 @xmath50 is pd . according to corollary [ cor : pd_simple ] , it is enough to show that the @xmath354 s are positive definite . this follows from claim [ cl : pd ] , using the group of points in @xmath308 with addition modulo @xmath114 . @xmath342 . the super - verifier outputs the triple @xmath355 with probability half . by the assumption of the theorem we thus know that @xmath356 note that by claim [ claim : purestate_correlation ] , for any @xmath357 , @xmath358 substituting equation ( [ eq : tr ] ) in ( [ eq:1 ] ) and multiplying by 2 we get , @xmath359 @xmath50 is @xmath343-gaussian approximating on the set @xmath344 . the super - verifier outputs a triple of the form @xmath360 with probability half . hence , with probability at least @xmath361 @xmath362 where the probability is taken on the choice of @xmath302 by the super - verifier . substituting equation ( [ eq : tr ] ) in equation ( [ eq:2 ] ) and multiplying by 2 , we get : @xmath363 recall that @xmath302 is chosen in two steps : we first choose @xmath300 and then choose @xmath302 to be either @xmath49 or @xmath303 . hence , with probability at least @xmath364 over the choice of @xmath49 , both @xmath365 and @xmath366 hold . hence , @xmath50 is @xmath343-gaussian approximating on the set @xmath301 . we obtain a contradiction by using the following lemma with @xmath367 . recall that the coefficients of @xmath13 in the lattice basis are multiples of @xmath368 . this implies that @xmath369 can be represented as an integer combination of the vectors @xmath370 . since @xmath267 was chosen to be at least @xmath371 , @xmath372 . the proof of the lemma appears in the next section . [ le : noconditions ] let @xmath373 such that @xmath374 is also in @xmath257 and @xmath70 . then , there is no positive definite function @xmath50 , @xmath52 , which is @xmath343-gaussian approximating on @xmath344 and @xmath375 . ( of lemma [ le : noconditions ] ) assume by contradiction that @xmath50 is a positive definite function , that @xmath376 and that @xmath50 is @xmath343-gaussian approximating on @xmath301 . we will derive a contradiction in two steps . first , we will find a short vector @xmath81 in @xmath68 s direction such that @xmath78 is much lower than the gaussian value of @xmath377 . this is done using the upper bound on @xmath378 and `` pulling '' it towards the origin using the pd conditions . we will then apply a lemma that shows that the same deviation from the gaussian occurs everywhere and not only in @xmath68 s direction . define @xmath79 , where @xmath379 is the minimal integer such that @xmath380 . notice that if @xmath381 then @xmath382 . hence , using @xmath70 , we get that @xmath383 . @xmath384 . since @xmath385 , @xmath81 is an integer multiple of @xmath374 and is therefore in @xmath257 . the gaussian at @xmath81 , @xmath377 , can be approximated by @xmath386 which is at least @xmath387 . the following claim shows that @xmath78 is strictly less than the gaussian at @xmath81 : let @xmath50 be pd such that @xmath52 , @xmath388 and @xmath70 . then @xmath389 . lemma [ pdf3x3 ] , using @xmath68 , @xmath390 , shows that @xmath391 . applying lemma [ pdf3x3 ] again gives @xmath392 , and applying it @xmath59 times gives @xmath393 . since @xmath383 , we have @xmath394 . to derive a contradiction , we will use the following claim : [ cl : psdapprox ] let @xmath50 be pd such that @xmath52 and @xmath389 for some @xmath395 . let @xmath396 be such that @xmath397 and @xmath398 . then at least one of the vectors @xmath399 is bad for @xmath50 . the proof uses the pd condition with @xmath83 matrices . it is quite technical , and is delayed to the appendix . we want to show that in the verifier s second test , it has a non negligible chance of picking @xmath49 which is equal to one of the vectors of the form @xmath400 satisfying the requirements in claim [ cl : psdapprox ] . this would mean it has a good chance of catching a `` bad '' vector , as we will see later . for this we define : @xmath401 @xmath402 by the triangle inequality , the norm of a vector in @xmath403 or @xmath404 is at most @xmath405 , because @xmath395 . hence , its norm is less than @xmath299 , the radius of @xmath298 . let @xmath49 be chosen uniformly at random from @xmath344 . the probability for @xmath49 to be in @xmath406 is at least @xmath407 , for all @xmath408 . first notice that @xmath409 and they are all subsets of @xmath298 . hence , the probability that @xmath49 is @xmath406 is the same for @xmath408 . therefore , in the following it will be enough to consider the set @xmath410 . let @xmath411 be a subset of @xmath410 . notice that since @xmath263 was chosen to be very small , any point @xmath412 such that @xmath413 must satisfy @xmath414 . similarly , if we define @xmath415 as the ball of radius @xmath416 then any point @xmath396 such that @xmath417 must satisfy @xmath418 . hence we obtain , @xmath419 we now lower bound this ratio of volumes . recall that the volume of an @xmath8 dimensional ball around the origin of radius @xmath420 is @xmath421 where @xmath106 is the volume of the unit @xmath8-ball . @xmath422 using @xmath423 , @xmath424 where in the last inequality we used @xmath425 . @xmath50 is not @xmath343-gaussian approximating on @xmath344 . for any @xmath426 , consider the triple @xmath87 and notice that @xmath427 , @xmath428 . by claim [ cl : psdapprox ] , at least one point in each triple is bad for @xmath50 . hence , at least a third of the points in one of the sets @xmath429 are bad for @xmath50 . since each of these sets contains @xmath407 of the points in @xmath301 , the fraction of bad points for @xmath50 in @xmath344 is at least @xmath430 . this is a contradiction and thus completes the proof of lemma [ le : noconditions ] . in this section we prove theorem [ thm : cvptosvp ] . we show how to construct a verifier @xmath217 for @xmath431 given a verifier @xmath145 for @xmath96 . by using amplification @xcite , we can assume without loss of generality that for @xmath135 instances there exists a witness such that @xmath145 accepts with probability at least @xmath432 and that for @xmath137 instances @xmath145 accepts with probability less than @xmath433 for any witness . let @xmath9 be the input lattice given by @xmath112 . the witness supplied to @xmath217 is supposed to be of the following form : @xmath434 each @xmath347 is supposed to be a witness for the @xmath96 instance given by the lattice @xmath435 spanned by @xmath436 and the target vector @xmath270 . the verifier @xmath217 applies @xmath145 to each @xmath347 with the instance @xmath437 . it accepts if and only if @xmath145 accepted in all the calls . first assume that @xmath9 is a @xmath135 instance to @xmath97 . in other words , the length of the shortest vector is at least @xmath136 . since @xmath435 is a sublattice of @xmath9 , its shortest vector is at least @xmath136 . in addition , since for any @xmath438 $ ] , @xmath439 this implies that @xmath440 . hence , @xmath437 is a @xmath135 instance of @xmath96 and there exists a witness @xmath347 such that @xmath145 accepts it with probability at least @xmath432 . therefore , the combined witness @xmath441 is accepted by @xmath217 with probability at least @xmath442 . it is left to consider the case where @xmath9 is a @xmath137 instance . in other words , if @xmath443 denotes the shortest vector , then its length is at most @xmath10 . notice that not all the @xmath444 s are even for otherwise the vector @xmath445 is a shorter lattice vector . let @xmath273 be such that @xmath446 is odd . then the distance of @xmath447 from the lattice @xmath448 is at most @xmath449 since @xmath450 . hence , the @xmath273th instance of @xmath96 is a @xmath137 instance and for any witness @xmath451 , @xmath145 accepts with probability at most @xmath433 and so does @xmath217 . we would like to thank hartmut klauck , alex samordnitsky , benny sudakov , umesh vazirani and john watrous for helpful discussions . or thanks martin schlather for sending a copy of his technical report . 10 d. aharonov and t. naveh . quantum np - a survey . in _ quant - ph/0210077 , http://xxx.lanl.gov_ , 2002 . d. aharonov and a. ta - shma . quantum adiabatic state generation and stataistical zero knowledge . in _ proc . 35th acm symp . on theory of computing _ , san diego , ca , june 2003 . d. aharonov , w. van dam , j. kempe , z. landau , s. lloyd , and o. regev . on the universality of quantum adiabatic computation on a 2d lattice ( temporary name ) . 2003 . manuscript . a. m. childs , r. cleve , e. deotto , e. farhi , s. gutmann , and d. a. spielman . exponential algorithmic speedup by quantum walk . in _ proc . 35th acm symp . on theory of computing _ , san diego , ca , june 2003 . m. ettinger and p. hyer . on quantum algorithms for noncommutative hidden subgroups . , 25(3):239251 , 2000 . k. friedl , g. ivanyos , f. magniez , m. santha , and p. sen . hidden translation and orbit coset in quantum computing . in _ proc . 35th acm symp . on theory of computing _ , 2003 . o. goldreich and s. goldwasser . on the limits of nonapproximability of lattice problems . , 60(3):540563 , 2000 . o. goldreich , s. micali , and a. wigderson . proofs that yield nothing but their validity , or all languages in np have zero - knowledge proof systems . , 38(3):691729 , 1991 . o. goldreich , d. micciancio , s. safra , and j .- approximating shortest lattice vectors is not harder than approximating closest lattice vectors . , 71(2):5561 , 1999 . m. grigni , l. schulman , m. vazirani , and u. vazirani . quantum mechanical algorithms for the non - abelean hidden subgroup problem . in _ proc . 33th acm symp . on theory of computing _ , pages 6874 , 2001 . s. hallgren . polynomial - time quantum algorithms for pell s equation and the principal ideal problem . in _ proc . 34th acm symp . on theory of computing _ , pages 653658 , 2002 . j. kempe and o. regev . 3-local hamiltonian is qma - complete . in _ quant - ph/0302079 , http://xxx.lanl.gov_ , 2003 . kitaev , a. h. shen , and m. n. vyalyi . , volume 47 of _ graduate studies in mathematics_. ams , 2002 . j. c. lagarias , h. w. lenstra , jr . , and c .- p . korkin - zolotarev bases and successive minima of a lattice and its reciprocal lattice . , 10(4):333348 , 1990 . d. micciancio . improving lattice based cryptosystems using the hermite normal form . in _ cryptography and lattices conference ( calc ) _ , volume 2146 of _ lecture notes in computer science _ , pages 126145 , providence , rhode island , march 2001 . springer - verlag . d. micciancio and s. goldwasser . , volume 671 of _ the kluwer international series in engineering and computer science_. kluwer academic publishers , boston , massachusetts , march 2002 . o. regev . quantum computation and lattice problems . in _ proceedings of the 43rd annual symposium on foundations of computer science ( focs ) 2002 _ , vancouver , canada , november 2002 . o. regev . new lattice based cryptographic constructions . in _ proc . 35th acm symp . on theory of computing _ , san diego , ca , june 2003 . z. sasvri . , volume 2 of _ mathematical topics_. akademie verlag , berlin , 1994 . m. schlather . introduction to positive definite functions and to unconditional simulation of random fields . technical report st 99 - 10 , lancaster university , 1999 . p. w. shor . polynomial - time algorithms for prime factorization and discrete logarithms on a quantum computer . , 26(5):14841509 , 1997 . a. shpilka and r. raz . on the power of quantum proofs . 2002 . unpublished . w. van dam , s. hallgren , and l. ip . quantum algorithms for some hidden shift problems . in _ proceedings of the acm - siam symposium on discrete algorithms _ , pages 489498 , 2003 . j. watrous . succinct quantum proofs for properties of finite groups . in _ proceedings of the 41st annual symposium on foundations of computer science _ , pages 537546 , 2000 . [ cl : mu_error ] for any two vectors @xmath452 , @xmath453 the derivative of @xmath100 is @xmath454 which is at most @xmath455 in absolute value . hence , for any @xmath456 , @xmath457 the claim follows since for any @xmath458 , @xmath459 and @xmath460 . [ cl : int_mu ] @xmath461 @xmath462 [ cl : tail_mu ] @xmath463 according to claim [ cl : int_mu ] , it is enough to show that @xmath464 since @xmath465 depends only on the norm of @xmath313 we can switch to polar coordinates and get @xmath466 using stirling s formula , @xmath467 hence , ( [ eq : tail_mu_1 ] ) is @xmath468 in this section we prove lemma [ le : integrals ] . recall that @xmath53 is defined as @xmath469 the function @xmath57 is periodic on the lattice @xmath9 . hence , @xmath470 furthermore , @xmath471 can be seen as a bijection between @xmath269 and @xmath472 . hence , the above is equal to , @xmath473 when @xmath474 , @xmath475 . also , if @xmath476 then @xmath129 because the shortest vector in the lattice is at least @xmath255 . therefore , the above sum is , @xmath477 notice that for @xmath476 , @xmath478 . also , if @xmath479 then @xmath480 and therefore @xmath481 . using the assumption that the shortest vector in the lattice is @xmath255 , this implies that the closest lattice point to @xmath482 is the same as the closest lattice point to @xmath49 . in other words , @xmath483 . let @xmath484 denote the set of all @xmath485 such that both @xmath486 and @xmath487 are at most @xmath488 . then the above sum is , @xmath489 for any @xmath490 , @xmath491 . using claim [ cl : mu_error ] , we see that for any @xmath492 , @xmath493 . hence , this translates to a multiplicative error of @xmath494 . a similar argument shows that @xmath495 . by combining the two equalities and taking the square root , we get that for any @xmath496 and for any @xmath497 , @xmath498 averaging the right hand side over all @xmath497 , @xmath499 we therefore obtain the following estimation of @xmath54 : @xmath500 recall that @xmath277 was chosen so that @xmath175 . hence , we get that @xmath501 since @xmath502 contains the ball of radius @xmath42 around the origin , @xmath503 where we used claim [ cl : tail_mu ] and claim [ cl : int_mu ] . hence , @xmath504 thus , the estimation of @xmath54 becomes @xmath505 this can be further approximated by @xmath506 where in the first equality we used @xmath507 . we can now upper bound @xmath54 by @xmath508 in particular , this means that for @xmath49 such that @xmath116 is greater than , say , @xmath509 , @xmath54 is indeed exponentially close to @xmath510 . therefore , it remains to consider the case @xmath511 . here , @xmath512 and therefore @xmath54 can be lower bounded by @xmath513 where we used claim [ cl : tail_mu ] . we assume by contradiction that the vectors @xmath399 are good for @xmath50 , that @xmath389 and that @xmath50 is a positive definite function . we will derive a contradiction by using the pd condition with a @xmath83 matrix . choose @xmath514 in definition [ def : positive_definite ] and choose the origin , the vector @xmath515 , the vector @xmath313 and the vector @xmath81 as the four vectors . by the assumption that @xmath50 is positive definite , and by corollary [ cor : pd_determinant ] , the following holds : let @xmath520 be the projection of @xmath313 on the subspace orthogonal to @xmath81 . according to claim [ cl : mu_error ] , by replacing @xmath313 with @xmath521 in the above estimations we introduce an error of at most @xmath522 : @xmath523 we can replace each entry of the above determinant by its estimation . by lemma [ pdf2x2 ] , all the entries of the determinant have an absolute value of at most one and therefore the error introduced is at most @xmath518 : let us now expand the determinant : @xmath530 hence , @xmath531 from the assumption that @xmath532 it follows that @xmath533 is in the order of @xmath534 . hence , dividing by @xmath533 which is a positive number , we get @xmath535 rearranging terms , @xmath536 from this we can obtain the following lower bound on @xmath78 , @xmath537 we show that the term under the square root is negligible , because it is @xmath538 : @xmath539 the term @xmath540 is of the order @xmath534 , the term @xmath541 is at most @xmath542 and the term @xmath543 equals @xmath544 which is of the order @xmath545 . hence , the above expression is of the order @xmath538 . after taking the square root it is of the order @xmath546 . we therefore get : @xmath547 we have @xmath548 where we used the taylor series expansion of the left hand side around @xmath10 .

in this paper we propose a time - independent _ equality _ and time - dependent _ inequality _ , suitable for an experimental test of the hypothesis of realism . the derivation of these relations is based on the concept of conditional probability and on bayes theorem in the framework of kolmogorov s axiomatics of probability theory . the equality obtained is intrinsically different from the well known ghz - equality and its variants , because violation of the new equality might be tested in experiments with only two microsystems in a maximally entangled bell state , while a test of the ghz - equality requires at least three quantum systems in a special state . the obtained inequality differs from bell s , wigner s , and leggett - garg inequalities , because it deals with spin projections onto only two non - parallel directions at two different moments of time , while a test of the bell and wigner inequalities requires at least three non - parallel directions , and a test of the leggett - garg inequalities at least three distinct moments of time . hence , the proposed inequality seems to allow one to avoid the `` contextuality loophole '' . violation of the proposed equality and inequality is illustrated with the behaviour of a pair of anti - correlated spins in an external magnetic field and also with the oscillations of flavour - entangled pairs of neutral pseudoscalar mesons .

1610.09905

one of the main questions that arise when comparing predictions of quantum theory with experiment is to what extent the real physical properties of micro - objects correspond to the observed values , measured with macro - devices . heisenberg noted in his book @xcite that this essential question of quantum theory is close to the general analysis of our perception of `` phenomenon '' of our world and the gist of the phenomenon , the `` noumenon '' , according to kant @xcite . from the point of view of the simplest `` orthodox '' version of the quantum mechanical formalism @xcite , a process of measurement corresponds to expansion of a microsystem state vector @xmath3 into a superposition of a macroscopically definitive states @xmath4 : @xmath5 where , according to the law of born , @xmath6 defines the probability to find the system in state @xmath4 after measurement . usually one takes @xmath4 as a set of eigenvectors of hermitian operator @xmath7 which corresponds to some physical characteristics ( observables ) @xmath8 of the microsystem studied . of course , the expansion ( [ superposition ] ) and law of born may be generalized in terms of povms ( positive operator - valued measures a description of a measurement using positively - defined operators ) and the projection postulate of dirac von neumann @xcite . however it is not important for the consequent arguments which approach is used . we will use the simplest one , i.e. the superposition principle ( [ superposition ] ) and the law of born . let a microsystem now have two distinct observables @xmath8 and @xmath9 which have spectra @xmath10 and @xmath11 accordingly . if physical characteristics @xmath8 and @xmath9 may be simultaneously measured ( i.e. may be measured with zero dispersion with a pair of macroscopic devices of the same type ) , then the vectors by which the state @xmath3 is expanded must be the common eigenvectors of the operators @xmath7 and @xmath12 , leading to the commutation condition @xmath13 = 0 $ ] . if the operators @xmath7 and @xmath12 do not commute , then they do not have a common system of eigenvectors . in this case the observables @xmath8 and @xmath9 can not be measured together by any macro - device . the simplest example of the observables that can not be measured together is the projection of a fermion spin onto two non - parallel directions , which are defined by unit vectors @xmath14 and @xmath15 . another example is the @xmath16-parity and flavour of a neutral pseudoscalar meson . the following question may be posed : do the physical characteristics @xmath8 and @xmath9 exist simultaneously and independently without the assumption of the possibility to measure them by some macro - devices ( this is the hypothesis of local realism ) . usually the terms `` hypothesis of local realism '' and `` concept of macroscopic realism '' are understood as the possibility to describe a physical system in the classical paradigm using some assumptions about the nature of `` classical reality '' . it might be for example locality or the negligible influence of a measurement device . all these assumptuons we will call together the `` hypothesis of realism '' by einstein @xcite . it does not make sense to talk about the physical properties of a micro - object without making a statement about the macro - devices used to measure these properties ( this is the copenhagen interpretation of quantum mechanics and the principle of complementarity of bohr @xcite ) . a natural ( but probably not unique ) way to write in mathematical terms the condition that a set of physical characteristics of a micro - system exists jointly regardless of the possibility to measure it with a macro - device , is that the joint probability of the set of observables under consideration is non - negative at any time . for example for the observables @xmath8 and @xmath9 that means that for any elements of the spectra @xmath17 and @xmath18 the probability of simultaneous existence of @xmath19 and @xmath20 the joint probability @xmath21 satisfies the following condition : @xmath22 the assumption of the existence of non - negative joint probabilities ( [ w(ab ) ] ) was implicitly used by bell in his pioneering works @xcite , as the density distribution @xmath23 of hidden variables @xmath24 is a direct corollary of ( [ w(ab ) ] ) . later bell s idea was developed by clauser , horne , shimony , and holt @xcite . a historical review of bell s inequalities may be found in @xcite , the idea of the existence of non - negative joint probabilities ( [ w(ab ) ] ) was used by wigner in @xcite . in @xcite the arguments of bell were translated into non - negative joint probabilities for the first time . in classical physics the joint probabilities @xmath25 always exist and are well defined for any physical system . in quantum theory if @xmath26 \ne 0 $ ] , the joint probabilities @xmath21 can not be direcly measured by macro - devices . however in this case it is possible to use an indirect procedure based on specific properties of entangled states and the notion of an `` element of physical reality '' introduced by einstein . the `` element of physical reality '' is defined as follows @xcite : _ `` if , without in any way disturbing a system , we can predict with certainty ( i.e. , with probability equal to unity ) the value of a physical quantity , then there exists an element of physical reality corresponding lo this physical quantity . '' _ and _ `` every element of the physical reality must have a counterpart in the physical theory''_. it is obvious that a given element may be identified with some property of a physical system ( for example with a spin projection onto some direction ) and that obtaining information about the element of physical reality differs from obtaining information about some observable only by measurement method . in the first case the measurement is indirect , in the second , direct , and accompanied by a reduction of the state vector or density matrix . because of the above we will not make any distinction in the current paper between the observables and the elements of physical reality . let us show how the indirect procedure works using the decay of a pseudoscalar meson to a fermion - antifermion pair . if the decay happens at time @xmath27 through the strong or electromagnetic interaction ( i.e. preserving @xmath28-parity ) , then the pair will be in a spin - singlet bell state @xmath0 . this fact follows from the general structure of hamiltonians @xmath29 which can be compared with similar decays in quantum field theory ( qft ) . here @xmath30 is the field of pseudoscalar particles , @xmath31 and @xmath32 are fermionic fields , @xmath33 is divergence , @xmath34 and @xmath35 are effective coupling constants . let us denote the antifermion with index `` 1 '' and the fermion with index `` 2 '' . let the spin projections of the fermion and the antifermion onto two directions exist simultaneously or jointly . note that the directions in space are defined by non - parallel unit vectors @xmath36 and @xmath37 , such that the spin projection operators do not commute . for brevity let us denote the spin @xmath38 projection of fermion @xmath39 onto any axis , specified by unit vector @xmath40 , as @xmath41 where @xmath42 . then the spin projections at the initial time @xmath27 onto each of the directions in the state @xmath0 satisfy the anticorrelation condition @xmath43 let us denote the spin projection operators of the fermion and antifermion onto direction @xmath36 as @xmath44 and @xmath45 accordingly . similarly @xmath46 and @xmath47 are the spin projection operators onto direction @xmath37 . as the vectors @xmath36 and @xmath37 are non - parallel @xmath48 \ne 0 , \quad \left [ \hat a^{(2 ) } , \hat b^{(2)}\right ] \ne 0 . \end{aligned}\ ] ] at the same time , according to eberhard s theorem @xcite , @xmath49 = 0 , \quad \left [ \hat a^{(2 ) } , \hat b^{(1)}\right ] = 0 . \end{aligned}\ ] ] equalities ( [ ns - conditions ] ) ensure locality of the quantum theory ( even non - relativistic ) at the level of macro - devices ( so called `` non - signaling conditions '' ) . the commutation conditions ( [ ns - conditions ] ) allow joint measurement for example of the projection of the fermion spin onto direction @xmath36 and the projection of the antifermion spin onto direction @xmath37 . hence the joint probability @xmath50 at any time is a well - defined value and it is possible to use for it probability theory based on kolmogorov s axiomatics . here let us apply to this probability the concept of the elements of physics reality and the anticorrelation condition ( [ pm = mp2 ] ) . then for the time @xmath27 it is possible to formally introduce the joint probability @xmath52 of the existence of physical characteristics of a microsystem ( in our case the projection of fermionic spins onto two non - parallel directions @xmath14 and @xmath15 ) , corresponding to simultaneously non - measurable observables @xmath53 and @xmath54 . so , despite condition ( [ no - commuting ] ) , the definition of the element of physical reality allows us to give operational meaning to the joint probability @xmath55 and its analogs . i.e. , formula ( [ w(ab)physreality ] ) might be considered as a possible expansion of the definition of the joint probability concept to the area where heisenberg s uncertainty principle prevents us from defining such a probability for direct measurements . it seems logical to assume that bayes theorem can be applied to probabilities like ( [ w(ab)physreality ] ) . using the concept of local realism , it is possible to derive not only bell- or wigner - like inequalities , but also equalities . such equalities , often called greenberger-horne-zeilinger ( ghz ) equalities , were introduced in @xcite . proofs of ghz equalities may be found in @xcite . using the concept of local realism and some additional assumptions , a system of equations has been obtained for distinct spin projections of three fermions in the ghz - state @xmath56 . this system is incompatible with calculations in non - relativistic quantum mechanics . beginning in 1972 @xcite there is much experimental evidence of violation of bell s and wigner s inequalities , i.e. evidence of unsoundness of the hypothesis of local realism and/or the concept of elements of physical reality . however until recently these experiments were not free from some loopholes , which cast doubt on the connection between the violations of bell s or wigner s inequalities and the soundness of the quantum mechanical description of the world . the first two are the locality loophole ( or communication loophole ) and fair - sampling loophole ( or detection loophole ) . let us consider the first loophole using the example with fermion - antifermion pairs . in this case the locality loophole appears due to the fact that during the measurement of the spin projections of each of the particles , the spacetime interval between them is time - like . hence in the process of measurement it is not possible to exclude a hypothetical exchange of information between macro - devices , which enables a quasi - non - local strong correlation between the two measurements . such a correlation may lead to the violation of bell s or wigner s inequalities . the `` locality loophole '' was first overcome in experiments by aspect @xcite using the idea by wheeler @xcite of delayed choice for a pair of two - channel polarizers . in quantum optics this loophole is closed by `` brute force '' @xcite , when pairs of correlated photons are separated by a significant distance . for example in photon experiments with fiber cables this distance is greater than a hundred kilometers @xcite . for spin - correlated fermions this distance is about 1.3 kilometers @xcite . when analyzing experiments @xcite one should by definition take into account the limit of the speed of signal exchange between spatially separated subsystems of a correlated quantum system . hence one should use qft where the signal exchange speed is finite , instead of non - relativistic quantum mechanics where this speed is formally infinite @xcite . the second loophole arises due to the fact that all the detectors have certain registration efficiencies , and hence there is a gedanken possibility to select from the whole set the pairs of correlated particles which lead to the violation of bell s or wigner s inequalities , ignoring the others . for exclusion of the detection loophole in bell s inequalities , one needs to have efficiency of the detector of the level 2/3 @xcite . in quantum optics this barrier was overcome only in 2013 @xcite . for wigner s inequalities the authors are not aware of any papers studying that efficiency value . the detection loophole should play an important role in particle physics , because typical efficiencies of detectors ( taking into account selection criteria ) do not exceed percents . up to now , in each of the experiments for testing bell s inequalities , it was possible to close only one of the two loopholes either locality or detection ( @xcite and @xcite ) . however in 2015 three successful experiments were conducted ( two with photon pairs @xcite , and one with correlated spins @xmath2 @xcite , which managed to avoid both loopholes . let us consider another few loopholes that have not been closed by contemporary experiments . also , as will be shown below , the experimental situation chosen to avoid the locality loophole @xcite , raises questions about applying these experiments to time - independent bell s inequalities @xcite and static wigner s inequalities @xcite . among the rest of the loopholes , the most quoted in the literature the freedom - of - choice loophole @xcite , where due to hidden interactions and unknown parameters an experiment itself causes the observer to select events with stronger correlations . in our opinion such a loophole is unfalsifiable and hence should not be considered scientifically according to popper s refutability criterion @xcite . we now consider the `` contextuality loophole '' @xcite , which we believe is as important as the two previously discussed . let us define `` context '' as the aggregate of all the experimental conditions . then , in order to obtain the values of correlators in bell s inequalities or joint probabilities in wigner s inequalities , it is necessary to conduct four or three independent experiments . there is no guarantee that the measurements are conducted under the same conditions . even if one uses the same experimental devices , it is still impossible to repeat exactly all the internal parameters of macro - devices , because , for example , in each of experiments one should select different pairs of spin directions . i.e. it is impossible to make sure that quantum ensembles of correlated particles are identical in all the experiments . all the more , if the experiments were conducted in different and non - controllable conditions , their results should not be summed up , subtracted , or compared with each other . if we suppose that every experiment has its own distribution of probabilities of the observable spectra , then it is possible to obtain a generalized bell s inequality which is never violated in experiment @xcite . to avoid the contextuality loophole it is necessary to conduct the measurements with an invariable state of the macro - devices , i.e. with only two non - parallel directions @xmath36 and @xmath37 . one can use time as an additional degree of freedom . the same inequality , which potentially allows the experimenters to avoid the contextuality loophole , will be introduced in section [ section : mainbayes ] of this paper . time - dependent generalizations of static bell s and wigner s inequalities may be justified in another way . the derivation of these inequalities strongly depends not only on the simultaneous existence of all of the physical characteristics of a micro - system , but also on the assumption of locality both on the macroscopic measurement level ( i.e. eberhard s theorem ) and on the microscopic level ( the hypothesis of local realism ) . local realism contradicts the mathematical structure of non - relativistic quantum mechanics ( nrqm ) . because of that , the violation of bell s inequalities is often considered to be experimental proof of the non - locality of quantum theory . however this is not true , as there are two inseparable potential causes of violation of the static bell s and wigner s inequalities : the absence of joint kolmogorov s probabilities for observables , and non - locality on the microscopic level . to exclude the second possibility , it is necessary to switch to calculations of probabilities and correlators in the framework of qft , which is local on the micro - level by definition , for example because of bogolyubov s principle of microcasuality @xcite . however in qft it is not possible to use static bell s or wigner s inequalities , because it is not possible to exclude interactions of quantum fields with each other and with vacuum fluctuations @xcite . i.e. , in qft any particle or system of particles is an open system . also the description of an entangled quantum system at different spacetime points should take into account relativistic effects , when the finite time of signal propagation between the two parts of the entangled microsystem is beyond the duration of macro - device response . this is undoubtedly true for @xcite . hence to have the possibility of a theoretical description of locality loophole - free spatially separated experiments it is necessary to include time evolution into any bell - like inequalities to assure their compatibility with the theory of relativity . the wigner s inequalities are more suited for relativistic generalization , because their intrinsic joint probabilities are well defined both in nrqm and qft . the correlators in the bell s inequalities are calculated from loop diagrams , part of which calculation depends on the renormalization procedure at each order of perturbation theory . it is not possible to get a definitive answer which does not depend on a renormalization technique @xcite . this paper is a logical continuation of a series @xcite in which we have studied possible relativistic corrections for static wigner s inequalities , and then introduced inequalities that generalized static wigner s inequalities for time - dependent ones required in qft . the main goal of the papers above was testing bohr s complementarity principle in the relativistic domain . the complementarity principle is directly related to the concept of realism . we believe that the statement of `` testing the concept of realism '' , most correctly reflects the gist and the results of @xcite . also the connection between the violation of leggett - garg inequalities and the complementarity principle does not appear to be that obvious . in the current paper we will talk about realism , not complementarity , yet this approach is fully compatible with kolmogorov s axiomatics of probability theory and with the concept of local realism ; it is a particular realization of the supposition of the independent time evolution of every physical ( micro)system . it would be nice to find a time evolution description that does not require the supposition of independence . in section [ section : mainbayes ] of the current paper we will propose such a description , which may be obtained in the framework of kolmogorov s probability theory using a bayesian approach and operates only with ( conditional ) probabilities . in addition to the works cited above , there are many proposals to test bell s inequalities in particle physics . most of these tests use oscillations of neutral pseudoscalar mesons . they were discussed in the work @xcite . note also the analysis of the test of static bell s inequalities in neutrino oscillations @xcite . time - dependent inequalities were proposed by a. leggett and a. garg in their pioneering work @xcite . the initial goal of this work was for testing whether quantum mechanics may be applied at the macroscopic scale for many - particle quantum systems in a coherent state . the inequality , which is satisfied by two - times correlators of one observable , assumes that this observable obeys the laws of classical physics ( this is the concept of `` macroscopic realism per se '' . the expression of the leggett - garg inequalities is similar to the bell s inequalities in form @xcite . because of that the leggett - garg inequalities are sometimes called `` temporal bell inequalities '' @xcite . this name is not precise , because bell s inequalities involve correlators of various observables at one moment of time , while the leggett - garg inequalities should involve different - time correlators of one observable . the name `` time - dependent bell s inequalities '' should be attributed to inequalities that contain probabilities or correlators of various observables at different moments of time . following this logic , the `` time - dependent wigner s inequalities '' were introduced @xcite . references to all the key works related to leggett - garg inequalities may be found in the review @xcite . in 2015 a successful experimental test of violation of the leggett - garg inequalities were conducted : first in an experiment with non - classical movement of a massive quantum particle over a lattice @xcite , and then in a violation of quantum coherence in macroscopic crystals @xcite . the leggett - garg inequalities may also be used to test for the existence of joint probabilities of observables which have non - commuting operators in nrqm , i.e. to test the hypothesis of realism . let us note that in the recent publication @xcite authors contested the well - known statement that the leggett - garg inequalities might be applicable to test the principle of macroscopic realism . but their applicability to test the principle of ( local ) realism is not disputed . the leggett - garg inequalities may be reproduced if one supposes that some hidden parameters @xmath24 whose probability density @xmath57 depends on time in markov s way , exist in a quantum system @xcite . such hidden parameters automatically lead to non - negative joint probabilities . while a test of macroscopic coherence requires soft measurements , a test of the existence of the joint probabilities ( i.e. the `` hypothesis of realism '' without the term `` macroscopic '' ) requires the use of sets of parallel measurements , each conducted at only two fixed moments of time @xcite . this approach related to the methodology of the measurement of four distinct correlators in bell s inequalities , so one might expect that this approach would not be free of the contextuality loophole . the use of projection measurements and sets of parallel experiments for testing the hypothesis of realism in particle physics is considered in @xcite , where entangled pairs of pseudoscalar mesons are used . in 2016 the neutrino experiment minos reported a test of the leggett - garg inequalities in neutrino oscillations @xcite . the current paper considers some generic experimental situations where one measures some properties of a physical system ; then constructs a kolmogorov s space of elementary outcomes and introduces into this space events , corresponding to these experimental situations ; we write some relations in terms of conditional probabilities , because they are well - defined in both classical and quantum physics ( in contrast to joint probabilities ) ; and finally we show that corollaries of bayes theorem are violated for correlated quantum systems . there are many works dedicated to studies of interconnections between the conditional probabilities in quantum and classical theories , starting from fundamental monography by von neumann @xcite and paper by lders @xcite , where rules for calculating conditional probabilities were introduced . important generalizations of the notion of the conditional probability on a generalized probability space of quantum mechanics were presented in @xcite . in the current paper the calculation of conditional probabilities will be based on @xcite . generalizations of lders rule for non - hermitian projection operators for entangled and open quantum systems were proposed in @xcite . based on this generalization , a quantum - bayesian interpretation of quantum mechanics was developed ( so called qbism ) @xcite , which is now , we believe , is one of the most elegant interpretations of quantum theory . it provides a unified approach to classical and quantum phenomena . qbism criticized in @xcite . a versatile analysis of quantum conditional probability and its relation to classics may be found in @xcite . the main conclusion of these works is that von neumann s formula can not be considered a good generalization of classical conditional probability for quantum phenomena , however there is no doubt in lder s rule @xcite and its extention @xcite . this paper consists of the following sections . in section [ section : mainbayes ] , using kolmogorov s approach , conditional probabilities , and bayes theorem , we obtain a static equality and a time - dependent inequality , which allow us to test the hypothesis of realism in time - dependent and open quantum systems . in section [ section : spininh ] we use an example of correlated spin-@xmath38 particles in an external stationary and homogeneous magnetic field to demonstrate that in the framework of nrqm for open quantum systems the relations obtained in section [ section : mainbayes ] are violated . section [ section : mbarm ] is devoted to the study of corollaries to violation of bayes theorem for systems of pseudoscalar neutral mesons . some experiments for testing the concept of realism are proposed . they might be applicable to experiments at the large hadron collider @xcite and belle ii @xcite . appendices [ sec : a][sec : dd ] contain all the auxiliary formulae , that are required for derivation of the results of sections [ section : spininh][section : mbarm ] . consider conditional probabilities in classical and quantum theories using the observables @xmath8 and @xmath9 . in contrast to joint probabilities ( [ w(ab ) ] ) , conditional probabilities like @xmath58 are well - defined in both classical and quantum theories . in the classical case the probability to measure value @xmath20 of the spectrum of observable @xmath9 assuming that value @xmath19 of observable @xmath8 was measured can be written as follows : @xmath59 as was noted above , the joint probability @xmath60 always exists . value @xmath61 is the probability to measure value @xmath19 from the spectrum of observable @xmath8 . from ( [ classicalconditionalprob ] ) a simplest case of bayes theorem can be derived : @xmath62 where @xmath63 : this is the probability to measure value @xmath20 from the spectrum of observable @xmath9 . in the quantum case for the calculation of the conditional probability , one should use von neumann s formula @xcite @xmath64 where @xmath65 is the density matrix of the quantum system in the initial state , @xmath66 is the projector onto the state related to the value @xmath19 of the spectrum of the observable @xmath8 , and @xmath67 is the analogous projector for the value @xmath20 of the spectrum of the observable @xmath9 . applicability of formula ( [ neumann_rule ] ) does not require commutation of @xmath7 and @xmath12 . it does not matter whether the state is entangled or not ( and hence to which subsystems the observables @xmath8 and @xmath9 correspond ) . it also does not matter whether the quantum system is open or not . it is obvious that if @xmath26 \ne 0 $ ] , then @xmath68 hence in the quantum case it is not always possible to obtain an analog of bayes theorem ( [ bayestheorem ] ) . moreover in the framework of qft and for open quantum systems , field operators and observable operators , which consist of fields operators , do not commute at distinct moments of time , i.e. @xmath69 \ne 0 $ ] . because of that the time can be treated as an additional degree of freedom together with spatial directions @xmath14 , @xmath15 and so on ( if we are talking about spins , for example ) . although we will refer to spin projections onto various directions in space , the static equality and time - dependent inequality obtained below are true for any set of dichotomic observables . select three space directions , which are defined by non - parallel unit vectors @xmath14 , @xmath15 , and @xmath70 . let the system of antifermion `` @xmath71 '' and fermion `` @xmath72 '' be in a singlet spin state at time @xmath27 . suppose that the concept of realism is true , i.e. spin projections @xmath73 , @xmath74 , and @xmath75 of the antifermion and fermion onto all three directions exist simultaneously at any time @xmath76 , despite the fact that they can not be measured by any macro - device . at the time @xmath27 these projections obey the anticorrelation condition ( [ pm = mp2 ] ) . for this hypothetical situation it is easy to introduce a classical probability model based on kolmogorov s axiomatics . let us define space @xmath77 of elementary outcomes @xmath78 . each of them is one of the possible sets @xmath79 of spin projections onto all three chosen directions @xmath14 , @xmath80 , and @xmath70 , where indices @xmath81 . the set of the elements of the space @xmath77 does not depend on time . denote `` elementary event '' @xmath82 as a subset of all elementary outcomes @xmath78 of the set @xmath77 ( i.e. @xmath83 and @xmath84 ) . realization of any of these events gives rise to an element of physical reality a concrete set of spin projections @xmath85 . the aggregate of the considered events forms an algebra ( @xmath86-algebra ) @xmath87 . more complicated events may be constructed by merging elementary events . it is possible to introduce a probability measure @xmath88 on @xmath89 , which is always real and non - negative . it is additive ( @xmath86-additive ) for non - intersecting events . using this measure we can define probabilities of joint and conditional events on @xmath77 . in order to derive a static equality let us consider three events @xmath90 , @xmath91 , and event @xmath92 , when the fermion - antifermion pair is in the spin singlet state at @xmath27 . events @xmath93 and @xmath94 can be easily constructed in @xmath77 using the elementary events and condition ( [ pm = mp2 ] ) : @xmath95 in @xmath77 space , event @xmath96 is defined as follows : @xmath97 this notation corresponds to the classical approach , which in this case is identical to the concept of local realism , and in essence differs from a description of event @xmath98 in nrqm using a maximally - entangled bell state @xmath99 where @xmath100 is any of the directions @xmath14 , @xmath15 , or @xmath101 . in qft the initial state is defined using a hamiltonian ( [ heff_for_ps2ff ] ) , which creates a corresponding entangled state when calculating the evolution operator matrix element . if the concept of realism is true , then we can consider non - negative joint and conditional probabilities for the events @xmath102 , @xmath94 , and @xmath98 . it is possible to apply to them a multiplication theorem on @xmath89 . then @xmath103 in analogy @xmath104 equating these results with each other , we obtain the following variant of bayes theorem : @xmath105 experiments that can test the hypothesis of realism using the static equality ( [ bayes - static - bell ] ) are fully identical to those that test static bell s @xcite and wigner s @xcite inequalities . however it is easier to check the violation of ( [ bayes - static - bell ] ) than the violation of bell s inequalities . also , ( [ bayes - static - bell ] ) has an advantage over the ghz - equality @xcite , because it allows the experimenter to check in a system of only two ( entangled ) subsystems , while an experimental check of violation of the ghz - equality requires at least three subsystems in an entangled state . the last condition makes it almost impossible to study the ghz - equality in particle physics . the static equality ( [ bayes - static - bell ] ) is one of two main results of this paper . in sections [ section : spininh ] and [ section : mbarm ] it will be shown that this equality is violated in the framework of quantum theory . we now derive a time - dependent inequality which follows from the hypothesis of realism and bayes theorem . in @xcite the authors have suggested a variant of the time - dependent wigner s inequality , based on kolmogorov s axiomatics and an assumption about the statistical independence of processes of evolution of dichotomic variables of each of two subsystems , which satisfy the condition ( [ pm = mp2 ] ) at the time @xmath27 ( and only at that time ) . for @xmath106 this inequality may be written as follows where the dichotomic variable of the first subsystem is measured at the time @xmath107 , while the dichotomic variable of the second subsystem is measured at the time @xmath108 . although the assumption of statistical independence of the evolution of classical observables is almost obvious in the framework of the hypothesis of local realism , it is quite hard to prove in some cases . for this reason we would like to write an inequality in which the time evolution is a consequence of a more common property of classical objects rather than the property of statistical independence , which is used in the derivation of ( [ w - b-4t1t2 ] ) . that more common property might be bayes theorem . consider two moments in time : the initial @xmath27 , and some @xmath109 . the anticorrelation condition ( [ pm = mp2 ] ) is supposed to be true only at the time @xmath27 . at any other moment of time it might not be satisfied . as the space of elementary outcomes @xmath77 does not depend on time , it is possible to select the following events : the event @xmath110 , the event @xmath111 , where @xmath112 , and the event @xmath113 for which at the time @xmath27 the fermion - antifermion pair were in a singlet spin state . under the hypothesis of realism we again use the bayes theorem , but now for the two moments of time . @xmath114 in this formula the conditional probability @xmath115 is badly defined mathematically in both nrqm and qft . however if we suppose that the hypothesis of realism is true then this conditional probability must satisty the following conditions : @xmath116 we obtain the time - dependent inequality : @xmath117 the time - dependent inequality ( [ bayes - dinamic - bell ] ) is the second main result of the paper . only two directions , @xmath14 and @xmath15 , were used in the derivation of this inequality , not three or more as in bell s and wigner s inequalities . potentially that allows experimenter to test inequality ( [ bayes - dinamic - bell ] ) using only one series of experiments and thus evade the contextuality loophole . it is suitable to write inequality ( [ bayes - dinamic - bell ] ) using the spin @xmath38 projections onto directions @xmath14 and @xmath15 : @xmath118 where , let us emphasize again , we do not suppose any specific dependence of the observables on time , and the event @xmath119 may correspond , in principle , to any initial condition of a system , not only the condition which satisfies ( [ pm = mp2 ] ) . event @xmath93 also might be selected in a generic way as @xmath120 ; however that generalization does not lead to any new types of violation of the inequality ( [ neravenstvobayesa ] ) . we first show that the static equation ( [ bayes - static - bell ] ) is violated in nrqm if we consider a positron and electron in a spin singlet bell state . the density matrix corresponding to this state is @xmath121 . for tests of static inequality the time is not important , so let @xmath122 . main formulae required for derivation of conditional probabilities in static equation ( [ bayes - static - bell ] ) are given in appendix [ sec : a ] . state vector @xmath123 is defined as ( [ correlation - t=0 ] ) . in plane @xmath124 , define three directions @xmath14 , @xmath80 , and @xmath70 . projectors onto the events @xmath125 , @xmath126 , and spin singlet @xmath92 are @xmath127 using ( [ p - t0 ] ) , ( [ wf - t0 ] ) , and ( [ neumann_rule ] ) we obtain @xmath128 where @xmath129 . for calculation of the conditional probabilities @xmath130 and @xmath131 it is necessary to define projectors onto states @xmath132 and @xmath133 . in the general case such a procedure might be non - trivial . however the isotropy of bell state @xmath123 allows us to obtain a simple calculation algorithm . let us find for example @xmath134 . rewrite ( [ psi-_t0 ] ) in terms of the projection onto the direction @xmath14 : @xmath135\ , -\ , { \big\vert\ , a^{(2)}_-\,\big\rangle}{\big\vert\ , a^{(1)}_+\,\big\rangle}\right ) \ , = \nonumber \\ & = & ... \ , + \ , { \big\vert\ , \psi _ { \mathcal{s}_1}\,\big\rangle}\ , + \ , ... \ , .\end{aligned}\ ] ] from this , according to the superposition principle and the results of @xcite , we obtain the non - normalized state vector of event @xmath136 : @xmath137 hence @xmath138 in analogy @xmath139 using the von neumann rule ( [ neumann_rule ] ) , we obtain the following result . if the @xmath140-pair is in a spin singlet state , @xmath141 we used the fact that for a bell state @xmath65 , @xmath142 the equality of the conditional probabilities ( [ ws1s2 ] ) is not general and is only related to the special choice of the initial state @xmath143 . substituting ( [ w_s1-and - w_s2 ] ) and ( [ ws1s2 ] ) into ( [ bayes - static - bell ] ) , if nrqm is compatible with the hypothesis of local realism ( and with probability theory in kolmogorov s axiomatics ) , then for any three directions @xmath14 , @xmath80 , and @xmath70 in plane @xmath124 , the following equation should be always satisfied : @xmath144 obviously this is not true . if the vector @xmath14 is perpendicular to the vector @xmath15 , while the vector @xmath70 is the bisector of the angle between @xmath14 and @xmath15 , then ( [ ab = bc ] ) is violated . we now show that the time - dependent inequality ( [ bayes - dinamic - bell ] ) may also be violated in nrqm . again consider an @xmath145-pair , which at the time @xmath146 is described by the density matrix @xmath121 . put the system into an external constant and homogeneous magnetic field @xmath147 aligned along the @xmath148-axis . from all possible decays , select only those where the leptons are propagated along the magnetic field . this is assumed for simplification of calculation of probabilities . choose two space directions @xmath14 and @xmath15 lying in the plane @xmath149 ; then it is more suitable to test the violation of ( [ neravenstvobayesa ] ) than of ( [ bayes - dinamic - bell ] ) . projectors onto the events @xmath150 , and @xmath92 may be written as @xmath151 the expression for the conditional probability @xmath152 is calculated in ( [ w_s1-and - w_s2 ] ) . in order to obtain two other conditional probabilities , which are included in the formulae ( [ bayes - dinamic - bell ] ) and ( [ neravenstvobayesa ] ) , it is necessary to consider four cases for different values of @xmath153 and @xmath154 for the event @xmath155 . * a ) * let at time @xmath156 the indices @xmath157 and @xmath158 , i.e. @xmath159 . considering that @xmath160 and that @xmath161 , using the von neumann rule ( [ neumann_rule ] ) , we find that @xmath162 the square of the modulus of the corresponding matrix element is calculated using expressions ( [ wf - t - e- ] ) , ( [ wf - t - e+ ] ) , and ( [ psi-_t ] ) . finally : @xmath163 using the von neumann rule , property ( [ proporc ] ) , and formulae ( [ wf - t - e- ] ) and ( [ wf - t - e+ ] ) , one can obtain @xmath164 we have used the standard properties of projection operators : @xmath165 and @xmath166 , where @xmath167 . combine results ( [ w_s1-and - w_s2 ] ) , ( [ s2(t)++ ] ) , and ( [ s2(t)++s1s3 ] ) , and substitute them into the inequality ( [ neravenstvobayesa ] ) , to get @xmath168 if the concept of realism is true then the inequality ( [ neravenstvobayesa++ ] ) should never be violated . however if we choose @xmath76 such that @xmath169 , then ( [ neravenstvobayesa++ ] ) becomes @xmath170 which is violated for most angles @xmath171 . the inequality ( [ neravenstvobayesa++ ] ) may be tested selecting events from one experiment without changing the internal state of a macro - device , to avoid the contextuality loophole . however the detection loophole in this case might still be open . * b ) * let at the time @xmath156 the indices @xmath172 and @xmath173 , i.e. let us consider the event @xmath174 . performing calculations analogous to the above , we find the following values for the conditional probabilities : @xmath175 substituting probabilities from ( [ w_s1-and - w_s2 ] ) and ( [ s2(t)s1s3 ] ) into inequality ( [ neravenstvobayesa ] ) , we find that @xmath176 if we choose @xmath177 , inequality ( [ neravenstvobayesa ] ) is almost always violated : @xmath178 * c ) * finally let us consider the situation when @xmath179 and @xmath180 . then @xmath181 . the corresponding conditional probabilities are equal to @xmath182 inequality ( [ neravenstvobayesa ] ) turns into @xmath183 if we set @xmath184 , we obtain : @xmath185 which like ( [ violate1 ] ) and ( [ violate2 ] ) , is true for almost all choices of directions @xmath14 and @xmath15 . in section [ section : mainbayes ] , equality ( [ bayes - static - bell ] ) and inequality ( [ bayes - dinamic - bell ] ) were obtained in terms of spin @xmath38 projections onto various directions in three - dimensional space . actually , relations ( [ bayes - static - bell ] ) and ( [ bayes - dinamic - bell ] ) are true for any dichotomic observables of any nature in any space . in the case of neutral pseudoscalar mesons @xmath186 , the dichotomic variables can be the flavour of the meson , its @xmath16-parity , and the lifetime ( or mass ) . pseudoscalar mesons are unstable particles , hence they may serve as a simple model of an open quantum system . formulae ( [ bayes - static - bell ] ) and ( [ bayes - dinamic - bell ] ) , in principle , may be tested in experiments at the large hadron collider @xcite , at @xmath9-factory belle ii @xcite and at @xmath187-factories . when decaying a neutral vector meson with the quantum numbers @xmath188 of a photon into a @xmath189-pair , the latter rests in the bell state @xmath0 by flavour , @xmath16-parity , or lifetime ( @xmath190 ) . in appendix [ sec : b ] the main properties of pseudoscalar mesons are presented as well as the formula for evolution of an entangled bell state ( here and below we use @xmath191 ) . as it is impossible to unambiguously relate the spin projections to the directions @xmath14 , @xmath15 , and @xmath70 and projections of states of pseudoscalar mesons onto `` directions '' of flavour , @xmath16-parity and `` directions '' with a definitive mass / lifetime , we need to consider some variants of these correspondences . note that @xmath192 , but @xmath193 . then it is suitable to use the following : @xmath194 , @xmath195 , @xmath196 , @xmath197 , @xmath198 , and @xmath199 , for the algorithms of calculation of projectors @xmath200 and @xmath201 in section [ section : spininh ] to be analogous to the ones from section [ section : mbarm ] . we now show that equality ( [ bayes - static - bell ] ) is violated in systems of neutral pseudoscalar mesons . at @xmath122 the state of the @xmath202-system is defined by density matrix @xmath203 , where the bell state @xmath123 is defined by formula ( [ correlationbbarb - t=0 ] ) . projection operators onto events @xmath204 , @xmath205 , and singlet state in the flavour space @xmath92 are @xmath206 using von neumann rule ( [ neumann_rule ] ) , formulae ( [ pm - t0 ] ) , and ( [ correlationbbarb - t=0 ] ) analogous to ( [ w_s1-and - w_s2 ] ) , one may write that ( see formula [ w - bbarb - i ] ) @xmath207 using a condition of orthogonality analogous to calculations from section [ section : spininh ] for non - normalized projection operators onto the states corresponding to events @xmath208 and @xmath209 , we have @xmath210 then calculation of conditional probabilities according to von neumann s rule ( [ neumann_rule ] ) leads to the equality @xmath211 which is analogous to formula ( [ ws1s2 ] ) . substituting ( [ w_s1m - and - w_s2 m ] ) and ( [ ws1s2 m ] ) into the equality ( [ bayes - static - bell ] ) , we have @xmath212 the equality should be satisfied if the hypothesis of realism is true . as shown in appendix [ sec : b ] for neutral @xmath213 and @xmath214mesons , the ratio @xmath215 is close to @xmath216 while for @xmath217 and @xmath218mesons it is almost always equals to @xmath219 . hence ( [ 2=0m01 ] ) implies false relations like @xmath220 and @xmath221 . if we use the correspondence @xmath194 , @xmath222 , @xmath223 , @xmath224 , @xmath198 and @xmath199 , which differs by interchanging @xmath225 and @xmath226 , then in the framework of the hypothesis of realism we come to the following equality : @xmath227 which , like ( [ 2=0m01 ] ) , is not true in flavour - entangled systems of neutral pseudoscalar mesons . examination of other correspondences leads to equalities which do not provide anything new . now consider the time - dependent inequality ( [ bayes - dinamic - bell ] ) , and let us show that it is also violated in systems of neutral pseudoscalar mesons . the most natural choice of `` directions '' for neutral @xmath214 and @xmath228mesons is related to states with a definite flavour and @xmath16parity . for example at @xmath9factories the flavour of neutral @xmath217meson is determined by a lepton sign in semileptonic decay , while @xmath16parity is determined using the decay @xmath229 . at hadron machines the task is much more complicated , as @xmath230pairs are mainly produced not through the @xmath231 decay but through the process of direct hadronization of @xmath232 quark pairs . in order to select states corresponding to @xmath231 , one needs to know the invariant mass of the @xmath233pair , i.e. to fully reconstruct the energy and momentum of each @xmath217meson . so at hadron machines it is not possible to use semileptonic decays with branching ratios of the order of @xmath234 for determination of the @xmath9-meson flavour . an alternative way to detect the flavour is to use the cascade decay @xmath235 , with a branching ratio of about @xmath236 . for @xmath218mesons at the lhc experiments , the flavour can be determined in the decay @xmath237 . at hadron machines , the statistics required for testing for violation of ( [ bayes - dinamic - bell ] ) should be higher by a few orders of magnitude than at the @xmath9factories . for @xmath214mesons the situation is slightly better because the flavour of the @xmath214meson may be determined in the decay @xmath238 , which has a branching ratio of about 4% . in the case of @xmath213mesons , the hadron machines are not suitable at all , and the test for violation of the inequality ( [ bayes - dinamic - bell ] ) must be performed only at @xmath187-factories . consider events @xmath239 , @xmath240 , and event @xmath92 , which corresponds to the singlet spin state of the @xmath189-pair . using formulae ( [ w - bbarb - i ] ) , ( [ w - bbarb - ii ] ) from appendix [ sec : d ] and the algorithm for calculation of projector @xmath241 described in section [ section : spininh ] , we find @xmath242 using the set of probabilities ( [ probm1 m ] ) , inequality ( [ bayes - dinamic - bell ] ) becomes @xmath243 to understand for which mesons inequality ( [ timem1 m ] ) may be violated , consider the case without @xmath16violation , i.e when @xmath244 . then for @xmath213 and @xmath214mesons formula ( [ timem1 m ] ) becomes @xmath245 which is not violated for any values of @xmath246 , as @xmath247 ( see table [ table : parameters ] ) . for @xmath228-mesons , inequality ( [ timem1 m ] ) becomes @xmath248 we estimate , for @xmath249-mesons , at which times @xmath246 the inequality ( [ timem1m - simple ] ) should be violated . from table [ table : parameters ] one can see that @xmath250 . hence for small adjustments of parameter @xmath76 , the argument of the cosine goes through its full period . hence the condition of guaranteed violation of inequality ( [ timem1m - simple ] ) is given by @xmath251 @xmath252 s@xmath253 and @xmath254 s @xcite , where @xmath255 is the average lifetime of the @xmath249meson , so the condition for guaranteed violation of inequality ( [ timem1m - simple ] ) is @xmath256 . due to the small magnitude of @xmath16violation effects , the exact inequality ( [ timem1 m ] ) should be violated at times of the same order . from the calculations above it is clear that for various choices of events @xmath93 and @xmath257 , inequality ( [ bayes - dinamic - bell ] ) will always transform into the following expression : @xmath258 where functions @xmath259 depend on dimentionless variables @xmath260 , @xmath261 , the modulus of @xmath262 and phase @xmath263 of the ratio @xmath215 ( see appendix [ sec : b ] ) . there are some experimental limits on the range of possible values of parameters @xmath262 and @xmath263 . in @xcite and @xcite concerning the modelling of the violation of time - dependent wigner s inequalities in systems of @xmath218-mesons the following values were selected : @xmath264 and @xmath265 . in the present paper we will also use these values . if we introduce function @xmath266 then inequality ( [ timem1 m ] ) transforms into ( [ f_n ] ) . a plot of the function @xmath267 is shown in figure [ fig : f1_and_f5_bs ] left . that taking into account the oscillations , the violation of inequality ( [ timem1 m ] ) holds for @xmath268 , which agrees with a nave estimate obtained from the simplified inequality ( [ timem1m - simple ] ) . one can also obtain inequality ( [ timem1 m ] ) by choosing events @xmath269 and @xmath270 in the form @xmath271 , @xmath272 . the event @xmath92 remains the same . 99 w. heisenberg , `` physics and philosophy . the revolution in modern science '' , `` harper & row '' , n .- y . , 1958 . i. kant , `` kritik der reinen vernunft '' , riga , 1781 . a. m. dirac , `` the principles of quantum mechanics '' , oxford university press , 1999 . a. peres , `` quantum theory : concepts and methods '' , kluwer , dordrecht ( 1993 ) . a. einstein , b. podolsky and n. rosen , phys . * 47 * ( 1935 ) 777 . n. bohr , phys . rev . * 48 * , 696 ( 1935 ) . j. s. bell , physics * 1 * , 195 ( 1964 ) . j. s. bell , rev . phys . * 38 * , 447 ( 1966 ) . j. f. clauser , m. a. horne , a. shimony , and r. a. holt , phys . * 23 * , 880 ( 1969 ) . , ed . by f. selleri ( plenum press , ney york and london , 1988 ) ; _ the einstein - podolsky - rozen paradox in atomic , nuclear and particle physics _ , ed . by a. afriat and f. selleri ( science , 1999 ) ; g. auletta , _ foundations and interpretations of quantum mechanics _ ( world scientific , singapore , 2001 ) ; _ quantum ( un)speakables _ , ed . by r. a. bertlmann and a. zeilinger ( springer , 2002 ) . m. d. reid , p. d. drummond , w. p. bowen , _ et al . _ , rev . phys . * 81 * , 1727 ( 2009 ) . d. rosset , j .- d . bancal , and n. gisin , arxiv:1404.1306 [ quant - ph ] . n. brunner , d. cavalcanti , s. pironio , _ et al . _ , rev . phys . * 86 * , 419 ( 2014 ) ; rev . phys . * 86 * , 839 ( erratum ) ( 2014 ) . conference `` quantum [ un]speakables ii : 50 years of bell s theorem '' , vienna ( 2014 ) , ` https://bellconference2014.univie.ac.at ` . e. p. wigner , am . j. phys . * 38 * , 1005 ( 1970 ) . w. m. de muynck , phys . a * 114 * , 65 ( 1986 ) . p. h. eberhard , nuovo cimento * * 46**b , 392 ( 1978 ) . d. m. greenberger , m. horne , a. zeilinger , _ in `` bell s theorem , quantum theory , and conceptions of the universe '' _ , edited by m. kafatos ( kluwer , dordrecht , 1989 ) d. m. greenberger , m. a. horne , a. shimony , a zeilinger , am . j. phys . 58 , pp.1131 - 1143 ( 1990 ) . n. d. mermin , am . j. phys . 58 , pp . 731733 ( 1990 ) . s. j. freedman , j. f. clauser , phys . * 28 * , 938 ( 1972 ) . a. aspect , p. grangier , g. roger , phys . lett . * 49 * , p.91 ( 1982 ) . j. a. wheeler , _ in `` quantum theory and measurement '' , ed . by j.a . wheeler and w.h . zurek _ , princeton university press , princeton , nj , pp . 182 - 213 ( 1984 ) . w. tittel , j. brendel , h. zbinden , n. gisin , phys lett . * 81 * , p.3563 ( 1998 ) . a. zeilinger et al . , nature * 489 * , pp . 269 - 273 ( 2012 ) . b. hensen et al . , nature * 526 * , pp.682 - 686 ( 2015 ) . n. nikitin and k. toms , phys . a * 82 * , 032109 ( 2010 ) . p. h. eberhard , phys . a * 47 * , r747 ( 1993 ) . m. giustina et al . , nature * 497 * , pp . 227 - 230 ( 2013 ) . b. g. christensen et al . lett . * 111 * , 130406 ( 2013 ) . a. aspect , physics * 8 * , p.123 ( 2015 ) . m. giustina et al . , phys . lett . * 115 * , 250401 ( 2015 ) . l. k. shalm et al . lett . * 115 * , 250402 ( 2015 ) . j. s. bell , _ in `` speakable and unspeakable in quantum mechanics '' _ , p.232 ( cambridge university press , cambridge , 2004 ) . k. popper , `` logik der forschung '' , 1934 . w. de baere , lett . nuovo cimento * 39 * , pp . 234 - 238 ( 1984 ) . w. de baere , lett . nuovo cimento * 40 * , pp . 488 - 498 ( 1984 ) . khrennikov , _ `` contextual approach to quantum formalism ( fundamental theories of physics ) '' _ , springer , 2009 . n. n. bogoliubov , d. v. shirkov , _ introduction to the theory of quantized fields _ ( john wiley , 1980 ) , 3rd ed . k. a. kazakov , v. v. nikitin , ann . phys . * 327 * , pp . 2914 - 2945 ( 2012 ) . m. e. peskin , d. v. schroeder , _ `` an introduction to quantum field theory '' _ ( addison - wesley advanced book program , 2005 ) . n. nikitin , v. sotnikov and k. toms , phys . a * 90 * , 042124 ( 2014 ) . n. nikitin , v. sotnikov and k. toms , phys . d * 92 * , 016008 ( 2015 ) . n. v. nikitin , v. p. sotnikov and k. s. toms , phys . * 78 * , 805 ( 2015 ) . m. bahrami , s. donadi , l. ferialdi , a. bassi , c. curceanu , a.di domenico and b.c . hiesmayr , scientific reports * 3 * , 1952 ( 2013 ) . s. banerjee , a. k. alok , r. srikanth , b. c. hiesmayr , european physical journal c * 75 * , p.487 ( 2015 ) . a. j. leggett and a. garg , phys . * 54 * ( 1985 ) 857 . g. waldherr , p. neumann , s. f. huelga , f. jelezko , and j. wrachtrup , phys . lett . * 107 * , 090401 ( 2011 ) ; erratum phys . lett . * 107 * , 129901 ( 2011 ) . c. emary , n. lambert and f. nori , rep . 77 * , 039501 ( 2014 ) . c. robens , w. alt , d. meschede , _ et al . _ , phys . x * 5 * , 011003 ( 2015 ) . zhou , s. f. huelga , c - f . li , g - c . guo , phys . lett . * 115 * , 113002 ( 2015 ) . l. clemente and j. kofler , phys . lett . * 116 * , 150401 ( 2016 ) . s. f. huelga , t. w. marshall , and e. santos , phys . a * 54 * , p.1798 ( 1996 ) . g. c. knee et al . , nature communications * 3 * , 606 ( 2012 ) . d. gangopadhyay , d. home , and a. s. roy , phys . a * 88 * , 022115 ( 2013 ) . j. a. formaggio , d. i. kaiser , m. m. murskyj , t. e. weiss , phys . lett . * 117 * , 050402 ( 2016 ) . j. von neumann , `` mathematische grundlagen der quantenmechanik '' , springer - verlag , berlin ( 1932 ) ; english translation : `` mathematical foundations of quantum mechanics '' , princeton university press , princeton , nj ( 1955 ) . g. lders , `` ber die zustandsanderung durch den messprozess '' , annalen der physik * 8 * , 322 - 328 ( 1951 ) . english translation by k. a. kirkpatrick , arxiv e - print : quant - ph/0403007 . g. cassinelli , n. zanghi , nuovo cimento * * 73**b , 237 ( 1983 ) . g. cassinelli , n. zanghi , nuovo cimento * * 79**b , 141 ( 1984 ) . c. a. fuchs , in _ proceedings of the nato advanced research workshop on decoherence and its implications in quantum computation and information transfer _ , edited by a. gonis ( plenum , new york , 2001 ) , arxiv : quant - ph/0106166 . c. a. fuchs , _ quantum theory : reconsideration of foundations _ , modeling * 2 * ( vxj university press , vxj ) , edited by a. khrennikov , 463 ( 2002 ) , arxiv : quant - ph/0205039 . c. a. fuchs and r. schack , rev . * 85 * , 1693 ( 2013 ) . n. d. mermin , nature * 507 * , 421 ( 2014 ) . a. khrennikov , arxiv:1604.07766 [ quant - ph ] . i. g. bobo , `` quantum conditional probability '' , ph.d . thesis , madrid 2010,i . g. bobo , `` on quantum conditional probability '' , an international journal for theory , history and foundations of science * 28 * , 115 ( 2013 ) . lhcb collaboration , journal of instrumentation * 3 * , s08005 ( 2008 ) . lhcb collaboration , int . phys . a 30 , 1530022 ( 2015 ) . atlas collaboration , jinst 3 ( 2008 ) s08003 . cms collaboration , jinst 3 ( 2008 ) s08004 . t. aushev et al . ( belle ii collaboration ) , arxiv:1002.5012 [ hep - ex ] . k. a. olive _ [ particle data group collab . ] , chin . c * 38 * , 090001 ( 2014 ) . f. uchiyama , phys . a * 231 * , 295 ( 1997 ) . r. a. bertlmann and b. c. hiesmayr , phys . a * 63 * , 062112 ( 2001 ) [ hep - ph/0101356 ] . r. a. bertlmann , w. grimus , and b. c. hiesmayr , phys . lett . a * 289 * , 21 ( 2001 ) [ quant - ph/0107022 ] . a. j. buras , s. jager , and j. urban , nucl . b * 605 * , 600 ( 2001 ) [ hep - ph/0102316 ] . p. del amo sanchez _ et al . _ [ babar collab . ] , phys . lett . * 105 * , 081803 ( 2010 ) [ arxiv:1004.5053 [ hep - ex ] ] . k. abe _ et al . _ [ belle collab . ] , phys . lett . * 99 * , 131803 ( 2007 ) [ arxiv:0704.1000 [ hep - ex ] ] .

the accuracy of indoor wireless localization systems can be substantially enhanced by _ map - awareness _ , i.e. , by the knowledge of the map of the environment in which localization signals are acquired . in fact , this knowledge can be exploited to cancel out , at least to some extent , the signal degradation due to propagation through physical obstructions , i.e. , to the so called _ non - line - of - sight _ bias . this result can be achieved by developing novel localization techniques that rely on proper map - aware statistical modelling of the measurements they process . in this manuscript a unified statistical model for the measurements acquired in map - aware localization systems based on _ time - of - arrival _ and _ received signal strength _ techniques is developed and its experimental validation is illustrated . finally , the accuracy of the proposed map - aware model is assessed and compared with that offered by its map - unaware counterparts . our numerical results show that , when the quality of acquired measurements is poor , map - aware modelling can enhance localization accuracy by up to 110% in certain scenarios . localization , map - aware , toa , rss , nlos .

1402.3783

in recent years significant attention has been devoted to the development of accurate and low cost wireless localization systems for indoor civilian applications , since they can be employed to provide a number of new services , like asset tracking and tracking of people with special needs @xcite . in these services estimated positions need always to be related to a surrounding infrastructure ( e.g. , rooms and corridors ) to be useful to their end users , so that the knowledge of the map of the environment where users are expected to move ( e.g. , the plan of a building floor ) is required . in principle , map knowledge ( i.e. , _ map - awareness _ ) can be also exploited to improve the estimation accuracy of a localization system . in fact , any wireless localization system first acquires a set of point - to - point measurements related to user position ( first step ; technology - dependent ) and then processes such measurements for _ bi - dimensional _ ( 2-d ) or three - dimensional position estimation by means of a proper localization technique ( second step ; technology - agnostic ) ( * ? ? ? 4 ) . maps can play a significant role in the second step , since they provide information about environmental obstructions ( e.g. , walls ) which interfere with signal propagation ; however , a full exploitation of these information requires a ) the availability of _ map - aware statistical models _ for the acquired measurements and b ) the development of localization techniques explicitly based on these models . at present the only available map - aware statistical models are the model ( * ? ? ? * eq . ( 6 ) ) and the model ( * ? ? ? 4.11.5 ) ( also known as _ wall - attenuation _ model or _ multi - wall _ model @xcite ) ; these models have been developed for and localization systems , respectively , and are based on experimental evidence . this preliminary work shows that map - awareness can significantly improve localization accuracy by compensating for the so called _ bias _ , which is a major source of error . however , as far as we know , the accuracy and validity of the above mentioned models in real world localization systems is under - explored and , generally speaking , there is a lack of experimental results supporting them in the technical literature . in fact , most of the state - of - the - art localization methods rely on _ map - unaware models_. for instance , the well known _ log - distance propagation model _ @xcite ( or , in some cases , models based on polynomial series expansions @xcite ) are adopted to relate rss to distance . similar comments hold for those models relating and to distance ; in this case additive error terms are usually represented by gaussian _ random variables _ ( rvs ) * eq . ( 6 ) ) , @xcite , although more refined models accounting for propagation are also available ( see @xcite and references therein ) . it is also worth mentioning that map - awareness is implicitly employed in _ fingerprinting_-based localization systems to select fingerprint locations @xcite . in those systems no statistical modelling of acquired measurements is needed ( even if combined fingerprinting / statistical approaches are possible @xcite ) ; however , extensive and time consuming measurement campaigns ( which may be very sensitive to environmental changes ) are necessary to achieve an acceptable accuracy , since the localization error of fingerprinting methods is roughly bounded by the spacing between calibration sites . the aim of this manuscript is twofold . first of all , a novel _ unified statistical map - aware model _ for , or measurements is proposed and is validated exploiting a set of rss and toa data acquired in indoor environments . secondly , the improvement in localization accuracy provided by optimal localization algorithms based on the novel map - aware modelling with respect to their counterparts relying on map - unaware modelling is quantified ; this unveils that , specially in rss systems , the accuracy improvement justifies the increased complexity of map - aware modelling . the proposed map - aware model has the following relevant features : a ) it relates the nlos bias affecting measurements to map geometrical features ; b ) it can be employed in localization systems based on ranging techniques ( i.e. , toa , tdoa and rss , but not angle of arrival @xcite ) provided that their radio signals mainly propagate through obstructions ; c ) it contains few parameters to be estimated from measurements ; d ) even if its validity is assessed for narrowband low - frequency rss measurements and for toa measurements , its use can be envisaged for other technologies , like _ wireless local area network _ toa / tdoa or even non - radio - based technologies ( e.g , ultra - sound ) , since it does not rely on technology - specific properties ; e ) likelihood functions based on it can be employed in navigation systems ( where mobile agents are considered ) . this manuscript is organized as follows . in section [ sec : statistical_model ] , the localization system we consider is described and general statistical models for map - aware and map - unaware scenarios are proposed . specific models for rss and toa measurements , based on our experimental data , are illustrated in section [ sec : experimental_results ] . in section [ sec : accuracy_results ] map - aware and map - unaware optimal localization algorithms are derived , their performance is assessed and compared , and some indications about their computational complexity are provided . finally , section [ sec : conclusions ] offers some conclusions . _ notations _ : the probability density function ( pdf ) of the rv @xmath0 evaluated at the point @xmath1 is denoted @xmath2 ; @xmath3 denotes the pdf of a gaussian rv having mean @xmath4 and variance @xmath5 , evaluated at the point @xmath1 ; @xmath6 denotes the pdf of an exponential rv translated by @xmath7 and having mean @xmath8 , evaluated at the point @xmath1 , so that @xmath9 , where @xmath10 denotes the heaviside step function ; @xmath11 denotes the cardinality of the set @xmath12 ; @xmath13 denotes the floor of the real parameter @xmath14 ; @xmath15 denotes the convolution integral . > 018inkscape -z -d file = fig1.svg export - pdf = fig1.pdf export - latex in the following we focus on a 2-d localization system operating in an indoor scenario and employing @xmath16 devices , called _ anchors _ and whose positions @xmath17 are known , to estimate the position @xmath18 of a single _ agent _ in a large indoor scenario ( i.e. , the floor of a large building ) . the system structure is exemplified by fig . [ fig : system_model ] for @xmath19 . formulating the problem of _ map - aware localization _ requires defining mathematical models for a ) its map , b ) the wireless connectivity between the agent and the anchors , and c ) the measurements acquired for position estimation . these models are illustrated in the following three paragraphs . our prior knowledge about the agent position @xmath20^{t}$ ] is expressed by the _ uniform _ pdf @xcite @xmath21 where @xmath22 denotes the region where the agent is constrained to lie and @xmath23 its area . note that ( [ eq : map_pdf ] ) describes a _ uniform map _ model , which is fully characterized by the map _ support _ @xmath24 ( whose _ bounding box _ is @xmath25 ) . in the following we assume that any _ map - aware _ ( _ map - unaware _ ) localization system is endowed with the knowledge of @xmath24 ( @xmath25 ) . note that the propagation of wireless signals ( and , consequently , localization performance ) is affected by the obstructions ( e.g. , walls ) shaping the map support @xmath24 ; in particular , an important role is played by the number of obstructions @xmath26 interposed between two arbitrary points @xmath27 and @xmath28 , both belonging to @xmath24 . we assume that an agent located at @xmath18 is connected with the @xmath29-th anchor when the of the wireless signal radiated by this anchor and received by the agent exceeds a given threshold ( i.e. , the agent s _ receiver sensitivity _ ) ; therefore , each anchor is characterized by a specific coverage region . even if the map of the environment is known , the prediction of the shape of each coverage region is not easy . for this reason , the following simple ( map - unaware ) connectivity model is adopted : the coverage region @xmath30 of the @xmath29-th anchor is a circle centered at @xmath31 and whose radius @xmath32 depends on the power radiated by the anchor itself and the propagation conditions represents a form of a priori knowledge ; when such a knowledge is unavailable , @xmath33 can be selected . ] . then , the signal coming from the @xmath29-th anchor is _ theoretically _ detectable inside @xmath30 and in such a region some localization information or , more precisely , an _ observation _ ( denoted @xmath34 ) can be acquired in the first step of any localization technique ( see section [ sec : intro ] ) . in practice , in the presence of harsh propagation conditions the number @xmath35 of observations available to the agent may differ from the number which can be predicted theoretically using the above mentioned connectivity model . let @xmath36 denote the set of indices associated with the anchors _ truly connected _ with the agent ; then , we have that @xmath37 and that , if @xmath38 , then @xmath39 . consequently , in a _ map - aware _ localization system the agent position has to be searched for inside the region @xmath40 whereas , in a _ map - unaware _ system , the agent position is expected to belong to @xmath41 finally , it is important to note that the wireless links referring to the anchors connected with an agent located at @xmath18 can be in or conditions . for this reason , the set @xmath36 can be partitioned into the subsets @xmath42 and @xmath43 containing the indices of the connected anchors in and conditions , respectively . the localization system we consider statistically infers the agent position @xmath18 from a set of @xmath35 observations ( in practice , estimated ranges ) @xmath44 extracted from the radio signals coming from the connected anchors . for an indoor environment , such signals are affected by _ multipath fading _ 5 ) which substantially complicates the problem of the statistical modelling of the available observations . to simplify this problem , we assume that : a ) the time interval needed to acquire all the observations is small with respect to the coherence time of the propagation scenario , so that the effects due to its _ time selectivity _ can be ignored ; b ) in toa - based systems the effects due to the _ time dispersion _ characterising the propagation scenario are mitigated transmitting wideband signals , so that multiple echoes of the signal radiated by a given anchor can be resolved by the agent ( in particular , the first path can be correctly detected ) . given such assumptions , the observation model @xmath45 characterized by additive bias and noise terms and commonly adopted in the localization literature ( e.g. , see ( * ? ? ? * eq . ( 3 ) ) , @xcite and ( * ? ? ? * eq . ( 3 ) ) ) can be also employed in our scenario for any @xmath38 ; here @xmath46 denotes the _ true distance _ between the agent position @xmath18 and the @xmath29-th anchor , @xmath47 represents the nlos _ bias _ originating from obstructions affecting the @xmath29-th wireless link and @xmath48 is a noise term modelling various sources of errors ( e.g. , the error introduced in extrapolating the observation @xmath34 from the continuous - time waveform collected by the agent antenna , the radio interference , and the noise generated by both transmitter and receiver hardware ) . it is important to point out that : 1 . the effects of _ spatial selectivity _ ( that may generate substantial fluctuations in the received power as the agent position changes ) are difficult to predict and hence have been included in the noise term , even if the availability of a simple and reliable model for spatial fading would certainly improve model accuracy . the @xmath29-th observation @xmath34 ( [ eq : z_model ] ) is generated by mapping , through a function denoted @xmath49 in the following , the physical quantity measured by the hardware ( e.g. , a rss or a toa ) into a _ range_. the function @xmath49 is derived under the assumption of los propagation , since nlos propagation conditions are accounted for by introducing the bias term @xmath47 in ( [ eq : z_model ] ) . in map - aware modelling such a term is modelled as @xmath50 for @xmath51 and @xmath52 for @xmath53 ; a different ( and not necessarily gaussian ) model is adopted , instead , in map - unaware modelling , as discussed in detail below . providing further details about the observation model ( [ eq : z_model ] ) requires a ) considering specific localization techniques ( in the following , quantities specifically referring to toa and rss techniques are identified by the superscripts @xmath54 and @xmath55 , respectively ) and b ) making a clear distinction between map - aware and map - unaware modelling , as illustrated below . toa__-based localization _ _ - in this case the observation @xmath34 is related to the measured toa ( expressed in seconds ) by the los mapping @xmath56 , where @xmath57 is the speed of light in vacuum . in nlos propagation conditions @xmath58 is affected by a _ bias _ ( due to the fact that obstacles slow down electromagnetic waves ) ; in map - aware _ _ systems , the nlos bias mean @xmath59 can be modelled as ( * ? ? ? 3.3.1 ) @xmath60 for any @xmath38 , where @xmath61 and @xmath62 are the thickness and the relative electrical permittivity , respectively , of the @xmath63-th obstruction . if all the obstructions have the same thickness @xmath64 and the same permittivity @xmath65 , ( [ eq : bias_mean_toa ] ) simplifies as @xmath66 ( * ? ? ? * eq . ( 6 ) ) . in map - unaware localization systems , instead , we assume that : a ) a state - of - the - art localization strategy relying on a los / nlos detector is adopted ; b ) such a detector is characterized by an error ( i.e. , false detection and missed detection ) probability @xmath67 over each link ; c ) it provides the map - unaware localization algorithm with the estimates @xmath68 and @xmath69 of the sets @xmath70 and @xmath71 , respectively . given these estimates , the model @xmath72 is employed for any @xmath73 , since the nlos bias is always positive for toa measurements * * @xcite . tdoa__-based localization _ _ - the results illustrated above for toa modelling can be easily exploited for tdoa - based localization systems . in fact , in this case @xmath34 is evaluated subtracting the toa provided by a _ reference _ anchor from that acquired by the @xmath29-th anchor and both measurements are modelled by ( [ eq : z_model ] ) ; further details are not provided here for space limitations . rss__-based localization _ _ - in this case the observation @xmath34 is related to the measured rss @xmath74 ( i.e. , the average power of the radio signal received by the agent from the @xmath29-th anchor ) according to the formula @xmath75 ; specific forms for @xmath76 are discussed later in sec . [ sub : rss_measurements ] . nlos propagation needs to be accounted for carefully in rss - based systems , since the attenuation due to obstructions results in an overestimated range @xmath77 . however , unlike the toa and tdoa cases , no simple and accurate theoretical formula is available to model the nlos bias mean in map - aware localization , so that models extracted from experimental data must be employed . in our work measurement - based models have been developed for both @xmath78 and @xmath79 ( see section [ sec : experimental_results ] ) ; such models are similar to the experimental model ( developed in ( * ? ? ? * sec . 4.11.5 ) and assuming a logarithmic dependence of the rss on the number of obstructions ) . in the map - unaware case , instead , we assume again the availability of a los / nlos detector characterized by the error probability @xmath67 and adopt * * the bias model @xmath80 . the statistical models illustrated above allow us to develop a novel _ unified _ ( i.e. , valid for toa , tdoa and rss ) statistical representation for the set of observations @xmath44 . in fact , given the trial agent position @xmath81 , the _ map - aware _ likelihood associated with these observations can be expressed as ( see ( [ eq : z_model ] ) ) @xmath82 where @xmath83^{t}$ ] . similarly , the _ map - unaware _ counterpart of ( [ eq : likelihood_with_maps ] ) , given the trial agent position @xmath84 , can be put in the form @xmath85 or @xmath86\label{eq : likelihood_no_maps_exp_exact}\end{aligned}\ ] ] if a gaussian or an exponential model is adopted for the nlos bias , respectively . in the latter case , the convolution integral appearing in the right - hand side of ( [ eq : likelihood_no_maps_exp_exact ] ) substantially complicates the likelihood ( [ eq : likelihood_no_maps_exp_exact ] ) ; for this reason in the technical literature ( e.g. , see @xcite ) it is usually assumed that the nlos bias dominates over the gaussian noise , so that ( [ eq : likelihood_no_maps_exp_exact ] ) can be approximated as @xmath87 finally , it is important to stress that ( [ eq : likelihood_with_maps ] ) and ( [ eq : likelihood_no_maps_gaussian])-([eq : likelihood_no_maps_exp ] ) represent _ general _ statistical models applying to toa , tdoa and rss measurements ( under the assumptions described above ) ; in the next section they will be _ specialized _ to fit our experimental results . in this section we provide a ) various details about the measurement campaigns supporting our proposed models , b ) some explicit expressions for the functions @xmath76 , @xmath78 , @xmath88 and @xmath89 , and c ) the exact form of the likelihood functions fitting our experimental data . rss data have been acquired in a measurement campaign performed by our research group in the first half of 2012 . two different types of radio devices operating in two distinct frequency bands ( namely , @xmath90 and @xmath91 ) have been employed . in this paragraph a detailed analysis of the results extracted from the data acquired at @xmath90 only is provided ; we comment on the differences between theses results and those referring to @xmath91 in the following paragraph . in our measurement campaign two emb - wmb169 t narrowband radio modules based on texas instruments transceivers have been employed @xcite . these modules have the following relevant features : a ) they transmit at @xmath92 employing a 2-gfsk modulation ( whose bandwidth is about 3 khz ) ; b ) they radiate @xmath93 at the carrier frequency @xmath94 ; c ) their antenna is a dipole omnidirectional in the azimuthal plane and @xmath95 long ( it is shorter than the wavelength @xmath96 for practical reasons ) and its gain is equal to @xmath97 . various measurements have been acquired in both los and nlos conditions . the procedure we adopted was similar to that described in @xcite . on the 2nd floor of our departmental building ( see fig . [ fig : map_dii ] ; the involved area was about @xmath98 ) @xmath99 different `` measurement sites '' have been selected ; let @xmath100 denote the position of the @xmath101-th site , with @xmath102 . then , the above mentioned radio modules were employed to acquire rss measurements referring to all @xmath103 location pairs . in four of these pairs no wireless connection was established , yielding 116 measured links ( the number depends on the coverage region of each anchor , which , in turn , is affected by both a ) the power radiated by the anchor itself and b ) the propagation environment ( see our connectivity model of paragraph [ sub : connectivity_model ] ) . ] of measurement sites `` visible '' from the @xmath101-th site will be denoted as @xmath104 ) for each link ( i.e. , couple of sites ) , @xmath105 separate rss measurements were acquired ( our experimental setup is shown in fig . [ fig : expsetup_picture ] ) . such measurements have been stored in a database and processed to estimate the functions @xmath78 , @xmath88 and @xmath89 , as illustrated in detail in appendix [ app : mapaware_fitting ] . note that the generation of such a database required a density of measurements ( 1 measurement site every @xmath106 ) much lower than that required by fingerprinting methods ( e.g. , in @xcite 1 measurement site every @xmath107 was necessary ) . > 018inkscape -z -d file = fig2.svg export - pdf = fig2.pdf export - latex , width=226 ] additional rss measurements have been acquired in los conditions to estimate the function @xmath108 ( see paragraph [ sub : observation_model ] ) . in particular , about 2200 rss measurements have been acquired in a long corridor of the same environment for 44 distinct values of @xmath109 . [ fig : rss_fitting_169mhz ] shows such experimental results and their regression fits referring to a ) the log - distance path loss model ( see @xcite ) @xmath110 where @xmath111 denotes the rss measured at a reference _ _ distance @xmath112 from the transmitter and @xmath113 is the path - loss exponent , and b ) the linear model @xmath114 in evaluating the regression curves , the values of the parameters @xmath111 , @xmath113 and @xmath112 ( @xmath111 and @xmath115 ) have been optimized in the first ( second ) case . the results illustrated in fig . [ fig : rss_fitting_169mhz ] evidence that , as already mentioned in some previous work ( e.g. , see @xcite and @xcite ) , in indoor los conditions a simple linear law may offer a better match with experimental data than the standard model ( [ eq : logdistance_path_loss ] ) ; this can be partly related to wave guiding effects characterizing indoor propagation along corridors ( e.g. , see @xcite ) . actually , the model ( [ eq : logdistance_path_loss ] ) might offer a better match with experimental data if additional rss measurements referring to distances greater than @xmath116 were available ; note , however , that in the considered indoor environment @xmath116 is the maximum los link distance . for these reasons , in the following , the model ( see ( ) ) @xmath117 is adopted for the acquired rss measurements ; the values selected for its parameters are listed in table [ tab : statistical_model_summary ] . in comparing our measurements at @xmath91 ( not shown here for space limitations ) with those referring to @xmath90 the following relevant differences were found : a ) a stronger attenuation due to obstructions ( and , in particular , an attenuation increasing faster with @xmath118 ) is experienced at @xmath91 ; b ) the log - distance path loss model ( [ eq : logdistance_path_loss ] ) is more accurate than the linear one ( [ eq : lineardistance_path_loss ] ) at @xmath91 ; c ) a lower sensitivity to the presence of people moving around the measurement sites is found at @xmath90 , since the corresponding wavelength @xmath119 m is comparable or larger than the typical size of a person ; d ) spatial variations in the rss are slower at @xmath90 than those experienced at @xmath91 ( this results in an increased rss stability for small displacements of the antennas of the employed radio devices ) . these results suggest that localization systems operating at low frequencies require a smaller number of anchors ( see point a ) and exhibit an higher stability , in a practical scenario ( see point d ) than the widely used counterparts exploiting @xmath91 radios ; for this reason , in this manuscript we focus on measurements acquired at @xmath90 . finally , it is also worth mentioning that very few studies about indoor low - frequency localization are available in the technical literature @xcite . a campaign for acquiring toa measurements has not been necessary , since various experimental data are already publicly available in the newcom++ eu database @xcite , whose measurements refer to @xmath120 different sites of the building floor shown in fig . [ fig : map_deis ] and have been acquired using a couple of timedomain pulseon 220 radios ( resulting in 35 nlos links and 33 los links ) . the newcom++ database @xcite has already been analysed in @xcite ; in particular , in @xcite it has been shown that its range measurements can not be described by a gaussian model if the bias is not accounted for . in appendix [ app : mapaware_fitting ] it is shown instead that , if the bias model described in sec . [ sub : observation_model ] is adopted , a good fit can be provided by the gaussian map - aware model ( [ eq : likelihood_with_maps ] ) . > 018inkscape -z -d file = fig5.svg export - pdf = fig5.pdf export - latex the procedure we adopted in processing the available rss and toa measurements for map - aware fitting consists of the following steps ( see fig . [ fig : fitting ] ) : 1 . randomly partitioning the acquired measurements into a _ training _ set , containing ~25% of the available experimental measurements , and a _ validation _ set , collecting the remaining part of the measurements ; the last set has been processed ( see paragraph [ sub : accuracy_test ] ) to validate the proposed models against statistically independent data @xcite ; 2 . processing map information and the measurements contained in the _ training _ set in order to extract a ) estimates of the bias mean and variance versus the number of obstructions and b ) ml estimates of the noise standard deviation versus the link distance . 3 . developing smooth parametric functions @xmath121 for the bias and a smooth parametric function @xmath122 for the noise variance by evaluating standard regression fits ( based on the ml estimates evaluated in the first step ) . in appendix [ app : mapaware_fitting ] a detailed description of these steps is provided ; in this paragraph we limit to summarising the main results . the _ map - aware _ _ likelihood _ model we propose for toa measurements iswhereas the _ map - aware likelihood _ model for rss measurements is @xmath123 where @xmath124 . it is important to point out that : a ) map - awareness is embedded in these likelihoods through the @xmath125 function ; b ) the mathematical structure of the bias and noise models contained in the last two expressions is based on experimental evidence , as illustrated in appendix [ app : mapaware_fitting ] ; c ) although simple models can be developed for fitting the mean of the nlos bias ( see fig . [ fig : rss_fitting_bias_mean ] ) , the same does not hold for both bias variance and noise variance ; d ) anomalies in bias variance and noise variance fitting can be related to various propagation mechanisms ( e.g. , reflections and diffractions ) not accounted for by the proposed models . [ cols="^,^,^,^,^,^,^ " , ] > 018inkscape -z -d file = fig6.svg export - pdf = fig6.pdf export - latex as far as map - unaware modelling is concerned , in appendix [ app : nomap_fitting ] a detailed description of the procedure for extracting data fits in this case is illustrated . based on the results provided by this procedure , the _ map - unaware likelihood _ model for the toa measurements @xmath126 and the _ map - unaware likelihood _ model for rss measurements @xmath127 have been proposed . estimates of all the parameters appearing in the likelihood models ( [ eq : likelihood_with_maps_toa_final])-([eq : likelihood_no_maps_rss_final_gaussian ] ) are listed in table [ tab : statistical_model_summary ] . note that : a ) the evaluation of the map - aware likelihoods ( [ eq : likelihood_with_maps_toa_final])-([eq : likelihood_with_maps_rss_final ] ) is more computationally demanding than that of their map - unaware counterparts ( [ eq : likelihood_no_maps_toa_final_exp])-([eq : likelihood_no_maps_rss_final_gaussian ] ) ) ; b ) the significant differences between map - aware and map - unaware models can be evidenced comparing a plot of ( [ eq : likelihood_with_maps_rss_final ] ) with that of ( [ eq : likelihood_no_maps_rss_final_gaussian ] ) referring to the same scenario ( see fig . [ fig : example_log_likelihood_mapaware ] and [ fig : example_log_likelihood_nomap ] , respectively ) . these results show that the presence of obstructions ( and , in particular , of walls ) introduces jump discontinuities in the map - aware likelihood function ; on the contrary , such discontinuities are not visible in the representation of its map - unaware counterpart . ^{t})$ ] ( [ eq : likelihood_with_maps_rss_final ] ) evaluated for the specific map shown in fig . [ fig : map_dii ] ( the parameter values listed in table [ tab : statistical_model_summary ] for rss radios have been employed ) and for the specific observation vector @xmath128 corresponding to the measurement site # 6 ( identified by a white plus sign ) , which represents the real agent position . the 5 measurement sites selected as anchors are identified by black triangles and the peak of the log - likelihood is evidenced by a green cross ( and is @xmath129 away from @xmath18).[fig : example_log_likelihood_mapaware],width=264 ] ^{t})$ ] ( [ eq : likelihood_no_maps_rss_final_gaussian ] ) referring to the same scenario as fig . [ fig : example_log_likelihood_mapaware ] , with @xmath130 . the 5 measurement sites selected as anchors are identified by black triangles and the peak of the log - likelihood is evidenced by a green cross ( and is @xmath131 away from @xmath18 ) . [ fig : example_log_likelihood_nomap],width=264 ] finally , we would like to point out that , in our opinion , a universal statistical model for rss and toa data is unlikely to exist and that some adjustments may be required in the models ( [ eq : likelihood_with_maps_toa_final])-([eq : likelihood_no_maps_rss_final_gaussian ] ) if different radio devices are employed . however , we also believe that our modelling approach ( starting from ( [ eq : likelihood_with_maps])-([eq : likelihood_no_maps_exp ] ) and then involving the procedures illustrated in the appendices ) is quite general and is useful for any reader interested in modelling experimental data acquired in real world localization systems . in this section map - aware and map - unaware estimators are formulated and compared in terms of accuracy . moreover , a brief analysis of their computational complexity is provided . the map - aware models ( [ eq : likelihood_with_maps_toa_final ] ) and ( [ eq : likelihood_with_maps_rss_final ] ) derived in paragraph [ sub : final_toa_rss_models ] can be exploited to derive two types of optimal localization algorithms : the and the . however , our computer simulations have evidenced that these algorithms perform similarly ; for this reason in the following we focus on the only , since its mathematical structure is simpler than that of the . the @xmath132 of the agent position @xmath18 can be evaluated as @xmath133 @xcite . exploiting ( [ eq : map_pdf ] ) and the bayes rule , this estimate can be expressed as @xmath134 where @xmath135 is the trial agent position , @xmath136 is the search space ( [ eq : mapaware_anchor_coverage_region ] ) and the pdf @xmath137 is given by and ( [ eq : likelihood_with_maps_toa_final ] ) and ( [ eq : likelihood_with_maps_rss_final ] ) for toa - based and rss - based systems , respectively . the evaluation of @xmath138 requires solving a constrained problem since , thanks to our unified modelling , the likelihood @xmath137 appearing in ( [ eq : map ] ) is always expressed as the product of gaussian pdfs . however , finding the global solution to this optimization problem is not easy since its cost function is non - convex . in principle , to avoid local minima the and the techniques could be exploited ; however , refined observation models like ( [ eq : likelihood_with_maps_toa_final ] ) and ( [ eq : likelihood_with_maps_rss_final ] ) hinder their use . it is also important to note that the cost function appearing in ( [ eq : map ] ) is non - differentiable because of the discontinuous behaviour of the functions ( [ eq : bias_mean_toa ] ) , ( [ eq : bias_mean_rss ] ) , ( [ eq : bias_variance_toa ] ) and ( [ eq : bias_variance_rss ] ) . this prevents the _ exact _ use of optimization methods based on the gradient / hessian matrix of cost functions ( e.g. , steepest descent variants ) . for these reasons , in our computer simulations the matlab routine ` fmincon ` , employing sequential quadratic programming and a numerical _ approximation _ of the hessian matrix , has been employed to solve the problem ( [ eq : map ] ) . the optimal map - unaware estimator is the @xcite . the @xmath139 of @xmath18 can be expressed as @xcite @xmath140 where @xmath135 is the trial agent position , @xmath141 is the search space ( [ eq : nomap_anchor_coverage_region ] ) and the pdf @xmath142 is given by ( [ eq : likelihood_no_maps_toa_final_exp ] ) and ( [ eq : likelihood_no_maps_rss_final_gaussian ] ) for toa - based and rss - based systems , respectively . note that the cost function characterising the ( [ eq : ml ] ) is discontinuous and non - linear because of the assumption of a los / nlos detection in identifying @xmath68 and @xmath69 . in addition , in the toa case , the use of exponential pdfs for modelling nlos links ( see ( [ eq : likelihood_with_maps_toa_final ] ) ) entails that @xmath143 for a large portion of the domain @xmath141 and this complicates the search for global minima . a simple approximation that may be employed to mitigate this problem is to replace the function @xmath6 appearing in the right - hand side of ( [ eq : likelihood_no_maps_toa_final_exp ] ) with the pdf mixture @xmath144 , which exhibits a decreasing exponential behaviour for @xmath145 and a small gaussian tail for @xmath146 . moreover , in order to further mitigate the problem of local minima , in our work the search domain @xmath141 has been partitioned into @xmath147 sub - rectangles ( whose sides have a length not exceeding @xmath148 ) ; then , the matlab ` fmincon ` routine has been run @xmath147 times over each of the sub - rectangles and the final has been selected on the basis of a minimum - cost criterion . the accuracy provided by the estimators ( [ eq : map ] ) and ( [ eq : ml ] ) has been assessed by means of a simulation tool implementing the following procedure : 1 . a measurement site , identified by the index @xmath149 , is randomly selected in the set @xmath150 referring to the measurement sites shown in fig . [ fig : map_dii ] ( rss case ) or in fig . [ fig : map_deis ] ( toa case ) . the position of this site represents the true position @xmath18 of the agent , so that @xmath151 . 2 . for the selected agent position , in the validation set introduced in sec . [ sub : final_toa_rss_models ] , @xmath152 sets of measurements associated with @xmath152 distinct sites are available . each of these sites can be employed as a `` virtual '' anchor , and in each selected set ( containing @xmath153 or @xmath154 measurements ) one measurement is randomly selected and placed in the observation vector @xmath128 . to make our simulation results realistic , if @xmath155 , only the observations associated with 5 random sites are chosen . 3 . the estimates @xmath132 ( [ eq : map ] ) and @xmath139 ( [ eq : ml ] ) and the associated _ localization errors _ @xmath156 and @xmath157 are evaluated . at the end of the simulation such errors are processed to assess the performance . steps 1 - 3 are repeated until at least @xmath158 iterations have been carried out . some rmse performance results @xmath159 and @xmath160 ( referring to the mapbe ( [ eq : map ] ) and the mle ( [ eq : ml ] ) , respectively ) are shown in fig . [ fig : model_comparison_rss ] and [ fig : model_comparison_toa ] for rss and toa localization , respectively . these results evidence that : 1 . in the rss case the always outperforms its map - unaware counterpart ( on the average @xmath161 for @xmath162 and @xmath163 for @xmath164 ; in other words , map - aware modelling enhance localization accuracy by 70% for @xmath162 and by 110% for @xmath164 ) . this is due to the fact that the former technique mitigates nlos bias ( which represents a major source of error ) more accurately than the latter one . in fact , maps provide a significant help whenever the bias mean predicted by ( [ eq : bias_mean_rss ] ) is significantly different from @xmath165 ( which accounts for nlos propagation in the ) , i.e. whenever there are several obstructions between the agent and the anchors ( see fig . [ fig : rss_fitting_bias_mean ] ) . 2 . in the toa case map - aware and map - unaware estimators perform similarly and offer good accuracy ; this is due to the fact that the toa measurements stored in the database @xcite are not affected by large nlos biases ( since all such measurements refer to close sites , often experiencing los conditions ) , so that map awareness does not play a significant role in this case . note also that localization errors for both and are smaller than @xmath166 for all measurement sites except # 3 , # 8 and # 12 . this result can be explained referring to the specific case of site # 3 , for which it has been found that the large error is due to a combination of a ) an anchor placement characterized by an high _ geometric dilution of precision _ @xcite and b ) undetected direct paths in the ranging phase ( see @xcite ) resulting in highly biased measurements . it is also interesting to point out that the is slightly more accurate than the in the measurement sites # 3 , # 17 , # 19 and # 20 ; such results can be related to propagation effects not accounted for by our models ( e.g. , reflections and corridor waveguiding ) ; however , in these cases , the difference between and rmse is very small ( few centimetres ) . 3 . the performance gap between the rss - based and the is large ( @xmath167 on the average ) , while toa - based algorithms perform similarly ( @xmath168 on the average ) . this shows that maps play a significant role in improving localization accuracy when the quality of available measurements is poor , i.e. , when measurements are affected by significant nlos bias and noise ( this occurs in rss - based systems ) ; on the contrary , if the quality is good ( like in toa uwb systems ) , accuracy is not significantly enhanced by map awareness ( some theoretical results about this can be found in @xcite ) . map - aware fitting provides good results even if a small number of links are used in the training set ( i.e. , in the fitting phase ) ; in fact , in the rss case , only 32 links ( ~25% of the 116 available links ) composed the training set and the performance shown in fig . [ fig : model_comparison_rss ] have been obtained using the validation set . our numerical results also confirm some well - known results and , in particular , that : 1 . rss - based localization algorithms are substantially outperformed by their uwb toa - based counterparts ( on the average , @xmath169 against @xmath170 ) , thanks to the superior ranging capabilities provided by uwb waveforms . 2 . significant variations in the rmse of both toa - based and rss - based estimators are found when different sites are considered ( similar results can be found in ( * ? ? ? 2 ) , although in the scenario we consider both and perform better than the and algorithms analysed in that reference ) . this is due , in the toa case , to errors in toa estimation originating from significant multipath @xcite ; in the rss case , instead , variations originate from the spatial selectivity experienced in indoor environments . finally , it is worth pointing out that some additional numerical results ( e.g. , referring to the robustness of the proposed statistical models to parametric inaccuracies ) have not been included here for space limitations , but are publicly available online ( see @xcite ) . in this paragraph a brief analysis of the computational complexity of the and is provided . to begin , we note that the complexity associated with the evaluation of @xmath171 is @xmath172 if the shape of map obstructions can be approximated by @xmath173 segments , since @xmath173 line intersection tests are involved in such a case . therefore , the complexity associated with the evaluation of the map - aware likelihood @xmath137 ( see ( [ eq : likelihood_with_maps_toa_final])-([eq : likelihood_with_maps_rss_final ] ) ) is approximately @xmath174 , where @xmath175 is the average number of segment intersection tests accomplished in processing each of the @xmath35 observations is non - linearly related to the shape of obstructions , to the trial agent position and to the anchor positions . ] . then , it can be inferred that the overall computational complexity of the ( [ eq : map ] ) is @xmath176 , where @xmath177 denotes the overall number of times the likelihood function ( [ eq : likelihood_with_maps_toa_final ] ) or ( [ eq : likelihood_with_maps_rss_final ] ) is computed . unluckily , the parameter @xmath177 can not be easily related to the other parameters of the proposed strategy ( [ eq : map ] ) , since it is non - linearly related to the shape of @xmath137 ( which is non - convex and non - differentiable ) and depends strongly on the exact type of optimization method employed ( see paragraph [ sub : mapbe ] ) . as far as the map - unaware ( [ eq : ml ] ) is concerned , its overall complexity can be expressed as @xmath178 since the complexity associated with the evaluation of the likelihood @xmath142 ( [ eq : likelihood_no_maps_toa_final_exp])-([eq : likelihood_no_maps_rss_final_gaussian ] ) is @xmath179 ; note that @xmath180 denotes the overall number of times the likelihood function ( [ eq : likelihood_no_maps_toa_final_exp ] ) or ( [ eq : likelihood_no_maps_rss_final_gaussian ] ) is evaluated and , just like @xmath177 , it can not be easily related to the other parameters of the . the constraint @xmath84 appearing in ( [ eq : ml ] ) makes even more difficult obtaining a closed expression for @xmath180 ; however , it can not be neglected , specially on large maps , because : a ) the map - unaware likelihood functions ( [ eq : likelihood_no_maps_toa_final_exp])-([eq : likelihood_no_maps_rss_final_gaussian ] ) are not defined for @xmath181 ; b ) even if the likelihood domains are extended to @xmath182 , minima different from the global one are substantially less likely to be contained in the domain @xmath141 than in the whole @xmath182 space . for these reasons , constrained optimization methods capable of handling the constraint @xmath84 must be selected ( eventually employing heuristic approximations ; see paragraph [ sub : mle ] ) . the considerations illustrated above led us to the conclusion that an accurate estimation of the computational complexity of both ( [ eq : map ] ) and ( [ eq : ml ] ) requires a mixed analytical / simulative approach ; for more details , the reader is referred to @xcite , which provides an in - depth computational complexity analysis together with reduced - complexity algorithms . in this manuscript novel statistical models for map - aware indoor localization have been developed . such models are based on experimental evidence and , in particular , rely on toa uwb and low - frequency narrowband rss measurements acquired in experimental campaigns . these measurements have been exploited to a ) develop suitable parametric models for nlos bias and noise , b ) estimate the values of the parameters appearing in such models and c ) assess the accuracy of map - aware localization techniques based on the proposed models . our results have evidenced that map - aware localization techniques may significantly outperform their optimal map - unaware counterparts , even when state - of - the - art map - unaware algorithms based on los / nlos detection and mitigation are considered . finally , we would like to point out that : 1 . the experimental campaign needed to estimate the parameters of the proposed model can be accomplished much more easily than those commonly required by fingerprinting methods . the likelihoods ( [ eq : likelihood_with_maps_toa_final ] ) and ( [ eq : likelihood_with_maps_rss_final ] ) and the parameters listed in table [ tab : statistical_model_summary ] can be exploited by localization algorithms operating in environments different from the ones considered here , provided that similar conditions ( in terms of bandwidth of localization signals , wall thickness and composition , etc . ) are experienced . 3 . even if the case of a _ static _ agent has been taken into consideration in this manuscript , the application of the developed models to navigation systems ( where a _ mobile _ agent is tracked ) is straightforward ; for instance , ( [ eq : likelihood_with_maps ] ) can be adopted as a _ measurement model _ for tracking filters ( e.g. , kalman or particle filters ) . in this paragraph the two steps of the fitting procedure ( see fig . [ fig : fitting ] and paragraph [ sub : final_toa_rss_models ] ) leading to the models ( [ eq : likelihood_with_maps_toa_final ] ) and ( [ eq : likelihood_with_maps_rss_final ] ) and to the parameter values listed in table [ tab : statistical_model_summary ] are described . _ data modelling _ - let @xmath183 denote the link between the @xmath101-th and the @xmath29-th measurement sites , where @xmath184 denotes the the set of the couples of indices identifying each available link . we assume that the residual rvs @xmath185 follow a gaussian distribution ( see the proposed data model ( [ eq : likelihood_with_maps ] ) ) @xmath186 the validity of the gaussian assumption ( [ eq : pdf_single_measurement ] ) has been analysed resorting to the anderson - darling normality test @xcite , which relies on the computation of a statistic , commonly denoted @xmath187 , and on its comparison with a fixed threshold @xmath188 ( if a significance level equal to @xmath189 is selected ) . we computed the @xmath187 statistic for the sets of measurements @xmath190 referring to the rss radios and associated with the links crossing @xmath191 , @xmath192 , @xmath193 , @xmath194 , 5 , 6 walls and for these sets it was found that @xmath187 is equal to @xmath195 , @xmath196 , @xmath197 , @xmath198 , @xmath199 and @xmath200 , respectively ; therefore , the gaussian assumption ( [ eq : pdf_single_measurement ] ) can be deemed valid in all the considered cases . 1 . statistical independence among different links is assumed . the rss bias mean @xmath202 and rss / toa / tdoa bias variance @xmath203 are assumed to depend on the number of obstructions @xmath204 associated with the link @xmath205 , so that @xmath206 and @xmath207 can be denoted @xmath208 and @xmath209 , respectively . note that @xmath210 is known at this stage . 3 . the noise term @xmath211 is assumed to depend on the distance @xmath212 ; such a dependence can be assessed estimating the function values @xmath213 associated with @xmath214 `` distance bins '' ; here @xmath215 represents the distance bin size ( in meters ) and @xmath32 is the maximum connectivity distance for the @xmath29-th anchor ( see sec . [ sub : connectivity_model ] ) . then , a staircase approximation for each of the continuous functions @xmath216 can be adopted ; the values associated with the steps are represented by the set @xmath217 , where @xmath218 is the noise standard deviation referring to the distance values that belong to the distance bin @xmath219 $ ] . for this reason , @xmath220 can be denoted @xmath221 , where @xmath222 is the index of the distance bin associated with the distance @xmath223 of the link @xmath205 and is known at this stage . in fact , if the assumptions listed above hold , the joint pdf of @xmath224 ( [ eq : pdf_single_measurement ] ) can be expressed as @xmath225 _ bias estimation _ - estimates @xmath226 of the parameters @xmath227 of ( [ eq : pdf_all_measurements ] ) can now be evaluated . this requires processing all the measurements @xmath228 such that @xmath229 , i.e. , the set @xmath230 , with @xmath231 , since this set represents a sufficient statistic for the estimation of @xmath232 using the neyman - fisher theorem ( * ? ? ? * sec . 5.4 ) and assuming that @xmath232 is a parameter independent from @xmath233 . this assumption is indeed valid at this stage since only after evaluating regression fits the random quantities @xmath234 are `` linked '' together . ] . the for , given the above mentioned set , is ( see ( [ eq : pdf_all_measurements ] ) ) @xmath235 for @xmath236 , where @xmath237 is the maximum number of obstructions affecting the links considered in our measurement campaign ; note that the right - hand side of ( [ eq : bias_mean_estimator ] ) represents the sample average of the residuals @xmath190 . similarly , the for is given by ( see ( [ eq : pdf_all_measurements ] ) ) @xmath238 for @xmath236 . unluckily , solving the last problem requires the knowledge of models for @xmath239 and for @xmath240 . to circumvent this , we assume that the bias mean @xmath239 is well approximated by its estimate @xmath241 and that @xmath242 , so that ( [ eq : bias_var_estimator1 ] ) can be simplified as @xmath243 the last result shows that is given by a ( biased ) variance of the residuals @xmath190 . _ noise estimation _ - the estimation of the noise standard deviation @xmath221 is based on the set @xmath244 with @xmath245 , since this set represents a sufficient statistic . the is ( see ( [ eq : pdf_all_measurements ] ) ) @xmath246 for @xmath247 . like in the previous case , we assume that the bias mean @xmath208 and variance @xmath248 are well approximated by their estimates @xmath249 and @xmath250 , so that ( [ eq : noise_var_estimator1 ] ) turns into @xmath251 which , unluckily , can not be put in a closed form . for this reason , the last problem has been solved numerically resorting to the matlab ` fmincon ` routine . in the second step of the procedure illustrated in fig . [ fig : fitting ] , the ml estimates acquired in the first step are processed to extract two regression fits , one for the bias , the other one for the noise model . _ bias model fit _ - as already discussed in paragraph [ sub : observation_model ] , in the toa / tdoa case a simple _ linear _ regression procedure can be employed for extracting the average parameters @xmath64 and @xmath65 appearing in @xmath252 ( [ eq : bias_mean_toa ] ) from the ml estimates @xmath253 . in the rss case , instead , the simple linear regression function , based on experimental evidence ( see fig . [ fig : rss_fitting_bias_mean ] ) , @xmath254 is proposed ; here @xmath255 and @xmath256 are expressed in meters and the factor @xmath124 forces @xmath78 ( [ eq : bias_mean_rss ] ) to zero when @xmath257 . note that the selected model differs from the af bias model , characterized by a logarithmic dependence on @xmath125 . as far as the _ _ standard deviation @xmath258 of the toa bias is concerned , the power - like model @xmath259 is proposed ; here @xmath260 is a dimensionless parameter whereas @xmath261 is expressed in meters . for rss systems , instead , our experimental results have evidenced that the bias variance is mostly constant or even decreases as the number of obstructions @xmath125 gets larger and that the linear model @xmath262 can be adopted , where both the parameters @xmath263 and @xmath264 are expressed in meters . _ noise model fit _ - in various technical papers ( e.g. , see @xcite ) the additive noise term @xmath265 is assumed to have the same variance for all los ( nlos ) links , regardless of other factors , like the link distance . this noise model accounts for the inaccuracy of the algorithm adopted in extrapolating the observations @xmath266 in nlos conditions , but does not account for the dependence of the performance of such an algorithm on the . in this work , based on experimental evidence , we have decided to employ , for both toa , tdoa and rss systems , the model adopted in @xcite and expressed by @xmath267 where @xmath268 is the noise variance characterizing the transmitter - receiver distance @xmath112 and @xmath269 is the noise path loss exponent . finally , substituting ( [ eq : bias_mean_toa ] ) , ( [ eq : bias_variance_toa ] ) and ( [ eq : noise_model ] ) ( or ( [ eq : bias_mean_rss ] ) , ( [ eq : bias_variance_rss ] ) and ( [ eq : noise_model ] ) ) in ( [ eq : likelihood_with_maps ] ) produces ( [ eq : likelihood_with_maps_toa_final ] ) ( or ( [ eq : likelihood_with_maps_rss_final ] ) ) . _ data modelling _ - on the basis of ( [ eq : likelihood_no_maps_gaussian ] ) , each of the residuals @xmath190 follows a gaussian distribution , i.e. , @xmath270 or a mixed gaussian - exponential distribution ( [ eq : likelihood_no_maps_exp ] ) , i.e. , @xmath271 where @xmath70 and @xmath71 are known in this fitting phase . _ bias model fit _ * - * on the basis of experimental data the model ( [ eq : pdf_single_measurement_nomap_gaussian ] ) has been adopted for modelling the rss measurements acquired by means of a couple of radio devices operating at @xmath90 . the model ( [ eq : pdf_single_measurement_nomap_exp ] ) , instead , has been employed to describe the toa data collected in the newcom++ database . the of the parameters @xmath272 and @xmath273 appearing in the model ( [ eq : pdf_single_measurement_nomap_gaussian ] ) ( see sec . [ sec : statistical_model ] ) are easily shown ( using a procedure similar to that adopted in the bias model fitting for the map - aware case ) to coincide with the sample average and the sample variance , respectively , of the residuals @xmath274 , where @xmath275 . * * similarly , if the model ( [ eq : pdf_single_measurement_nomap_exp ] ) is used , the of @xmath276 coincides with the sample average of the residuals @xmath274 . _ noise model fit _ - following the approach adopted in the map - aware case , the noise is assumed to depend on the link distance and the ml estimate @xmath277 of the noise standard deviation is evaluated for those links identified by a couple of indices @xmath205 such that @xmath278 . unluckily , no closed - form solution is available for this estimate , which can be expressed as ( note the similarity with ( [ eq : noise_var_estimator_final ] ) ) @xmath279 for @xmath247 . it is also important to mention that @xmath280 and @xmath281 are defined as : a ) @xmath282 and @xmath283 , respectively , if the nlos bias is modelled as a gaussian rv ( see ( [ eq : likelihood_no_maps_gaussian ] ) and ( [ eq : pdf_single_measurement_nomap_gaussian ] ) ) ; b ) @xmath284 and @xmath285 , respectively , if the nlos bias is modelled as an exponential random variable ( see ( [ eq : likelihood_no_maps_exp ] ) and ( [ eq : pdf_single_measurement_nomap_exp ] ) ) . in both cases we have found that the quantities @xmath286 do exhibit a monotonic dependence on the distance ( due to all propagation mechanisms not accounted for by a map - unaware model ) ; despite this , the model ( [ eq : noise_model ] ) has been selected as a regression function for @xmath287 . finally , substituting ( [ eq : noise_model ] ) in ( [ eq : likelihood_no_maps_gaussian ] ) produces the likelihood function ( [ eq : likelihood_no_maps_rss_final_gaussian ] ) proposed for map - unaware rss systems . similarly , _ _ substituting ( [ eq : noise_model ] ) in ( [ eq : likelihood_no_maps_exp ] ) yields the likelihood function ( [ eq : likelihood_no_maps_toa_final_exp ] ) referring to map - unaware toa systems . n. patwari , j. ash , s. kyperountas , a. o. hero , iii , r. moses , and n. correal , `` locating the nodes : cooperative localization in wireless sensor networks , '' _ ieee signal processing mag . _ , vol . 22 , no . 4 , pp . 5469 , jul . 2005 . s. venkatesh and r. m. buehrer , `` non - line - of - sight identification in ultra - wideband systems based on received signal statistics , '' _ iet microw . antennas propag . _ , vol . 1 , no . 6 , pp . 1120 1130 , dec . 2007 . s. mazuelas , a. bahillo , r. m. lorenzo , p. fernandez , f. a. lago , e. garcia , j. blas , and e. j. abril , `` robust indoor positioning provided by real - time rssi values in unmodified wlan networks , '' _ ieee j. of sel . topics in signal process . _ , vol . 3 , no . 5 , 821831 , oct . 2009 . f. montorsi , f. pancaldi , and g. m. vitetta , `` statistical characterization and mitigation of nlos bias in uwb localization systems , '' _ advances in electronics and telecommunications _ , vol . 2 , no . 4 , pp . 1117 , dec . 2011 . c. figuera , i. mora - jimenez , a. guerrero - curieses , j. l. rojo - alvarez , e. everss , m. wilby , and j. ramos - lopez , `` nonparametric model comparison and uncertainty evaluation for signal strength indoor location , '' _ ieee trans . on mobile computing _ , vol . 8 , no . 9 , pp . 12501264 , sep . m. bshara , u. orguner , f. gustafsson , and l. v. biesen , `` fingerprinting localization in wireless networks based on received - signal - strength measurements : a case study on wimax networks , '' _ ieee trans . _ , vol . 59 , no . 1 , 283294 , jan . 2010 . p. closas , j. arribas , and c. fern , `` testing for normality of uwb - based distance measurements by the anderson - darling statistic , '' in _ future network and mobile summit conf . _ , florence , 2010 , pp . 18 . m. r. gholami , e. g. strom , f. sottile , d. dardari , a. conti , s. gezici , m. rydstrom , and m. a. spirito , `` static positioning using uwb range measurements , '' in _ future network and mobile summit conf . _ , florence , 2010 , pp . 110 . f. montorsi , s. mazuelas , f. pancaldi , g. m. vitetta , and m. z. win , `` on the impact of a priori information on localization accuracy and complexity , '' in _ ieee int . on commun . _ , budapest , 2013 , pp . 57925797 . francesco montorsi ( s06 ) received both the laurea degree ( _ cum laude _ ) and the laurea specialistica degree ( _ cum laude _ ) in electronic engineering from the university of modena and reggio emilia , italy , in 2007 and 2009 , respectively . he received the ph.d degree in information and communications technologies ( ict ) from the university of modena and reggio emilia in 2013 . he is employed as embedded system engineer for an ict company since 2013 . in 2011 he was a visiting phd student at the wireless communications and network science laboratory of massachusetts institute of technology ( mit ) . his research interests are in the area of localization and navigation systems , with emphasis on statistical signal processing ( linear and non - linear filtering , detection and estimation problems ) , model - based design and model - based performance assessment . dr . montorsi is a member of ieee communications society and served as a reviewer for the ieee transactions on wireless communications , ieee transactions on signal processing , ieee wireless communications letters and several ieee conferences . he received the gtti award for phd theses in the field of communication technologies in 2013 from the italian telecommunications and information theory group ( gtti ) . fabrizio pancaldi was born in modena , italy , in july 1978 . he received the dr . degree in electronic engineering ( cum laude ) and the ph . d. degree in 2006 , both from the university of modena and reggio emilia , italy . from march 2006 he is holding the position of assistant professor at the same university and he gives the courses of telecommunication networks and ict systems . he works in the field of digital communications , both radio and powerline . his particular interests lie in the wide area of digital communications , with emphasis on channel equalization , statistical channel modelling , space - time coding , radio localization , channel estimation and clock synchronization . giorgio m. vitetta ( s89-m91-sm99 ) was born in reggio calabria , italy , in april 1966 . he received the dr . ing . degree in electronic engineering ( _ cum laude _ ) in 1990 , and the ph.d . degree in 1994 , both from the university of pisa , pisa , italy . from 1995 to 1998 , he was a research fellow with the department of information engineering , university of pisa . from 1998 to 2001 , he was an associate professor with the university of modena and reggio emilia , modena , italy , where he is currently a full professor . his main research interests lie in the broad area of communication theory , with particular emphasis on detection / equalization / synchronization algorithms for wireless communications , statistical modelling of wireless and powerline channels , ultrawideband communication techniques and applications of game theory to wireless communications .

we study the limited data problem of the spherical radon transform in two and three dimensional spaces with general acquisition surfaces . in such situations , it is known that the application of filtered - backprojection reconstruction formulas might generate added artifacts and degrade the quality of reconstructions . in this article , we explicitly analyze a family of such inversion formulas , depending on a smoothing function that vanishes to order on the boundary of the acquisition surfaces . we show that the artifacts are orders smoother than their generating singularity . moreover , in two dimensional space , if the generating singularity is conormal satisfying a generic condition then the artifacts are even orders smoother than the generating singularity . our analysis for three dimensional space contains an important idea of lifting up a space . we also explore the theoretical findings in a series of numerical experiments . our experiments show that a good choice of the smoothing function might lead to a significant improvement of reconstruction quality .

1507.00082

let @xmath2 be a convex closed smooth hyper - surface . we consider the following spherical radon transform @xmath3 of a function @xmath4 defined in @xmath5 @xmath6 here , @xmath7 is the sphere centered at @xmath8 of radius @xmath9 , and @xmath10 is its surface measure . this transform appears in several imaging modalities , such as thermo / photoacoustic tomography ( e.g. , @xcite ) , ultrasound imaging ( e.g. , @xcite ) , sonar ( e.g. , @xcite ) , and inverse elasticity ( e.g. , @xcite ) . for example , in thermo / photoacoustic tomography ( tat / pat ) , @xmath4 is the initial ultrasound pressure generated by the thermo / photo - elastic effect . it contains useful information about the inner structure of the tissue , which can be used , e.g. , for cancer detection . on the other hand , the knowledge of @xmath11 can be extracted from the ultrasound signals collected by a transducer located at @xmath12 , which is called the * observation * surface . one , therefore , can concentrate on finding @xmath4 given @xmath13 . the same problem also arises in other aforementioned image modalities . it is commonly assumed that @xmath4 is supported inside the bounded domain @xmath14 whose boundary is @xmath15 . let us discuss an inversion formula under this assumption . let @xmath16 be the pseudo - differential operator defined by @xmath17 and @xmath18 be the back - projection type operator @xmath19 when @xmath15 is an @xmath20-dimensional ellipsoid , one has the following inversion formula @xcite @xmath21 we note here that formula ( [ e : inversion ] ) was written in other forms in the above references . the above form , presented in @xcite , is convenient to analyze from the microlocal point of view . another advantage of the above form is that it can be implemented straight forwardly : 1 ) @xmath22 can be computed fast by using fast fourier transform ( fft ) and 2 ) @xmath23 only involves a simple integration on @xmath15 . [ fig : full - circle ] is the result of our implementation when @xmath24 and @xmath15 is the unit circle . the image size is @xmath25 pixels . the sampling data has the resolution of @xmath26 for the spatial ( angular ) variable @xmath27 and radial variable @xmath28 $ ] . the reconstruction is almost perfect . when @xmath15 is a general convex surface , the operator on the right hand side of ( [ e : inversion ] ) might not be the identity . however , it only differs from the identity by a compact operator ( see @xcite ) , which is a pseudo - differential operator of order @xmath29 ( as shown in @xcite ) . moreover , from the numerical experiments , we observe that @xmath30 and @xmath4 are very close , even when @xmath15 is not any ellipse / ellipsoid . for example , in fig . [ fig : full - polar ] we present the reconstruction using @xmath31 for @xmath15 being the polar curve defined by @xmath32 the reconstruction is almost perfect . in this article , we are interested in the limited data problem . that is , the knowledge of @xmath3 is only available on a closed proper subset @xmath33 with smooth boundary @xmath34 and nontrivial interior @xmath35 ( see , e.g. , @xcite ) . it is natural to consider the following formula @xmath36 where @xmath37 is the characteristic function of @xmath38 . as we observe from fig . [ fig : limited - cir ] , formula ( [ no_smoothing ] ) reconstructs some singularities and also smoothens out some singularities of the original image . moreover , it also creates some artifacts ( added singularities ) into the picture . more detailed discussion will be presented in sections [ s:2d ] and [ s:3d ] ( see also section [ s : num ] for numerical demonstrations ) . the operator @xmath39 reconstructs all the visible singularities . however , the artifacts it generates are quite strong . we now introduce a generalization of @xmath39 in order to reduce the artifacts . let us consider @xmath40 here , @xmath41 and @xmath42 on @xmath43 . moreover , we assume further that @xmath44 on the interior @xmath35 of @xmath38 and @xmath45 vanishes to order @xmath0 on the boundary @xmath34 . of course , @xmath46 if @xmath47 . in this article , we will study formula ( [ e : mt ] ) . namely , we will analyze which singularities are reconstructed and how strong the reconstructed singularities are , compared to the original ones . moreover , we will describe how the artifacts are generated by ( [ e : mt ] ) and how strong they are . let us discuss here the main idea of our analysis . we first follow the approach in @xcite to represent @xmath48 as an oscillatory integral . to that end , let @xmath49 be the schwartz kernel of @xmath48 . then , it can be written as ( see @xcite ) @xmath50 by the simple change of variables @xmath51 , @xmath52 here , @xmath53 are the intersections of @xmath15 with the positive and negative rays @xmath54 to illuminate the idea , let us at the moment assume that @xmath45 vanishes to * infinite * order at the boundary of @xmath55 , then the functions @xmath56 are ( smooth ) symbols of order zero . then , due to ( * ? ? ? * theorem 3.2.1 ) , @xmath48 is a pseudo - differential operator ( @xmath57do ) of order zero whose principal symbol is ( see @xcite , and also @xcite , for the same result in a more general framework ) : @xmath58.\ ] ] one important consequence of @xmath48 being a @xmath57do is that it does not generate the artifacts into the picture . moreover , let us denote @xmath59 , the line passing through @xmath60 along direction @xmath61 . then , the following discussions hold . * let us denote @xmath62 then , any singularities @xmath63 of @xmath4 in this zone generates a corresponding singularity of on the observed data @xmath64 . it is , therefore , called the * visible zone * ( see , e.g. , @xcite ) . let @xmath65 , then either @xmath66 or @xmath67 . therefore , @xmath68 . that is , @xmath48 is an elliptic pseudo - differential operator of order @xmath69 near @xmath63 . due to the standard theory of pseudo - differential operators ( see lemma [ l : pet ] ) , @xmath70 if and only if @xmath71 . that is , all the visible singularities of @xmath4 are reconstructed by @xmath48 with the correct order . we notice that one visible singularity may be either visible in two directions , when both @xmath72 and @xmath73 belong to @xmath35 , or in one direction , when only one of @xmath74 or @xmath73 belongs to @xmath38 . we will say that they are * doubly * visible and * singly * visible , respectively . * on the other hand , let us denote @xmath75 then , @xmath76 is called * invisible zone * , since any singularity of @xmath4 in this zone does not generate any singularity of @xmath77 . a singularity of @xmath4 in this zone is called * invisible*. we notice that for each @xmath78 , then @xmath79 . that is , the full symbol of @xmath49 is zero near @xmath63 . therefore , due to the standard theory of pseudo - differential operators , @xmath80 , even if @xmath81 . that is , all the invisible singularities are completely smoothen out by @xmath48 . let us now consider the case of our interest : @xmath45 only vanishes to a * finite * order @xmath0 on the boundary of @xmath82 ( @xmath0 can be zero as in case of @xmath39 ) . then , @xmath48 is not a @xmath57do anymore , since @xmath83 are not smooth . therefore , @xmath48 may generate artifacts into the picture , as shown fig . [ fig : limited - cir ] where @xmath46 . our results ( see sections [ s:2d ] and [ s:3d ] ) , show that the effect of @xmath48 on the zones @xmath84 and @xmath76 is exactly what we discuss above for the case of infinitely smooth @xmath45 . this can be seen from the facts that @xmath85 are smooth on these two zones . the artifacts , on the other hand , come from the * boundary zone * where at least one of @xmath86 or @xmath87 is non - smooth : @xmath88 we will characterize how these artifacts are generated and how strong they are . to that end , we will make use of the formula ( [ e : mu - ori ] ) . compared to ( [ e : mu ] ) , the formula ( [ e : mu - ori ] ) still keeps track of the geometric information of @xmath89 , which is helpful to understand the generation artifacts . moreover , we will study ( [ e : mu - ori ] ) as a fourier integral operator , whose order determines the strength of artifacts ( compared to the original singularity generating them ) . in term of geometric description of the artifacts , we will take advantage of the technique developed in @xcite ( see also @xcite ) . in @xcite , the third author studied the strength of the artifacts in @xmath15 is a hyperplane ( that is , @xmath38 is flat ) . in this article , we study the problem for any convex smooth surface @xmath15 . we will restrict ourselves to the two and three dimensional problems ( i.e. , @xmath90 ) , since they are the most practical ones . we will follow the microlocal analysis technique in @xcite . however , due to the generality of the geometry involved , the arguments are more sophisticated . moreover , for the three dimensional problem , we introduce a new idea of lifting up the space , which is simple but interesting . it is worth mentioning that similar problem has been studied for the x - ray ( or classical radon ) transform @xcite . the article is organized as follows . in section [ s:2d ] , we state and prove the main results for the two dimensional problem . in section [ s:3d ] , we state and prove the results for three dimensional problem . we then present some numerical experiments for the two dimensional problem in section [ s : num ] . finally , we recall some essential background in microlocally analysis in the appendix . let us consider @xmath24 . we assume that @xmath15 is a smooth convex curve , and @xmath38 is a connected piece of @xmath15 . for our convenience , we assume that @xmath38 is arc - length parametrized by the smooth function @xmath91 \to { \mathbb{r}}^2,\ ] ] and its end points are @xmath92 and @xmath93 . let us define the function @xmath94 \to { \mathbb{r}}$ ] by @xmath95 then , @xmath96 vanishes to order @xmath0 at @xmath97 and @xmath98 , and @xmath99 for @xmath100 . we define the following canonical relations in @xmath101 @xmath102 and @xmath103 @xmath104 we notice that @xmath105 iff @xmath106 is in the boundary zone , corresponding to @xmath107 , and @xmath63 is obtained from @xmath106 by rotating around the corresponding boundary point @xmath107 . similar description holds for @xmath108 . [ p : wavefront ] we have @xmath109 here , @xmath110 is the twisted wave front set of @xmath49 : @xmath111 proposition [ p : wavefront ] was proved in @xcite when @xmath38 is a line segment . the proof carries naturally to the general curve @xmath38 without any major changes . we present it here for the sake of completeness and convenience in later discussion . due to the composition rule for wave front sets ( see theorem [ t : compose ] in appendix ) , we obtain is a pseudo - differential operator , it does not generate new wave front set elements . this well - known pseudo - locality property of a pseudo - differential operator is recalled in appendix [ a : pdo ] , see ( [ e : wave - front - pdo ] ) . ] @xmath112 where @xmath113 and @xmath114 are the schwartz kernel of @xmath23 and @xmath89 , respectively . let us now proceed to analyze the right hand side of the above inclusion . we note that @xmath89 is an fio with the phase function ( see , e.g. , @xcite ) @xmath115 for the sake of simplicity , for an each @xmath116 , we use the notation @xmath117 for @xmath118 . due to theorem [ t : wave - fio ] ( see appendix ) , we obtain , by letting @xmath119 @xmath120 also considering @xmath121 as a function of @xmath122 , we have @xmath123 applying the product rule for wave front sets ( see theorem [ t : product ] in appendix ) , we obtain @xmath124 where @xmath125 { \mathcal{c}}_a & = & \{\big({{\bf a}},r , \ , \tau \ , \left < x-{{\bf a } } , z'(a ) \right>+ \tau ' , - \tau \ , r ; \ , x,\tau \ , ( x-{{\bf a } } ) ) : |x-{{\bf a}}|=r,\,\tau ' \neq 0\ } , \\[4 pt ] { \mathcal{c}}_b & = & \{\big({{\bf b}},r , \ , \tau \ , \left < x-{{\bf b } } , z'(b ) \right>+ \tau ' , - \tau \ , r ; \ , x,\tau \ , ( x-{{\bf b } } ) ) : |x-{{\bf b}}| = r , \ , \tau ' \neq 0\}.\end{aligned}\ ] ] on the other hand , we notice that @xmath23 is a fio with the same phase function @xmath126 as @xmath89 ( but the order of variables is switched ) , see e.g. , @xcite . therefore , @xmath127 where @xmath128 is the transpose relation of @xmath129 @xmath130 from ( [ e : wfmu ] ) , we arrive to @xmath131 we notice that @xmath132 therefore , @xmath133 [ r : wave ] assume that @xmath134 in a neighborhood of @xmath135 and @xmath136 such that @xmath137 . then , due to ( [ e : wf - pro ] ) , @xmath138 where @xmath139 is the right projection operator . from ( [ e : wfmu ] ) , we obtain : @xmath140 this observation will be used later in the proof of theorem [ t : main1 ] a ) . let us now employ proposition [ p : wavefront ] to describe the geometric effects of @xmath48 on the wave front set of @xmath4 ( see also the discussion in @xcite ) . we first keep in mind the following inclusion , coming from theorem [ t : cal - wave ] : @xmath141 therefore , due to proposition [ p : wavefront ] , @xmath142 \cup \big[{\lambda}_{{\bf a}}\circ { \mbox{wf}}(f)\big ] \cup \big[{\lambda}_{{\bf b}}\circ { \mbox{wf}}(f ) \big].\ ] ] the first part on the right hand side contains all the singularities that may be possibly reconstructed by @xmath48 . the other two contain all the possible artifacts generated by @xmath48 . we now discuss the implications of ( [ e : poss ] ) in more details . * let @xmath81 be an invisible singularity . we observe that @xmath143 therefore , from inclusion ( [ e : poss ] ) , @xmath63 is not reconstructed and does not generate any artifacts . * let @xmath81 be a visible singularity . then , @xmath144 + from inclusion ( [ e : poss ] ) , @xmath63 may be reconstructed and does not generate any artifacts . * let @xmath81 be a boundary singularity pointing through @xmath107 , that is @xmath145 for some @xmath146 . then , @xmath147 from the inclusion ( [ e : poss ] ) , we observe that @xmath63 may generate artifacts @xmath106 by rotating around @xmath107 . conversely , assume that @xmath148 be an artifact . then , there is @xmath149 that generates @xmath106 . + similar description holds for a boundary singularity pointing through @xmath150 . the strength of the reconstructed singularities , described in b ) , will be obtained by analyzing the singularities of @xmath49 near @xmath151 . this will be done by the standard theory of pseudo - differential operators . in order to analyze the strength of the artifacts , described in c ) , we will make use of a class of fios associated to a point , introduced in section [ s : fio - point ] . here is our main result of this section : [ t : main1 ] we have * microlocally on @xmath151 , we have @xmath152 . moreover , its principal symbol is @xmath153,\ ] ] where @xmath154 is the intersection of the ray @xmath155 with @xmath15 . * we can write @xmath156 such that : * * @xmath157 . moreover , microlocally near @xmath158 , @xmath159 . * * @xmath160 . moreover , microlocally near @xmath161 , @xmath162 . will present the proof of theorem [ t : main1 ] in section [ s : proof-2d ] . we now describe some of its consequences . * let @xmath81 be a visible singularity . then , due to theorem [ t : main1 ] b ) , microlocally near @xmath163 , @xmath49 is a fourier distribution of order zero with positive principal symbol . applying lemma [ l : pet ] , we obtain @xmath70 if and only if @xmath71 . that is , all the visible singularities are reconstructed with the correct order . + moreover , the formula @xmath164 provides the magnitude of the main part of reconstructed singularities . for example , if @xmath63 is a jump singularity across a curve @xmath165 with the jump equal to @xmath166 , then @xmath63 is also a jump singularity across @xmath165 with the jump equal to @xmath164 . this explains the difference in the magnitude of the reconstructed singularities , that is demonstrated in section [ s : num ] . * now , assume that @xmath71 is an artifact pointing through @xmath107 . then , each of its generating singularities @xmath167 satisfies @xmath168 then , due to theorem [ t : main1 ] b ) and lemma [ l : ho ] , at least one generating singularity @xmath106 satisfies @xmath169 . that is , all the artifacts are at least @xmath0 order(s ) smoother then their strongest generating singularities . + if we assume further that @xmath63 has finitely many generating singularities @xmath106 , each of them is a conormal singularity of order @xmath9 along a curve whose order of contact with the circle @xmath170 is exactly @xmath166 . , since both curves are perpendicular to @xmath171 at @xmath172 . therefore , the condition on the contact order is quite generic . ] then , due to theorem [ t : main1 ] b ) and lemma [ l : spread ] , @xmath173 is a conormal singularity of order @xmath174 along the circle @xmath175 . that is , the artifacts are at least @xmath176 order(s ) smoother than the strongest generating singularity . in our numerical experiments in section [ s : num ] , we will demonstrate this fact . let us now discuss the proof of theorem [ t : main1 ] . it is similar to that of ( * ? ? ? * theorem 2.2 ) . however , we need to employ more sophisticated microlocal arguments due to the generality of the geometry involved . as in ( * ? ? ? * theorem 2.2 ) , the proofs for a ) and b ) require two different oscillatory integral representations for @xmath49 . the idea is similar to the case of infinitely smooth @xmath45 presented in @xcite ( see also @xcite for more general framework ) . the main point here is to microlocalize the argument to stay away from @xmath177 . let @xmath178 be such that @xmath179 . we need to prove that there is @xmath180 such that @xmath181 and the principall symbol of @xmath182 at @xmath183 equals @xmath184 $ ] , where @xmath185 . since @xmath179 , @xmath45 is smooth near both @xmath186 and @xmath187 . let us define a cut - off function @xmath188 such that @xmath189 near @xmath190 and zero elsewhere such that @xmath191 . let us write @xmath192 where @xmath193 let @xmath194 be the schwartz kernel of @xmath195 . since @xmath196 at @xmath190 , using remark [ r : wave ] , we obtain @xmath197 on the other hand , similarly to ( [ e : mu ] ) ( see also @xcite for the derivation ) , we obtain the formula for the schwartz kernel of @xmath198 @xmath199 we notice that @xmath200 is smooth and homogenous of degree @xmath69 in @xmath61 . therefore , @xmath201 moreover , the principal symbol of @xmath49 is ( see , e.g. , ( * ? ? ? * theorem 3.2.1 ) ) @xmath202 since @xmath203 at @xmath190 , we obtain @xmath204 this finishes the proof ( [ e : diff ] ) where @xmath205 . * proof of b ) . * let us decompose @xmath45 into the form @xmath206 where @xmath207 such that @xmath208 vanishes near @xmath150 and @xmath209 vanishes near @xmath107 . we then can write @xmath210 where @xmath211 repeating the argument in proposition [ p : wavefront ] ( see also remark [ r : wave ] ) , we obtain @xmath212 we now prove that microlocally near @xmath213 , @xmath159 . let @xmath214 . we need to prove that there is @xmath215 such that @xmath216 indeed , we notice that @xmath217 and @xmath218 belong to the same circle centered at @xmath107 . therefore , @xmath219 otherwise , the tangent line of @xmath38 at @xmath107 passes through the midpoint of the line segment connecting @xmath217 and @xmath218 . this would be a contradiction to the assumption that @xmath14 is convex and @xmath220 . thus , there is a neighborhood @xmath221 of @xmath222 and @xmath223 such that @xmath224 . \end{aligned}\ ] ] let @xmath225 such that @xmath226 near @xmath227 and @xmath228 for @xmath229 such that @xmath230 . let us write @xmath231 where @xmath232 we denote by @xmath233 and @xmath194 the schwartz kernels of @xmath198 and @xmath195 , respectively . an argument similar to the proof of proposition [ p : wavefront ] shows that @xmath234 therefore , @xmath235 similarly to ( [ e : mu - ori ] ) , we can write : @xmath236 the phase function of @xmath233 can be written as @xmath237 therefore , @xmath238 let us denote @xmath239 applying lemma [ l : sym ] , we obtain @xmath240 { \lambda}}\,|{\lambda}| \ , \left [ e^{- i\left < x - y,\ , 2 \,{{\bf a}}\right > { \lambda } } \,a(x , y,{\lambda } ) \right ] \ , d\,{\lambda}\\&= & \frac{1}{2 \ , \pi^2 } { \int\limits}_{{\mathbb{r } } } e^{i ( |x-{{\bf a}}|^2 - |y-{{\bf a}}|^2 ) { \lambda } } \,b(x , y,{\lambda})\ , d\,{\lambda},\end{aligned}\ ] ] where @xmath241 is a symbol of order @xmath242 on @xmath221 . we notice that @xmath243 is the canonical relation associated to the phase function of @xmath233 ( see appendix [ a : fios ] ) . therefore , @xmath244 let @xmath182 be obtained from @xmath233 by multiplying with a cut - off function near @xmath245 . we arrive to @xmath246 and @xmath247 we conclude that @xmath159 microlocally near @xmath248 . the proof for @xmath249 is similar . in the above proof , we have used the following result : [ l : sym ] let @xmath250 $ ] and define @xmath251 we have @xmath252 . moreover , * assume that @xmath253 near @xmath254 and @xmath255 vanishes to order @xmath0 at @xmath256 . then , on the set @xmath257 \mbox { such that } z(s ) \in { \operatorname{supp}}(h ) \},\ ] ] we have @xmath258^{(k + 1 ) } } \ , \big[1 + r(x , y,{\lambda})\big ] . \end{aligned}\ ] ] here , @xmath259 is the normal vector of @xmath15 at @xmath92 . * assume that @xmath253 near @xmath97 , and @xmath255 vanishes at @xmath260 to order @xmath0 . on the set @xmath261,~ z(s ) \in { \operatorname{supp}}(h ) \},\ ] ] we have @xmath262^{(k + 1 ) } } \ , \big[1 + r(x , y,{\lambda})\big].\end{aligned}\ ] ] here , @xmath263 is the normal vector of @xmath15 at @xmath93 . in both a ) and b ) , @xmath264 is a symbol of order at most @xmath29 . the lemma is proved by successive integration by parts . it is very similar to ( * ? ? ? * lemma 2.3 ) . we skip it here for the sake of brevity . let us now consider @xmath265 . we assume that @xmath266 is a convex domain with the smooth boundary @xmath15 . we assume that @xmath38 is a connected and simply connected subset of @xmath15 with the smooth boundary @xmath267 . we will analyze @xmath48 when @xmath45 vanishes to a finite order @xmath0 on @xmath267 . with a slight abuse of notation , we arc - length parametrize @xmath267 by the function @xmath268 ( with @xmath269 , where @xmath270 is the length of @xmath267 ) . similarly to the case @xmath24 , we define @xmath271 we denote by @xmath272 the following canonical relation in @xmath273 : @xmath274 we notice that @xmath275 if and only if @xmath63 and @xmath106 are boundary elements corresponding to a common boundary point @xmath276 and they are obtained from the other by a rotation around the tangent line of @xmath267 at @xmath8 . the following result gives us a geometric description of the singularities of the schwartz kernel @xmath49 of @xmath48 . [ p : wavefront-3 ] we have @xmath277 the proof of proposition [ p : wavefront-3 ] is similar to that of proposition [ p : wavefront ] ( see also ( * ? ? ? * proposition 3.1 ) and @xcite ) . we skip it for the sake of brevity . similarly to proposition [ p : wavefront ] we obtain the following implications of proposition [ p : wavefront-3 ] : * @xmath48 smoothens out all the visible singularities . * @xmath48 may reconstruct the visible singularities , and * @xmath48 may generate artifacts by rotating a boundary singularity around the tangent line of @xmath267 at the corresponding boundary point . the following result tells us the strength of the reconstructed singularities , explained in b ) , and artifacts , described in c ) : [ t : main2 ] the following statements hold : * microlocally on @xmath278 , we have @xmath152 with the principal symbol @xmath279,\ ] ] where @xmath154 is the intersection of the ray @xmath155 with @xmath15 . * microlocally on @xmath280 , we have @xmath281 . similarly to theorem [ t : main1 ] , we obtain the following implications of theorem [ t : main2 ] ( see also lemma [ l : ho3d ] for the discussion on artifacts ) : * if @xmath81 is a visible singularity , then @xmath63 is reconstructed with correct order . * the artifacts are at least @xmath0 order(s ) smoother than the strongest generating singularity . let us now proceed to prove theorem [ t : main2 ] . we will need the following two lemmas : [ l : f ] let @xmath282 be defined by @xmath283 where @xmath284 . then , @xmath285 . we first notice that the above formula @xmath286 does not directly define an fio , since the phase " function @xmath287 involves an extra variable @xmath288 , which is neither a variable of @xmath4 nor a phase variable . in @xcite , where @xmath267 is a line segment , the above result was proved by a change of variables . when @xmath267 is a general curve , such change of variable seems to be complicated . we , instead , introduce a simple idea of lifting up the space . for the notational ease , let us denote @xmath289 and @xmath290 . then , @xmath291 and @xmath292 are smooth manifolds of dimensions @xmath293 and @xmath294 , respectively . let us define the operator @xmath295 by @xmath296 then , @xmath297 can be written in the following form @xmath298 that is , @xmath297 is an fio of order ( see ( [ e : ord ] ) in appendix [ a : fios ] ) is the dimension of the phase variable @xmath61 . ] @xmath299 with the canonical relation @xmath300 let us define @xmath301 by the formula @xmath302 then , @xmath303 is an fio of order ( see ( [ e : ord ] ) in appendix [ a : fios ] ) is the dimension of the phase variable @xmath304.]@xmath305 with the canonical relation @xmath306 we observe that @xmath307 , and @xmath308 . therefore , see @xcite , @xmath309 the proof of a ) is similar to that of theorem [ t : main1 ] a ) . we skip it for the sake of brevity . we now proceed to prove b ) . let @xmath310 . that is @xmath311 and there is @xmath312 such that @xmath313 here , @xmath314 is the unit tangent vector of @xmath267 at @xmath315 . let @xmath316 be the unit normal vector of @xmath267 which is tangent to @xmath15 and points inward to @xmath38 . let @xmath317 be the metric on @xmath15 and @xmath318 be a small ( bounded ) neighborhood of @xmath315 such that for each @xmath319 there exists uniquely @xmath288 such that @xmath320 that is , each @xmath319 can be unique parametrized by @xmath321 . by narrowing down @xmath322 , if necessary , we may assume @xmath323 defines a smooth map from @xmath324 to @xmath325 $ ] , for some @xmath326 , whose jacobian @xmath327 is bounded from below . let @xmath328 such that @xmath329 near @xmath330 . let us write @xmath192 where @xmath193 let us denote by @xmath331 the schwartz kernels of @xmath198 and @xmath195 , respectively . then @xmath332 . an argument similar to the proof of proposition [ p : wavefront-3 ] ( see also proof of proposition [ p : wavefront ] ) shows that @xmath333 it now remains to analyze @xmath233 . it can be written in the form : @xmath334 by changing the variable @xmath335 , we obtain @xmath336 here , @xmath337 satisfies @xmath338 for @xmath339 and @xmath340 vanishes to order @xmath0 at @xmath341 . moreover , notice that by choosing @xmath322 small enough , we can assume that @xmath342 as small as we wish . let us show that there is a neighborhood @xmath343 of @xmath344 such that @xmath345 indeed , consider the phase function of the left hand side @xmath346 taking the derivative with respect to @xmath347 , we obtain @xmath348 therefore , @xmath349 where @xmath350 is the unit vector tangent to @xmath15 and normal to @xmath267 ( pointing inward to @xmath38 ) . from and the fact that @xmath14 is convex , we easily observe that @xmath351 therefore , by choosing @xmath352 small enough , we may assume that @xmath353 \times ( { \mathbb{r}}\setminus 0),\ ] ] where @xmath343 is a small neighborhood of @xmath354 . taking integration by parts , we obtain for all @xmath355 : @xmath356 continuing the integration by parts @xmath357 here , @xmath358 is homogeneous of degree @xmath359 with respect to @xmath304 . that is , @xmath360 here , @xmath361 and @xmath362 \ , { \lambda}}\ , { \partial}_s h_{k+1}(x , y , z_{\gamma},\delta ) \ , d\delta.\ ] ] using standard integration by parts as above , one can obtain @xmath363 moreover , from the definition of @xmath358 we get @xmath364 for all @xmath365 . therefore , @xmath366 this finishes the proof of ( [ c : claim1 ] ) . now let us write : @xmath367 applying lemma [ l : f ] , we obtain @xmath368 . in this section we investigate reconstructions from limited view circular mean data in a series of numerical experiments . here , we focus on the two dimensional problem for illustration reasons . we will show that our theoretical results from section [ s:2d ] directly translate into practice and , in particular , that they can be used to significantly improve the reconstruction quality . [ [ experimental - setup . ] ] experimental setup . + + + + + + + + + + + + + + + + + + + in all of our experiments we consider the phantom consisting of a disc with radius @xmath369 centered at the origin , see fig . [ fig : fig1 ] , where the image size of the phantom was chosen to be @xmath370 . in our experiments , we will numerically illustrate the visible and invisible singularities , as well as boundary singularities for different angular range . we will also numerically investigate the difference between singly and doubly visible singularities . it is therefore useful to keep in mind that all the singularities of the original phantom are located on the circle of radius @xmath369 centered at the origin , and the directions of all singularities are given by normal ( orthogonal ) directions to the circle at the location of the singularity . in what follows , we assume that the limited view data of this phantom are collected on a circular arc of the form @xmath371 in all of our experiments we computed the circular means of the phantom analytically and , to obtain limited view data , we sampled the data on @xmath372 ( first experimetn ) and @xmath373 ( second experiment ) . in each experiment , we chose the number of angular samples @xmath374 as well as the number of radial samples @xmath375 independently of the angular range as @xmath376 . given this data we then implemented and applied the reconstruction formula @xmath377 in matlab ( cf . ) where @xmath378 is the characteristic function of @xmath379 . in this situation , no artifact reduction was performed . to incorporate artifact reduction into the reconstruction formula , we also implemented the modified reconstruction formula @xmath380 where @xmath45 was constructed to be smooth in the interior @xmath381 of @xmath379 and , at the same time , to vanish to some order @xmath382 at the end points of @xmath379 ( cf . and subsequent discussion ) . more precisely , in our experiments we consider the following construction . let @xmath383 then , @xmath384 and * @xmath385 for @xmath386 , @xmath387 , * @xmath255 vanishes to order @xmath166 at @xmath388 . let @xmath45 be defined as @xmath389 , then @xmath48 is smoothing of order one . the parameter @xmath390 controls how close the function @xmath45 is to the constant function @xmath166 . the closer @xmath391 is to @xmath69 , the closer @xmath255 is to the constant function @xmath166 on @xmath392 , see fig . [ fig:1q - simple_smoothing_eps_var_power_var ] . that is , the smaller @xmath391 , the closer is the function @xmath45 to the constant function @xmath166 on @xmath38 . we also consider smoothing of orders @xmath393 and @xmath394 . the corresponding smoothing functions are defined through @xmath395 and @xmath396 , respectively . in general , we set @xmath397 the graphs of the functions @xmath398 for @xmath399 with @xmath400 are shown in fig . [ fig:1q - simple_smoothing_eps_var_power_var ] . [ [ experiment-1 ] ] experiment 1 : + + + + + + + + + + + + + in our first experiment we assume the data are collected on the acquisition surface @xmath372 ( quarter of the unit circle ) and consider reconstructions without as well as with artifact reduction ( using the smoothing function with parameter @xmath401 ) . we numerically investigate the effect of changing the reconstruction parameters @xmath391 as well as the effect of changing the smoothing order @xmath0 at the ends of the angular range . the results of this experiment are presented in fig . [ fig : fig1]-[fig:1q - order_varies ] . before examining the reconstructions , let us first note that the visible singularities at the considered angular range @xmath402 $ ] are located on two open quarters of the circle : first ( upper right ) and third ( lower left ) . accordingly , the invisible singularities are located on the other two quarters of the circle , whereas the boundary singularities correspond to the boundary of the acquisition surface @xmath372 and , therefore , are located at the four points @xmath403 , @xmath404 . these characterizations are due to the analysis and explanations that we presented in section [ s:2d ] . to numerically verify those theoretical findings , let us now examine the reconstruction in fig . [ fig : fig1 ] . our observations are as follows : * all visible singularities are reconstructed sharply . they visually appear to be of the same order as the original singularities ( jump from red to blue ) . * the invisible singularities are smoothed away and , hence , not present in the reconstruction . this can be seen from the fact that there are no sharp boundaries ( intensity jumps ) along invisible directions . * added singularities ( artifacts ) are generated along four circles , each of them touches the disc ( phantom ) tangentially . moreover , we observe that two circles are concentric and centered on the x - axis , and the other two are concentric and centered on the y - axis . more precisely , the added artifacts are located on circles that are centered at the boundary points of the acquisition surface @xmath405 ( which is illustrated by the green curve in fig . [ fig : fig1 ] ) and that are tangent to a singularity of the original phantom . that is , the artifacts are generated by the boundary singularities at @xmath406 , @xmath407 , @xmath408 , and @xmath409 . by further examining the artifacts in fig . [ fig : fig1 ] , we also observe that the jumps along the added artifact circles are not as sharp as in the case of visible singularities . this indicates that the added artifacts are weaker than original ( generating ) singularities . in fact , our theoretical analysis shows that they are @xmath410-order weaker . summing up , this experiment shows that the above observations correspond to our theoretical findings stated in propostion [ p : wavefront ] and theorem [ t : main1 ] . in the next step we investigate performance of the modified ( artifact reduction ) reconstruction operator @xmath48 . to that end , we use the operator @xmath48 with the cutoff function @xmath411 defined in - ( cf . also fig . [ fig:1q - simple_smoothing_eps_var_power_var ] ) , and apply it to the limited view data . note that the cutoff function @xmath411 is smooth in the interior of @xmath405 and vanishes to an order @xmath0 at the end points of @xmath405 . according to theorem [ t : main1 ] , the reconstructions obtained through @xmath48 , will exhibit added artifacts that are @xmath1 orders smoother than the original singularities . therefore , the degree of artifact reduction is linked to the order @xmath0 and we expect the operator @xmath48 to mitigate artifacts more when the bigger the order @xmath0 is . in addition to that , we expect that the strength of artifacts is influenced by the parameter @xmath391 ( see and the definition of @xmath398 ) . to investigate the practical performance , we computed a series of artifact reduction reconstructions by varying the parameters @xmath391 and @xmath0 . the results are shown in fig . [ fig:1q - simple ] and [ fig:1q - order_varies ] . first , we observe that in all reconstructions shown in fig . [ fig:1q - simple ] and [ fig:1q - order_varies ] most of the visible singularities are well reconstructed . in fig . [ fig:1q - simple ] , we have displayed some reconstructions using smoothing order @xmath412 and varying the parameter @xmath391 . here we observe that for @xmath413 almost no artifact reduction happens . this is due to the fact that , in the discretization regime , @xmath45 changes very fast near the endpoints of @xmath405 and , hence , @xmath45 behaves like a discontinuous function . the artifact reduction gets clearer as we increase the value of @xmath391 . next , we consider the effect of varying the smoothing parameter @xmath0 for a fixed @xmath400 . the corresponding reconstructions are shown in fig . [ fig:1q - simple_smoothing_eps_var_power_var ] . as expected , the artifacts get weaker ( are better reduced ) as the order increases . this is in accordance with our theoretical characterizations in theorem [ t : main1 ] . [ [ experiment-2 . ] ] experiment 2 . + + + + + + + + + + + + + our second experiment follows the lead of our first experiment . here , we only consider a larger angular range where we use limited view data collected on @xmath414 ( three quarters of the unit circle ) . again , we compute a series of reconstructions using the standard as well as the modified reconstruction operators , @xmath415 and @xmath48 , respectively . the results of this experiment are shown in fig . [ fig:3q - no_smoothing ] - [ fig:3q - order_original_smoothing ] . before we start , let us first remark that in this example all singularities of the phantom image are visible ( they are located on the circle centered at the origin of radius @xmath369 ) , and the locations of all boundary singularities are the same as in experiment 1 , namely @xmath404 and @xmath403 . in contrast to experiment 1 where all of the visible singularities were singly visible , we now have both types of visible singularities , doubly and singly visible ones . those on the first and third quarters are doubly visible , while those are on the second and fourth quarters are singly visible . see section [ s:2d ] for theoretical explanation . by examining the reconstructions using the standard reconstruction formula @xmath39 ( i.e. @xmath47 ) in fig . [ fig:3q - no_smoothing ] we easily observe that indeed all singularities of the phantom are reconstructed reliably . here , the doubly visible singularities have more contrast than the singly visible ones . this is due to the fact that for each doubly visible singularity there are two positions on the acquisition arc @xmath414 from which the singularity is visible , whereas there is only one position on this arc for a singly visible singularity . mathematically , this is reflected by the different values of the principal symbol @xmath164 of the reconstructions operator @xmath48 , cf . theorem [ t : main1 ] , where we can see that the principal symbol @xmath416 if @xmath63 is doubly visible and @xmath417 if @xmath63 is singly visible . in fig . [ fig:3q - simple_original_smoothing ] , we further observe that added artifacts are generated on circles that are centered at the boundary points of @xmath414 and tangent to the boundary singularities . these artifacts , however , are not as strong as the reconstructed singularities , which is again in accordance with our theoretical results , see theorem [ t : main1 ] and the discussion below . we again studied the performance of artifact reduction by using the modified reconstruction operator @xmath48 with the smoothing function @xmath411 for @xmath418 ( cf . - ) . the reconstruction results for varying parameters @xmath391 and for varying smoothing orders are shown in fig . [ fig:3q - simple_original_smoothing ] and fig . [ fig:3q - order_original_smoothing ] , respectively . not surprisingly , we observe here the same behavior as in experiment 1 . [ [ experiment-3 . ] ] experiment 3 . + + + + + + + + + + + + + in our last experiment we investigate how the choice of the smooth cutoff function for @xmath48 influences the artifact reduction . to that end , we consider the same limited view situation as in the experiment 2 where the data are collected on @xmath414 and define a new smoothing function @xmath45 which is equal to @xmath166 in the interior of @xmath414 and smoothly decreases to @xmath69 in transition regions of length @xmath391 at the boundary of @xmath414 . to that end , we let @xmath419 and @xmath420 1 , \hspace{55pt } \epsilon < s < 1-\epsilon , \\[5pt ] h_0(-s+1 ) , \hspace{10pt } 1 -\epsilon \leq s \leq 1 . \end{array } \right.\ ] ] the parameter @xmath390 again controls how close the function @xmath45 is to the constant function @xmath166 . moreover , this function @xmath255 vanishes to order 1 at the endpoints of @xmath414 . to obtain higher order smoothness at the endpoints we again consider integer powers of @xmath255 and set @xmath421 a plot of the functions @xmath422 is depicted in fig . [ fig : function_h ] for different values of @xmath391 and @xmath0 . the corresponding limited view reconstructions are presented in fig . [ fig:3q-2sm_perc_varies ] and [ fig:3q-2sm_order_varies ] . the advantage of such a choice of the function @xmath45 lies in the fact that it is exactly ( not approximately ) equal to @xmath166 in the range @xmath423 . therefore , if @xmath391 is very small , most of the visible singularities are reconstructed up to the factor @xmath166 ( if it is doubly visible ) or @xmath410 ( if it is singly visible ) . however , the ( theoretical ) disadvantage of such a choice is due to singularities of @xmath45 at the interior points @xmath424 and @xmath425 . according to our analysis in section [ s:2d ] , this may lead to the generation of added artifacts ( located on circles that are centered around these points ) . however , since @xmath411 is @xmath0 order smoother at these inner points than at the endpoints , those new artifacts will be weaker than those rotating around the endpoints @xmath426 and @xmath427 . indeed , as can be seen in fig . [ fig:3q-2sm_perc_varies ] and [ fig:3q-2sm_order_varies ] , the new added artifacts are too weak to be recognized in the reconstructions . concerning the influence of parameters @xmath391 and @xmath0 , we arrive at similar observations as in experiments 1 and 2 . comparing the reconstructions that were computed with different smoothing functions in fig . [ fig:3q - order_original_smoothing ] and fig . [ fig:3q-2sm_order_varies ] we observe that the new smoothing function leads almost always to a significantly better artifact reduction . for example , in fig . [ fig:3q-2sm_order_varies ] , the artifacts almost completely vanish and the phantom as well as the background are reconstructed very well . this examples shows that the choice of the smoothing function might influence artifact reduction performance significantly . [ [ conclusion . ] ] conclusion . + + + + + + + + + + + the above numerical experiments show that our theoretical results directly translate into practical observations . in particular , the proposed artifact reduction technique can lead to a significant improvement of the reconstruction quality if the smoothing function is chosen appropriately . we will explore more experiments and report in - depth results in a future publication . the work of l.l.b . was supported by the national science foundation major research instrumentation program , grant 1229766 . j. f. thanks eric todd quinto for many enlightening discussions about limited data tomography and microlocal analysis over the years as well as for the warm hospitality during several visits at tufts university . he also acknowledges support from the hc @xmath428rsted postdoc programme , co - funded by marie curie actions . l.v.n.s research is partially supported by the nsf grant # dms 1212125 . he is thankful to professor g. uhlmann for introducing him to the theory of pseudo - differential operators with singular symbols presented in @xcite , whose spirit inspires this and the previous works @xcite . he also thanks professor t. quinto for the encouragement and helpful comments / suggestions . the authors would like to thank g. ambartsoumian for sharing his codes in spherical radon transform , some of them are reused in this article s numerical implementations . they also thank professor p. stefanov for pointing out some missing references in the initial version of the article . let @xmath429 be an open set , and @xmath430 be the cotangent bundle of @xmath14 . for simplicity , we can consider @xmath430 as @xmath431 . we also denote @xmath432 let @xmath433 and @xmath434 be the standard spaces of test functions and distributions on @xmath14 . in this section , we briefly introduce some basic concepts in microlocal analysis , such as wave front set , pseudo - differential operators ( @xmath57do ) , and fourier integral operators ( fios ) . extensive presentations can be found in @xcite . the use of microlocal analysis in geometric integral transforms are pioneered in @xcite . its extensive uses in the studies of spherical mean transform can be found in many works , see @xcite , just to name a few . [ wave front set @xcite ] let @xmath435 and @xmath436 . then , @xmath4 is microlocally smooth at @xmath437 if there is a function @xmath438 satisfying @xmath439 and an open cone @xmath440 containing @xmath441 , such that @xmath442 is rapidly decreasing in @xmath440 . that is , for any @xmath443 there exists a constant @xmath444 such that @xmath445 the * wavefront set * of @xmath4 , denoted by @xmath446 , is the complement of the set of all @xmath447 where @xmath4 is microlocally smooth . an element @xmath81 indicates not only the location @xmath60 but also the * direction * @xmath61 of a singularity of @xmath4 . for example , if @xmath4 is the characteristic function of an open set @xmath448 with smooth boundary @xmath449 , then @xmath81 if and only if @xmath450 and @xmath61 is perpendicular to the tangent plane of @xmath451 at @xmath60 . detailed discussion can be found in @xcite and , more briefly , in @xcite . [ t : compose ] let @xmath465 and @xmath466 be linear transformations whose schwartz kernels are @xmath467 and @xmath468 . we assume that @xmath469 and @xmath470 . then , the schwartz kernel @xmath49 of @xmath471 satisfies : @xmath472 let @xmath435 and @xmath473 . then @xmath4 is in the space @xmath474 microlocally at @xmath437 if there is a function @xmath438 satisfying @xmath439 and a function @xmath475 homogeneous of degree zero and smooth on @xmath476 with @xmath477 , such that @xmath478 the * @xmath474-wave front set * of @xmath479 , denoted by @xmath480 , is the complement of the set of all @xmath447 where @xmath479 is not microlocally in the space @xmath474 . one can use the sobolev orders to compare the singularities @xmath481 and @xmath482 , where @xmath483 are two distributions , not necessarily defined on the same set . for example , @xmath484 is stronger than @xmath485 if there is @xmath347 such that @xmath486 but @xmath487 . assume that @xmath488 is a smooth surface of co - dimension @xmath0 . let @xmath489 be a defining function for @xmath165 with @xmath490 on @xmath165 . the class @xmath491 consists of the distributions which locally can be written down as a finite sum of oscillatory integrals of the form @xmath492 where @xmath493 . let @xmath429 be an open set . the space @xmath494 consists of all functions @xmath495 such that for any multi - indices @xmath496 and @xmath497 , there is a positive constant @xmath498 such that @xmath499 the elements of @xmath494 are called symbols of order @xmath500 . one can use the order @xmath9 to compare two conormal singularities @xmath507 ( along the surface @xmath508 ) and @xmath509 ( along the surface @xmath510 ) , where @xmath483 are two distributions on @xmath14 . for example , @xmath63 is * weaker * than @xmath106 , if there is @xmath511 such that @xmath63 is of order @xmath9 while @xmath106 is not . since the above integral may not converge in the classical sense , the expression in needs to be properly defined , see , e.g. , ( * ? ? ? * proposition 1.1.2 ) . given this proper definition , @xmath48 extends continuously to @xmath516 . in particular , it can be shown that a pseudo - differential operator @xmath48 does not generate new singularities . that is , ( * ? ? * page 131 ) @xmath517 moreover , if @xmath4 is in the space @xmath474 microlocally at @xmath437 then @xmath518 is in space @xmath519 microlocally at the same element @xmath437 , see @xcite . [ d : pdo ] let @xmath521 be a conic set that is open in the induced topology of @xmath522 . we say that * near @xmath523 , @xmath49 is microlocally in the space @xmath524 with the symbol @xmath525 * if the following holds : for each element @xmath526 there exist @xmath527 such that @xmath528 and the symbol of @xmath529 is equal to @xmath530 in a conic neighborhood if @xmath183 . [ l : pet ] let @xmath531 be a linear operator whose schwartz kernel @xmath532 satisfies @xmath533 . assume that @xmath49 is microlocally in @xmath524 near @xmath523 with the symbol @xmath530 . let @xmath534 such that @xmath526 and in a conic neighborhood of @xmath183 @xmath535 assume further that @xmath536 then , for any @xmath537 , @xmath538 in this section , we introduce some special fourier distributions which are needed in this article . the reader is referred to , e.g. , @xcite for the general theory of the topic . let @xmath291 and @xmath292 be two manifolds of dimension @xmath539 and @xmath540 , respectively , and @xmath272 be a homogeneous canonical relation in @xmath541 . then , there is an associated class of fourier distributions of order @xmath500 , denoted by @xmath542 . each element of @xmath542 , called a fourier integral distribution of order @xmath500 , is a distribution @xmath532 such that it can be locally written down in the form @xmath543 here , of @xmath544 specified here is due to , e.g. , @xcite . ] @xmath545 and @xmath126 is a phase function associated to @xmath272 . that is , @xmath546 satisfies [ d : fio ] let @xmath555 be an open conic set in the induced topology of @xmath272 . we say that * near @xmath523 , @xmath49 is microlocally in the space @xmath542 * if the following holds : for each element @xmath556 there exists @xmath557 such that @xmath558 [ t : ho ] let @xmath429 and @xmath559 be a homogeneous canonical relation such that both of its left and right projections on @xmath14 have surjective differentials . assume that the differentials of the left and right projections @xmath560 have rank at least @xmath561 . then , every @xmath562 maps continuously from @xmath563 to @xmath564 . let @xmath566 such that @xmath567 . we define the following homogeneous canonical relation in @xmath568 @xmath569 that is , @xmath570 is defined by rotating @xmath571 , which pass through @xmath572 , around @xmath572 . [ l : ho ] let @xmath531 be a linear operator whose schwartz kernel @xmath532 satisfies @xmath579 . assume that @xmath49 is microlocally in @xmath542 near an open conic set @xmath555 . let @xmath580 such that @xmath581 if @xmath582 , then there is @xmath583 such that @xmath584 the following result is useful to analyze the artifacts when the original singularities are conormal . its proof is almost exactly the same as that of ( * ? ? ? * theorem 2.16 ) . we skip it for the sake of brevity . * there are at most finitely many @xmath585 such that @xmath586 * each such @xmath587 is a conormal singularity of order @xmath9 along a curve @xmath129 whose contact order with @xmath588 is exactly @xmath166 . , since both curves are perpendicular to @xmath589 at @xmath218 . therefore , the condition on the contact order is quite generic . ] let us consider @xmath266 . we introduce a class of fourier distributions , whose canonical relation is defined by the rotations around tangent lines of a smooth curve . this class of fourier distributions appears in the statement and proof of theorem [ t : main2 ] b ) . let @xmath267 be a closed smooth curve in @xmath592 parametrized by the parameter @xmath347 . assume that @xmath593 . we define the following homogeneous canonical relation in @xmath101 @xmath594 that is , @xmath595 is defined by rotating an element @xmath596 , that passes through @xmath597 , around the the tangent line of @xmath267 at @xmath8 . in section [ s:3d ] , we make use of this class @xmath598 . we state here some needed basic facts of this class . we note that : @xmath602 the following result is a microlocal version of the above result , which is used in section [ s:3d ] to analyze the strength of artifacts . its proof is almost exactly the same as that of ( * ? ? ? * corollary 2.15 ) . we skip it for the sake of brevity . [ l : ho3d ] let @xmath531 be a linear operator whose schwartz kernel @xmath532 satisfies @xmath579 . assume that @xmath49 is microlocally in @xmath542 near an open conic set @xmath555 . let @xmath580 such that @xmath581 if @xmath582 , then there is @xmath603 such that @xmath584 jos l. antoniano and gunther a. uhlmann . a functional calculus for a class of pseudodifferential operators with singular symbols . in _ pseudodifferential operators and applications ( notre dame , ind . , 1984 ) _ , volume 43 of _ proc . pure math . _ , pages 516 . soc . , providence , ri , 1985 . d. finch , i .- r . lan , and g. uhlmann . microlocal analysis of the x - ray transform with sources on a curve . in _ inside out : inverse problems and applications _ , volume 47 of _ math . _ , pages 193218 . cambridge univ . press , cambridge , 2003 . a. greenleaf and g. uhlmann . microlocal techniques in integral geometry . in _ integral geometry and tomography ( arcata , ca , 1989 ) _ , volume 113 of _ contemp . _ , pages 121135 . soc . , providence , ri , 1990 . l. hrmander . , volume 256 of _ grundlehren der mathematischen wissenschaften [ fundamental principles of mathematical sciences]_. springer - verlag , berlin , 1983 . distribution theory and fourier analysis .

we examine signals for sparticle production at the cern large hadron collider ( lhc ) within the framework of gauge mediated supersymmetry breaking models with a low susy breaking scale for four different model lines , each of which leads to qualitatively different signatures . we first examine the reach of the lhc via the canonical and multilepton channels that have been advocated within the msugra framework . next , we examine special features of each of these model lines that could serve to further enhance the susy signal over standard model backgrounds . we use isajet to evaluate the susy reach of experiments at the lhc . we find that the susy reach , measured in terms of , is at least as large , and sometimes larger , than in the msugra framework . in the best case of the co - nlsp scenario , the reach extends to tev , assuming 10 of integrated luminosity .

hep-ph0004001

the search for supersymmetric particles has become a standard item for on - going as well as future experiments at high energy colliders . no signal has as yet been found , and the absence of evidence has been translated to lower limits on the masses of various sparticles @xcite . since the qualitative features as well as the size of experimental signals are determined by sparticle production cross sections and decay patterns , any such interpretation can strictly speaking only be done within the framework of a particular model , most often taken to be the minimal supergravity ( msugra ) model . while it is true that in some cases the non - observation of a signal can be translated into a sparticle mass bound in a relatively model - independent manner ; even here it is tacitly assumed that the decay of @xmath4 is prompt . ] , the extraction of signals ( or bounds ) becomes increasingly complicated if several sparticles are expected to be simultaneously produced as , for instance , at hadron colliders such as the cern lhc . this may be due to significant differences of sparticle mass and decay patterns from the msugra expectations @xcite , or due to new decay modes being accessible as in gauge - mediated susy breaking ( gmsb ) models @xcite with a low susy breaking scale , so that the gravitino is by far the lightest supersymmetric particle ( lsp ) . in recent years , there has been a flurry @xcite of theoretical as well as experimental activity @xcite on the detection of susy signals within the gmsb framework , where susy breaking is communicated from the hidden sector to the observable sector via standard model ( sm ) gauge interactions of messenger particles ( which , we assume , can be classified into @xmath5 complete vector representations of @xmath6 ) with quantum numbers of @xmath7 doublets of quarks and leptons whose mass scale is characterized by @xmath8 . the soft supersymmetry breaking masses for the susy partners of sm particles are thus proportional to the strength of their gauge interactions , so that squarks are heavier than sleptons , while the gaugino masses satisfy the usual `` grand unification '' mass relations , though for very different reasons . within the minimal version of this framework , the couplings and masses of the sparticles in the observable sector are determined ( at the messenger scale @xmath8 ) by the parameter set , @xmath9 the parameter @xmath10 sets the scale of sparticle masses and is the most important of these parameters . the model predictions for soft - susy breaking parameters at the scale @xmath8 are evolved to the weak scale . the parameter @xmath11 @xcite and enters only into the partial width for sparticle decays to gravitino . the decay to a gravitino is phenomenologically important only for the decay of the next to lightest supersymmetric particle ( nlsp ) . we fix @xmath12 which corresponds to the fastest possible decay of the nlsp into the gravitino : for larger values of @xmath13 , the nlsp may decay with an observable decay gap ( possibly improving the sensitivity to the signal ) , or may even pass all the way through the detector . the phenomenological implications of the model are only weakly sensitive to @xmath8 . aside from the scale of sparticle masses , the single item that most sensitively determines sparticle signals at colliders is the identity of the nlsp . in the simplest version of model , the nlsp is either the lightest neutralino ( @xmath14 ) with the hypercharge gaugino as its dominant component , or the lighter stau ( @xmath15 ) . in the latter case , depending on the value of @xmath16 , @xmath17 and @xmath18 may be essentially degenerate with @xmath15 . if the nlsp is a @xmath19 gaugino - like neutralino , it dominantly decays via @xmath20 , and susy events always contain two hard , isolated photons and @xmath0 ( from the escaping gravitinos ) in addition to jets and/or leptons . in the case where the nlsp is a stau which is significantly lighter than other sleptons , it is produced in most susy decay chains , and susy events in this scenario typically contain several tau leptons . other light sleptons decay via @xmath21 if the decay is allowed ; otherwise as discussed in ref . @xcite , these mostly decay via @xmath22 or @xmath23 which ( even for @xmath12 ) , as long as these are not kinematically strongly suppressed , have a much larger branching ratio than for the decay @xmath24 . for the case where @xmath25 ( the co - nlsp scenario ) , only the gravitino mode is allowed , so that we may then expect a significant multiplicity of isolated @xmath26 s and @xmath27 s in susy events . in a previous paper @xcite , we examined susy signals and the reach of the fermilab tevatron main injector ( for integrated luminosities of 2 and 25 @xmath3 ) for representative cases for each one of these three model lines discussed in the previous paragraph . we also examined the reach for the so - called higgsino model line where the nlsp is a higgsino - like neutralino . although this does not occur in the minimal version of the gmsb framework because @xmath27 tends to be large compared to the hypercharge gaugino mass @xmath28 , we were motivated to examine this case since experimental signals tend to be quite different because the nlsp dominantly decays via @xmath29 ( rather than @xmath20 ) unless this decay is suppressed by phase space . our purpose here is to perform a similar study for the reach of the cern large hadron collider ( lhc ) , a @xmath30 collider with a center of mass energy @xmath31 tev , currently scheduled to begin operation in 2005 . we assume throughout this paper that an integrated luminosity of 10 fb@xmath32 is collected , which corresponds to a year of running at the lower value of the design luminosity . toward this end we examine the susy reach of the lhc for , * model line a : @xmath33 , @xmath34 tev , @xmath35 , @xmath36 , which gives a @xmath19 gaugino - like nlsp ; * model line b : @xmath37 , @xmath38 , @xmath39 , @xmath36 , for which the nlsp is a stau ; * model line c : @xmath40 , @xmath41 , @xmath42 , @xmath43 , which yields the co - nlsp scenario ; * model line d : @xmath37 , @xmath38 , @xmath42 , @xmath44 , which gives a higgsino - like neutralino as the nlsp , which if heavy enough mainly decays via @xmath45 ; * model line e : @xmath37 , @xmath38 , @xmath42 , @xmath46 , which also gives a higgsino - like neutralino as the nlsp , but which mainly decays via @xmath47 as long as the decay is not suppressed by phase space . since we have already examined the lhc reach for model line a before @xcite , we will confine our attention to model lines b - e in this paper . the tevatron reach for model lines b , c and d was studied in ref . the reach for a gaugino - like nlsp model line , but with a slightly different set of parameters ( which should have no qualitative effect on the results ) was also examined in this same study . model line e is new . the difference between d and e is that while @xmath14 mainly decays via @xmath48 for model line d , the decay @xmath47 dominates @xcite in the case of model line e ( assuming , of course , that these decays are kinematically unsuppressed ) . for each of the model lines b - e , we first examine the reach using the strategy @xcite for susy searches within the msugra framework : _ i.e. _ the reach via the @xmath0 plus @xmath49 , @xmath50 and @xmath51 inclusive channels . we expect that this ( or something like it ) will be the canonical strategy @xcite for susy searches in the initial data samples from the lhc . for each of these model lines , we then examine whether and by how much the reach can be improved using the special features characteristic of the particular model line , _ e.g. _ the presence of hard , isolated photons in model line a @xcite . the remainder of this paper is organized as follows . in sec . ii , we briefly discuss our simulation using the program isajet @xcite . in section iii , we discuss the lhc reach for the four model lines b - e . we end with a summary and some concluding remarks in sec . iv . we use the program isajet v 7.43 @xcite for our simulation of susy events . this incorporates decay matrix elements for three body decays of gluinos , charginos and neutralinos in the event generator . the implementation of the gmsb framework into isajet v 7.43 has been described elsewhere @xcite and will not be repeated here . isajet uses the gmsb model predictions for sparticle masses as initial conditions valid at the messenger scale @xmath8 , then evolves these down to the weak scale relevant for phenomenology , and finally calculates the ` mssm parameters ' that are then used in the evaluation of sparticle cross sections and decay widths . for detector simulation at the lhc , we use the isajet toy calorimeter simulation package calsim . we simulate calorimetry covering @xmath52 with a cell size given by @xmath53 , and take the hadronic calorimeter resolution to be @xmath54 for @xmath55 , where @xmath56 denotes addition in quadrature , and to be @xmath57 for @xmath58 , to model the effective @xmath59 resolution of the forward calorimeter including the effects of shower spreading , which is otherwise neglected . we take electromagnetic resolution to be @xmath60 . although we have included these resolutions , which are typical of atlas @xcite and cms @xcite , we have made no attempt to estimate the effects of cracks , edges , and other problem regions . much more detailed detector simulations are needed to understand the effects of such regions and of the resulting non - gaussian tails , particularly on the @xmath0 resolution . jets are defined as hadronic clusters with @xmath61 gev within a cone of @xmath62 with @xmath63 . muons and electrons with @xmath64 gev and @xmath65 are considered to be isolated if the the scalar sum of electromagnetic and hadronic @xmath66 ( not including the lepton , of course ) in a cone with @xmath67 about the lepton to be smaller than 5 gev . isolated leptons are also required to be separated from one another by @xmath68 . photons are identified as isolated with @xmath69 and @xmath64 gev if the additional @xmath66 within a cone of @xmath70 about the photon is less than 2.5 gev . for @xmath71-jets , we require a jet ( satisfying the above jet criteria ) to contain a @xmath72-hadron with @xmath73 gev and @xmath74 . jets with @xmath75 gev are tagged produced via @xmath45 for model line d , we assume that softer @xmath71-jets can also be tagged , but with a reduced efficiency as detailed in sec . iii c. ] as a @xmath71-jet with a probability of @xmath76 . qcd jets with @xmath77 ( @xmath78 ) gev are misidentified as @xmath71-jet with a probability of @xmath79 ( @xmath80 ) with a linear interpolation in between @xcite . in this section we identify strategies to enhance the lhc susy reach for gmsb models beyond what would be obtained via canonical multijet + multileptons + @xmath0 analyses @xcite advocated for the msugra framework . we examine the reach for model lines b - e , for which the nature of the nlsp , and hence the characteristic susy signals , vary with the model line . the reach of model line a has already examined previously . we use isajet v 7.43 to compute signal cross sections , after incorporating acceptances to simulate the experimental conditions at the lhc together with additional cuts that serve to separate the susy signal from sm backgrounds . to realize the case of the stau nlsp , we take @xmath37 , @xmath38 , @xmath36 , and choose @xmath39 to ensure that the stau is light enough for other sleptons to be able to decay to it . it is worth noting that although @xmath81 , the sleptons are all sufficiently close in mass so that generally speaking the phase space for chargino and neutralino decays to staus and other sleptons is not very different . for @xmath82 tev , @xmath83 , but for @xmath84 tev , @xmath85 is kinematically forbidden , and @xmath14 would decay via the four body decay @xmath86 ( which is not yet included in isajet ) or via @xmath87 . in our study , we consider @xmath88 tev , the range safe from lep constraints . in fig . [ csectionb ] , we show cross sections for various sparticle production processes at the lhc . for @xmath89 tev ( @xmath90 tev ) , squark and gluino production ( solid line ) dominates , while for larger values of @xmath10 electroweak production of various charginos and neutralinos ( dashed line ) are the dominant susy process . for cross sections in the range of interest at the lhc , direct slepton and sneutrino production occurs at @xmath91 the rate for chargino and neutralino production . cascade decays of squarks and gluinos naturally lead to events with @xmath92 leptons + @xmath93 jets + @xmath0 . these event topologies may also arise from the production of gluinos and squarks in association with a chargino or a neutralino , and also from chargino and neutralino production . the decay patterns of charginos , neutralinos and sleptons have been discussed in detail in ref . @xcite and we will only summarize the results here . the two body decay @xmath94 is always accessible , and dominates for @xmath95 tev ( @xmath96 gev ) . the decay rate for @xmath97 becomes comparable to that for @xmath94 for @xmath98 tev , while for very heavy charginos ( @xmath99 gev ) , the branching ratio for @xmath97 is around 75% . turning to neutralinos , @xmath100 is the main decay mode of @xmath101 if @xmath102 gev . the decay @xmath103 is only important for @xmath104 tev ( @xmath105 gev ) and becomes dominant for @xmath106 tev ( @xmath107 gev ) . branching fractions for @xmath108 ( @xmath109 ) may also be significant and vary from 20% each for @xmath110 tev to @xmath111% for @xmath112 tev . the branching fraction for the decay @xmath113 is @xmath114% if this decay is kinematically accessible . for small enough @xmath10 , the lightest neutralino @xmath14 mainly decays via @xmath115 , but for large enough values of @xmath10 its decay to other sleptons are also important , and together exceed the decays to stau when for @xmath116 tev ( @xmath117 gev ) . finally , for small values of @xmath10 accessible at tevatron upgrades , @xmath118 can only decay via @xmath21 . for larger @xmath10 , this channel is closed , the neutralino is virtual and then the decays @xmath119 and @xmath120 dominate . however , the decay @xmath121 also becomes significant ( branching fraction @xmath12210% for @xmath123 tev ) if @xmath10 is large enough . we begin our discussion of the lhc reach via canonical searches in the multijet plus multilepton plus @xmath0 channels as discussed in ref . @xcite . for squark and gluino production , the @xmath59 of the primary jets from gluinos ( squarks decay to gluinos ) , as well as the @xmath0 , are expected to scale with @xmath1 . since the signal decreases with increasing @xmath1 , its separation from sm backgrounds can be optimized by requiring cuts on jets and @xmath0 that scale with the gluino mass . the momenta of leptons , produced far down in the cascade decay chain from chargino and neutralino daughters , do not scale in energy the same way as @xmath0 or energies of jets which are produced in the first step of the decay cascade . following ref . @xcite , we introduce the running cut variable @xmath124 , and require that the signal events satisfy : * @xmath125 ( with @xmath126 gev ) ; * the transverse sphericity , @xmath127 ; * @xmath128 , @xmath129 and @xmath130 ; * for @xmath131 signal , @xmath132 gev and @xmath133 gev ; * for @xmath134 lepton signal , @xmath135 gev , in addition to the basic acceptance cuts . we classify signal events according to their lepton topology into the nonleptonic @xmath0 + jets channel , the opposite - sign ( os ) dilepton + jets + @xmath0 channel , the same - sign ( ss ) dilepton + jets + @xmath0 channel , and the multilepton + jets + @xmath0 channel with @xmath136 . throughout this paper , we consider a signal to be observable in a particular channel if there are at least 5 signal events and also the signal has a statistical significance @xmath137 , assuming 10 fb@xmath32 of integrated luminosity . we study the observability by varying @xmath124 between 100 and 500 gev in steps of 100 gev . the dominant sm backgrounds are @xmath138 , @xmath139 or @xmath140 + jet production , @xmath141 production and vector boson pair production . since our analysis cuts are identical to those in ref . @xcite , we do not re - compute these backgrounds but instead use the background cross sections as a function of @xmath124 given in these papers in our analysis . the results of our computation of the reach of the lhc using the msugra cuts are summarized in fig [ reachmlb ] where we show the maximum @xmath142 as a function of @xmath124 , requiring further that there be at least 5 signal events per 10 @xmath3 . for each value of @xmath10 that we studied , the number on the various lines denotes the @xmath143 value for which the signal is observable according to our criteria , and for which the signal significance is the greatest ( subject to the five event minimum ) , i.e. a 2 denotes @xmath144 gev yields the best significance , _ etc . _ in other words , this number provides information about the optimal strategy for the particular value of @xmath10 . the different lines correspond to the different event topologies : @xmath145 ( lower solid ) , @xmath146 ( upper solid ) , ss dileptons + @xmath0 ( dashed ) , os dileptons + @xmath0 ( dotted ) and trileptons ( dot - dashed ) . the horizontal line denotes the @xmath147 level for an integrated luminosity of 10 @xmath3 . we see from the figure that unlike the case of the msugra framework , the inclusive @xmath51 channel provides the greatest reach . this is presumably because of additional leptons that come from slepton decays to the gravitino , or as secondaries of daughter taus produced via the decay of the stau nlsp . the @xmath0 signal ( with a lepton veto ) gives a rather poor reach for essentially the same reason . we can probe @xmath10 values beyond 150 gev which corresponds to a gluino in excess of 2 tev in this channel . notice also that if @xmath89 tev ( @xmath148 tev ) confirmatory signals should appear in several other channels . for comparison , we note that even for an integrated luminosity of 25 @xmath3 , the reach of the tevatron was projected @xcite to be about 50 tev . next , we examine whether we can extend the reach by requiring the presence of additional tagged @xmath149 leptons . in table i , we show the signal cross section for @xmath150 and 160 tev , with events classified first by the number of identified taus , and then by the lepton multiplicity . here , a jet with @xmath151 gev is identified as a tau if it has one or three charged prongs with @xmath152 gev in a 10@xmath153 cone about the jet axis , with no other tracks in the corresponding 30@xmath153 cone . if the number of charged prongs is three , we require their net charge to be @xmath154 and the invariant mass to be smaller than @xmath155 . qcd fakes are included with the same algorithm used in ref . the cross sections shown in table i are obtained with just @xmath156 gev in addition to the basic lepton and @xmath149 identification cuts . we see that even with this minimal requirement , the cross section for @xmath157 events is just 3.25 ( .92 ) @xmath158 for @xmath159 tev . since these channels all suffer from physics , and more importantly instrumental , backgrounds from sm sources , we conclude that tagging hadronically decaying taus is unlikely to improve the reach for this model line : cuts to remove these will reduce the signal to below the observable level for @xmath10 values where the canonical strategy does not lead to a detectable signal . we also attempted to extract the signal via `` clean '' multilepton plus multi - tau channels where we veto events with jets that are not identified as taus . we found that typically the signal reduces by an order of magnitude ( or more , depending on the transverse momentum threshold for the jet veto ) and falls well below the one event level in all the channels . we thus conclude that the canonical susy search strategy @xcite via multilepton channels provides the largest reach for model line b. for this model line , @xmath17 , @xmath18 and @xmath15 are approximately degenerate , and @xmath17 and @xmath18 can not decay to @xmath15 . to realize this , we choose @xmath40 and @xmath42 with other parameters as before . various sparticle production cross sections are shown in fig . [ csectionc ] . we see that while production of squarks and gluinos dominates for @xmath89 tev ( @xmath160 tev , @xmath161 tev ) , slepton pair production is dominant for @xmath162 tev ( corresponding to @xmath163 gev ) . in this model line , gluinos decay directly to squarks . the left - squarks typically cascade to @xmath164 and @xmath101 which then decay to the lightest neutralino , while @xmath165 mostly decays directly to @xmath14 . the sparticle decay patterns have been discussed in ref . the lightest neutralino decays via @xmath166 , ( @xmath167 , @xmath27 , and @xmath149 ) , with branching fractions essentially independent of the lepton flavour . sleptons and sneutrinos decay via @xmath168 while @xmath169 ( @xmath170,@xmath171 ) . for very small @xmath10 ( excluded by lep 2 since the slepton is then very light ) , the charginos can only decay via @xmath172 but over much of the allowed range of @xmath10 the decays @xmath173 dominate and have ( approximately ) flavour - independent branching fractions . the decay @xmath174 also becomes significant for charginos heavier than @xmath175 gev , and has a branching fraction that decreases from a maximum of @xmath176% to @xmath177% as @xmath10 increases to 130 tev . the second lightest neutralino @xmath101 decays to sleptons with branching fractions more or less independent of the lepton flavour . decays to the heavier ( mainly left - handed ) sleptons and sneutrinos dominate when these are not kinematically suppressed . the branching fraction for the decay @xmath103 is also significant once this decay is allowed , and the branching fraction reduces from a maximum of 25% to below 15% for @xmath178 tev . neutralino decays to @xmath140 are negligible . an inspection of these decay patterns shows that it would be natural to expect events with a large multiplicity of isolated leptons ( @xmath26 and @xmath27 ) from sparticle production , so that multilepton plus @xmath0 events should provide the best reach . the naturally high lepton multiplicity in these events led us to focus on the essentially background - free @xmath179 lepton plus @xmath0 channel . we have not attempted to make a quantitative calculation of the background but have instead contented ourselves with trying to bound cross sections for physics sources of such events . these include , * @xmath180 production with @xmath181 @xmath158@xcite , which yields @xmath182 @xmath158 ( neglecting isolated leptons s is 0.25 @xmath158 ; since the branching fraction for leptonic @xmath71 decays is about 0.22 , we conclude that as long as the lepton from @xmath71 is isolated less than @xmath183 of the time , this background is smaller than the @xmath184 level estimated above . ] from @xmath71 decays ) ; * @xmath185 production with @xmath186 @xmath158 @xcite , which yields @xmath187 @xmath158 ; * @xmath188 production with @xmath189 @xmath158 @xcite , which yields @xmath190 @xmath158 ( here we assume that events @xmath191 can be removed by a mass cut , and retain leptons only from @xmath192 ) ; * @xmath193 production with @xmath194 pb , which yields @xmath195 @xmath158 where , once again , we assume that the leptons come only via @xmath196 decays . these events tend not to have a large @xmath0 and softer leptons , and the background can be greatly reduced by a cut on @xmath0 and the lepton @xmath59 . turning to the signal , we show the distributions of the two hardest leptons in @xmath197 susy events in fig . [ mlc_lep12pt ] for four choices of @xmath10 . we see that requiring these leptons to be harder than 50 gev reduces the signal by just a few percent for @xmath198 tev . even for @xmath199 tev , the reduction is just 40% , which is very affordable because of enormous susy cross sections at the lhc . although we do not show this , we have also checked that the signal loss from requiring @xmath200 gev is again only a few percent for @xmath198 tev . to obtain the lhc reach via the @xmath197 signal for model line c , we thus require : * @xmath201 gev ; * @xmath202 , @xmath203 gev ; * a @xmath140 veto we remove events where with @xmath204 pairs that reconstruct the @xmath140 mass to within 10 gev . we assume that these cuts reduce the sm physics backgrounds already found to be below 0.2 @xmath158 to completely negligible levels . assuming that instrumental backgrounds are under control , we then estimate the reach for this model line by requiring at least five signal events in 10 @xmath3 as illustrated in fig . [ reachmlc ] where we show the signal cross section with @xmath205 versus @xmath10 . we see that the reach extends up to @xmath206 tev , which corresponds to squarks ( gluinos ) just below ( above ) 3 tev , to be compared with the reach of @xmath207 tev @xcite for an integrated luminosity of 25 @xmath3 at the tevatron . since the sleptons are the co - nlsps in this scenario , we have also examined the range of parameters where these can be directly observed at the lhc . we follow the general strategy for slepton searches at the lhc detailed in ref . @xcite , with some modifications to optimize for the gmsb scenario . the analysis cuts that we impose are , * exactly two isolated same flavor os leptons , each with @xmath208 gev ; * @xmath209 gev ; * number of jets with @xmath210 gev is zero ( jet veto ) ; * @xmath211 . these cuts above are exactly the same as in ref . @xcite . for the final cut , however , we choose instead , * @xmath212 gev , @xmath213 gev and @xmath214 . the major sm background at the lhc come from @xmath141 production and @xmath215 production . the jet veto is crucial to reduce the top quark background . in our simulation , we have assumed that jets can be vetoed with an efficiency of 99% . after these cuts , the @xmath141 background is 0.06 @xmath158 , while the @xmath215 background is 0.038 @xmath158 , so that the total sm background @xmath216 and the minimum detectable cross section corresponds to 5 events per 10 @xmath3 . the signal cross section for opposite sign dilepton from just slepton production ( slepton , chargino and neutralino production ) is shown by the dashed ( solid ) line in fig . [ reachslepc ] . we see that except for the @xmath217 tev case ( corresponding to @xmath218 gev , just beyond the current bound from lep 2 ) the signal dominantly comes from slepton pair production because @xmath219 . the reason the dashed curve falls for low values of @xmath10 is the inefficiency of our cuts ( which have been optimized for heavier sleptons ) for such light sleptons . nonetheless , even with these cuts there is ample signal for susy not to evade detection at the lhc . we see that the reach extends up to @xmath220 tev corresponding to @xmath221 gev . the reason that the reach in slepton mass is significantly smaller than the corresponding reach within the msugra framework is because the production of @xmath222 and @xmath223 , which was the main contributor to the slepton cross section in msugra , is now kinematically suppressed : _ i.e. _ within the gmsb , @xmath224 production forms the bulk of the slepton cross section . within the minimal gmsb framework , the lightest neutralino is dominantly the hypercharge gaugino . however , since the phenomenology is very sensitive to the nature of the nlsp , we entertain the possibility that in other models , the nlsp may be higgsino - like . we do not attempt to construct any model , but take a purely phenomenological approach , by letting @xmath27 to be a free parameter instead of fixing it via radiative electroweak symmetry breaking conditions . and also the @xmath72-parameter in this framework could conceivably alter the relation between @xmath27 and the gaugino masses . ] if @xmath27 is smaller than the hypercharge gaugino mass @xmath28 , the nlsp will indeed contain a large higgsino component . we take @xmath37 , @xmath42 , @xmath226 , @xmath12 but fix @xmath227 rather than the value obtained from radiative electroweak symmetry breaking . within this ` small @xmath27 ' model line , the two lightest neutralinos and the lighter chargino are higgsino - like and close in mass , while the heavier charginos and neutralinos are gaugino - like . the @xmath14 is the nlsp and the @xmath101 and @xmath164 are only little heavier . the fermions from @xmath164 and @xmath101 decays to @xmath14 will be rather soft . for this sign of @xmath27 , the nlsp dominantly decays via @xmath45 if the @xmath14 is heavy enough ( @xmath106 tev ) so that the decay is not suppressed by phase space . the mass @xmath228 of the lighter higgs boson is just above 100 gev , independent of @xmath10 . if @xmath229 tev ( smaller values of @xmath10 would have charginos and neutralinos accessible at lep 2 ) , we see from fig . [ csectiond ] that the production of charginos and neutralinos is the dominant source of sparticles at the lhc . gluinos and squarks are relatively heavy so that their production becomes rapidly suppressed as @xmath10 is increased . slepton and sneutrino production forms only a percent of the total susy cross section . as already noted , the branching fraction for the two body decay @xmath230 exceeds 50% if @xmath106 tev , and becomes @xmath23175% for @xmath232 tev @xcite . the decays @xmath47 account for essentially all the remaining decays of @xmath14 since @xmath233 with a branching fraction of at most a few if @xmath234 tev . the heavier neutralino @xmath101 as well as the lighter chargino @xmath164 decay via three body decays , but their decay products are soft since they are only a little heavier than @xmath14 . as a result , signatures for @xmath235 or @xmath236 production closely resemble those for @xmath237 production . we first consider the reach via the canonical missing @xmath66 @xcite and multilepton analyses @xcite . the results of our computation are shown in fig . [ reachmld ] . this figure corresponds to fig . [ reachmlb ] for model line b and , in fact , is qualitatively quite similar to it . despite the fact that the @xmath14 decays mainly to the light higgs boson for @xmath238 tev , the best reach , which is obtained in the @xmath239 leptons channel , extends out to @xmath240 tev corresponding to @xmath241 tev , with squarks only slightly heavier . the reach in 1 lepton and 2 os leptons channels are comparable , up to @xmath242 tev , while the zero lepton channel again gives the poorest reach . to understand this , we first note that although for these large values of @xmath10 chargino and neutralino production dominates the total cross section , gluino and squark production events ( which can be detected much more efficiently ) form a significant portion of the cross section after the selection cuts : events from the lighter chargino and @xmath243 production ( these have masses of just @xmath244 gev ) pass the hard selection cuts with very low efficiency since @xmath245 gev . gluinos decay to @xmath246 or @xmath247 with a branching fraction of 60% , with the bulk of the remaining decays into the heavier chargino and to the heaviest neutralino . squarks mainly decay to the gaugino - like @xmath248 and @xmath249 ( @xmath165 mainly decays to @xmath250 ) ; @xmath251 and @xmath249 mainly decay via two body decays to vector bosons or the lightest higgs scalar @xmath252 and a lighter chargino or neutralino , and at a few percent into left sleptons and sneutrinos . the @xmath250 , on the other hand , mostly decays via @xmath253 . for @xmath254 gev , @xmath222 and @xmath255 mainly cascade decay to @xmath250 which then mostly decays into right sleptons . notice that if @xmath222 , @xmath255 or @xmath250 are produced at any step , their decays are by themselves likely to give two hard leptons , resulting in multiple leptons from squark and gluino events . for the higgsino nlsp case , the cascade decay chains are more complicated than in the canonical msugra case , because of the dominance of gluino and squark decays to the heavier charginos and neutralinos , and in the gluino case , also to top quarks . for values of @xmath10 close to the lhc reach in fig . [ reachmld ] , the nlsp dominantly decays via @xmath230 . in this case , since @xmath252 mainly decays via @xmath256 , the susy signal will contain @xmath71-jets in addition to leptons , @xmath0 and possibly photons ( if one of the nlsps decays via the photon mode which has a branching fraction of a few percent @xcite ) . we were , therefore , led to examine whether requiring tagged @xmath71-jets in susy events can increase the reach via the observation of lighter chargino and neutralino events which have a production cross section @xmath257 @xmath158 . the dominant sm background to multi-@xmath71 plus large @xmath0 events presumably comes from @xmath141 production . we examined several event topologies as well as distributions from the signal and from the top background and concluded that the most promising channel was the one with just two tagged @xmath71-jets channel , with the following cuts to reduce the top background : 1 . @xmath258 gev ; 2 . @xmath259 gev ; 3 . @xmath260 gev , where the sum extends over untagged jets . the results of our computation are shown in table ii . we show the signal cross sections for @xmath261 and 160 tev and compare these with the top background for two choices for the @xmath71 mistagging probability . the set of columns labelled 2% assume the @xmath71 mis - tagging probability introduced in sec . ii . the columns labelled 1% assume the mis - tagging probability to be half this . the statistical significance @xmath262 is for 10 @xmath3 . we see from the table that for the canonical mis - tag rate even the @xmath263 tev case falls just short of observability but appears to pass our observability criteria with the more optimistic assumptions about @xmath71-mistagging . in view of the fact that @xmath141 is the only background we have included in our computation , and further , that the reach is only marginally increased above this background , we conclude that the canonical multilepton search is sufficient to probe model line d at the lhc , and further that the lhc reach extends out to @xmath264 tev ( corresponding to @xmath265 tev , with squarks a little heavier ) . we have also examined the @xmath266 tagged @xmath71-jet channel which without the hardness cut 3 has a much better @xmath267 ratio . the signal cross section , in this channel , is small and it is difficult to make other cuts ( to reduce the top background ) without reducing the signal to below the level of observability . it is conceivable though that with an integrated luminosity of 100 @xmath3 this channel may prove more promising ; the background from @xmath268 production would then have to be carefully examined . finally , we examine whether it is possible to identify the higgs boson , which is the characteristic feature of model line d , in susy events . it has been known for many years that the presence of higgs bosons can be inferred either via the multiplicity of @xmath71-jets in susy events @xcite , or more directly , via the reconstruction of a mass bump in the @xmath269 distribution @xcite . we follow the latter approach , and attempt to reconstruct the lighter higgs boson ( which has a mass of about 104 gev ) in susy events in channels with identified @xmath71-jets for @xmath270 tev and @xmath271 tev . again top quark pair production is the major sm background . to reduce this , we require that @xmath272 where the sum extends over untagged @xmath71-jets with @xmath273 gev . we adjust the cut variable @xmath274 depending on @xmath10 and on the event topology . in particular , we choose @xmath275 ( 900 ) gev for two ( three ) tagged @xmath71 events if @xmath276 tev , and 1500 ( 1100 ) gev if @xmath277 tev . we found that we were unable to obtain a mass bump ( with these as well as several other selection cuts that we examined ) even though the cross section was quite large . we traced this to the fact that most of the time one of the two @xmath71-jets from @xmath252 decay usually has @xmath278 gev , and so fails our tagging criteria . in other words , the two @xmath71-jets from the decay of @xmath252 are almost never simultaneously tagged . to enable the identification of @xmath252 in susy events , we include in our analysis that @xmath71-jets with @xmath66 between 25 and 50 gev may also be tagged , but with a reduced efficiency which we take to be @xcite @xmath279 . we extrapolate our previous mistagging rate down to 25 gev . the @xmath269 distribution in the two and three tagged @xmath71 channels after the cuts above is shown in fig . [ mbb_mld ] for @xmath276 and 110 tev . the dashed histograms are the @xmath141 background while the solid histograms are the sum of the signal and background . we see that there is a clear peak close to @xmath228 in the @xmath269 distributions in the two tagged @xmath71 channel . the distribution cuts off at the low end because each tagged jet has at least 25 gev , while the continuum beyond the peak corresponds to combinations where one or both @xmath71-jets come from sources other than higgs decay or are qcd jets that have been mistagged . the corresponding peaks for the three @xmath71 channel , though present , is not as distinctive , presumably because the combinatorial background also contributes to the continuum . we also examined the distribution of @xmath59 of the higgs boson reconstructed from two tagged @xmath71s with @xmath280 ; _ i.e. _ in the peak . we might expect that this distribution would scale with @xmath281 , and hence provide a measure of @xmath10 . we found , however , that the distributions for @xmath10 in the 80 - 110 tev range overlapped quite substantially , and conclude that it is difficult to measure @xmath10 this way . we attribute this to the fact that the change in @xmath281 is rather limited ( 160 - 220 gev ) , and further , that @xmath252 can also be produced at an earlier stage in the cascade , thereby smearing this distribution . for larger values of @xmath10 the event rate appears to be too low for this to be feasible . if the nlsp contains a significant fraction of the susy partner of the would - be neutral goldstone boson ( that becomes the longitudinal component of the @xmath140 by the higgs mechanism ) , it would preferentially decay via @xmath47 provided that this decay is not kinematically suppressed . it has recently been pointed out that this situation is indeed realized @xcite for a higgsino model line where the sign of @xmath27 is flipped relative to that in model line d above . we thus consider a final model line with @xmath37 , @xmath42 , @xmath226 , @xmath12 but fix @xmath283 . the decay pattern of the neutralino nlsp is shown in fig . [ nlspedecay ] . we see that for @xmath284 tev ( corresponding to an nlsp heavier than 140 gev ) the decay @xmath47 dominates , while for smaller masses , the photon decay is the main mode . search discussed at the end of this section . ] the branching fraction for the decay @xmath225 becomes significant only when @xmath10 is rather large . the main sparticle production cross sections are shown in fig . [ csectione ] which is qualitatively very similar to fig . [ csectiond ] . this should not be surprising because changing the sign of @xmath27 qualitatively affects only the nlsp decay pattern , though of course there is some effect on the chargino and neutralino masses and mixing angles . again , as in model line d , the mass gap between @xmath164 or @xmath101 and @xmath14 is not large , so that @xmath164 and @xmath101 decays are similar to those from the decay of just the nlsp because the remaining decay products tend to be soft . the reach of the lhc via the canonical @xmath0 and multilepton channels is illustrated in fig . [ reachmle ] . as before , the number on the curves denotes the @xmath124 value that optimizes the signal . again , the upper ( lower ) solid curve denotes the trilepton ( zero lepton ) plus @xmath0 channel . the main difference from fig . [ reachmld ] is that the greatest reach , which again extends out to 140 tev ( corresponding to @xmath285 tev , with squarks marginally heavier ) , is now obtained via the single lepton channel . it may seem surprising that the reach via multilepton signals degrades relative to model line d where nlsp decays are not expected to lead to isolated leptons . to understand this , we examined the decay patterns of sparticles ( for @xmath254 tev ) , and found that , except for @xmath14 , these are very similar for these two model lines . the big difference is that the branching fraction for @xmath253 which was 26% per lepton flavour for model line d is now about 19% , while there is a corresponding increase in @xmath286 decay . since the presence of @xmath250 in a cascade decay chain frequently results in two leptons ( see the corresponding discussion for model line d ) , it is not surprising that the inclusive trilepton signal is reduced relative to model line d. indeed we have checked that for @xmath287 gev , this signal , although it has a large statistical significance , falls below the five event level ( for @xmath288 tev ) , so that we are forced to lower @xmath124 to 100 gev for which the sm background @xcite is much larger . the trilepton channel would again yield the best reach . ] we now examine whether the lhc reach can be extended by using the feature that nlsp mostly decays via @xmath289 , so that a significant fraction of susy events would contain a real @xmath140 boson in addition to jets , leptons and @xmath0 @xcite . since the @xmath140 can be identified via its leptonic decay , we focus on events with an identified @xmath140 together with either two jets with @xmath290 gev or additional leptons . to further suppress backgrounds , we found it was necessary to require a large @xmath0 in these events . the signal cross sections for three values of @xmath10 just beyond the reach in fig . [ reachmle ] , together with the main sources of sm backgrounds , are shown in table iii for @xmath291 gev . here @xmath204 pairs with @xmath292 gev are identified as a @xmath140 . we see from this table that the dominant background comes from @xmath141 production where the leptons from top decay accidentally reconstruct the @xmath140 mass . we also see that with these cuts the signal falls below the observable level , though it should be noted that all these cases would be observable for an integrated luminosity @xmath293 @xmath3 . two comments are worth noting . * the signal cross section for two identified @xmath140 bosons falls below the observable level . this can even be seen from fig . [ csectione ] and fig . [ nlspedecay ] where the total susy production cross section ( @xmath294 @xmath158 ) and branching ratio ( 75% ) for @xmath47 for @xmath263 tev yields in @xmath295 events even before any acceptance cuts . * the dominant background from tops in table iii mainly contributes to the @xmath140 plus two jet channel , and is completely negligible for @xmath140 plus lepton channels . we have checked though that the @xmath140 plus two jet channel accounts for about 70 - 80% of the signal , so that the total signal in just the @xmath140 plus lepton channels again falls below the observable level . motivated by the fact that the branching fraction for @xmath296 is @xmath23110% for @xmath297 tev , we have also examined whether inclusive @xmath298 signals could extend the lhc reach . toward this end we computed sm backgrounds from @xmath299 , @xmath300 and @xmath141 production , where a photon from radiation or the decay of a hadron is accidently isolated , using isajet . we also estimated the background from @xmath298 production ( this process is not included in isajet ) by introducing the @xmath301 @xmath302 process into a private version of isajet , and ignoring spin correlations for leptons from @xmath140 decay . we search for events in the inclusive @xmath298 channel where the @xmath140 is identified by its leptonic decay and the isolated photon has @xmath303 gev . we also require @xmath200 gev . this last cut completely eliminates the physics background from @xmath298 events ( which essentially cuts off by @xmath304 gev ) and reduces backgrounds from other sources . the results of our calculation are summarized in table iv . we see that the physics backgrounds that we have computed are completely negligible . we expect though that the dominant background could well come from instrumental effects when a jet or lepton is misidentified as a photon . although a calculation of this detector - dependent background is beyond the scope of the present analysis , we attempt to obtain a rough estimate by computing the inclusive @xmath140 background for @xmath200 gev and then multiplying this by the probability for a jet to fake a photon . we found this inclusive @xmath140 background cross section to be 192 @xmath158 from @xmath141 production , 36.7 @xmath158 from @xmath299 production and 10.2 @xmath158 from @xmath300 production , giving a total of 239 @xmath158 . assuming that the probability for a jet to fake a photon is @xmath305 @xcite , and further that there are typically @xmath114 ( 10 ) jets with @xmath306 gev in these lhc events , we obtain the `` fake @xmath307 '' background cross section of @xmath308 @xmath158 . we see from table iv that this reducible background can potentially be comparable to the signal . furthermore , since this estimate does not include the effects of cracks and other real - world detector effects , it is likely to be increased once these effects are incorporated . we conclude that _ unless these reducible backgrounds can be controlled _ it is unlikely that the inclusive @xmath298 channel will increase the reach . but with excellent jet-@xmath309 rejection , the lhc reach for model line e could be as large as @xmath310 tev . moreover , if the signal is truly rate - limited ( which does not seem to be the case with our assumptions about the instrumental backgrounds ) , the reach would grow significantly with the size of the data sample . the cern lhc is generally considered as the accelerator facility at which weak scale supersymmetry will either be discovered or definitely excluded . this is because the lhc experiments are expected @xcite to probe @xmath1 out to 2 tev , at least within the msugra framework . this extends well beyond the generally accepted bounds from naturalness considerations . moreover , experiments at the lhc should be able @xcite to probe gluinos and squarks beyond 1 tev even in the experimentally unfavourable scenario where @xmath311-parity violating interactions cause the neutralino lsp to decay hadronically , thereby reducing the @xmath0 as well as degrading the lepton isolation in susy events . although not in themselves a `` proof '' , it is precisely such studies that have led to the general belief that the lhc will be the definitive machine that will test susy . in this study , we have added to this evidence by examining the susy reach of the lhc within the gmsb framework with susy broken at a scale of a few hundred tev . in this case , the gravitino is then the lsp and susy signatures are governed by the nature of the nlsp which then decays into the gravitino and sm particle(s ) . in our analysis , we conservatively assume that the nlsp decay is prompt , and do not attempt to use the possible presence of a displaced vertex to enhance the susy signal . we divide our study along various model lines introduced in section i. for each of these model lines , the nlsp ( which is almost always produced as the penultimate product of the sparticle decay cascade ) decays via characteristic modes , which can then be used to enhance the susy signal , or even to narrow down the model parameters . we do not attempt the latter in the present study . for model line a which was studied in ref . @xcite , for instance , @xmath87 , so that the presence of hard , isolated photons in susy events served to distinguish signal from sm backgrounds , resulting in a susy reach that corresponding to @xmath312 tev , to be compared with a reach of about 2 tev expected within the msugra framework . in this paper we have examined the lhc reach for four other model lines corresponding to ( b ) the stau nlsp scenario , ( c ) the co - nlsp scenario where all three flavours of light charged sleptons are essentially degenerate , and two higgsino - like nlsp scenarios where nlsp decays to ( d ) the light higgs boson and ( e ) the @xmath140 boson , dominate . the usual strategy for susy searches via @xmath0 and multilepton events will probe gluino masses up to about 2 tev , except in the co - nlsp model line where the fact that selectrons and smuons are produced in many cascade decay chains leads to a plethora of @xmath0 events with @xmath313 isolated leptons ( @xmath26 or @xmath27 ) which have very small sm backgrounds . in this case the reach of the lhc extends to @xmath314 tev ! for model lines b and d , however , we found that whereas the decays of the nlsp may result in events that are characteristic of the particular model line , signals with identified @xmath149 leptons or tagged @xmath71 jets do not , in general , extend the susy reach beyond what is obtained via standard multilepton analyses . for model line e , some enhancement of the reach might be possible via the inclusive @xmath298 channel which is rate - limited , but only if superb jet-@xmath309 discrimination is possible . nevertheless , it should be kept in mind that it will be worthwhile to search for these characteristic events in any lhc data sample enriched in susy events , as they will point to the underlying theory , and possibly also lead to a measurement of some of the model parameters @xcite . this research was supported in part by the u. s. department of energy grants de - fg02 - 97er41022 and de - fg03 - 94er40833 . during his stay at the university of hawaii where this work was begun , p.m. was partially supported by fundao de amparo pesquisa do estado de so paulo ( fapesp ) . @xmath315 & @xmath316 & @xmath317 & @xmath318 & @xmath315 & @xmath319 @xmath49 & 1.75 & 1.55 & 0.62 & 0.02 & 0.16 & 0.21 & 0.09 & 0.01 @xmath320 & 5.55 & 4.37 & 1.24 & 0.16 & 0.69 & 0.75 & 0.26 & 0.04 @xmath321 & 2.09 & 1.47 & 0.27 & 0.00 & 0.30 & 0.29 & 0.09 & 0.02 @xmath322 & 1.56 & 1.19 & 0.39 & 0.01 & 0.31 & 0.23 & 0.09 & 0.01 @xmath323 & 4.23 & 2.08 & 0.34 & 0.00 & 0.75 & 0.51 & 0.13 & 0.01 @xmath324 & @xmath325(2% ) & @xmath326 & @xmath327 & @xmath325(1% ) 140 & 2.11 & 0.89 & 7.09 & 2.05 & 0.72 & 7.62150 & 1.34 & 0.89 & 4.49 & 1.36 & 0.72 & 5.36 160 & 0.70 & 0.89 & 2.34 & 0.73 & 0.72 & 2.72 .the susy signal cross sections ( in @xmath158 ) for inclusive @xmath140 plus @xmath328 gev events for model line e together with sm backgrounds from @xmath299 , @xmath300 and @xmath141 production . for the case of the @xmath141 background , the ` @xmath140 ' is a fake from two leptons accidently reconstructing the @xmath140 mass . the signal cross sections are shown in the first three columns for @xmath261 and 160 tev along with the statistical significance for an integrated luminosity of 10 @xmath3 . [ cols="^,^,^,^,^,^,^ " , ] v. ruhlmann - kleider , _ plenary talk presented at the 1999 lepton - photon symposium _ , slac , stanford , hep - ex/0001061 ( 2000 ) ; a. savoy - navarro , in _ proc . society meeting _ , tampere , finland , july ( 1999 ) , fermilab - conf-99/281-e ( 1999 ) . h. baer , m. daz , p. quintana and x. tata , hep - ph/0002245 ( 2000 ) . m. dine and a. nelson , phys . * d48 * , 1277 ( 1993 ) ; m. dine , a. nelson , y. shirman , phys . rev . * d51 * , 1362 ( 1995 ) ; m. dine , a. nelson , y. nir and y. shirman , phys . rev . * d53 * , 2658 ( 1996 ) . s. dimopoulos , m. dine , s. raby and s. thomas , phys . rev . lett . * 76 * , 3494 ( 1996 ) ; s. dimopoulos , s. thomas and j. wells , phys . * d54 * , 3283 ( 1996 ) and nucl b488 * , 39 ( 1997 ) ; s. ambrosanio _ et al . * 76 * , 3498 ( 1996 ) and phys . rev . * d54 * , 5395 ( 1996 ) ; k. s. babu , c. kolda and f. wilczek , phys . lett . * 77 * , 3070 ( 1996 ) ; h. baer , m. brhlik , c. h. chen and x. tata , phys . rev . * d55 * , 4463 ( 1997 ) ; j. bagger , k. matchev and d. pierce , phys . rev . * d55 * , 3188 ( 1997 ) ; r. rattazzi and u. sarid , nucl . phys . * b501 * , 297 ( 1997 ) ; d. dicus , b. dutta and s. nandi , phys . lett . * 78 * , 3055 ( 1997 ) and phys . * d56 * , 5748 ( 1997 ) ; k. cheung , d. dicus , b. dutta and s. nandi , phys . rev . * d58 * , 015008 ( 1998 ) ; b. dutta , d. j. mueller and s. nandi , nucl . phys * b 544 * , 451 ( 1999 ) ; d. j. mueller and s. nandi , phys . rev . * d60 * , 015008 ( 1999 ) ; s. martin and j. wells , phys . rev . * d59 * , 035008 ( 1999 ) ; s. ambrosanio and eur . j. phys . * c12 * , 287 ( 2000 ) ; see g. giudice and r. rattazzi , phys . rep . * 322 * , 419 ( 1999 ) for a review . h. baer , p.g . mercadante , x. tata and y. wang , phys . lett . * b435 * , 109 ( 1998 ) . h. baer , p.g . mercadante , x. tata and y. wang , phys . * d60 * , 55001 ( 1999 ) . f. paige and i. hincliffe , phys.rev . * d60 * , 095002 ( 1999 ) . k. matchev and s. thomas , hep - ph/9908482 ( 1999 ) . b. abbott _ et al . * 80 * , 442 ( 1998 ) ; f. abe _ et al . lett . * 81 * , 1791 ( 1998 ) and phys . rev . * d59 * , 092002 ( 1999 ) . r. barate _ b429 * , 201 ( 1998 ) and cern - ep/99 - 171 ( 1999 ) ; m. acciarri _ et al . * b470 * , 268 ( 1999 ) . h. baer , c. chen , f. paige and x. tata , phys . rev . * d52 * , 2746 ( 1995 ) . h. baer , c. chen , f. paige and x. tata , phys . rev . * d53 * , 6241 ( 1996 ) . h. baer , f. paige , s. protopopescu and x. tata , hep - ph/9810440 ( 1998 ) . atlas technical proposal , cern / lhcc/94 - 43 ( 1994 ) . cms technical proposal , cern / lhcc 94 - 38 ( 1994 ) . s. abdullin _ et al . _ cms note 1998/006 , hep - ph/9806366 _ , ( cteq collaboration ) , phys . rev . d*51 * , 4763 ( 1995 ) . v. barger , a. stange and r.j.n . phillips , phys . rev . * d44 * , 1987 ( 1991 ) . v. barger , t. han and h. pi , phys . rev . * d41 * , 824 ( 1990 ) . v. barger and t. han phys . lett . * b212 * , 117 ( 1988 ) . h. baer , c. chen , f. paige and x. tata , phys . rev . * d49 * , 3283 ( 1994 ) . h. baer , m. bisset , x. tata and j. woodside , phys . rev . * d46 * , 303 ( 1992 ) . i. hinchliffe , f. paige , m. shapiro , j. soderqvist and w. yao , phys . rev . * d55 * , 5520 ( 1997 ) . h. baer , x. tata and j. woodside , phys . rev . * d42 * , 1450 ( 1990 ) . f. gianotti , private communication . h. baer , c. chen and x. tata , phys . rev . * d55 * , 1466 ( 1997 ) .

ballistic transport through a collection of quantum billiards in undoped graphene is studied analytically within the conformal mapping technique . the billiards show _ pseudodiffusive _ behavior , with the conductance equal to that of a classical conductor characterized by the conductivity , and the fano factor . by shrinking at least one of the billiard openings , we observe a _ tunneling _ behavior , where the conductance shows a power - law decay with the system size , and the shot noise is poissonian ( ) . in the crossover region between tunneling and pseudodiffusive regimes , the conductance . the degeneracy for the corbino disk , which preserves the full symmetry of the dirac equation , for billiards bounded with smooth edges which break the symplectic symmetry , and when abrupt edges lead to strong intervalley scattering . an alternative , analytical or numerical technique , is utilized for each of the billiards to confirm the applicability of the conformal mapping for various boundary conditions . = 1

0906.1393

the isolation of single layers of carbon ( graphene ) whose low - energy spectrum is described by the dirac - weyl hamiltonian of massless spin-@xmath7 fermions @xcite , has offered physicists the unique possibility to test the predictions of relativistic quantum mechanics in a condensed phase . a particular attention focuses on ballistic transport @xcite , as the unusual band structure of a carbon monolayer @xcite leads simultaneously to a divergent fermi wavelength @xmath8 in the undoped graphene limit , and to a zero bandgap . for these reasons , the quantum - mechanical wave character of an electron plays an essential role in transport even through a macroscopic graphene sample , provided that the influence of disorder is negligible @xcite . a separate issue concerns the fact that dirac fermions in graphene occur in two degenerate families , resulting from the presence of two different valleys in the band structure . this valley degree of freedom offers conceptually new possibilities to control charge carriers the so - called `` valleytronics '' @xcite . so far , extensive theoretical studies of ballistic transport , based on mode - matching analysis for the dirac equation @xcite , are available for a _ rectangular _ graphene sample of width @xmath9 , length @xmath10 , and various types of boundary conditions . in the regime of large aspect ratios @xmath11 , the conductance of an undoped sample scales as @xmath12 , with the universal conductivity @xmath0 , regardless of boundary conditions @xcite . moreover , as shown by tworzydo _ @xcite , the fano factor in this case coincides with that of a diffusive wire ( @xmath1 ) . also , the transmission eigenvalues of these two systems display the same distribution . this analogy coined the term of _ pseudodiffusive transport _ , which describes ballistic graphene properties in the universal conductivity limit . recent experiments report an agreement with the theoretical predictions of refs . @xcite for either the conductance @xcite or the fano factor @xcite . furthermore , the temperature dependence of the conductivity @xcite also shows an approximate agreement with the ballistic theory generalized to finite temperatures @xcite . however , even for low temperatures , the convergence with @xmath13 is much slower than predicted . in particular , for the largest aspect ratio @xmath14 studied in ref . @xcite the deviations from the limiting values @xmath15 and @xmath1 are close to @xmath16 , whereas results of ref . @xcite show the convergence should be already reached for moderate aspect ratios @xmath17 . a clear explanation of this discrepancy is missing , but it is usually attributed to the fact that boundary conditions used in theoretical works , which describe either an abrupt termination of a perfect honeycomb lattice or an infinite mass confinement @xcite , may not model the real - sample edges correctly @xcite . in this work , we consider ballistic graphene systems of geometries for which the boundary effects are absent or suppressed . the paper is organized as follows : in sec . [ conmap ] we briefly recall the mode - matching analysis for a graphene strip , and show how to employ the conformal symmetry of the dirac equation for undoped graphene @xcite to obtain analytically the transmission eigenvalues for other systems . then , in sec . [ corbdi ] the method is applied to the corbino disk . the results are compared with those obtained by direct mode - matching for angular momentum eigenstates , a relation with the nonrelativistic electron gas in the disk setup is also discussed . in sec . [ bimass ] we study two basic billiards bounded with mass confinement : a finite section of the corbino disk and a quantum dot with circular edges . the results obtained with the conformal mapping technique are confirmed by the computer simulation of transport using the tight - binding model on a honeycomb lattice . finally , in sec . [ ribb ] we study numerically the transport _ across _ an infinitely long nanoribbon by utilizing the @xmath18-terminal recursive green s function algorithm @xcite , as well as across a finite section of a nanoribbon with an abrupt lattice termination . all the systems show pseudodiffusive transport behavior in a wide range of geometrical parameters . a further analogy between them appears when ( at least ) one of the leads is narrow in comparison to the characteristic length of the conducting region @xmath19 . namely , the conductance in such a _ quantum tunneling _ regime shows a power - law decay @xmath20 , where @xmath21 is a nonuniversal ( geometry - dependent ) exponent . moreover , it is related to the shot noise by @xmath22 , so that the poissonian limit ( @xmath2 ) is approached for large @xmath19 . the symmetry - dependent factor @xmath4 in the presence of full spin , valley and symplectic degeneracy ( the case of the corbino disk ) , or @xmath5 when the mass confinement breaks symplectic symmetry of the dirac equation . a final reduction to @xmath6 may be reached by adding abrupt ( i.e. armchair ) edges , which scatter the valleys . the original feature of the geometries studied in this article is that the influence of sample edges are eliminated ( for the corbino setup ) or irrelevant , as the spatial current distribution is not uniform , but concentrated far away from the edges . this is why we believe that our theoretical findings could be confirmed experimentally with better precision than that for rectangular samples , as the role of boundary conditions is strongly suppressed . quantum billiards in undoped graphene studied analytically ( schematic ) . ( a ) the corbino disk with inner radius @xmath23 , and outer radius @xmath24 . ( b ) generic section of the disk characterized by the spanning angle @xmath25 . ( c ) quantum dot with circular edges . a voltage source , shown on panel ( a ) only , drives the current through each of the devices . shadow areas on all panels mark heavily - doped ( so highly - conducting ) graphene leads , white dots are the poles of conformal transformation mapping a given system onto a strip of fig . [ interf12 ] . thick lines on panels ( b , c ) indicate infinite - mass confinement . ] the compact derivation of transmission eigenvalues of a weakly - doped ( or undoped ) graphene sample coupled to heavily - doped graphene leads is known due to sonin @xcite , who pointed out that one can first calculate the reflection and transmission amplitudes for an interface between weakly - doped and heavily - undoped regions , and then employ the double - contact formula @xcite . here we show that the derivation of ref . @xcite can be easily adopted to the corbino disk , a finite section of it , and to a quantum dot with circular edges ( all shown in fig . [ corbcido ] ) , as these systems can be obtained from a strip by appropriate conformal transformations . scattering on interfaces ( 1 ) and ( 2 ) between weakly - doped ( white area ) and heavily - doped ( shaded area ) regions in graphene . horizontal dashed - dot lines mark symbolically generic boundary conditions applied to a strip . ] let us first consider an electron crossing from the weakly - doped region ( @xmath26 ) to the heavily - doped one ( @xmath27 ) , as depicted in fig . [ interf12 ] . the dirac hamiltonian for graphene has the well - known form @xcite @xmath28 where @xmath29 is the fermi velocity , @xmath30 is the vector operator build of pauli matrices for the sublattice - pseudospin degree of freedom , and @xmath31 is the in - plane momentum operator . due to translational invariance along the @xmath32-axis , the solution of the dirac equation with energy @xmath33 may be written as @xmath34 @xcite , with the transverse momentum @xmath35 ( @xmath36 and @xmath37 with @xmath38 denote momentum components in the weakly- and heavily - doped regions , respectively ) , and the spinor @xmath39 where @xmath40 , and the limit of an infinite doping ( @xmath41 ) is imposed . the continuity of the two spinor components on both sides of the interface leads to expressions for the reflection and transmission amplitudes @xmath42 the amplitudes @xmath43 and @xmath44 depend solely on the angle of incidence @xmath45 ( see fig . [ interf12 ] ) , what illustrates the generic feature of ballistic transport in graphene that is insensitive to the lead details @xcite . the reflection and transmission amplitudes for an electron crossing from the undoped region to the second heavily - doped lead are @xmath46 and @xmath47 ( up to a phase factor ) , as the angle of incidence @xmath48 in this case . thus , employing the double - contact formula of ref . @xcite the total transmission probability for phase - coherent transport through the system of fig . [ interf12 ] is @xmath49 where @xmath50 is the phase - shift earned by an electron when passing from the first interface to the second one . the above expression holds true for either propagating modes ( for which @xmath51 ) or , as an analytic continuation , for evanescent modes ( @xmath52 , with @xmath53 ) . for a confined geometry , the quantization of transverse momenta is determined by boundary conditions @xcite . namely , @xmath54 ( with @xmath55-integer ) may be written in a compact form @xmath56 where @xmath57 for the closed and ( generalized ) periodic b.c . , respectively . @xmath58 for either mass confinement or antiperiodic b.c . studied in this paper . ( for a nanotube - like geometry as considered in ref . @xcite , @xmath59 corresponds to periodic b.c . ) here we limit ourselves to the case of undoped graphene ( @xmath60 ) , in which the charge transport is carried fully by evanescent modes . an analytic continuation yields @xmath61 and @xmath62 in eq . ( [ tthphi ] ) , where we use the quantization ( [ kyquant ] ) and define the half - integer @xmath63 . as pointed out by katsnelson and guinea @xcite , the zero - energy solution of the dirac equation may be obtained via conformal transformation that links the considered geometry to a simple one , for which the wavefunction is known @xcite . in particular , if the conformal transformation @xmath64 turns the system under consideration into a rectangle of width @xmath9 and length @xmath10 ( fig . [ interf12 ] ) , the transmission probability for the @xmath65-th evanescent mode may be written as @xmath66 } = \frac{4}{\left(\lambda^{gj } + \lambda^{-gj}\right)^2},\ ] ] where @xmath67 ( with the degeneracy @xmath68 ) . notice that the amplitudes ( [ rt1th ] ) remain unchanged after applying an arbitrary conformal transformation to the coordinate system of fig . [ interf12 ] , so the only term in eq . ( [ tthphi ] ) affected by the transformation @xmath64 is the phase - shift @xmath69 . the real functional @xmath70 is defined by @xmath71 the explicit form of @xmath70 depends on the geometry , and is given below for the examples of conformal transformation @xmath64 having one and two poles in a complex plane , which allows us to obtain expressions for transmission probabilities through a finite section of the corbino disk and through a quantum dot with circular edges , respectively . but first , we discuss the two basic physical regimes of quantum transport in graphene , which are described by opposite limits of eq . ( [ tjlambda ] ) . the conductance of undoped graphene @xcite is given by the landauer formula @xmath72 with the degeneracy @xmath5 ( spin and valley ) for smooth mass confinement , and @xmath4 for antiperiodic b.c . due to an additional ( symplectic ) symmetry @xcite . the universal conductivity is @xmath73 . the fano factor also follows from summing over the modes @xmath74 but is affected by the symmetry - dependent factors @xmath75 only via @xmath76-s ( [ tjlambda ] ) . for the limit @xmath77 , we can replace summation in eq . ( [ gsumtj ] ) by integration , and get @xmath78 where we use the relation @xmath79 , valid for the two classes of b.c . studied here . in the @xmath77 limit , the relevant information about transmission probabilities is given by their statistical distribution @xmath80 as the distribution @xmath81 coincides with the known distribution @xcite for diffusion modes in a disordered metal , @xmath77 constitutes a _ pseudodiffusive _ regime of transport through graphene billiards . notice that the generic conformal transformation @xmath64 affects @xmath81 only via the prefactor @xmath82 . in particular , the fano factor @xmath83 regardless of the particular form of @xmath64 . this observation may also help to understand why experimental results @xcite generally show better agreement with theory for the fano factor rather than for the conductance . for instance , various geometrical defects ( such as a corrugation of the lead - graphene interface ) may affect @xmath82 strongly , but not affect @xmath84 at all . in the opposite limit ( @xmath85 ) , we find from ( [ tjlambda ] ) that @xmath86 , leading to @xmath87 these expressions constitute a _ quantum - tunneling _ regime for ballistic graphene , in which the transport is governed by a single electronic mode with the four - fold ( spin and valley ) degeneracy . below , we provide examples illustrating how the power - law dependence of @xmath88 on @xmath89 may be followed by a power - law decay of @xmath88 with the characteristic length - scale of the system . the corbino setup , in which the graphene sample formed as an annulus is attached to coaxial leads , as shown schematically in fig . [ corbcido](a ) , seems to be the simplest way to eliminate boundary effects , which are claimed to strongly affect experimental results for rectangular samples with small and moderate aspect ratios @xcite . in this section , we first utilize the conformal mapping technique to find transmission eigenvalues for an _ undoped _ disk , and then compare the results with that obtained by a direct wave - function matching , possible also for a _ doped _ disk . conductance and fano factor for the undoped corbino disk in graphene , as a function of the radii ratio ( a , b ) and the shot noise vs conductance diagram ( c ) . the curves calculated from eqs . ( [ gsumtj ] ) and ( [ fsumtj ] ) are plotted with solid lines on all panels . dashed lines show the pseudodiffusive limit ( [ gfdiffdsk ] ) on panels ( a , b ) and the tunneling limit ( [ gftunndsk ] ) on panel ( c ) . ] the conformal transformation that changes the corbino disk with the inner radius @xmath23 and the outer radius @xmath24 , shown in fig . [ corbcido](a ) , into a rectangle of the width @xmath9 and the length @xmath10 ( see fig . [ interf12 ] ) , is given by @xcite @xmath90 ( hereinafter , we use the symbol @xmath91 to denote the natural logarithm in a complex domain . ) for the complex variable @xmath92 , with @xmath93 , and @xmath94 , transformation ( [ contradsk ] ) leads to @xmath95 and @xmath96 provided the condition @xmath97 is satisfied . using ( [ lambdadef ] ) , such a condition implies the functional @xmath70 to have the form @xmath98 as the conformal mapping is known , the only part to be explained now are the boundary conditions applied to a strip of fig . [ interf12 ] . to define them , one needs to notice that after a rotation by @xmath99 in the coordinate system of fig . [ corbcido](a ) , the spinor part of the wavefunction acquires the berry phase @xcite @xmath100 . within the mapping ( [ contradsk ] ) , a rotation by @xmath99 turns into a shift along the @xmath32-axis in fig . [ interf12 ] by a strip width @xmath9 . this is why the spinor - rotational invariance of the original wavefunction implies antiperiodic boundary conditions @xmath101 for a strip . such boundary conditions , together with the functional @xmath70 given by ( [ lacontradsk ] ) , lead the formula ( [ tjlambda ] ) for transmission probabilities to a form @xmath102},\ \ \ \ \ \ j={\textstyle\frac{1}{2},\frac{3}{2},\frac{5}{2},\dots}.\ ] ] a generalization for the setup with circular , but not coaxial contacts is presented in appendix [ appcir ] . the dependence of the conductance ( [ gsumtj ] ) and the fano factor ( [ fsumtj ] ) on the radii ratio @xmath103 is plotted in fig . [ gfdiscr ] ( solid lines ) . the limiting behavior for @xmath104 , corresponding @xmath77 ( [ lacontradsk ] ) is characterized by @xmath105 ( [ gdifflam ] ) , with @xmath106 the formula for @xmath82 coincides with the well - known classical conductance of the corbino disk @xcite . the asymptotic values ( [ gfdiffdsk ] ) are depicted with dashed red lines on fig . [ gfdiscr](a , b ) . in the opposite limit ( @xmath107 ) , eq . ( [ gftunnlam ] ) takes the form @xmath108 the second formula from above is shown in fig . [ gfdiscr](c ) with dashed black line . the results presented in fig . [ gfdiscr ] show that the pseudodiffusive formulas ( [ gfdiffdsk ] ) for @xmath88 and @xmath84 match the exact expressions ( [ gsumtj ] ) and ( [ fsumtj ] ) with @xmath76 given by ( [ tjlambdadsk ] ) in a relatively wide range of ratios @xmath103 . namely , the agreement becomes better than @xmath109 if @xmath110 for the conductance , and if @xmath111 for the fano factor . for smaller @xmath103 , one can identify the crossover from the pseudodiffusive to quantum - tunneling behavior . in particular , the exact values of @xmath88 are closer to the tunneling formula ( [ gftunndsk ] ) than to the pseudodiffusive formula ( [ gfdiffdsk ] ) below @xmath112 . the same is observed for @xmath84 below @xmath113 . the most characteristic feature of the tunneling regime is the relation @xmath114 , following from ( [ gftunndsk ] ) . it is satisfied with an accuracy better than @xmath16 for @xmath115 ( or @xmath116 ) , corresponding to @xmath117 . in this range , we also find that the conductance decays ( at fixed @xmath23 ) as @xmath118 , where @xmath119 is the characteristic length of the sample area . large graphene flake probed by two circular leads of radius @xmath120 separated by the distance @xmath55 ( a ) , and by a lead placed in the distance @xmath121 from a straight interface between undoped and heavily - doped regions ( b ) . ] a similar , power - law decay of the conductance with the sample length is predicted for geometries with non - coaxial leads , considered in appendix [ appcir ] . in the two limiting situations , the mbius transformation ( [ mobius ] ) maps an infinite plane ( hemiplane ) with two ( one ) narrow circular openings onto the corbino disk . physically , these two situations correspond to the setup consisting of two circular leads probing a large graphene sample ( see fig . [ flak2prob]a ) , and of one circular lead and a long straight interface between the undoped and the heavily - doped region playing the role of a second lead ( see fig . [ flak2prob]b ) . in the first case , the mapping ( [ mobius ] ) leads to @xmath122 and , subsequently , to the quadratic decay of the conductance @xmath123 ( with the radius of each lead @xmath120 and the distance between leads @xmath55 ) . in the second case , the functional @xmath124 and the conductance @xmath125 shows reciprocal decay with the sample area length , similarly as observed for the corbino disk . the approximate relation between the conductance and the fano factor @xmath114 holds true for both situations of fig . [ flak2prob ] , showing the tunneling - transport regime in graphene appears generically for a setup consisting of ( at least ) one narrow circular lead . chemical potential dependence of the conductance ( a ) and fano factor ( b , c ) at a fixed radii ratio @xmath103 . solid and dashed lines on panels ( a , b ) correspond to @xmath126 and @xmath127 , respectively . the dotted line on panel ( a ) is the semi - classical approximation for the conductance . panel ( c ) shows the fano factor as a function of the chemical potential in the units of @xmath128 , with @xmath103 specified for each curve on the plot . ] we complement the study of the corbino disk in graphene with its transport properties at finite doping , characterized by the chemical potential @xmath129 ( where @xmath130 and @xmath131 refers to electron and hole doping , respectively ) . the analysis is closely related to that for the electronic levels of graphene rings @xcite . the single valley hamiltonian for the doped disk reads @xmath132 where @xmath133 is given by ( [ dirham ] ) , the electrostatic potential @xmath134 if @xmath135 , and @xmath136 otherwise . the chemical potential @xmath137 in the disk , or @xmath138 in the leads . the rotational invariance of the problem allows us to perform the mode - matching for each eigenstate of the total angular momentum @xmath139 ( with @xmath140 the orbital angular momentum ) separately . the eigenstate of the hamiltonian ( [ dirhamcir ] ) corresponding to the @xmath65-th eigenvalue of @xmath141 can be written as @xmath142 where @xmath65 is a half - odd integer @xmath67 . for the electron doping ( @xmath143 ) , the radial components @xmath144^t$ ] for the incoming and outgoing waves are given ( up to the normalization ) by @xmath145 where @xmath146 is the hankel function of the ( first , second ) kind , and the dimensionless radial coordinate is @xmath147 ( so @xmath148 in the disk and @xmath149 in the leads , with @xmath150 ) . the radial current density is @xmath151 , where the upper ( lower ) sign is for @xmath152 ( @xmath153 ) , @xmath154 , and we use the identity @xmath155= 2/(\pi\rho)$ ] . for the hole doping ( @xmath156 ) , the wavefunctions are @xmath157^{\star}$ ] , where we use the relation @xmath158^{\star}$ ] . the transmission and reflection amplitudes are obtained by wavefunction matching at @xmath159 and @xmath160 . ( note that the @xmath161 dependence of the spinor ( [ jeigen ] ) plays no role for the mode - matching analysis . ) details of the calculations are given in appendix [ appmod ] . for @xmath162 ( the heavily - doped leads limit ) , the transmission probability for the @xmath65-th mode @xmath163 reads @xmath164 with @xmath165 . \mathfrak{d}_j^\pm=\mbox{im}\left[h_{j-1/2}^{(1)}(kr_1)h_{j\mp 1/2}^{(2)}(kr_2)\right . \\ h_{j+1/2}^{(1)}(kr_1)h_{j\pm 1/2}^{(2)}(kr_2)\right].\end{gathered}\ ] ] eqs . ( [ gsumtj],[fsumtj ] ) for @xmath88 and @xmath84 remain unchanged , since we again observe the symmetry @xmath166 . in addition , the particle - hole symmetry @xmath167 allows us to limit the discussion to @xmath130 . numerical values for the conductance and fano factor of the doped disk are presented in fig . [ gfdisckr12 ] . following the idea of kirczenow @xcite , we compare ( in fig . [ gfdisckr12]a ) the exact quantum conductance given by eqs . ( [ gsumtj ] ) and ( [ tjmodma ] ) with the semiclassical approximation for large angular momenta @xmath168 where @xmath169 is the maximal value of @xmath65 such that @xmath170 . surprisingly , the quantization steps of @xmath171 ( dotted black line ) are missing in the actual data even for an extremely small radii ratio ( solid blue and dashed red line for @xmath126 and @xmath127 , respectively ) . instead , weak modulation with a period @xmath172 is observed when varying @xmath173 . earlier , conductance quantization ( with the steps of @xmath174 ) was predicted to appear for a graphene strip with a moderate aspect ratio @xmath175 @xcite . the quantization with the steps of @xmath176 was found theoretically for a bipolar junction in graphene , which shows the goos - hnchen effect @xcite . the lack of conductance quantization observed here for the corbino disk shows the role of evanescent modes showing a slow ( power - law ) decay with distance is crucial also far away from the dirac point , what exhibits a striking consequence of the angular - momentum conservation . similar to the strip geometry @xcite , the conductance minimum at @xmath177 corresponds to the maximum of the fano factor ( see figs . [ gfdisckr12]b ) . the peak width shrinks approximately as @xmath178 ( for more datasets , plotted as a function of @xmath179 , see fig . [ gfdisckr12]c ) . from an analytical treatment of the limit @xmath180 for angular - momentum eigenstates ( see appendix [ appmod ] ) , we find that eq . ( [ tjlambdadsk ] ) for @xmath76 , obtained within the conformal mapping technique , is reproduced . chemical potential dependence of the conductance ( a ) and the fano factor ( b ) for the corbino disk in a 2deg . different lines in each panel correspond to different values of the radii ratio : @xmath126 ( solid blue lines ) , @xmath181 ( dashed red lines ) , and @xmath182 ( black dash - dotted lines ) . the electrostatic potential step is fixed at @xmath183 . ] as the corbino disk containing dirac fermions , described by the hamiltonian ( [ dirhamcir ] ) , has not been studied in the literature yet , a comparison with the corresponding schrdinger system is desirable for the sake of completeness . the existing theoretical works @xcite , however , focus on the model with a special , angular - momentum dependent effective potential which simplifies the analysis , but makes a relation to the dirac system studied here unclear . for this reason , we now present a mode - matching analysis for two - dimensional nonrelativistic electron gas ( 2deg ) arranged in a corbino setup with the same potential profile as applied to dirac fermions earlier in this paper . the schrdinger equation for the corbino disk in a 2deg reads @xmath184\psi = e\psi,\ ] ] where @xmath185 is the effective mass , and the electrostatic potential @xmath186 is chosen identically as in the hamiltonian ( [ dirhamcir ] ) . the solutions are written in the form of orbital - momentum @xmath187 eigenstates @xmath188 with @xmath55 integer , and the radial wavefunction @xmath189 a complex scalar . the propagating modes in the leads exist only for @xmath190 , and have the form @xmath191 and @xmath192 , where @xmath193 , and we assume scattering from the outer lead . for the disk area , two linearly independent solutions are given by @xmath194 and @xmath195 ( with @xmath196 ) for @xmath130 . otherwise , the solutions are given by modified bessel functions @xmath197 and @xmath198 . the mode - matching analysis is carried out separately for each value of @xmath55 @xcite , leading to the transmission propabilities @xmath199 where @xmath200 @xmath201\mathcal{c}_l^{(j)}(kr)\\-(k / k)h_{l}^{(i)}(kr)\left[\mathcal{c}_{l-1}^{(j)}(kr)\mp\mathcal{c}_{l+1}^{(j)}(kr)\right],\end{gathered}\ ] ] with @xmath202 , and the upper ( lower ) sign corresponding to @xmath130 ( @xmath131 ) . we further define @xmath203 with the step function @xmath204 for @xmath26 or @xmath205 otherwise . numerical values of the conductance and the fano factor following from eq . ( [ ttl2deg ] ) are presented in fig . [ gf2deg ] @xcite for a large but finite value of the doping in the leads , adjusted such that @xmath206 . both @xmath88 and @xmath84 are plotted as functions of @xmath207 for fixed values of the radii ratio @xmath126 , @xmath181 and @xmath127 ( solid , dashed , and dash - dot lines , respectively ) ; @xmath88 is additionally rescaled by a factor @xmath208 to illustrate its asymptotic behavior for @xmath209 , which is insensitive to the ratio @xmath103 @xcite . we also limit the discussion to @xmath130 , as the propabilities @xmath210 given by eq . ( [ ttl2deg ] ) decay rapidly for @xmath131 , due to lack of propagating modes in the sample area . although the values of @xmath88 shown in fig . [ gf2deg](a ) are close to the semiclassical result @xcite , the quantization steps are absent in the data . instead , we observe fabry - prot oscillations with the amplitude increasing with @xmath211 , for either @xmath88 or @xmath84 . we attribute the conductance quantization , reported in earlier works @xcite to the particular choice of the effective radial potential ( note that the existing experiments for the corbino disk in a 2deg @xcite found no conductance quantization ) . the main difference in transport through the corbino disk geometry between massless fermions in graphene and massive fermions in a 2deg , is the reduced backscattering at the contacts and the absence of details of the leads ( i.e. the doping ) in the former case @xcite . this is a direct consequence of the energy - independent velocity in graphene , which is also responsible for the klein - tunneling phenomena @xcite . moreover , we note a suppression of the fabry - prot oscillations for the relativistic system . a simple generalization of the formula ( [ contradsk ] ) leads to the conformal transformation that changes a finite section of the corbino disk with the inner radius @xmath23 , the outer radius @xmath24 , and the spanning angle @xmath25 ( shown in fig . [ corbcido]b ) into a rectangle of the width @xmath9 and the length @xmath10 , which is given by @xmath212 for @xmath92 , where @xmath93 and @xmath94 , we get @xmath95 and @xmath213 ( with @xmath214 ) , under the condition that @xmath215 . using ( [ lambdadef ] ) , such a condition leads to the functional @xmath70 in the form @xmath216 thus , substituting ( [ lacontra1 ] ) into ( [ gsumtj ] ) leads to the exact expression for the conductance of a section of the corbino disk . notice that the transmission probabilities @xmath76 for the full disk ( [ tjlambdadsk ] ) are _ not _ reproduced for @xmath217 , as the mass confinement is now present in the system . instead , they are equal for @xmath218 , what causes the conductance of such a half - disk to be equal to half of the full disk conductance for arbitrary @xmath103 . the pseudodiffusive limit @xmath219 is realized for @xmath220 , and the conductance ( [ gdifflam ] ) is @xmath221 the above formula coincides with ( [ gfdiffdsk ] ) for @xmath217 . the opposite , quantum tunneling limit ( @xmath85 ) is reached for @xmath107 where formula ( [ gftunnlam ] ) gives @xmath222 in this case , the conductance decays ( at fixed @xmath23 ) with the characteristic length @xmath223 as @xmath224 . the reciprocal decay , observed in section [ corbdi ] for the full disk , now appears at @xmath218 . as a next example , we consider the conformal transformation , which changes the quantum dot shown in fig . [ corbcido](c ) into a rectangle of the width @xmath9 and the length @xmath10 . the transformation is given by the formula @xcite @xmath225 with the condition @xmath226 , which leads to @xmath227 the origin of the coordinate system of fig . [ interf12 ] is now shifted to @xmath228 . the poles of the transformation ( marked by white dots in fig . [ corbcido]b ) are placed at @xmath229 , with @xmath230 . the angle @xmath231 , at which the dot edges intersect each other , is @xmath232 @xmath233 and @xmath234 . again , substituting ( [ lacontra2 ] ) into ( [ gsumtj ] ) provides one with the exact expression for the system conductance , which reaches the pseudodiffusive limit for @xmath220 , where @xmath105 [ see eq . ( [ gdifflam ] ) ] and @xmath235},\ ] ] whereas for the quantum tunneling limit @xmath107 , the formula ( [ gftunnlam ] ) reads @xmath236 this leads to an asymptotic form @xmath237 , as @xmath238 and @xmath239 for @xmath240 , while the fano factor approaches the poissonian value @xmath241 . the half - corbino disk ( a ) and quantum dot with circular edges ( b ) realized on a honeycomb lattice . shadow areas mark heavily - doped graphene leads . thick black lines indicate the mass confinement . ] conductance ( a , b ) and fano factor ( c , d ) for the half - corbino disk ( left ) and circular quantum dot ( right ) . solid lines show the results obtained by numerical summation of ( [ gsumtj ] ) and ( [ fsumtj ] ) over the modes , dashed lines show the pseudodiffusive limits ( [ gdifflam ] ) and ( [ favertt ] ) . datapoints on left / right panels are obtained from a computer simulation of transport through the system of fig . [ set2term](a ) and [ set2term](b ) . ] we now test the analytical predictions reported earlier in this section by comparing them with the results of a computer simulation of electron transport in graphene . the discussion starts from the tight - binding model of graphene , with hamiltonian @xmath242 the hopping matrix element @xmath243 if the orbitals @xmath244 and @xmath245 are nearest neighbors on the honeycomb lattice ( with @xmath246 ) , otherwise @xmath247 . the single - particle potential @xmath248 is arranged such that the chemical potential @xmath249 in the leads marked by shadow areas in fig . [ set2term ] , whereas between the leads ( white area ) @xmath250 , except for the small regions , where we put @xmath251 ( with @xmath252 , depending whether the atom belong to the @xmath253 or @xmath254 sublattice ) to model a mass - confinement on a honeycomb lattice @xcite . such regions are : the outermost edge atoms in the case of the half - corbino disk ( fig . [ set2term]a ) , and the atoms placed out of the dot edge ( thick lines in fig . [ set2term]b ) for the case of a quantum dot with circular edges . we have calculated the transmission matrix numerically by adapting the method developed by ando for a square lattice @xcite to the honeycomb lattice . the results of our computer simulation @xcite , depicted by datapoints in fig . [ gf2term ] , match theoretical predictions ( solid blue lines ) as long as the number of modes in the narrow lead @xmath255 . moreover , the formulas ( [ gcordif ] ) and ( [ gcirdif ] ) for the pseudodiffusive conductance ( dashed red lines in fig . [ gf2term]a , b ) reproduce the full expression ( [ gsumtj ] ) with @xmath109 accuracy for @xmath110 in the case of the half - corbino disk , and for @xmath256 in the case of the quantum dot with circular edges . analogously , the pseudodiffusive value of the fano factor @xmath257 ( see fig . [ gf2term]c , d ) matches the full expression ( [ fsumtj ] ) with @xmath109 accuracy for @xmath111 and @xmath258 , respectively . in other words , the half - corbino disk , attached to one narrow and one wide lead , represents the case in which electron transport demonstrates the pseudodiffusive character in a surprisingly wide range of the system s geometrical parameters . on the contrary , in the case of the circular quantum dot attached to two narrow leads , both the conductance and the shot noise show strong deviations from the pseudodiffusive predictions , as the transport is dominated by a single mode in a relatively wide range of parameters . the latter represents an example of a graphene system for which our predictions on quantum - tunneling transport ( such as an approximately _ quadratic _ decay of the conductance with @xmath211 ) seem to be particularly feasible for an experimental verification , also because similar systems have already been fabricated @xcite suggesting that the role of mass confinement is crucial when discussing the electronic structure of _ closed _ quantum dots in graphene . moreover , a recent numerical study shows that the mass confinement leads to a strong suppression of weak localization in such systems @xcite , as observed earlier in experiment @xcite . below , we extend our numerical analysis to _ open _ systems that can not be obtained from a strip by conformal transformation , to illustrate the generic character of the quantum - tunneling transport in undoped graphene . nanoribbon attached _ perpendicularly _ to the semicircular ( a ) and rectangular ( b ) leads . each system is characterized by the lead width @xmath259 and the sample area length @xmath19 . shadow areas mark heavily - doped graphene leads . ] conductance ( a , b ) and fano factor ( c , d ) obtained numerically for the system of fig . [ set4term](a ) and fig . [ set4term](b ) ( open and solid symbols in all panels ) compared with analytical predictions ( lines ) . ( a ) the pseudodiffusive conductance ( [ gcirdif ] ) ( solid red line ) and ( [ gsqudif ] ) ( dashed blue line ) . solid and dashed lines in panels ( b , c ) : the tunneling conductance ( [ gciltun ] ) and ( [ gsqutun ] ) , and the corresponding values of the fano factor @xmath260 . [ the relation depicted by the solid line in the shot noise vs conductance diagram ( d ) . ] the pseudodiffusive limit @xmath1 is shown by the black dotted line ( c , d ) . ] in this section , we present the results obtained from computer simulations of transport _ across _ a long nanoribbon attached to the semicircular ( fig . [ set4term]a ) and rectangular ( fig . [ set4term]b ) leads , which demonstrate a striking analogy between these systems and the circular quantum dot studied in the previous section . each of the systems in fig . [ set4term ] is modeled by the tight - binding hamiltonian ( [ hami ] ) . the simulation parameters @xcite are chosen to grasp the basic features of recently fabricated graphene nanoribbons @xcite , which have zigzag edges and are insulating , as the weak staggered potential placed at the ribbon edge opens a band gap in the electronic spectrum @xcite . a similar effect was observed in recent numerical studies of long nanoribbons with weak edge disorder @xcite or irregular edges @xcite . we utilize the @xmath18-terminal recursive green s function algorithm @xcite , which allows us to analyze directly the electron transport across an _ infinitely _ long nanoribbon in graphene . namely , we attached two extra leads ( one from the top and one from the bottom , not shown ) to each of the systems in fig . [ set4term ] , that are undoped and thus contain the evanescent modes only . ( notice that the chemical potential for the outermost edge - atoms @xmath261 @xcite . ) the results are shown in fig . [ gf4term ] . the conductance of a nanoribbon attached perpendicularly to circular leads ( top panel in fig . [ gf4term ] , open symbols ) approaches the asymptotic formula for the circular quantum dot ( [ gcirdif ] ) with @xmath262 , @xmath263 , @xmath218 for @xmath264 ( solid red line ) . for instance , a @xmath265 agreement is reached at @xmath266 . this is a consequence of the fact that in the absence of propagating modes in a ribbon , most of the current flows via the central region of the device , and the system of fig . [ set4term]a becomes effectively identical to the circular quantum dot in the pseudodiffusive limit , where the role of boundary conditions is negligible . for the opposite , quantum - tunneling limit @xmath267 the corresponding formula ( [ gcirtun ] ) may be written as @xmath268 what agrees surprisingly well with the actual data shown in fig . [ gf4term ] ( see inset in the top panel ; solid red line and open symbols , respectively ) . such an agreement can be understood when looking at the current - density distribution , shown in fig . [ jxy4te ] . even for an aspect ratio as small as @xmath269 , over @xmath270 of the current does not leave the area of a circular quantum dot ( bounded symbolically with dashed lines ) . for the case of a nanoribbon attached perpendicularly to rectangular leads ( fig . [ set4term]b ) , the pseudo - diffusive conductance ( for @xmath271 ) is given by @xcite @xmath272 which is depicted in top panel of fig . [ gf4term ] ( dashed blue line ) and matches the numerical data ( solid symbols ) within @xmath265 accuracy for @xmath273 . an identically good agreement with the numerics is observed for the asymptotic form of the formula ( [ gsqudif ] ) @xmath274 , showing that the infinite ribbon attached perpendicularly to the leads has an extra @xmath275 conductance in comparison with the rectangular geometry considered in refs . @xcite . a brief comparison between the formula ( [ gsqudif ] ) and the generic form of the pseudodiffusive conductance ( [ gdifflam ] ) allows us to consider the functional @xmath276 in an approximate form given by @xmath277.\ ] ] subsequently , an approximate form of the quantum - tunneling conductance ( [ gftunnlam ] ) for @xmath267 is @xmath278 again , the formula ( [ gsqutun ] ) shows a surprisingly good , approximately @xmath16 agreement with the numerical data presented in fig . [ gf4term ] ( see the inset in the top panel ; dashed blue line and solid symbols , respectively ) , suggesting that the power - law ( approximately quadratic ) decay of @xmath88 for large @xmath19 is a generic feature for transport across the nanoribbon , unrelated to the particular shape of the leads @xcite . the numerical results for the shot - noise power are presented in the bottom panel of fig . [ gf4term ] . the approximative formulas ( [ gciltun ] ) and ( [ gsqutun ] ) are substituted in the relation @xmath279 , which produces the analytical predictions depicted by solid red and dashed blue lines , respectively . in both cases , the agreement with numerical results is better than @xmath280 when the fano factor @xmath281 . an additional insight into the nature of the crossover from the poissonian to the pseudodiffusive regime is provided by an @xmath84 versus @xmath88 plot ( see the inset ) . in particular , values of @xmath84 are very close to @xmath282 even for a relatively large conductance @xmath283 , which indicates that electron transport is governed by a single , valley - degenerated mode in a wide range of the geometrical parameters ( @xmath284 for the circular leads , and @xmath285 for the rectangular leads ) . mode - participation ratio for transport through the system of fig . [ set4term ] with @xmath286 and finite width of the undoped region @xmath287 . three curves on each panel correspond to @xmath288 ( squares ) , @xmath289 ( circles ) , and @xmath290 ( triangles ) , with @xmath291-integer . top : circular leads , bottom : rectangular leads . lines are drawn as a guide for the eye only . ] so far , we have analyzed the transport across an _ infinitely _ long , zigzag nanoribbon attached perpendicularly to the leads . to find out how the results change for the realistic case of a long but _ finite _ nanoribbon , we consider now the system of fig . [ set4term ] with a central ( undoped ) region of finite width @xmath287 . the system is terminated from the top and the bottom by armchair boundaries which mix valley degrees of freedom @xcite , so the fourfold ( spin and valley ) degeneracy of transmission eigenvalues @xmath76 is expected to be replaced by the twofold ( spin only ) degeneracy . to trace the effect of armchair boundaries in a quantitative manner , we define the mode - participation ratio @xmath292 where we assume spin - only degeneracy in summations . in particular , for the quantum - tunneling limit @xmath267 , the mode - participation ratio is @xmath293 if the lowest mode , that governs the electronic transport , has an approximate valley degeneracy . otherwise , in this limit @xmath294 . the numerical values of the mode - participation ratio ( [ pmoddef ] ) are presented in fig . we took @xmath295 ( providing @xmath296 propagating modes for @xmath297 ) , @xmath298 ( so @xmath286 ) and vary @xmath287 . the remaining parameters are identical as in the case of an infinite ribbon , studied before . the datapoints in fig . [ pmod ] illustrate a smooth crossover from the transport dominated by a single mode with spin - only degeneracy ( @xmath299 ) , to the situation with full fourfold degeneracy ( @xmath300 ) . the details of the evolution depends on whether the width @xmath287 corresponds to the metallic ( @xmath301 ) or to one of the two insulating armchair boundary conditions ( @xmath288 , @xmath290 ) . in all cases , the valley degeneracy is approximately restored ( @xmath293 , marked with thin black line ) for @xmath302 , when the role of armchair edges becomes negligible , as the current is flowing predominantly via the central area of the system ( see the current distribution shown in fig . [ jxy4te ] , right panel ) . these two graphene billiards both have the same conductance and shot noise in the pseudodiffusive regime @xmath303 . ] for the sake of completeness , we analyze now the transport through a graphene billiard attached to two different leads , one narrow and semicircular , and the other wide and rectangular , as shown in fig . [ cirsqu ] . as before , mass confinement ( thick black lines ) is applied for the edges not connected to the leads ( shadow areas ) . the system shown in the left panel of fig . [ cirsqu ] can be exactly mapped onto a strip ( see fig . [ interf12 ] ) by the conformal transformation ( [ contra2 ] ) with the condition @xmath304 , where @xmath305 . this implies the functional @xmath306 where @xmath307 $ ] is the angle with which edges intersect each other in the poles of conformal transformation ( see fig . [ corbcido]c , with @xmath218 ) . the pseudodiffusive ( [ gdifflam ] ) and the quantum - tunneling conductance now take on the forms @xmath308},\ ] ] with @xmath309 for @xmath310 , and @xmath311 where the last asymptotic expression refers to the limit @xmath312 , for which the system behaves effectively like the half - corbino disk with the inner radius @xmath262 and the outer radius @xmath313 . as a consequence , the pseudodiffusive conductance ( [ gdiffcisq ] ) reproduces the values obtained from the exact expression @xcite with 1% accuracy for @xmath314 . another striking feature of the pseudodiffusive regime @xmath303 , which coincides with findings presented earlier in this section , is related to the fact that poles of the conformal transformation ( [ contra2 ] ) approach the circular lead tip for @xmath315 . this is why the current is flowing mainly through the central area of the system , and the two billiards shown in fig . [ cirsqu ] become equivalent in such a limit . the earlier findings for the circular quantum dot and the long nanoribbon attached perpendicularly to the leads , allow us to expect that the conductance of the nanoribbon - like system shown in the _ right _ panel of fig . [ cirsqu ] will not deviate significantly from the expression @xcite also in the tunneling limit @xmath267 . the numerical results presented in fig . [ gfcisq ] confirm such an expectation , as the conductance obtained by a computer simulation for the nanoribbon - like system @xcite match again the analytical predictions for the system with circular edges in surprisingly wide range of the parameters . namely , @xmath16 agreement is reached for @xmath316 , whereas for @xmath317 the deviation drops below @xmath265 . conductance and shot noise of the two systems shown in fig . [ cirsqu ] . main panel : the exact ( solid line ) and the pseudodiffusive ( dashed line ) value of conductance for a billiard with circular edges , and the results of computer simulation for a nanoribbon - like billiard ( datapoints ) . inset : shot noise vs conductance diagram for a nanoribbon - like billiard . solid and dashed lines mark the asymptotic values for the tunneling and the pseudodiffusive limit , respectively . ] we predict that the pseudodiffusive conductance ( [ gdiffcisq ] ) remains unchanged for a wide class of irregular graphene billiards of shapes fitting _ between _ the two limiting cases shown in fig . [ cirsqu ] . moreover , an approximate agreement should be observed even for the tunneling conductance ( [ gtunncisq ] ) . we believe that such an extra flexibility in device setups will facilitate experiments with better agreement with the theory as achieved so far for rectangular samples @xcite . in particular , the setup consisting of the narrow semicircular lead on one side and the straight graphene - lead interface on the other side , eliminates the difficulty of manufacturing the two _ parallel _ interfaces one of the main problems that have limited the number of experimental samples , suitable for both ballistic conductance and shot - noise measurements , to just a few so far . in conclusion , we have identified a novel type of quantum tunneling effect , which appears in transport through the corbino disk and quantum billiards in undoped graphene , provided that at least one of the leads ( or billiard openings ) is much narrower than the distance between openings @xmath19 , which defines the length - scale of the sample . in such a tunneling limit , the conductance @xmath88 shows a slow power - law decay with @xmath19 characterized by a geometry - dependent exponent . the fano factor @xmath84 exhibits a crossover from the pseudodiffusive ( @xmath1 ) to poissonian ( @xmath2 ) shot noise , with a relation @xmath318 in a surprisingly wide range of @xmath19 . this is because electron transport in the tunneling limit is effectively governed by a single mode , having the full spin , valley , and symplectic degeneracy ( @xmath4 ) in the absence of boundaries ( corbino geometry ) , spin and valley degeneracy ( @xmath5 ) if the boundary conditions do not scatter valleys , or the spin - only degeneracy ( @xmath6 ) otherwise . in particular , for the case of a ribbon which contains either infinite - mass or armchair boundaries , the valley degeneracy is restored when armchair endings are shifted away from the area where the main current flows . we would like to stress that the relation between @xmath88 and @xmath84 allows an experimental verification of the degeneracy @xmath319 without referring to any geometrical parameters . we have explored the idea of katsnelson and guinea @xcite , that transmission eigenvalues could be obtained analytically for any undoped graphene flake of a geometry linked via conformal transformation to a strip , for which the solution is known due to tworzydo et al . @xcite . in the pseudodiffusive limit , we show that the eigenvalue distribution is affected by an arbitrary conformal transformation only via a multiplicative prefactor , and the value @xmath1 is unchanged for any _ closed _ setup , provided that @xmath320 . we test the approach for the corbino disk , by comparing transmission probabilities obtained by a conformal mapping , and within the mode - matching analysis for angular momentum eigenstates . to analyze a crossover from the tunneling to the pseudodiffusive limit in a confined system , we focus on two particular billiards ( a section of the corbino disk and the quantum dot with circular edges ) confined by a mass . the results of our numerical simulation of transport through a lattice consisting of ca . @xmath321 carbon atoms match the expressions for @xmath88 and @xmath84 obtained by conformal mapping . moreover , we generalize the approach to obtain an approximate formula for @xmath88 ( which reproduces either the pseudodiffusive values or the tunneling - limit exponent resulting from the simulation ) in the case of an infinite ribbon attached perpendicularly to the leads an _ open _ billiard , not linked to the strip via conformal transformation . we thank bjrn trauzettel for helpful discussions . acknowledges the support from the alexander von humboldt stiftung - foundation , the polish ministry of science ( grant no . n202128736 ) , and the polish science foundation ( fnp ) . we further acknowledge financial support from the german research foundation ( dfg ) via tr950/11 ( p.r . ) and sfb 689 ( m.w . ) . generic circular contacts ( thick lines ) of radii @xmath43 and @xmath322 misplaced by a distance @xmath323 or @xmath324 ( dashed line depicts the _ inner _ contact interface for the latter case ) . white dots mark poles of the transformation ( [ mobius ] ) , thin lines are perpendicular to each of the interfaces ( and mapped onto the radiant lines by the transformation ) . ] we consider here a generic setup , containing two circular , but not coaxial interfaces splitting undoped and heavily doped graphene interfaces , as depicted schematically in fig . [ nocoacir ] . in the first case , an infinite graphene plane is probed by the leads of radii @xmath43 and @xmath325 ( thick solid lines ) , misplaced by the distance @xmath323 ( we further suppose @xmath326 ) and heavily doped . in the second case , a disk - like sample area is limited by the inner lead of radius @xmath43 ( dashed circle ) and the outer lead of radius @xmath325 , misplaced by @xmath324 ( and for @xmath327 the perfect corbino geometry is restored ) . in both situations , conformal mapping onto the corbino disk is provided by the mbius transformation @xmath328 where @xmath329 belongs to the disk area ( with the edges radii @xmath23 and @xmath24 , @xmath330 ) and @xmath331 belongs to the sample area of fig . [ nocoacir ] . the real parameter @xmath332 is adjusted such that @xmath333 which leads to the useful relation @xmath334 ( notice that we are using one form of @xmath64 to describe the two distinct situations , in each of which only one of the displacements @xmath335 is a physical parameter . ) the explicit form of the functional @xmath336 follows from the condition that the transformation ( [ mobius ] ) always maps the first contact ( of radius @xmath43 ) onto the inner edge of the disk , whereas the second contact is mapped onto the outer edge . for the two situations studied here @xmath337 where we define the variables @xmath338 , @xmath339 ( with @xmath340 , @xmath341 ) . the remaining one of the parameters @xmath335 , not listed explicitly as an argument of @xmath89 , is determined by eq . ( [ ddr12 ] ) . in particular , for the case of two identical circular leads probing a large graphene plane ( see fig . [ flak2prob]a ) @xmath342 , @xmath343 , and the functional @xmath344 where the approximation refers to the @xmath345 limit . defining @xmath346 we obtain eq . ( [ gtunnflak2a ] ) of the main text . similarly , taking the limit @xmath347 such that @xmath348 , we find from eq . ( [ lamdpm ] ) that @xmath124 for @xmath349 , what leads to eq . ( [ gtunnflak2b ] ) for the conductance . here , we derive the transmission and reflection amplitudes for scattering eigenstates of the hamiltonian ( [ dirhamcir ] ) for the corbino setup , as shown in fig . [ corbcido](a ) . without loss of generality , we suppose electron doping in the leads @xmath350 , but an arbitrary doping in the sample area @xmath351 . the radial component @xmath352 of the eigenstate @xmath353 ( [ jeigen ] ) corresponding to the total angular momentum @xmath354 ( @xmath65 half - odd integer ) and energy @xmath355 can be divided into three regions . for @xmath356 ( the outer lead ) , @xmath357 , with @xmath358 where @xmath359 with @xmath360 @xcite , and @xmath361 the reflection coefficient . next , for @xmath362 ( the disk area ) , @xmath363 , with @xmath364 where @xmath365 . finally , for @xmath366 ( the inner lead ) , @xmath367 , with @xmath368 + solving the matching conditions @xmath369 and @xmath370 we find @xmath371,\end{gathered}\ ] ] @xmath372 where @xmath373 ( with @xmath341 ) , @xmath374 , and we have defined with @xmath376 [ @xmath377 the bessel functions of the first [ second ] kind . the reflection and transmission amplitudes are @xmath378-h_{j-1/2}^{(1)}(\rho_2)\left[h_{j-1/2}^{(2)}(\rho_1)+i{\eta}h_{j+1/2}^{(2)}(\rho_1)\right]-\eta\mathfrak{d}_{j}^--i\mathfrak{d}_{j}^+\right\},\end{gathered}\ ] ] and @xmath379 defining @xmath380 , we obtain eq . ( [ tjmodma ] ) of the main text . notice that @xmath76 depends solely on @xmath137 , as @xmath381 is affected by @xmath138 only via a phase factor . it is also insensitive to the doping sign @xmath382 , what corresponds to the particle - hole symmetry . for the undoped - disk limit ( @xmath60 ) , eq . ( [ djdef ] ) leads to the asymptotic form @xmath383.\ ] ] substituting the above into eq . ( [ tjeta ] ) we obtain @xmath384 as given by eq . ( [ tjlambdadsk ] ) of the main text . hence , the correspondence between the mode matching for angular - momentum eigenstates and the conformal mapping technique for the disk in undoped graphene is established . a. rycerz , j. tworzydo , and c.w.j . beenakker , nature phys . * 3 * , 172 ( 2007 ) ; w. yao , s.a . yang , and q. niu , phys . rev . lett . * 102 * , 096801 ( 2009 ) ; g. tkachov and m. hentschel , phys . b * 79 * , 195422 ( 2009 ) . the translational invariance along the @xmath32-axis is satisfied for an infinitely - wide strip or for periodic boundary conditions of ref . @xcite . in case of a confined geometry that does not mix the valleys , one can use @xmath385 for the mode - matching , and find that eq . ( [ tthphi ] ) remains unchanged . zero - energy solution of the dirac equation in two dimensions has the general form @xmath386^{t}$ ] , where the spinor components @xmath387 and @xmath388 are holomorphic functions . for the strip geometry , infinite - mass boundary conditions ( @xmath389 , @xmath390 ) are satisfied after applying the transformation @xmath391 , @xmath392 , with @xmath393 an analytical function that turns the coordinate system @xmath394 into @xmath395 . the case of antiperiodic b.c . is discussed in section [ corbdi ] . katsnelson , k.s . novoselov , and a.k . geim , nature phys . * 2 * , 620 ( 2006 ) ; v.v . cheianov and v.i . falko , phys . b * 74 * , 041403 ( 2006 ) ; a.v . shytov , m.s . rudner , and l.s . levitov , phys . rev . lett . * 101 * , 156804 ( 2008 ) ; n. stander , b. huard , and d. goldhaber - gordon , phys . rev . * 102 * , 026807 ( 2009 ) . for the half - corbino disk ( @xmath218 ) we have fixed the width of the wide lead attached from the right ( see fig . [ set2term]a ) at @xmath415 ( corresponding to @xmath416 propagating modes for @xmath297 ) . the inner radius is varied in the range @xmath417 ( @xmath418 modes ) . the remaining parameters are the outer radius @xmath419 ( the diameter @xmath420 is noticeably smaller than @xmath287 to model the bulk - lead regime ) , and the edge - atoms chemical potential @xmath421 . for the quantum dot with circular edges ( see fig . [ set2term]b ) , we have fixed the lead radius at @xmath422 ( @xmath423 for @xmath297 ) , the edge radius is varied in the range @xmath424 , and the staggered - potential amplitude outside the dot is varied in the range @xmath425 . we find that the results are insensitive to the confinement strength . the chemical potential @xmath250 in the undoped region ( white areas in fig . [ set4term ] ) , except for the outermost edge atoms , where we put @xmath251 , with @xmath426 . in the leads ( shadow areas ) @xmath427 , what corresponds to @xmath428 propagating modes ( for the lead width fixed at @xmath429 ) . the sample area length @xmath19 is varied , but kept equal to the width of undoped leads . x. li , x. wang , l. zhang , s. lee , and h. dai , _ science _ * 319 * , 1229 ( 2008 ) ; m.y . han , b. zyilmaz , y. zhang , and p. kim , phys . lett . * 98 * , 206805 ( 2007 ) ; zh . chen , y .- m . lin , m.j . rooks , and p. avouris , physica e * 40 * , 228 ( 2007 ) . we suppose the current density @xmath430 , where @xmath431 is an electrostatic potential . the conductance @xmath432 is found by analytical integration of the current passing the vertical symmetry axis of the system ( @xmath433 ) . @xcite reports an exponential decay for a similar geometry , but it considers a far different leads arrangement , for which the current is effectively flowing _ along _ a narrow , insulating nanoribbon in graphene .

using a flexible galactic model with variable stellar velocity anisotropy , i apply the classical jeans mass modeling approach to the five dwarf spheroidal galaxies with the largest homogeneous datasets of stellar line - of - sight velocities ( between 330 and 2500 stars per galaxy ) carina , fornax , leo i , sculptor , and sextans . i carry out an exhaustive model parameter search , assigning absolute probabilities to each parameter combination . my main finding is that there is a well defined radius ( unique for each galaxy ) where the jeans analysis constraints on the enclosed mass are tightest , and are much better than the constraints at previously suggested radii ( e.g. 300 pc ) . for carina , fornax , leo i , sculptor , and sextans the enclosed dm mass is ( at 410 pc ) , ( at 925 pc ) , ( at 390 pc ) , ( at 435 pc ) , and ( at 1035 pc ) , respectively ( two - sigma uncertainties ; in m units ) . local dm density has the tightest constraints at smaller ( and also unique for each galaxy ) radii . the largest central dm density constraint is for sculptor : pc ( at two - sigma level ) . i show that the dm density logarithmic slope is totally unconstrained by the jeans analysis at all the radii probed by the data ( and not just at the center , as was demonstrated before ) . stellar velocity anisotropy has only very weak constraints . in particular , pure central tangential anisotropy is ruled out at better than two sigma level for three dwarfs , and the data are consistent with the global stellar velocity isotropy for all the five galaxies .

1504.08273

dwarf spheroidals ( dsphs ) are very faint galaxies ( some of the recently discovered dsph satellites of milky way are the faintest known galaxies , @xcite ) which tend to cluster around large galaxies @xcite . among the 50 or so local group dsphs discovered so far , only two ( tucana and cetus ) are relatively isolated ; the rest are located in the vicinity of either milky way or m31 , and are believed to be their satellites . despite their dull appearances ( no ongoing star formation ; little or no interstellar medium ; @xcite ) , dsphs are fascinating objects largely due to the fact that they contain significant quantities of dark matter ( dm ) , and as such represent the smallest observed scale of dm clustering . state of the art cosmological simulations predict that large galaxies should contain many smaller sub - halos which have not had time to be completely destroyed by the tidal field of the host halo . dsphs are believed to represent at least some of this substructure @xcite . dsphs present two major challenges to the standard ( @xmath10cdm ) cosmological model . first , despite the recent advances on both theoretical and observational sides , there still appears to be a factor of a few discrepancy between the observed and predicted numbers of milky way satellites so called `` missing satellites problem '' @xcite . second , some indirect evidence ( such as the `` timing paradox '' for the five globular clusters of fornax dsphs ; @xcite ) suggests that dm distribution in dsphs has a flat core , which is in line with what is deduced for larger dwarf and low surface brightness galaxies @xcite , but is at odds with the results of many cosmological simulations , predicting central divergent dm cusps with the logarithmic slope @xmath11 @xcite . different theoretical mechanisms suggested to rectify this `` cusp core '' problem produce different velocity anisotropy profiles for stars . for example , our stellar feedback mechanism @xcite , where the dm cusp is flattened via gravitational heating from the interstellar gas sloshed around by supernovae , results in isotropic stellar cores . on the other hand , the dynamical friction mechanism of @xcite , where the cusp is heated by massive baryonic clumps passively spiraling in towards the galactic center , should produce noticeable central tangential anisotropy @xcite . constraining stellar anisotropy profiles in spheroidal galaxies can hence be a valuable indirect way of identifying the correct theoretical mechanism of dm core formation . @xcite was the first who deduced the presence of significant amounts of dm in a dsph ( draco ) , based on the measurements of the line - of - sight velocities of only three stars . since then , the number of stars in dsphs with known line - of - sight velocities has grown dramatically : recently published homogeneous datasets have up 2500 stars per galaxy ( for fornax ; @xcite ) . mass modeling of dsphs has dramatically improved as well : from simple global estimates based on the virial theorem @xcite , to spatial models with strong simplifying assumptions ( such as `` mass follows light '' ; @xcite ) , to more recent modeling efforts with fewer assumptions @xcite . the classical method of dsph mass modeling is to solve the spherical jeans equation for a range of models with different dm density profiles , and then to compare the predicted stellar line - of - sight velocity dispersion profiles with the observed ones using either @xmath12 or maximum likelihood techniques to find the best models . the spherical symmetry assumption is reasonably accurate given the spheroidal appearances of dsphs ( ellipticity @xmath13 is usually @xmath14 or less ) , absence of disk - like structures , and negligible rotation in these systems @xcite . to solve the spherical jeans equation ( that is , to derive the stellar velocity dispersion as a function of radius ) , one has to specify two functions : total ( dm @xmath15 stars ) density , and stellar velocity anisotropy ( @xmath16 ) profiles . to reduce the dimensionality of the problem , one usually assumes a certain shape of the @xmath16 radial profile either a constant , or the osipkov - merritt profile ( purely isotropic at the center , purely radially anisotropic in the infinity ; @xcite ) . there are no reasons to believe that stellar anisotropy is constant in dsphs . for example , simple dynamical models which start with an initially non - equilibrium stellar configuration inside the dm halo ( `` cold collapse '' , @xcite , and `` hot collapse '' , @xcite ) tend to produce the stellar velocity isotropy at the center and a variable degree of radial anisotropy in the outskirts of the relaxed stellar body . as already mentioned , the central parts of dwarf galaxies can be significantly affected by dense and violent baryons , which may either randomize stellar velocities @xcite or induce a significant tangential anisotropy @xcite . on the observational side , dsphs are known to be non - homogeneous objects , with younger , more metal - rich and kinematically colder star populations concentrated towards the center @xcite ; it would be strange if the stellar velocity anisotropy would be the same across these different populations of stars . given the expectation that @xmath16 is not constant across a dsph and that the important properties of dsphs ( such as the central logarithmic slope of the dm density ) are strongly degenerate with respect to the unknown stellar velocity anisotropy @xcite , proper mass modeling of dsphs has to include radially variable stellar anisotropy . to the best of my knowledge , the only example of a full - fledged jeans mass modeling of dsphs with variable @xmath16 is the work by @xcite . here i outline the main differences of my approach with that of @xcite . 1 . in this paper , i carry out mass modeling of the limited number ( five ; carina , fornax , leo i , sculptor , and sextans ) of galactic dsphs with the highest quality , homogeneous data : large ( @xmath17 stars ) homogeneous datasets of stellar line - of - sight velocities @xcite and accurate stellar surface brightness profiles derived in a homogeneous manner in this paper . this is in contrast to the work of @xcite , who analyzed a much larger set of local group dsphs , with heterogeneously derived data and a wide range of data quality . i employ a `` brute force '' optimization while searching for the best fitting models . this approach generated a wealth of statistical data , which can be used for testing a variety of different hypotheses about the distribution of stars and dm in dsphs . the data are available online . the `` brute force '' approach coupled with a flexible dsph model allowed me to find the dsph parameters which the jeans analysis can constrain very well . specifically , i found that the total enclosed mass in a dsph can be constrained to a high accuracy ( two - sigma uncertainty as good as @xmath1815% , for fornax ) at a certain radius , which is different for different dwarfs . this information can be valuable for matching the galactic dsphs to the predictions of cosmological simulations . similarly , i derived tight constraints on the local dm density at a certain radius ( unique for each galaxy ) , which can be used in the research aimed at detecting dm in dsphs via its annihilation signal . 4 . unlike @xcite , i properly account for the self - gravity of stars . this can be important for dsphs with dense stellar cores ( e.g. sculptor ) . i use an advanced ( with the variance due to random locations of stars within one radial bin removed ; see [ good ] ) @xmath12 model fitting approach . unlike @xcite , who employed a maximum likelihood technique , my approach is insensitive to the ( unknown ) shape of the line - of - sight star velocity probability distribution function ( pdf ) . i had two main goals for this paper . first , i wanted to find out which ( if any ) dsph parameters can be meaningfully constrained via the classical , jeans mass modeling approach , if such an analysis is pushed to the extreme ( large homogeneous observational datasets of stellar line - of - sight velocities ; a very flexible galactic model ; careful numerical analysis consuming a significant @xmath19 cpu hours in my case amount of supercomputing time ; a `` brute force '' optimization which ensures that no hidden `` valleys '' and local and global minima in the multi - dimensional likelihood manifold are missed ) . the second goal was to generate high - quality data which can be used to dramatically reduce computational time ( by restricting the model parameter space to be explored ) in future post - jeans ( analyzing the full shape of the stellar velocity pdf ) mass modeling efforts which will be the subject of paper ii in this series . as the dynamical state of the outer parts of dsphs is still a matter of debate ( specifically , some authors , e.g. @xcite , argue that the observed outskirts of these galaxies are undergoing a severe tidal disruption and hence are not in dynamic equilibrium , which would invalidate the jeans analysis for those parts ; but see @xcite for the opposite example ) , in the present study i put the main emphasis on recovering the _ inner _ structure of dsphs . unfortunately , the most interesting aspect of the inner structure of these galaxies the `` dm cusp or core '' question can not be addressed via the jeans analysis alone , as the central logarithmic slope of the dm density profile @xmath20 is known to be degenerate in this method . this follows from the properties of the spherical jeans equation @xcite , and is clearly visible even in the mass modeling with a constant stellar anisotropy @xcite . in this paper i obtain a more general result that @xmath21 is completely unconstrained by the jeans analysis not only at the center ( @xmath22 ) , but also at any other radius @xmath23 within the stellar body of the dwarf . this paper is organized as follows . section [ model ] describes the dsph model . section [ obs_data ] discusses the observational data ( stellar line - of - sight velocities and stellar surface brightness profiles ) . section [ fitting ] gives a detailed description of my @xmath12 optimization technique . section [ results ] presents the main results of this study . the conclusions are presented in section [ conclusions ] . assuming a spherical symmetry , the jeans equation for a two - component ( dm @xmath15 stars ) system can be written as @xmath24 here @xmath23 is the distance from the center , @xmath25 , @xmath26 , and @xmath27 are the stellar density , radial velocity dispersion , and anisotropy , and @xmath28 is the total ( dm @xmath15 stars ) gravitational potential . we introduced the anisotropy parameter @xmath27 in @xcite . it is defined as @xmath29 , and is related to the more conventional anisotropy parameter @xmath16 through @xmath30 and @xmath31 . ( here @xmath32 is the one - dimensional tangential velocity dispersion . ) unlike @xmath16 , the parameter @xmath27 is conveniently symmetric : it is equal to @xmath33 , 0 , and 1 for purely tangential , isotropic , and purely radial orbits , respectively . ( the corresponding @xmath16 values are @xmath34 , 0 , and 1 , respectively . ) i assume that the stellar anisotropy @xmath27 smoothly varies between the two asymptotic values , @xmath35 at the center and @xmath36 in the infinity : @xmath37 @xcite . here @xmath38 is the anisotropy transition radius , and @xmath39 . the dm halo is assumed to have the following density profile : @xmath40 ( here @xmath41 and @xmath42 are the dm scaling radius and density , respectively . ) this is a generalization of the ( * ? ? ? * nfw ) profile , with an arbitrary central logarithmic density slope @xmath20 and the outer slope of @xmath43 . in the limit @xmath44 we recover the theoretical nfw profile ; when @xmath45 , we obtain a flat - cored profile , which is almost identical to the observationally inferred @xcite profile . stars are assumed to have a generalized plummer density profile ( see eq . ( [ rho _ * ] ) in [ sb ] ) . the dm contribution to the total gravitational potential @xmath28 is obtained by numerical integration of the corresponding poisson equation ; the stellar contribution has simple analytical solutions ( different for different values of the stellar outer density slope @xmath46 ) . to compare the solution of the jeans eq . ( [ jeans ] ) with observations , one has to compute the stellar line - of - sight velocity dispersion , @xmath47 , as a function of the projected radius , @xmath48 : @xmath49 @xcite here @xmath50 is the stellar surface density . overall , the model has six free parameters : @xmath41 , @xmath42 , @xmath20 ( characterize the dm distribution ) , and @xmath35 , @xmath36 , @xmath38 ( characterize the stellar anisotropy profile ) . for jeans mass modeling it is essential to know stellar density profiles of dsphs as accurately as possible . these can be derived from surface brightness profiles if one assumes a spherical symmetry and a certain value for the stellar mass - to - light ratio . as galactic dsphs are located within 300 kpc from the sun and are fully resolved into stars , one has to employ star count techniques to derive their surface brightness profiles . traditionally , a few simple analytical stellar surface density profiles have been used for mass modeling of dsphs : empirical @xcite , plummer @xcite , and @xcite . we showed in @xcite that the surface density profile of the form @xmath51^{-(\alpha-1)/2},\ ] ] corresponding to the `` generalized plummer law '' , @xmath52^{-\alpha/2},\ ] ] describes the surface brightness profile of draco dwarf spheroidal extremely well . ( here @xmath48 and @xmath23 are the projected and spatial distances from the dwarf s center , @xmath53 and @xmath54 are the central stellar surface and volume densities , @xmath55 is the `` core radius '' , and @xmath46 is an integer which should be @xmath56 for finite total stellar mass models . ) the classical plummer law is recovered when @xmath57 . for the five dsphs analyzed in this paper , i obtained the generalized plummer profile parameters , @xmath55 and @xmath46 , by @xmath12 fitting of eq . ( [ sigma ] ) to the best available star count data : ( * ? ? ? * their figure 15 ) for fornax , ( * ? ? ? * their figure 9 ) for leo i , ( * ? ? ? * their figure 6 ) for sculptor , and @xcite for sextans . for carina , there were significant discrepancies between the published star count profiles , so i carried out a joint @xmath12 fitting for three different datasets : @xcite , ( * ? ? ? * their table 3 ) , and ( * ? ? ? * their figure 4b ) . as the fitting was done in the linear space , all star count data points ( even those which became formally negative after subtracting the assumed contribution from the background sources ) were used figure [ fig1 ] shows the star count data and the best fitting generalized plummer models for the five dsphs . lcccc galaxy & @xmath58 & @xmath59 & @xmath60 & @xmath61 + carina & 2.18 & 2.18 & 2.83 & 2.92 + fornax & 2.35 & 9.81 & 2.86 & 2.55 + leo i & 3.12 & 47.8 & 11.5 & 6.66 + sculptor & 3.50 & 3.50 & 4.86 & 4.98 + sextans & 0.87 & 1.03 & 0.84 & 0.89 + draco & 0.65 & 1.69 & 1.25 & 1.10 + average & 2.11 & 11.0 & 4.02 & 3.18 + table [ tab1 ] presents the best @xmath12 values ( normalized by the number of degrees of freedom ) for the five dsphs for four different profiles ( generalized plummer , classical plummer , empirical king , and srsic ) ; i also included the fitting results for the draco data from ( * ? ? ? * their sample s2 ) the same data we used in @xcite for mass modeling of this galaxy . as you can see in table [ tab1 ] , the generalized plummer profile is the best one overall : for each galaxy , it produces the same or better fit than the other profiles , and is significantly better in terms of the average @xmath12 for all the six dwarfs . i warmly recommend this profile for all future dsph mass modeling work , as it is simple ( surface and volume densities and gravitational potential are simple analytical expressions as long as @xmath46 is an integer ) and appears to be the best fitting profile among 23 parameter models . one complication is that all the published star count profiles were computed inside elliptical annuli ( usually with the fixed ellipticity @xmath13 and position angle ) , with the quoted radius measured along the major axis of the galaxy . for the purposes of this study , spherically averaged ( in the plane of the sky ) surface brightness profiles are required . for each galaxy , i computed the spherically averaged profile by solving numerically the following integral for 100 values of the projected radius @xmath62 : @xmath63 here @xmath13 is the ellipticity of the galaxy . the resultant , averaged profile is very close ( but not identical ) to the generalized plummer profile with the same exponent @xmath46 and a smaller core radius @xmath55 . i derived this spherically averaged value of @xmath55 by @xmath12 fitting the above profile with the generalized plummer profile . the correction factor @xmath64 is a function of both @xmath13 and @xmath46 , and ranges from 0.816 ( sextans ) to 0.895 ( leo i ) . ( here @xmath65 is the core radius obtained from fitting the generalized plummer model to the original data , before averaging . ) lccccccccccc galaxy & @xmath66 & @xmath53 & @xmath13 & ( m / l)@xmath67 & @xmath68 & @xmath46 & @xmath65 & @xmath55 & @xmath54 & @xmath69 + & kpc & mag arcsec@xmath70 & & m@xmath6/l@xmath6 & km s@xmath71 & & arcmin & arcmin & m@xmath6 pc@xmath9 & arcmin + carina & 101 & 25.5 & 0.33 & 0.845 & 222.9 & 5 & 10.89 & 9.00 & 0.0055 & 3.97 + fornax & 138 & 23.4 & 0.31 & 0.935 & 55.2 & 7 & 27.92 & 23.61 & 0.0147 & 7.67 + leo i & 255 & 22.6 & 0.21 & 0.74 & 282.9 & 13 & 10.21 & 9.14 & 0.0490 & 1.92 + sculptor & 79 & 23.7 & 0.32 & 1.23 & 111.4 & 5 & 9.95 & 8.28 & 0.0584 & 3.66 + sextans & 86 & 26.2 & 0.35 & 1.15 & 224.3 & 4 & 22.18 & 18.09 & 0.00195 & 10.45 + the results of the generalized plummer profile fitting and other input galactic parameters for the five dsphs are presented in table [ tab2 ] . as this study properly takes into account the self - gravity of stars , i had to assume certain values for the stellar mass - to - light ratios in @xmath72-band , ( m / l)@xmath67 . for each galaxy , i used the average of the salpeter and composite ( m / l)@xmath67 estimates of ( * ? ? ? * their table 6 ) . knowing the distance to the galaxy , @xmath66 , the @xmath72-band central surface brightness , @xmath53 , the generalized plummer model parameters @xmath46 and @xmath55 , and ( m / l)@xmath67 allowed me to compute the central stellar density , @xmath54 ( see table [ tab2 ] ) . the table also lists the estimated half - light radii ( in arc minutes ) , @xmath69 , for spherically averaged models . i used the stellar line - of - sight velocities catalogs of @xcite for carina , fornax , sculptor , and sextans , and the catalog of @xcite for leo i. these are the largest homogeneous catalogs available for dwarf spheroidals , with the number of stars per galaxy ranging from @xmath73 ( leo i ) to @xmath74 ( fornax ) . i used the merged version of the data . the median error of the velocity measurements is @xmath75 km s@xmath71 . in all the dwarfs ( except for leo i ) some contamination from foreground galactic stars is expected to be present in the data . @xcite quantified this contamination by computing individual `` membership probabilities '' @xmath76 for each star . for this study , i only used stars with @xmath77 for carina , sculptor , and sextans , and stars with @xmath78 for fornax . ( despite the less conservative selection criterion for fornax , the resultant line - of - sight dispersion profile looks very smooth and un - contaminated , and the normalized @xmath12 for the best fitting model is @xmath79 see figure [ fig2 ] . ) due to the leo i s large heliocentric velocity and distance from the sun , the data for this dwarf should have essentially zero contamination in the range of heliocentric velocities from 260 to 315 km s@xmath71 ( * ? ? ? * their fig . 1 ) . accordingly , i used these simple velocity cuts to select the leo i star candidates . carina was the only galaxy in our sample for which the best fitting model deviated more than two sigmas from the data . this could be due to unaccounted foreground contamination ( which would not be surprising , given that it is the closest to the galactic plane dsph and has a very low surface brightness ) . for this galaxy only , i removed the six stars ( after removing the stars with @xmath80 ) , for which the deviation from the best fitting model was more than three sigmas , and then repeated the full mass modeling for the reduced star list . ( here one sigma corresponds to the total , model @xmath15 observational uncertainty , velocity dispersion . ) the removed stars have the following designations in the @xcite catalog : car-0056 , 0200 , 0465 , 0547 , 0680 , 1892 . this operation boosted the absolute probability of the best fitting model for carina from 0.04 to 0.26 ( when @xmath81 ; see [ method ] ) . it is important to note that all the six stars are located in the outer half of the radial bins , so this operation should have had very little impact on the analysis of the central parts of the galaxy which is the main focus of the current study . two popular statistical methods used in mass modeling of dsphs are @xmath12 fitting and maximum likelihood analysis . both methods have advantages and disadvantages . the @xmath12 method complements naturally the jeans analysis , as both operate only with the second moment ( dispersion ) of the stellar velocity distribution , whereas for the maximum likelihood approach one has to assume a certain ( usually gaussian ) form for the line - of - sight velocity probability distribution function . this factor makes the @xmath12 fitting less sensitive to outliers , and hence a much better approach when one is interested in absolute probabilities of different models . the maximum likelihood method , on the other hand , is more appropriate for post - jeans mass modeling , where the full model line - of - sight velocity pdf is computed and compared to the data . the main disadvantages of the @xmath12 technique are that 1 ) one has to try a few different values for the number of the radial bins , @xmath82 but that only means longer computing time , and 2 ) in its simplest implementation , the random locations of individual stars within each radial bin is an additional source of variance in the analysis . to address the latter issue , for both the observational data and model i estimate the following quantity ( for each radial bin ) : @xmath83 here @xmath84 is the line - of - sight velocity dispersion at the specific location ( projected distance from the dwarf s center ) of the @xmath85-th star , and @xmath86 is the number of stars in the current radial bin . the quantity @xmath87 will converge to the square of the local line - of - sight velocity dispersion when the radial extent of the bin shrinks to zero and the number of stars ( @xmath86 ) grows to infinity . the important point is that such a convergence is not required when all one wants to do is to measure the statistical deviation of the model from the data . all is needed is to estimate the mean ( for the data and the model ) and standard deviation ( only for the data ) of the quantity @xmath87 for each radial bin , and then compute the @xmath12 deviation for @xmath87 from all the radial bins . as the @xmath87 measurements for both observational data and model are carried out for the same projected distances from the center of the dwarf ( corresponding to the distances of actual stars ) , the variance due to random locations of stars within radial bins is completely removed . of course , we do not know from the observations the values of @xmath84 at the position of each star . instead , we know the individual line - of - sight velocities ( corrected for the systemic velocity of the galaxy ) , @xmath88 , and the associated measurement uncertainties , @xmath89 . one can use these known quantities to write down the following unbiased estimator of @xmath87 ( without any assumptions on the shape of the velocity pdf ) : @xmath90 this expression is valid even for the cases when the line - of - sight velocity dispersion changes dramatically within a single radial bin . the standard deviation of @xmath91 can be estimated from the following expression , @xmath92^{1/2},\ ] ] which does require the velocity dispersion to be approximately constant within a single bin . numerical experiments showed that eq . ( [ sgm_std ] ) noticeably deviates from the true answer only when the velocity dispersion changes significantly ( by a factor of two or more ) within a single radial bin , which is never the case in my best fitting models . ( [ sgm_mod][sgm_std ] ) are written for a single radial bin . it is now straightforward to write down the expression for the @xmath12 deviation of the model from the data for all the @xmath82 radial bins : @xmath93 here @xmath94 , @xmath95 , and @xmath96 are computed for the @xmath97-th radial bin using eqs . ( [ sgm_mod ] ) , ( [ sgm_obs ] ) , and ( [ sgm_std ] ) , respectively , and @xmath98 is the number of the statistical degrees of freedom , where @xmath99 the effective dimensionality of the free parameter space ( see [ test ] ) . the only drawback of the above procedure is that it can be substantially more computationally expensive than the standard @xmath12 analysis , as one has to compute the model line - of - sight velocity dispersion ( numerical integration using eq . ( [ sigma_los ] ) ) for each model , each star , and each value of @xmath82 . one can already use eq . ( [ chi2 ] ) from the previous section to compute the _ relative _ goodness of fit for different models . to go one step further and convert the @xmath12 values to the _ absolute _ probabilities of the models , @xmath100 one has to estimate the effective number of dimensions of the free parameter space , @xmath99 . ( here @xmath101 is the regularized gamma function . ) to estimate @xmath99 for my six - parameter model described in [ model ] , i generated a dataset of 2499 fake stars located at the same projected distances from the dwarf s center as the real 2499 stars in fornax with known line - of - sight velocities ( see [ vlos ] ) . i used my best fitting fornax model ( for the @xmath102 case ) and actual individual star velocity observational uncertainties , @xmath103 , to generate random values of the `` observed '' line - of - sight velocities for the artificial stars . ( i assumed that both the model line - of - sight velocities and the observational velocity uncertainties have gaussian distributions . ) next , i split my artificial dataset into 21 equal size sub - sets ( each one consisting of 119 randomly selected fake stars ) . in a sense , i generated 21 random realizations of the fornax galaxy . then i carried out full mass modeling ( as described in [ method ] , with @xmath102 ) and derived the @xmath104 value ( eq . ( [ chi2 ] ) ) for the best fitting model , separately for each sub - set . assuming that the resultant 21 random numbers are drawn from the same @xmath104 distribution , their mean should be an unbiased estimator for @xmath105 . in this numerical experiment , i derived @xmath106 , corresponding to the effective number of dimensions in my free parameter space @xmath107 . i used this number in the rest of the paper . this number is significantly ( almost by a factor of two ) lower than the formal number of the model free parameters ( six ) reflecting significant degeneracies ( from the jeans analysis point of view ) present in the model . as one will see in [ results ] , the most degenerate is the parameter @xmath20 ( the logarithmic slope of the dm density ) . post - jeans mass modeling ( dealing with the full line - of - sight velocity pdf , and not just with the dispersion ) is expected to be less degenerate . it is clear that @xmath99 will not be exactly the same for different data and for different number of radial bins ( because the radial extent of the binned data depends on @xmath82 ) , but as long as the number of bins is relatively large ( so that @xmath108 ) , the computed absolute probabilities should not be very sensitive to the errors in @xmath99 . to be more quantitative , i computed numerically the relative error for the absolute probabilities @xmath109 resulting from the @xmath99 uncertainty , @xmath110 , for @xmath111 and two different cases : @xmath112 ( corresponding to @xmath113 , as in my best fitting models ) , and @xmath114 ( corresponding to @xmath115 , my lower cutoff for `` good models '' ) . purely statistical ( poissonian ) errors in estimating @xmath105 ( and hence @xmath99 ) from the limited sample of the size @xmath116 is of the order of @xmath117 . assuming a larger uncertainty ( to account for a variance in data ) @xmath118 , the resulting relative errors in @xmath109 are only @xmath119 and @xmath120 , respectively , which i consider to be acceptable . if @xmath121 , as is the case with most of my galaxies , the relative errors in @xmath109 should be significantly smaller . in this section i outline my overall mass modeling procedure . i thoroughly explored ( `` brute force optimization '' ) the four out of the six free model parameters : the central logarithmic dm density slope @xmath20 , and the three stellar anisotropy parameters , @xmath35 , @xmath36 , and @xmath38 ( see [ model ] ) . for @xmath20 , i used the discrete values ( 0 , @xmath122 , @xmath123 , , @xmath124 ) , which covered the whole plausible range from the observations - suggested flat core ( @xmath45 ) to the cdm prediction , @xmath11 . in these low stellar density objects the presence of baryons is not expected to make the dm distribution `` cuspier '' via the adiabatic contraction of dm . a presence of a sufficiently massive central black hole could have steepened the dm cusp , but there is no evidence for super - stellar mass black holes in dsphs @xcite . both @xmath35 and @xmath36 had the following discrete values : ( @xmath33 , @xmath125 , @xmath126 , , 1 ) , which covered the full range of anisotropies from purely tangential ( @xmath127 ) to purely radial ( @xmath128 ) . in addition , for the central anisotropy @xmath35 i only explored the range allowed by the central anisotropy theorem of @xcite , @xmath129 . the anisotropy radius @xmath38 had the following values ( in the units of the stellar core radius @xmath55 ) : ( 0.1 , 0.147 , 0.215 , 0.316 , 0.464 , 0.681 , 1.47 ) . they are equidistant in the logarithmic space , and span the full range of radial distances resolved by the data . overall , i explicitly tested 22,344 different models with the above discrete values of the four model parameters . the remaining two model parameters , dm halo scaling radius @xmath41 and density @xmath42 , are very well constrained by the data when the rest of the parameters are fixed , which allowed me to use implicit optimization technique for these two parameters . this dramatically reduced the number of models to be tested and hence the total computational time , as @xmath41 and especially @xmath42 span a very large range of values ( four orders of magnitude for @xmath42 , see table [ tab3 ] ) . the only prior was for @xmath41 to stay within the interval 0.05 5 ( in @xmath55 units ; for sculptor , i used a different range 0.1 15 , as this galaxy has a very compact core ) . there was no prior for @xmath42 . i used the one - dimensional brent s method @xcite separately for @xmath41 and @xmath42 optimization ( finding global minimum in @xmath104 ; @xmath42 optimization is the inner loop , and @xmath41 optimization is the outer loop ) . the method showed a good numerical convergence , requiring only of order of a hundred models with different combinations of ( @xmath41 , @xmath42 ) to be tested ( when the other four explicit parameters were fixed ) . including both explicitly and implicitly explored parameters , i had to compute the @xmath104 deviation from the data for @xmath130 different models per galaxy and per @xmath82 value . for a given galaxy and a chosen combination of the six model parameters and the value of @xmath82 , i solved the jeans eq . ( [ jeans ] ) numerically , and then solved numerically the integral in eq . ( [ sigma_los ] ) to derive @xmath47 at all the projected radii @xmath48 corresponding to the locations of the dwarf s stars with known line - of - sight velocity . then i used eqs . ( [ sgm_mod][chi2 ] ) to compute the @xmath12 deviation of the model from the data , and eq . ( [ prob ] ) to assign absolute probabilities to the models . absolute probabilities of the best fitting models , @xmath131 , range from 0.23 for sculptor to 0.60 for sextans ( table [ tab3 ] ) , corresponding to @xmath132 sigma deviations . i used one global lower cut - off value of @xmath133 ( a two sigma deviation ) to select good models for all the galaxies . i used the following approach to determine the optimal value of the number of radial bins , @xmath82 , for each galaxy . in the limit of large @xmath82 ( small number of stars in one bin , @xmath86 ) , @xmath134 ( see eq.([sgm_std ] ) ) and the @xmath12 analysis becomes dominated by the poissonian noise and hence `` fuzzier '' ( less discriminatory ) . in the opposite limit , @xmath135 , the signal to noise ratio improves , but the @xmath12 analysis becomes again less accurate because with fewer radial bins we start loosing radial resolution and may overlook some sharp @xmath47 features near the dwarf s center or in the galactic outskirts . in addition , when @xmath82 approaches @xmath99 ( the effective dimensionality of the free parameter space ) , the errors in measuring @xmath99 become more important , which makes the absolute model probabilities less reliable ( see [ test ] ) . hence it appears there should be a certain value of @xmath82 when the @xmath12 analysis is the most accurate ( or discriminatory ) . i found this optimal value separately for each galaxy by running the full set of models for @xmath81 ( minimum acceptable number of bins for this model , see [ test ] ) , 15 , and 30 , and then choosing the value which produced the smallest fraction of good models ( @xmath136 in table [ tab3 ] ) . for fornax , sculptor , and sextans the optimal value of @xmath82 was found to be 15 ; for carina and leo i it is equal to 7 ( table [ tab3 ] ) . i applied the mass modeling procedure described in the previous sections to all the five dwarfs . the full results are available online . as good models found in the current analysis occupy only a small fraction of the total six - dimensional free parameter space , the data presented here can be used for post - jeans mass modeling projects ( utilizing the full pdf of the observed stellar line - of - sight velocities ; the subject of paper ii ) to dramatically reduce the required computational time . figure [ fig2 ] shows the radial @xmath47 profiles for the two best - fitting models ( with @xmath45 , solid lines , and @xmath44 , dotted lines ) for each galaxy . as one can see , the overall quality of the model fits is excellent , with the @xmath12 values ( normalized by the number of degrees of freedom @xmath98 ) being close to a unity . it is remarkable that these two dramatically different values of @xmath20 ( 0 and @xmath124 ) make very little difference in the quality of the model fit . this is most obvious for the galaxy with the highest quality data fornax where the two @xmath47 profiles are essentially identical way beyond the radial range covered by the data . for the galaxies with the best @xmath45 and @xmath44 profiles apparently diverging near the center ( sculptor and sextans ) , one could navely think that with a larger @xmath82 ( and hence with the binned data covering a wider range of radial distances ) the two models would become more statistically distinct . but this is not the case : for all the values of @xmath82 i tried ( 7 , 15 , and 30 ) , and for all the five galaxies , the difference between the best fitting models with different values of @xmath20 is statistically insignificant , to a similar degree . lccccccccccccccccccc name & @xmath137 & @xmath82 & @xmath136&@xmath138 & @xmath35 & @xmath41 & @xmath42 & @xmath139 & @xmath140 + & & & & & & & & @xmath8 pc@xmath9 & @xmath141 + carina & 740 & 7 & 0.680 & 0.262 & @xmath142 & @xmath143 & @xmath144 & 0.048 & 8.2 + fornax & 2499 & 15 & 0.253 & 0.468 & @xmath145 & @xmath143 & @xmath146 & 0.028 & 2.7 + leo i & 328 & 7 & 0.210 & 0.302 & @xmath145 & @xmath143 & @xmath147 & 0.080 & 1.95 + sculptor & 1349 & 15 & 0.230 & 0.228 & @xmath148 & @xmath149 & @xmath150 & 0.087 & 3.1 + sextans & 397 & 15 & 0.906 & 0.598 & @xmath142 & @xmath143 & @xmath151 & 0.007 & 5.3 + lccccccccccccccccccc name & @xmath152 & @xmath153 & @xmath154 & @xmath155 & @xmath156 & @xmath157 & @xmath158 & @xmath27 & @xmath159 & @xmath160 + & pc & @xmath161 & @xmath141 & pc & @xmath8 pc@xmath9 & @xmath141 & pc & & @xmath161 & @xmath161 + carina & 409 & 0.74 1.14 & 26 39 & 187 & 0.032 0.074 & 14 32 & & @xmath33 0.3 & 0.35 0.78 & 0.85 2.3 + fornax & 924 & 6.1 8.0 & 5.5 6.9 & 453 & 0.021 0.040 & 3.7 6.2 & 232 & @xmath162 0.05 & 0.27 3.4 & 1.90 5.1 + leo i & 389 & 1.55 1.94 & 3.4 4.1 & 204 & 0.067 0.115 & 2.5 3.8 & 72 & @xmath163 0.26 & 0.76 1.43 & 1.94 5.7 + sculptor & 435 & 2.17 3.01 & 22 30 & 256 & 0.067 0.107 & 20 31 & 83 & @xmath164 0.19 & 0.79 1.63 & 2.9 5.8 + sextans & 1034 & 1.4 3.2 & 19 42 & 451 & 0.004 0.013 & 11 32 & & @xmath33 0.3 & 0.06 1.1 & 0.37 1.7 + tables [ tab3 ] and [ tab5 ] summarize the main results of this study . in particular , i list there the ranges spanned by good ( @xmath165 ) models separately for the free model parameters @xmath35 , @xmath41 , and @xmath42 . as one can see , these parameters are very poorly constrained by the jeans analysis . the full allowed range for @xmath35 is @xmath142 ( due to the @xcite central anisotropy theorem ) , and only three galaxies ( fornax , leo i , and sculptor ) have a ( very weak ) constraint on this parameter , in the sense that the pure central tangential anisotropy is excluded . similarly , only one galaxy ( sculptor ) has any constraints on the dm halo scaling radius : @xmath166 ( in @xmath55 units ) . for the rest of the dwarfs , good models span the full explored range of @xmath41 , from 0.05 to 5 . parameter @xmath42 is the only one in the current analysis which is allowed to take any values ( no priors ) , so it does appear to be constrained in a meaningful way . one has to remember though that there is a strong degeneracy between @xmath42 and @xmath41 . e.g. , if @xmath41 were allowed to go to even smaller values ( @xmath167 ) , the range for @xmath42 would expand to even larger values . i do not list global constraints for the three remaining model parameters , @xmath20 , @xmath36 , and @xmath38 , as good models span the full ranges for these parameters : @xmath168 , @xmath169 , and @xmath170 , respectively . the conclusion here is that in mass modeling of dwarf spheroidal galaxies the jeans analysis is not capable of placing meaningful constraints on global model parameters ( for models which are flexible enough , as in the current study ) . the analysis does constrain the minimum central dm density , @xmath171 , and the corresponding mass to light ratio , @xmath172 ( see table [ tab3 ] ) . sculptor and leo i have the strongest constraints on @xmath171 ( @xmath173 m@xmath6 pc@xmath9 and @xmath174 m@xmath6 pc@xmath9 , respectively , at the two - sigma level ) . carina , on the other hand , has the strongest @xmath172 constraint : @xmath175 m@xmath176l@xmath6 . more useful constraints can be derived for radius - dependent ( computed at different distances from the dwarf s center ) quantities . figure [ fig3 ] plots the range of the averaged dm density @xmath177 spanned by good models as a function of the enclosed radius @xmath23 , separately for each galaxy . the averaged stellar density profiles ( dash - dotted lines ) are shown for comparison . there has been a debate in the literature @xcite regarding the value of the radius at which the jeans analysis provides the tightest constraints on the enclosed mass ( or equivalently the averaged density ) . as one can see in figure [ fig3 ] , such radius indeed exists ( vertical dotted lines ) and is very well defined . in table [ tab5 ] i list these radii ( @xmath152 ) , the range of the enclosed mass spanned by good models at those radii ( @xmath153 ) , and also the corresponding range of the mass - to - light ratio ( @xmath154 ) , for all the dwarfs . for comparison , in table [ tab5 ] i also list the good model ranges of the enclosed mass at 300 and 600 pc @xmath159 and @xmath160 , respectively . as one can see , @xmath152 ranges widely , from 390 pc for leo i to more than 1000 pc for sextans , and at neither of the previously suggested radii ( 300 and 600 pc ) the enclosed mass is well constrained by the jeans analysis for all the dwarfs . to be more quantitative , the mean - squared two - sigma error for all the five dwarfs is @xmath178 dex , @xmath179 dex , and @xmath180 dex ( @xmath181 ) for the enclosed mass measured within 300 pc , 600 pc , and @xmath152 ( unique for each galaxy ) , respectively . the accurately determined values of the enclosed mass presented here can be very useful for identifying the dsph counterparts in cosmological simulations , and is probably the most important result of the current study . the conjecture of @xcite that the enclosed mass is well constrained at the half - light radius , @xmath69 , is not corroborated by the current study . in fact , enclosed masses measured at this radius result in the worst mean - squared two - sigma error , @xmath182 dex . the conclusion of @xcite ( whose analysis is hard to interpret , as dm and stars are coupled in their model ) that the enclosed mass has the best constraint at the radius where @xmath183 is also not borne out in the current study : for the five dwarfs i obtained @xmath184 at my best radii ( @xmath152 ) , with the full range from 3.22 to 4.20 . my sample of galaxies is too small to try to find a good estimator for @xmath152 , but i believe it should be a function of both the stellar density profile ( @xmath185 ) and the quality and quantity of the stellar line - of - sight velocity measurements . this issue needs to be explored further once a larger sample of dwarfs is subjected to a similar analysis . to estimate how sensitive the above analysis is to particular values of @xmath82 , i repeated it for all the three values of @xmath82 : 7 , 15 , and 30 . for the less optimal values of @xmath186 , the radii where the enclosed mass is measured to the highest possible accuracy ( @xmath152 ) does not deviate from the value obtained with the optimal @xmath82 by more than 7% ( for sextans ; for fornax it is essentially zero ) . the averaged dm density measured at @xmath152 is also very close for all the values of @xmath82 ( less than one sigma deviation ) , with one notable exception : for leo i , moving from @xmath81 ( the optimal value ) to @xmath187 results in @xmath188 becoming smaller by 0.18 dex , or @xmath189 . ( this may be related to the fact that leo i has the smallest number of stars with a known line - of - sight velocity : @xmath190 . ) overall , this analysis seems to be largely insensitive to how well the optimal value of @xmath82 is chosen . another proposition discussed in the literature is that there is a radius ( 300 pc ) at which the enclosed mass is the same for all dwarf spheroidals around @xmath5 m@xmath191 @xcite . this is barely ( at @xmath192 level ) consistent with the data in the current analysis , as there is a tension between carina on one side , and leo i and sculptor on the other side ( see table [ tab5 ] ) . the galaxies may have a similar mass at this radius , but it is unlikely to be identical . if one considers even smaller radii ( say , 200 pc ) , then formally there will be a common enclosed mass for all the five dwarfs consistent with the data at better than two sigma level , but only because the uncertainties in measuring the enclosed mass increase dramatically at smaller radii ( see figure [ fig3 ] ) . another interesting radius - dependent quantity , local dm density , is more model dependent , and is not as well constrained by the jeans analysis as the enclosed mass . figure [ fig4 ] shows the ranges for @xmath193 spanned by good models . i also plot there the stellar density profiles , @xmath194 , as dash - dotted lines . as with the averaged density case , each galaxy has a well defined radius , @xmath155 , ( vertical dotted lines ) where the local dm density is best constrained by the data . table [ tab5 ] lists the values of @xmath155 and the corresponding ranges of @xmath195 and the local mass - to - light ratio @xmath157 for each galaxy . @xmath155 tends to be a factor of two smaller than @xmath152 . the uncertainty in @xmath195 ranges from 0.20 dex for sculptor to 0.5 dex for sextans , which is substantially worse than the constraints on the enclosed mass . still , knowing the local dm density at a certain radius ( which happens to be quite small ) to a reasonably good accuracy can be very useful for example , for studies aimed at direct search of dm in dwarf spheroidals via its annihilation signal . both figures [ fig3 ] and [ fig4 ] clearly demonstrate that these dsphs are dm dominated at all radii . the last radius - dependent quantity which has useful ( albeit very weak ) constraints is the local stellar velocity anisotropy , @xmath196 . figure [ fig5 ] demonstrates that better than two sigma models span a wide range of @xmath27 at each radius @xmath23 , which is only slightly narrower than the full allowed range ( the space between the two dotted lines ) . the best radius , @xmath158 , is not as well defined as with @xmath197 and @xmath193 . i list the values of @xmath158 and the corresponding good model ranges for @xmath198 in table [ tab5 ] . sculptor has the tightest constraints on @xmath198 ; carina and sextans are essentially unconstrained . given how poor the constraints on @xmath196 are , very little can be learned from the analysis of figure [ fig5 ] . one interesting conclusion is that at any given radius the data are consistent with stellar velocities being isotropic ( @xmath199 ) . one can not conclude from this radius - dependent analysis that the _ global _ stellar isotropy is also consistent with the data . to address the latter issue , one has to analyze the full results of the current study ( available online ) . specifically , one has to test if all the globally isotropic models ( with @xmath200 ) are good ones ( deviate less than @xmath192 from the data ) . i carried out such an analysis , with the conclusion being that indeed the data is consistent ( at better than @xmath192 level ) with the global stellar velocity isotropy hypothesis for all the dwarfs . it has already been demonstrated @xcite that in dsphs the central logarithmic dm density slope , @xmath20 , is not constrained by the jeans analysis ( when no stellar proper motion data are available ) . my analysis generalizes this result for all radii @xmath23 : the local logarithmic dm density slope , @xmath201 , is found to be unconstrained by the jeans analysis . i carried out the classical jeans mass modeling of the five dwarf spheroidal galaxies with the highest quality observational data carina , fornax , leo i , sculptor , and sextans . my primary goal was to push the analysis to its limits , by developing a flexible enough galactic model ( with variable stellar velocity anisotropy ) , refining the traditional @xmath12 fitting algorithm , and investing a significant amount ( @xmath19 cpu hours ) of supercomputing time to carry out an exhaustive model parameter optimization . the main results of this study are as follows . * my galactic model with the six free parameters ( @xmath41 , @xmath42 , @xmath20 , @xmath35 , @xmath36 , and @xmath38 ) gives a good description of the observational data : the normalized @xmath12 for the differences between the model and observed velocity dispersion profiles are close to a unity , which is consistent with the deviations being purely due to the observational errors . * i show that the jeans mass modeling approach ( even with high quality data and modeling ) can not place meaningful constraints on most galactic parameters , including the central dm density logarithmic slope ( @xmath20 ) , dm halo scaling radius ( @xmath41 ) and density ( @xmath42 ) , and central stellar velocity anisotropy ( @xmath35 ) . moreover , i show that the local dm density logarithmic slope , @xmath202 , is unconstrained at all the radii probed by the data . as a consequence , one must resort to much more computationally expensive post - jeans mass modeling techniques ( which deal with the full probability distribution function for stellar line - of - sight velocities , and not just the dispersion ) to be able to solve the `` cusp core '' issue for dsphs , or to identify theoretical mechanisms responsible for flattening dm cusps in dwarf galaxies ( by constraining the stellar velocity anisotropy profiles ) . * my most important finding is that there is a certain radius ( @xmath152 ; different for each galaxy ) where jeans mass modeling provides the tightest constraints on the enclosed mass in dsphs . for carina , fornax , leo i , sculptor , and sextans my two - sigma constraints for the enclosed dm mass are @xmath203 m@xmath6 ( at 409 pc ) , @xmath204 m@xmath6 ( at 924 pc ) , @xmath205 m@xmath6 ( at 389 pc ) , @xmath206 m@xmath6 ( at 435 pc ) , and @xmath207 m@xmath6 ( at 1034 pc ) , respectively . these constraints are much tighter than the constraints at the previously suggested `` good '' radii 300 pc , 600 pc , and the half - light radius . the tight constraints can be very valuable for placing the dsphs in the proper cosmological context . * i also derive useful constraints on the local dm density . similarly to the enclosed mass , each galaxy has a certain radius ( @xmath155 ) where my jeans analysis places the tightest constraints on this quantity . for example , i show that in sculptor @xmath208 @xmath8 pc@xmath9 ( two - sigma interval ) at 256 pc from its center . also , the analysis produced useful minimum central dm density ( @xmath139 ) constraints ( the largest one is for sculptor : @xmath7 @xmath8 pc@xmath9 ) . these constraints can be used in projects aimed at detecting dm in dsphs via its annihilation signal . * i show that stellar anisotropy profiles are very poorly constrained in jeans mass modeling . the only useful results here are that the pure central tangential anisotropy is excluded at better than two - sigma level for three out of the five dsphs , and that the data are consistent with the global stellar velocity isotropy for all the five dwarfs . * a significant advantage of the exhaustive search through the multi - dimensional model parameter space employed in this study , with all the intermediate results stored and available online , is that the results can be used for many other projects . most importantly , this analysis can be used as the first preliminary step ( which eliminates a vast majority of models which are incompatible with the data ) in post - jeans mass modeling projects ( which model the full pdf for the stellar line - of - sight velocities ) the subject of my paper ii in this series . the latter approach should be able to overcome most of the degeneracies present in jeans mass modeling ( as exposed in the current work ) , hopefully settling once and for all whether dsphs have flat dm cores . the simulations reported in this paper were carried out on facilities of the shared hierarchical academic research computing network ( sharcnet : www.sharcnet.ca ) . this research was supported in part by sharcnet . aaronson , m. 1983 , , 266 , l11 an , j. h. , & evans , n. w. 2006 , , 642 , 752 baes , m. , & van hese , e. 2007 , , 471 , 419 binney , j. , & tremaine , s. 1987 , galactic dynamics ( princeton : princeton univ . press ) burkert , a. 1995 , , 447 , l25 coleman , m. g. , da costa , g. s. , & bland - hawthorn , j. 2005 , , 130 , 1065 coleman , m. g. , da costa , g. s. , bland - hawthorn , j. , & freeman , k. c. 2005 , , 129 , 1443 de blok , w. j. g. , & bosma , a. 2002 , , 385 , 816 el - zant , a. , shlosman , i. , & hoffman , y. 2001 , , 560 , 636 gentile , g. , burkert , a. , salucci , p. , klein , u. , & walter , f. 2005 , , 634 , l145 goerdt , t. , moore , b. , read , j. i. , stadel , j. , & zemp , m. 2006 , , 368 , 1073 irwin , m. , & hatzidimitriou , d. 1995 , , 277 , 1354 king , i. 1962 , , 67 , 471 lora , v. , snchez - salcedo , f. j. , raga , a. c. , & esquivel , a. 2009 , , 699 , l113 majewski , s. r. , ostheimer , j. c. , patterson , r. j. , kunkel , w. e. , johnston , k. v. , & geisler , d. 2000 , , 119 , 760 marchesini , d. , donghia , e. , chincarini , g. , firmani , c. , conconi , p. , molinari , e. , & zacchei , a. 2002 , , 575 , 801 martin , n. f. , ibata , r. a. , chapman , s. c. , irwin , m. , & lewis , g. f. 2007 , , 380 , 281 mashchenko , s. , & sills , a. 2004 , , 605 , l121 mashchenko , s. , & sills , a. 2005 , , 619 , 243 mashchenko , s. , sills , a. , & couchman , h. m. 2006 , , 640 , 252 mashchenko , s. , couchman , h. m. p. , & wadsley , j. 2006 , , 442 , 539 mashchenko , s. , wadsley , j. , & couchman , h. m. p. 2008 , science , 319 , 174 mateo , m. l. 1998 , , 36 , 435 mateo , m. , olszewski , e. w. , pryor , c. , welch , d. l. , & fischer , p. 1993 , , 105 , 510 mateo , m. , olszewski , e. w. , vogt , s. s. , & keane , m. j. 1998 , , 116 , 2315 mateo , m. , olszewski , e. w. , & walker , m. g. 2008 , , 675 , 201 merritt , d. 1985 , , 90 , 1027 moore , b. , ghigna , s. , governato , f. , lake , g. , quinn , t. , stadel , j. , & tozzi , p. 1999 , , 524 , l19 muoz , r. r. , et al . 2006 , , 649 , 201 navarro , j. f. , frenk , c. s. , & white , s. d. m. 1997 , , 490 , 493 odenkirchen , m. , et al . 2001 , , 122 , 2538 osipkov , l. p. 1979 , pisma astron . , 5 , 77 press , w. h. , teukolsky , s. a. , vetterling , w. t. , & flannery , b. p. 1992 , cambridge : university press , |c1992 , 2nd ed . , sgall , m. , ibata , r. a. , irwin , m. j. , martin , n. f. , & chapman , s. 2007 , , 375 , 831 srsic , j. l. 1963 , boletin de la asociacion argentina de astronomia la plata argentina , 6 , 41 smoli , v. , zucker , d. b. , bell , e. f. , coleman , m. g. , rix , h. w. , schinnerer , e. , ivezi , . , & kniazev , a. 2007 , , 134 , 1901 strigari , l. e. , bullock , j. s. , kaplinghat , m. , diemand , j. , kuhlen , m. , & madau , p. 2007 , , 669 , 676 strigari , l. e. , bullock , j. s. , kaplinghat , m. , simon , j. d. , geha , m. , willman , b. , & walker , m. g. 2008 , , 454 , 1096 tolstoy , e. , et al . 2004 , , 617 , l119 tonini , c. , lapi , a. , & salucci , p. 2006 , , 649 , 591 van den bosch , f. c. , & swaters , r. a. 2001 , , 325 , 1017 walker , m. g. , mateo , m. , olszewski , e. w. , gnedin , o. y. , wang , x. , sen , b. , & woodroofe , m. 2007 , , 667 , l53 walker , m. g. , mateo , m. , & olszewski , e. w. 2009 , , 137 , 3100 walker , m. g. , mateo , m. , olszewski , e. w. , sen , b. , & woodroofe , m. 2009 , , 137 , 3109 walker , m. g. , mateo , m. , olszewski , e. w. , pearrubia , j. , wyn evans , n. , & gilmore , g. 2009 , , 704 , 1274 wolf , j , et al . 2009 , , submitted ( arxiv:0908.2995 )

ram pressure stripping is an important process in the evolution of both dwarf galaxies and large spirals . large spirals are severely stripped in rich clusters and may be mildly stripped in groups . dwarf galaxies can be severely stripped in both clusters and groups . a model is developed that describes the stripping of a satellite galaxy s outer h i disk and hot galactic halo . the model can be applied to a wide range of environments and satellite galaxy masses . whether ram pressure stripping of the outer disk or hot galactic halo occurs is found to depend primarily on the ratio of the satellite galaxy mass to the mass of the host group or cluster . how the effectiveness of ram pressure stripping depends on the density of the inter - group gas , the dark matter halo concentrations , and the scale lengths and masses of the satellite components is explored . the predictions of the model are shown to be well matched to h i observations of spirals in a sample of nearby clusters . the model is used to predict the range of h i gas fractions a satellite of mass can lose orbiting in a cluster of mass .

astro-ph0610088

ram pressure stripping was first proposed by @xcite to explain the observed absence of gas - rich galaxies in clusters . they noted that galaxies falling into clusters feel an intracluster medium ( icm ) wind . if this wind can overcome the gravitational attraction between the stellar and gas disks , then the gas disk will be blown away . they introduced a simple analytical condition to determine when gas is lost : @xmath2 the right - hand side , @xmath3 , is a gravitational restoring pressure , @xmath4 is the density of the icm , and @xmath5 is the orbital speed of the satellite galaxy . using this condition they concluded that spirals should lose their gas disks when they pass through the centers of clusters . galaxies near cluster centers have earlier type morphologies , are redder , and form fewer stars than galaxies in the field of similar luminosity @xcite . stripped galaxies have low h i masses for their size and morphology . such h i deficiency correlates with lower star formation rates ( sfrs ) @xcite , and lowered sfrs lead to redder colors . therefore , ram pressure stripping , in conjunction with processes like harassment that can alter the structure of a galaxy , may play a role in transforming galaxy morphologies . several sets of observations indicate that stripping occurs to spirals in clusters . @xcite define a deficiency parameter that compares a galaxy s observed h i mass to the expected h i mass in a field galaxy with the same morphology and optical size . galaxies in virgo , @xcite , coma , @xcite , and other nearby clusters , @xcite are observed to be h i deficient . @xcite observe that the average deficiency increases with decreasing distance to the clusters centers and that deficient galaxies have , on average , more eccentric orbits . @xcite observe indications of stripping in the h i distributions of individual galaxies in the virgo cluster . compared to an average profile for a field galaxy of similar morphology and optical extent , these galaxies have normal gas densities in their inner disks but are sharply truncated . more detailed observations of several virgo spirals have since been carried out . ngc 4522 is observed to have an undisturbed optical disk , a severely truncated h i disk , and extra planar h@xmath6 and h i on one side of the disk @xcite . ic 3392 , ngc 4402 , ngc 4419 , and ngc 4388 all have a truncated h i disk , and the first three have extraplanar h i @xcite . ngc 4569 has a truncated h i disk and extra planar h@xmath6 @xcite . observing asymmetric , extraplanar gas in combination with an undisturbed optical disk is a strong indication that the galaxy s gas is interacting with the icm . detailed simulations of ram pressure stripping confirm that spirals in clusters should be stripped . for some simulations the final radii of the galaxies are compared to the predictions of the gunn and gott condition . in addition , the morphologies of simulated galaxies have been compared to observations of spirals in virgo . simulations by @xcite show that galaxies in a face - on wind are stripped to radii near those predicted by the gunn and gott condition , but galaxies in an edge - on wind are only mildly stripped . in contrast , other simulations show a two - step stripping process that strips galaxies at all inclination angles . ram pressure stripping first quickly strips gas from the outer disk , and then the remaining disk is slowly viscously stripped @xcite . @xcite point out that , unlike ram pressure stripping , viscous stripping is more effective in an edge - on wind . they find that the final radii of the face - on galaxies match those predicted using the gunn and gott condition and that edge - on galaxies are stripped to slightly larger radii . @xcite run a series of simulations to test the dependence of the stripping radius on the speed of the icm wind , the mach number , and the vertical structure of the gas disk . they find that the most important parameters are the ram pressure and surface density of the gas disk and that the effects of varying the mach number and vertical structure are minor . they find that the original gunn and gott condition does a fair job predicting the stripping radius . however , when they use an adjusted ram pressure , they find that using the thermodynamic pressure in the central plane of the disk more accurately predicts the stripping radius . simulations of ram pressure stripping can also match the range of morphologies of stripped galaxies . @xcite track the density , velocity , and velocity dispersion of a simulated galaxy s cold disk gas . in ngc 4522 , the kinetically continuous extraplanar gas matches simulations of a galaxy that is currently being stripped . however , the gas in ngc 4569 is less pronounced , not kinetically continuous with the disk gas , and has a large velocity dispersion , more like simulations of a galaxy that has been stripped in the past . in their simulations @xcite observe a process they term `` annealing '' . the icm wind compresses the gas disk and triggers the formation of large spiral arms . the interaction of these arms with the wind leads to the contraction of the galaxy and the formation of a dense truncated gas disk . a burst of star formation may occur at the edge of the truncated disk . a ring of star formation is seen at the truncation radius of the disk in ic 3392 . some groups have studied stripping in groups and cluster outskirts where the ram pressure is lower than in cluster centers . while simulations and observations both support the idea that large spirals in clusters are stripped of their h i disks , it is not clear that stripping occurs in poorer environments . two types of stripping may occur in groups . the ram pressure in groups is lower than in clusters . however , dwarf galaxies have lower masses than large spirals and therefore lower restoring pressures , and dwarf galaxies may be stripped of most of their h i. in addition , large spirals have restoring pressures that decrease with radius and may be stripped of their outer disks . @xcite run simulations of dwarf - like galaxies in winds typical of the ram pressure in groups and find that the simulated dwarfs are stripped . the @xcite simulations include runs with low ram pressure . their large and medium - sized model spirals are stripped of some of their outer h i in these runs . the observational evidence for stripping in low - mass systems is circumstantial . the gas content and star formation histories of local group dwarfs correlate with their distances from the dominant spirals @xcite . this project aims to use the clear observations of stripping in clusters to predict when stripping occurs in groups . this is done by examining how the ram and restoring pressures depend on the masses of the cluster or group and the satellite galaxy . two ways in which the masses of the host system and satellite galaxy affect the ram and restoring pressures are identified . satellites orbital speeds are set by the depth of the cluster s potential , and the restoring force on the gas disk is determined by the depth of the satellite s potential . in addition , both cluster and galaxy morphologies change with mass . a model is developed and used to study both types of mass dependence . this is done in the following steps . in the next section a model for the mass and gas distributions is presented both in terms of physical parameters and in terms of scale - free parameters . in section 3 the dependence of ram pressure stripping on the masses of the satellite and cluster is presented assuming that the scale - free parameters are constant . in section 4 the dependence of the scale - free parameters on the masses of the satellite and cluster and the sensitivity of the model to these changes are explored . in section 5 the results are discussed and the model is compared to observations of stripping . section 6 concludes . the analytical model of ram pressure stripping developed here uses the gunn and gott condition to predict the extent to which galaxies are stripped . this condition is based on a simplified picture of stripping that ignores the details of the hydrodynamical processes and balances forces on large scales . however , simulations of galaxies that are stripped by a face - on wind result in stripping radii that match the predictions of the gunn and gott condition @xcite . in the simulations galaxies in an edge on wind are eventually viscously stripped to radii similar to those of face - on , ram pressure stripped galaxies . this is discussed further in section [ sec inclination ] . in light of the general agreement between gunn and gott s condition and simulations of ram pressure stripping , it is assumed that a model based on this condition can be used to make general statements about where and to what extent galaxies are stripped . the model for the group or cluster has two components , a gravitational potential in which the satellite orbits and an icm against which the satellite shocks . the gravitational potential is modeled using an nfw profile with mass @xmath1 , radius , @xmath7 , and concentration , @xmath8 . the scaled radius @xmath9 . @xmath10 the nfw profile is discussed in more detail in appendix [ nfw profile ] . the icm is modeled using a @xmath11 profile with scale length @xmath12 , and central density , @xmath13 . @xmath14 the @xmath11 profile can be rewritten as @xmath15 the parameter @xmath6 relates @xmath13 to the characteristic density of the nfw profile , and @xmath16 is the average dark matter density within @xmath7 . [ sec . restoring pressure ] the gravitational potential of the satellite and the surface density of the gas disk are needed to model the restoring pressure . the gravitational potential combines a dark matter halo , a stellar disk , and a stellar bulge . the dark matter halo is modeled as an nfw profile with mass @xmath0 and concentration @xmath17 . the disk , of mass @xmath18 , is described using a miyamoto and nagai potential @xcite . @xmath19 this potential is easily differentiated , and with the appropriate choice of disk scale length , @xmath20 , and height , @xmath21 , many of the properties of the potential of an exponential disk can be matched @xcite . this potential can be rewritten as @xmath22 where @xmath23 , @xmath24 , @xmath25 , and @xmath26 . the bulge , of mass @xmath27 , is described using a hernquist potential @xcite . @xmath28 this can be written as @xmath29 where @xmath30 , @xmath31 , and @xmath32 is the distance from the center of the satellite . by introducing @xmath33 and @xmath34 , the full gravitational potential of the satellite can be written in the form @xmath35 gas in galaxies can be found in three spatial components : an exponential molecular gas disk with a scale length comparable to the stellar disk , a nearly flat atomic disk that extends beyond the stellar disk and shows a sharp cutoff , and a hot galactic halo . beyond the disk cutoff , the atomic disk may continue as an ionized gas disk @xcite . the h i is modeled as a thin flat disk with mass @xmath36 , surface density , @xmath37 , and a sharp cutoff at radius @xmath38 @xmath39 @xmath40 where @xmath41 is the fractional mass of the gas and @xmath42 is the scaled size of the disk . the inner gas disk is dominated by the @xmath43 disk . the molecular disk is significantly more difficult to strip than the h i disk both because the @xmath43 disk is more compact and because the @xmath43 is found in molecular clouds . these clouds do not feel the effect of stripping as strongly as the diffuse h i. how stripping of @xmath43 occurs and how the @xmath43 clouds and diffuse h i in the inner disk interact are not known . the two phases may be tied together by magnetic fields ; in which case the inner h i disk will remain until the ram pressure can remove the entire inner disk . it may also be possible for the wind to remove the h i from the inner disk while leaving the @xmath43 behind . because of this uncertainty , the stripping of the inner disk will not be studied here . the model is only used to discuss stripping beyond 1.5 stellar scale lengths . if the ram pressure is not strong enough to affect the @xmath43 , the @xmath43 will contribute to the restoring pressure in the outer disk . therefore , the mass of the @xmath43 disk is added to the stellar mass . the emphasis of this project is to study the stripping of the h i disk . however , the fate of the hot galactic halo can also influence the evolution of the stripped galaxy . it is expected that the hot halo will be easily stripped both because it is diffuse and because the restoring potential of the satellite is strongest in the disk . to check this a hot halo is included in the model galaxy . @xcite give an analytic condition for the complete stripping of a hot galactic halo . @xmath44 where @xmath45 and @xmath46 are the central thermal pressure and density of the hot galactic halo , @xmath47 is the boltzmann constant , @xmath48 is the temperature of the galactic halo , and @xmath49 is the average molecular mass . the thermal pressure replaces the gravitational restoring pressure in the original stripping condition . in their simulations , this condition does a fair job of predicting the mass of a galaxy that can be completely stripped . this condition is adapted for the current model and all gas outside the radius where @xmath50 is assumed to be stripped . the hot galactic halo is modeled by assuming that the gas is at the virial temperature , @xmath51 , of the satellite and is in hydrostatic equilibrium in the satellite s dark matter potential . the self - gravity of the gas is ignored . the density of the hot halo is @xmath52 where scaled density profile @xmath53 is defined in appendix [ ap satellite halo ] and @xmath54 is the ratio of the mass of gas in the hot galactic halo to @xmath0 . a dimensionless temperature , @xmath55 is also introduced in appendix [ ap satellite halo ] . it is defined by @xmath56 it is a function of only @xmath57 . in this section the dependence of stripping on the masses of the large and satellite halos and on the model s parameters is explicitly determined . the maximum ram pressure , the ram pressure at the pericenter of the galaxy s orbit , is given by @xmath58 where @xmath59 is the distance of closest approach , @xmath60 is the orbital energy per unit mass , @xmath61 , @xmath62 , and @xmath63 is the orbital angular momentum per unit mass . both @xmath64 and @xmath65 are expected to be independent of the masses of the satellite and the cluster . at any point in the orbit , @xmath66 , the ram pressure is @xmath67 where dimensionless pressure , @xmath68 is defined . @xmath69\left(1 + \frac{s^2}{(r_c / r_{v , gr})^2 } \right)^{(-3/2)\beta}\ ] ] equations for the orbital speeds are derived in appendix [ ap orbits ] . assuming that @xmath6 , @xmath11 , @xmath70 , and @xmath71 do not depend on the mass of the halo , the @xmath1 dependence of both @xmath72 and @xmath73 is @xmath74 . for a potential of the form in eq . [ eqn . full restoring pressure ] , the force per unit gas mass in the @xmath75 direction is found as follows . @xmath76 @xmath77 assuming that the @xmath78 and @xmath79 introduced in section [ sec . restoring pressure ] and @xmath57 are constant with mass , the mass dependence of the restoring force per unit gas mass is given by @xmath80 . for the gas disk , if the fractional mass of the gas , @xmath81 , and the scaled size of the disk , @xmath82 , are both constant , then @xmath83 combining the above , @xmath84 for any radius along the disk , @xmath85 , there is a maximum restoring pressure . if the maximum restoring pressure is greater than the ram pressure , the satellite holds the gas at this radius . the condition for the satellite holding its gas can be written as @xmath86 @xmath87^{3/2}\ ] ] the restoring pressure of the galactic hot halo is @xmath88 the condition for retaining this gas within @xmath89 is @xmath90^{3/2}\ ] ] if the model s parameters : @xmath71 , @xmath57 , @xmath91 , @xmath92 , @xmath11 , @xmath70 , @xmath6 , the @xmath78 , and the @xmath79 ; are all independent of the mass of the satellite and cluster , then the fraction of gas that is stripped from a satellite depends on the ratio @xmath93 . in this case , @xmath94 and for both the gas disk and hot galactic halo @xmath95 [ model ind ] assuming that this set of parameters is constant is equivalent to assuming that the physical parameters scale with mass in the most obvious way . it assumes that component masses scale as @xmath96 , lengths scale as @xmath97 , and the central densities of the icm and the satellite s hot galactic halo are constant . for any model in which these scalings hold , the result that the extent of stripping depends on @xmath93 will hold . for any cluster mass , the scaled radius will set the icm density and the orbital velocity will be proportional to @xmath98 . the restoring pressure depends on the depth of the satellite galaxy s gravitational potential well and the density of the gas disk . for a generic potential , @xmath99 and the restoring force per unit gas mass is proportional to @xmath100 . if the mass of the gas disk scales with @xmath0 and the disk length with @xmath101 , then the density of the gas disk scales as @xmath100 and the restoring pressure as @xmath102 in this section a reference set of values for the model s parameters is introduced and the effect of varying these parameters is discussed . the reference parameters determine the values of @xmath93 at which stripping occurs . they are taken from observations , where possible , and @xmath103cdm simulations , otherwise . two types of parameter variations are discussed . first , several of the model parameters vary systematically with @xmath0 or @xmath1 . second , at fixed masses the parameters have scatter . the scatter in the parameters is discussed for several reasons . by combining an estimate for the scatter in each parameter with the model , the parameters that have the most effect on the radius to which an individual galaxy is stripped can be identified . second , it is useful to know the range of ratios across which stripping occurs . finally , if the overall scatter is not too large , the relationship between @xmath85 and @xmath93 could be used to search for signs of stripping in a large galaxy survey . the results of this section are discussed in the terms above in section [ predictions ] . in figures [ figure 1 ] and [ figure 2 ] , the @xmath0 at which a satellite is stripped is plotted versus @xmath104 and @xmath105 for a satellite orbiting at a variety of @xmath66 values in @xmath106 , @xmath107 , and @xmath108 clusters . the mass of the stripped gas is denoted as @xmath109 . table 1 lists the model parameters used for three models , a large satellite in a @xmath110 cluster , the reference model ; a large satellite in a @xmath111 cluster , the middle - mass model ; and a small satellite in a @xmath108 group , the low - mass model . the large satellite model is based on the milky way . physical length scales and masses for the stellar disk and bulge are from @xcite . the stellar disk is modeled using @xmath112 , @xmath113 kpc , and @xmath114 kpc . the equivalent @xmath20 for an exponential disk is 5 kpc . the bulge is modeled using @xmath115 and @xmath116 kpc . a flat disk with a radius of 30 kpc gives @xmath117 similar to that seen in observations . the h i gas mass is chosen to result in a disk density of @xmath118 @xmath119 @xmath120 . the virial mass is chosen such that @xmath121 . setting @xmath0 also sets @xmath101 . the concentrations used are @xmath122 and @xmath123 , 7.5 , and 10 . the concentrations are based on @xcite and assume an average over - density within @xmath97 , @xmath124 , of 337 . the assumed over - density affects the concentrations , which are measured using @xmath125 , and @xmath6 , because it relates the icm and dark matter densities . note that @xmath124 does not appear in equation [ disk m ratio ] or [ halo m ratio ] . the icm parameters , @xmath6 , @xmath11 , and @xmath126 , are set by x - ray observations of clusters and groups , and the orbital parameters are taken from simulations . both are discussed below . table 1 also lists the assumed scatter in each model parameter and resulting scatter in the mass at which a satellite is stripped . the profile of the icm is determined by the parameters @xmath6 , @xmath127 , and @xmath11 ( eq . [ beta ] ) . @xcite compile a set of average icm profiles for clusters with temperatures ranging from 0.3 to 17 kev . they find that the icm of groups and poor clusters is both less dense and more extended than in rich clusters . @xcite present x - ray observations of a set of clusters and @xcite and @xcite present x - ray observations of groups . the three groups present @xmath11 and @xmath12 ( in units of kpc ) for a best - fit @xmath11 profile and the x - ray temperature , @xmath128 , for each group or cluster . the average @xmath11 and @xmath126 for the cluster observations are approximately 0.67 and 12 . the averages for the group observations are approximately 0.45 and 170 . to convert @xmath12 ( in kpc ) to @xmath70 , virial masses for the groups and clusters are found using the @xmath129 relationship from @xcite . the mass @xmath130 is the mass contained within the radius at which the average over - density reaches 200 . it is converted to @xmath131 using equations [ nfwmass ] and [ mv of rv ] . the profiles in @xcite are scaled using @xmath132 as determined in @xcite . the differences in the @xmath70 determined using the two @xmath129 relationships are within the intergroup scatter in @xmath70 . the average profiles for the hottest and coolest groups from @xcite are well matched by using @xmath133 and 20 respectively and the appropriate @xmath11 and @xmath126 . in figures [ figure 1 ] and [ figure 2 ] , plots are made for stripping in @xmath134 , @xmath107 , and @xmath110 clusters . a cluster with @xmath135 falls into the hottest temperature bin from @xcite while a group with @xmath136 falls into the coolest . the icm parameters for the @xmath111 cluster are @xmath137 , @xmath138 , and @xmath139 . scatter in these parameters is estimated using the three sets of observations . the observed scatter in @xmath11 is @xmath140 . the majority of clusters have @xmath126 between 7 and 20 , and groups have @xmath126 between 75 and 270 . for both groups and clusters , @xmath6 is assumed to vary by a factor of @xmath141 in either direction . varying @xmath11 varies the ram pressure at @xmath142 by a factor of 2 for the clusters and a factor of 7.4 for the groups . the ram pressure varies by a factor of 6 in the clusters and 5.4 in the groups when @xmath126 is varied . varying @xmath6 by a factor of 6 results in an increase in the ram pressure by the same factor . the orbital parameters are important for determining both satellites pericenters and orbital speeds . in theory , the pericenter of a satellite s orbit is determined by @xmath143 and @xmath92 ( eq . [ turningpoint ] ) . however , @xcite present pericenter distributions for subhalos at @xmath144 in cluster - sized dark matter halos with @xmath145 . they find that the @xmath59 distribution can be fit with a gaussian with @xmath146 and @xmath147 . this distribution shows little variation with either @xmath1 or @xmath0 . the maximum ram pressure in a cluster with icm parameters corresponding to a few @xmath111 and @xmath148 increases by a factor of 5.4 when the pericenter decreases from @xmath149 0.23 . some satellites have not yet passed through the pericenter of their orbits , and the maximum ram pressure they have experienced is determined by their current @xmath66 . increasing @xmath66 from 0.35 to 1 results in a decrease of the mass at which a satellite can be stripped by two orders of magnitude . therefore , unless it is known that a group of satellites have all passed through their pericenters , the effect of varying @xmath66 can not be treated like the scatter in the other parameters . instead , when comparing satellites in different environments or with different masses , @xmath66 must be specified . at a given @xmath66 , the ram pressure is dependent on @xmath150 , but not @xmath151 . in simulations @xcite find that the speeds of satellites entering dark halos can be fit with a maxwell - boltzmann distribution with @xmath152 , @xmath153 . the total energy per unit mass of an object at the virial radius with this speed is @xmath154 , and is a weak function of @xmath8 . this distribution has a large scatter , @xmath155 , corresponding to @xmath91 between 0.2 and 2.8 . at @xmath142 , when @xmath156 is varied between 0.55 and 1.75 , the @xmath93 needed for stripping vary by a factor of @xmath157 . @xcite also find that for @xmath158 the angular momenta of satellites entering halos can be fit with a maxwell - boltzmann distribution with @xmath159 , @xmath160 . as expected , the distributions of @xmath91 and @xmath92 do not depend on @xmath0 and @xmath1 . however , unless the correlation between @xmath156 and @xmath151 is understood , these parameters can not be used to determine the distribution of @xmath59 . the gunn and gott condition assumes that all galaxies orbit in a face - on wind . in reality , galaxies orbit at all inclinations . simulations show that galaxies are stripping at all inclinations , but edge - on galaxies are stripped to larger radii than face - on galaxies . @xcite point out that while galaxies that are hit face - on by the icm wind experience the most ram pressure stripping , galaxies in an edge - on wind experience the most viscous stripping . they derive an expression for the ram pressure needed to viscously strip an edge - on galaxy to radius @xmath161 and treat this pressure as a restoring pressure for the edge - on case . when they compared the new condition to the gunn and gott condition , they find that any ram pressure that can completely strip a face - on galaxy will viscously strip an edge - on galaxy and that any ram pressure that can not strip a face - on galaxy will not strip an edge - on galaxy . the two models differ when a galaxy is partially stripped . in this case edge - on galaxies are still stripped , but to larger radii than face - on galaxies . they run simulations at a variety of inclination angles , @xmath162 , and find that the final stripping radii match the predictions of the two conditions . when galaxies at intermediate angles are partially stripped , they are stripped to intermediate radii . when marcolini et al . plot the two restoring pressures for three galaxy masses , 0.76 , 7.4 , and 77.2 @xmath163 , their figure 3 , the fractional difference between the two restoring pressures is close to constant throughout the outer disk but increases with the mass of the galaxy . for the largest galaxy the restoring force in the edge - on case is @xmath164 times larger . small dark matter halos tend to be more concentrated than large dark matter halos . @xcite use an analytical model of the evolution of dark matter halos to describe this dependency . for the standard @xmath103cdm model , near @xmath165 they find @xmath166 the @xmath8 used for the cluster and group models are determined using this relation . @xcite also see a scatter that is as large as the evolution of the concentration over the range @xmath167 . to explore the scatter introduced through @xmath8 , @xmath8 is varied between 5 and 16 . the cluster concentration affects both @xmath59 and the ram pressure at a given @xmath66 ( eqs . [ turningpoint ] and [ v sat ] ) . when @xmath8 is increased from 5 to 16 , for @xmath168 and @xmath169 , the pericenter decreases from @xmath170 to 0.5 . this is smaller than the scatter in the pericenter . at a given @xmath66 , the orbital speed increases slightly as @xmath8 increases . this difference increases as @xmath66 decreases . for @xmath171 , the change is not noticeable . for @xmath172 , the @xmath93 at which a satellite is stripped increases by a factor of 1.2 when @xmath8 is increased from 5 to 16 . in the outer h i disk the restoring pressure of the dark matter halo makes a non - negligible contribution to the restoring pressure . more concentrated satellites have higher restoring pressures than other satellites of the same mass . varying @xmath57 between 12 and 20 varies the @xmath93 at which stripping occurs by a factor of @xmath173 . the concentration also alters how the hot galactic halo is stripped . in a more concentrated dark matter halo the gas is denser in the center and more diffuse elsewhere . stripping can therefore occur down to a smaller radius , but a smaller fraction of the gas is lost . how disk scale lengths vary with @xmath0 is an open question . theory suggests that when cold gas disks form in dark matter halos , their initial radii are proportional to the virial radii of the halos they form in @xcite . there is a large scatter in this relationship that is due to a large scatter in the disk angular momenta . however , cold disks are composed of a combination of stars and gas , and how these components arise from the original cold disk is not understood . the fraction of the h i that is found in the outer disk , where the restoring force from the stellar disk is weakest , is set by the ratio @xmath174 ( eqs . [ stellar disk 2 ] and [ sigma not ] ) . @xcite observed the h i disks of 73 local dwarfs . they found @xmath175 for the dwarfs . for late - type spirals @xmath176 . these observations suggest that @xmath174 contributes to the scatter in @xmath93 , but not to the mass dependence . varying @xmath174 changes both the size and the shape of the restoring force . when @xmath82 is held at 5.7 and @xmath177 is decreased to 3.5 , the restoring force decreases within @xmath178 and increases at larger @xmath161 . the fractional change in @xmath179 is less than 1.1 throughout the disk . holding @xmath82 constant and increasing @xmath177 to 8.5 , results in decreasing the restoring force outside of @xmath180 by a factor of less than 1.13 . varying @xmath82 while holding @xmath174 constant changes the density of both the gas and stellar disks . at a given @xmath181 , the density of both disks @xmath182 and the restoring pressure @xmath183 . the scatter in @xmath82 and how @xmath82 evolves with mass are not known . however , because of the heavy dependence of the restoring pressure on it , the scatter in @xmath82 is likely to make an important contribution to the scatter in the @xmath93 at which stripping occurs . increasing @xmath82 from @xmath184 to @xmath185 decreases the @xmath93 needed for stripping by a factor of 20 . the presence of molecular gas complicates the physics of stripping in the inner disk . therefore , this model is only applied to disk radii beyond 1.5 stellar disk scale lengths . at these radii , @xmath186 , the bulge acts as a point mass , and changing @xmath187 ( eq . [ bulge 2 ] ) has no effect . in particular , the radius of the bulge changes with the mass of the bulge . below , the mass of the bulge is allowed to vary by 40% . assuming that @xmath188 , this corresponds to allowing @xmath189 to vary by 12% . this is not noticeable in the outer h i disk . both @xmath190 and @xmath81 ( eqs . [ bulge 2 ] and [ sigma not ] ) are correlated with the mass of the satellite . while the majority of large spiral galaxies have a significant fraction of their stars in a stellar bulge , low - mass late - type galaxies do not . very low mass spiral galaxies exist that do not contain any bulge @xcite and dwarf galaxies are observed to be either de / dsph or dirr . in addition , low - mass late - type galaxies have a higher fraction of their mass in the gas disk than large spirals @xcite . decreasing @xmath190 reduces the restoring pressure in the outer disk . in contrast , increasing @xmath81 increases the restoring pressure . in the bottom right panels of figures [ figure 1 ] and [ figure 2 ] , the satellite is bulgeless , the gas mass fraction , @xmath81 , is doubled , and the stellar mass fraction , @xmath191 , is reduced by 20% . the original model corresponds to a large late - type spiral , while the bulgeless model is closer to a late - type dwarf galaxy . because the restoring force due to the bulge is small , the net effect is to increase the restoring pressure . most galaxies will lie between the two plotted models . the @xmath78 can be split between those that set the depth of the potential well , @xmath192 and @xmath190 , and @xmath81 which sets the gas density . in late - type galaxies , bulges can contribute up to 50% of the stellar light @xcite . in the reference model , 25% of the stars are in the bulge . varying @xmath190 between 15% and 35% of the satellite s stars , varies the restoring force in the outer disk by a factor of @xmath193 . neither the mass dependence nor the scatter in @xmath192 are known . varying @xmath192 by 20% , varies the restoring force by a factor of @xmath194 . the effect of scatter in the gas density parameter , @xmath81 , is straight forward . the @xmath93 at which stripping occurs varies as @xmath195 . varying @xmath81 by a factor of 1.5 , similar to the scatter in @xmath196 in figure 9 of swaters et . al . , varies the @xmath93 needed for stripping by a factor of 1.8 . variations in the @xmath78 values should be correlated because the variation in the mass in baryons in a satellite is likely smaller than the variation in the mass of a given component . the correlation is such that it will reduce the overall scatter . in this section , the @xmath93 values at which the model predicts stripping should occur are compared to observations of h i disks from @xcite . this sample is useful for making comparisons between the model and observations because the number of spirals is large and the observations span several clusters . the average temperature of the clusters is 3.5kev . this corresponds to @xmath197 . the average gas profile for this temperature can be described using the middle mass cluster model . in figure 4 from solanes the fraction of spirals with @xmath198 and the average deficiency are plotted versus @xmath199 , where @xmath200 is the abel radius . in figure [ figure 4 ] of the present paper @xmath0 is plotted versus the deficiency in a @xmath201 cluster at several @xmath66 . the deficiency is defined as @xmath202 , where @xmath203 is the observed h i mass and @xmath204 is the average h i mass for field spirals with the same morphology and optical diameter . for the model @xmath205 . the observed fraction of spirals with @xmath198 reaches 0.3 at @xmath206 and 0.45 at @xmath207 . the model predicts that almost all galaxies within @xmath208 have @xmath198 . however , only @xmath209 of the galaxies viewed within @xmath210 actually reside within @xmath210 . this estimate assumes that the galaxy distribution follows an nfw profile and that all galaxies within @xmath211 are included in the projected cluster . at @xmath212 , the model predicts that an average galaxy traveling face - on to the wind is stripped to @xmath198 for @xmath213 . this estimate can be roughly adjusted for inclination by multiplying by @xmath214 , @xmath215 . approximately 45% of the galaxies seen within @xmath212 actually reside within @xmath212 . some galaxies beyond the @xmath66 under consideration will be stripped to @xmath198 . therefore , the observed fraction of spirals with @xmath198 should be greater than that predicted by multiplying the volume fraction with the fraction of stripped galaxies within @xmath66 . this condition is met for @xmath212 if at least a third of the observed spirals have @xmath216 , corresponding to @xmath217 . for @xmath208 , the volume fraction is lower than the observed fraction and this condition is easily met . the difference between the observed value and product of the volume fraction and stripped fraction should decrease as @xmath66 increases . a @xmath218 spiral traveling face - on to the wind is stripped to @xmath219 at @xmath220 . a @xmath221 spiral traveling face - on to the wind is stripped to @xmath219 at @xmath222 . spirals with @xmath219 are not observed in the @xcite sample outside of @xmath223 . the model does a good job of matching the observations of h i disks from @xcite . if the model over - predicted the degree of stripping , then it would over - predict the fraction of spirals with @xmath198 and predict deficiencies at large radii greater than those observed . if the model under - predicted the degree of stripping , then it might not predict that spirals in the center of the cluster can have @xmath219 and would require that a larger fraction of the observed spirals reside in the cluster center . both @xcite and @xcite run simulations of disk galaxies in winds typical of the outskirts of a group , equivalent to @xmath212 and 0.7 , and the center of a group , @xmath224 and 0.25 . in marcolini the two lowest mass simulated dwarfs are completely stripped in the high ram pressure . in the other cases , the dwarfs are partially stripped to a range of radii . when the gas surface density in the current model is adjusted to the exponential gas density in the dwarf models , the current model and the simulations match well for both @xmath66 . the only disparity is that for @xmath224 the high - mass model is stripped to @xmath225 in the simulations and to @xmath226 in the model . this is probably due to differing dark matter halo profiles . the high - mass , @xmath227 , galaxy model of roedinger and hensler is stripped to @xmath228 in the lower ram pressure and @xmath229 in the higher ram pressure . after the gas density is adjusted , the stripping radii of their high - mass model in the two winds are matched well by the model . in this model there exist three stripping regimes , stripping of the hot galactic halo , the flat h i disk , and the inner gas disk . the focus is on stripping of the flat h i disk . however , the model also includes a hot galactic halo . this hot gas is easily stripped in all environments . a satellite galaxy that is orbiting at @xmath230 and has a hot galactic halo consisting of 0.5% of its mass , is stripped of at least @xmath231 of its hot halo . at @xmath172 , a typical pericenter , all satellite galaxies are stripped of practically their entire halo in both groups and clusters . the stripping of the inner disk has not been studied . however , this gas should be difficult to strip and any spiral that has not been stripped of its entire outer h i disk should retain this gas . the rest of this section focuses on the outer disk . the model does a reasonable job of matching h i observations in large cluster spirals . therefore , it can be used to predict the extent of stripping at different @xmath0 and @xmath1 . the model s basic prediction , that the stripping radius is determined by the ratio @xmath93 is modified by the mass dependence of the model s parameters . therefore , the usefulness of the model s predictions depends in part on how well this dependence is understood . fortunately , the mass dependence of many of the parameters is based on observations . the model parameters that most alter @xmath232 in low - mass systems are the icm profile parameters and the distribution of mass in the galaxy . the scaled ram pressure , @xmath233 , that a satellite experiences decreases with @xmath1 because the icm density throughout most of the cluster decreases . the scaled restoring pressure , @xmath234 , is larger in low - mass disks because a higher fraction of the disk mass is found in the h i disk . the changes in both the scaled ram and restoring pressures act to decrease the @xmath93 at which stripping occurs in low - mass systems . for example , the local group dwarfs have @xmath93 smaller than the bright spirals in virgo but should experience similar amounts of stripping . the model makes three basic predictions . dwarf galaxies are stripped of their entire outer h i disks in clusters and lose varying fractions of their outer disk in groups . massive spirals can also be stripped of significant fractions of their h i disk in groups if they travel to small @xmath66 . in a @xmath108 group , a @xmath235 dwarf is stripped of 50% of its h i disk at @xmath236 , a @xmath218 satellite is similarly stripped at either @xmath208 or @xmath172 , depending on the satellite model used , and a large @xmath237 satellite is similarly stripped at @xmath238 . the corresponding orbits from stripping 80% of the h i disk are @xmath239 , 0.35 , 0.25 , and 0.2 . it should be kept in mind that these numbers are for an average satellite traveling face - on to the icm wind . the degree of stripping that a particular satellite of mass @xmath0 experiences orbiting in a cluster of mass @xmath1 can vary greatly . the parameters that are responsible for most of the variation in @xmath109 are the icm parameters , @xmath6 , @xmath11 , @xmath126 , the galaxy s inclination , @xmath162 , the pericenter , @xmath59 , and the extent of the disk , @xmath82 . the variation in the stellar and gas surface densities is largely contained in @xmath82 . the icm parameters are set by the choice of cluster , but clusters contain galaxies with a variety of orbits , morphologies , and inclinations . within the same cluster , a galaxy of mass @xmath240 with a low surface density on a radial orbit traveling face - on to the wind can be as severely stripped as a galaxy with @xmath241 that has a high surface density and is traveling edge - on . at the same time , the ram pressure can vary significantly between clusters with the same mass because of variations in the icm profile . two identical satellite galaxies on identical orbits in different clusters of the same mass can experience ram pressures that vary by a factor of @xmath242 . an effective mass , @xmath243 , can be defined for each individual galaxy . this is the mass of a satellite that the model predicts is stripped to the same @xmath85 as the given galaxy . the effect of the scatter in the parameters can then be discussed in terms of the scatter in @xmath243 at a given @xmath0 . the scatter in the @xmath93 at which a galaxy is stripped of @xmath109 , or in @xmath243 , can best be expressed as a multiplicative factor , @xmath244 . the source of scatter in @xmath243 can be grouped into three types , that due to the icm parameters , @xmath6 , @xmath11 , @xmath126 , the orbital parameters , @xmath91 , @xmath162 , @xmath8 , and the restoring pressure parameters , @xmath17 , @xmath78 , @xmath79 . the overall scatter in the high- and low - mass models , the result of varying all of the model parameters , are @xmath245 and @xmath246 . these can be broken down into @xmath247 , @xmath248 , @xmath249 , and @xmath250 . the discussion in the paragraph above is based on these estimates . the uses of this model are twofold . first , it makes general predictions about the degree of stripping that occurs to galaxies of different masses in a range of environments . second , it links the degree of stripping , @xmath109 or @xmath85 , to the ratio @xmath93 . this may be useful for verifying whether stripping is occurring within a galaxy catalog . the size of the scatter in @xmath243 determines the size of the galaxy catalog that is needed to see a dependence of @xmath109 on @xmath93 . the scatter in the @xmath11 profile parameters is a major contributor to the scatter in @xmath243 . therefore , such a catalog must either contain many clusters , or the icm profile of the clusters must be known . as an estimate of the catalog size necessary , if 200 satellite galaxies with approximately the same @xmath0 orbiting in 50 groups or clusters with the same @xmath1 are combined , then @xmath251 . if the range in @xmath0 and @xmath1 of such a catalog is large compared to this scatter , then the effects of stripping should be seen in the relationship between @xmath109 and @xmath93 . the model developed here relates the degree of ram pressure stripping a satellite galaxy experiences to the galaxy and cluster masses and can be used to quickly determine the extent to which a galaxy is likely to be stripped . all galaxies lose most of their hot galactic halo . in clusters galaxies at moderate mass ratios , @xmath252 , are moderately stripped , @xmath253 , at intermediate distances , @xmath254 , from the cluster center and severely stripped closer to the cluster center . satellites with lower @xmath93 are severely stripped even at intermediate distances . stripping also occurs in groups . however , the same degree of stripping occurs for lower @xmath93 in groups than in clusters . dwarf galaxies are moderately to severely stripped at intermediate distances in groups , and large spiral galaxies can be moderately stripped if they travel to small @xmath66 . the model is simple and motivated by observations . dark matter profiles are well matched to nfw profiles outside a possible core , observations of x - ray gas in clusters can be fit using @xmath11 profiles , and the average gravitational potential of most galaxies should be well matched by the model potential . the model has a large number of parameters , but the values for many can be taken from observations . however , the model assumes that clusters are static . from an evolutionary standpoint , it is only valid after the group or cluster has acquired an icm . on the other hand , in dynamic clusters bulk icm motions may cause more stripping to occur than is predicted by the model . this is observed in the virgo cluster @xcite . ram pressure stripping is not the only way to remove gas from galaxies . in groups and clusters tidal stripping of both gas and stars can occur ( e.g. , @xcite ) . for dwarf galaxies it has been proposed that supernova winds associated with bursts of star formation may expel gas @xcite , and the efficiency of this mechanism may depend on the environment @xcite . however , ram pressure stripping is capable of removing gas from galaxies across a large range of galaxy masses and environments . in particular ram pressure stripping can act when supernova - driven ejection is likely to be inefficient and can remove gas from galaxies that are either not tidally interacting or that are experiencing only weak tidal interactions . the tidal radius , @xmath255 , at the physical orbital radius , @xmath256 , as estimated using @xmath257 , can be compared to the stripping radius , @xmath85 , as determined using the model . here @xmath258 is the cluster mass enclosed within @xmath256 . in scaled coordinates , @xmath259^{-1/3}$ ] , where @xmath260 . the h i fraction lost to tidal stripping depends only on the scaled size of the gas disk and @xmath66 and not on either mass . because the gas disk resides in the center of the satellite s dark halo , tidal stripping rarely effects the h i disk and practically never competes with the effect rps . for example , in order for the h i disk to be tidally stripped to @xmath261 , a satellite must travel to @xmath262 . by this @xmath66 the entire outer h i disk has been ram pressure stripped for almost all satellites . the effect of varying the model s parameters was studied both to identify the parameters that most effect the extent to which an individual galaxy is stripped and to determine the range of satellite and cluster masses that result in the same @xmath85 . the extent to which an individual galaxy is stripped depends most strongly on the galaxy s @xmath66 or @xmath59 , inclination , @xmath162 , and disk scale length , @xmath82 , and on the cluster s icm profile . galaxies that reach smaller @xmath66 , are face - on to the wind , have denser disks , or orbit through a denser icm are more effectively stripped . galaxies in the same cluster with @xmath0 that differ by as much as a factor of 25 can be stripped to the same @xmath85 . in different clusters , galaxies on identical orbits with identical morphologies , identical @xmath78 , @xmath79 , @xmath17 , can be stripped to the same @xmath85 with @xmath0 values that differ by as much as a factor of 30 . ram pressure stripping of satellite galaxies outer gas disks and hot galactic halos is occurring frequently in both groups and clusters . in general , removing gas from a galaxy reduces star formation . the gas in the outer disk and hot galactic halo is not currently involved in star formation , but may feed future star formation if it is not stripped . therefore , it would be interesting to study in more detail how the removal of this gas affects galaxy evolution and to determine if , to what extent , and how quickly star formation declines after a galaxy is stripped . this project was advised by d. n. spergel and funded by nasa grant award # nng04gk55 g . i d like to thank the anonymous referee for useful comments . abadi , m. g. , moore , b. & bower , r. g. 1999 , mnras , 308 , 947 , astro - ph/9903436 binney , j. & merrifield m. 1998 , galactic astronomy ( princeton : princeton univ . press ) binney , j. & tremaine , s. 1988 , galactic dynamics ( princeton : princeton univ . press ) blitz , l. & robishaw , t. 2000 , apj , 541 , 675 , astro - ph/0001142 bravo - alfaro , h. , cayatte , v. , van gorkom , j. & balkowski , c. 2000 , aj 119 , 580 bullock , j. s. , kolatt , t. s. , sigad , y. , somerville , r. s. , kravtsov , a. v. , klypin , a. a. , primack , j. r. & dekel , a. 2001 , mnras , 321 , 559 bureau , m. , walter , f. , van gorkom , j. & carignan , c. 2004 , in iau symp . , 217 , recycling intergalactic and interstellar matter , ed . p .- a . duc , j. braine , & e. brinks ( san fransisco : asp ) , 452 , astro - ph/0311539 cayatte , v. , kotanyi , c. , balkowski , c. & van gorkom , j. h. 1994 , aj , 107 , 1003c cole , s. & lacey , c. , 1996 , mnras , 281,716c , astro - ph/9510147v2 ferrara , a. & tolstoy , e. 2000 , mnras , 313 , 291 , gavazzi , g. , boselli , a. , cortese , l. , arosio , i. , gallazzi , a. , pedotti , p. , & carrasco , l. 2006 , a&a , 446 , 839 gill , s. p. d. , knebe , a. , gibson , b. k. , & dopita , m. a. 2004 , mnras 351 , 410 giovanelli , r. , & haynes , m. p. 1983 , aj , 88 , 881 g gomez , p. l. , et al . 2003 , apj , 584 , 210 goto , t. , yamauchi , c. , fujita , y. , okamura , s. , sekiguchi , m. , smail , i. , bernardi , m. , & gomez , p. l. 2003 , mnras , 346 , 601 g , astro - ph/0312043 grebel , e. k. 2002 , in iap colloq . 17 , gaseous matter in galaxies and intergalactic space , ed . r. ferlet , m. lemoine , j .- desert , & b. raban ( paris : frontier group ) , 171 , astro - ph/0110471 gunn , j. e. & gott , j. r. 1972 , apj 176 , 1 helsdon , s. f. & ponman , t. j. 2000 , mnras , 315 , 356 hernquist , l. , 1990 , apj , 356 , 359 hogg , d. w. et . 2003 , apj , 585 , l5 hogg , d. w. et al . 2003 , apj , 601 , l29 johnston , k. v. , spergel , d. n. , & hernquist , l. 1995 , apj , 451 , 598 kenney , j. d. p. , crowl , h. , van gorkom , j. & vollmer , b. 2004 , in iau symp . , 217 , recycling intergalactic and interstellar matter , ed . duc , j. braine , & e. brinks ( san fransisco : asp ) , 217 , 370 kenney , j. d. p. & koopmann , r. a. 1999 , aj , 117 , 181k kenney , j. d. p. , van gorkom , j. , vollmer , b. 2004 , aj , 127 , 3361k lokas , e. w. & mamon , g. a. 2001 , mnras , 321 , 155 , astro - ph/0002395 marcolini , a. , brighenti , f. , & dercole , a. 2003 , mnras , 345 , 1329 , astro - ph/0309026 matthews , l. d. & gallagher , j. s. 1997 , aj , 114 , 1899 miyamoto , m. & nagai , r. 1975 , pasj , 27 , 533 mo , h. j. , mao , s. , & white , s. d. m. 1998 , mnras , 295 , 319 , astro - ph/9707093 mohr , j. j. , mathiesen , b. , evrard , a. e. 1999 , apj , 517 , 627 mori , m. , burkert , a. 2000 , apj , 538 , 559 m , astro - ph/0001422 murakami , i. , babul , a. 1999 , mnras , 309 , 161 m navarro , j. f. , frenk , c. s. , & white , s. d. m. 1997 , apj , 490 , 493 osmond , j. p. f. & ponman , t. j. 2004 , mnras , 350 , 1511 patterson , r. j. & thuan , t. x. 1992 , apj , 400 , l55 popesso , p. , biviano , a. , bhoeringer , h. , romaniello , m. , & voges , w. 2005 , a&a , 433 , 431p quilis , v. , moore , b. , & bower , r. 2000 , science 288 , 1617 roediger , e. hensler , g. 2005 , a&a , 433 , 875r sanderson , a. j. r. & ponman , t. j. 2003 , in clusters of galaxies : probes of cosmological structure and galaxy evolution , ed . j. s. mulchaey , a. dressler , & a. oemler ( pasadena : carnegie observatories ) , 50 , astro - ph/0303374 sanderson , a. j. r. , ponman , t. j. , finoguenov , a. , lloyd - davies , e. j. , & markevitch , m. 2003 , mnras , 340 , 989 schulz , s. & struck , c. 2001 , mnras , 328 , 185 silich , s. & tenorio - tagle , g. 2001 , apj , 552 , 91 , astro - ph/0101317 solanes , j. m. , manrique , a. , garcia - gomez , c. , gonzalez - casado , g. , giovanalli , r. & haynes , m. p. 2001 , apj , 548 , 97 swaters , r. a. , van albada , t. s. , van der hulst , j. m. , & sancisi , r. 2002 , a&a , 390 , 829 , astro - ph/0204525 vitvitska , m. , klypin , a. a. , kravtsov , a. v. , wechsler , r. h. , primack , j. r. , & bullock , j. s. 2002 , apj , 581 , 799 , astro - ph/0105349 vollmer , b. , marcelin , m. , amram , p. , balkowski , c. , cayatte , v. , & garrido , o. 2000 , a&a , 364 , 532v vollmer , b. , balkowski , c. , cayatte , v. , van driel w. & huchtmeier , w. 2004 , a&a , 419 , 35v when simulations are run that allow cold , non - interacting particles to evolve gravitationally from a flat power spectrum at high redshift , the dark matter halos that emerge at low @xmath75 are well fit by a universal density profile over many orders of magnitude in mass . this profile was originally parameterized by @xcite . both the satellite and group / cluster dark matter halos are described using an nfw profile , the relevant properties of which are summarized here . the nfw profile is @xmath263 where @xmath264 is the average density of the universe at @xmath144 , @xmath265 is an inner scale radius , and @xmath266 is a characteristic over - density . it can be rewritten in terms of a concentration , @xmath267 , an overdensity , @xmath124 , and a scaled radius , @xmath268 @xmath269 @xmath270 @xmath271 @xmath272 the overdensity , virial mass , @xmath131 , and virial radius , @xmath97 , are related by @xmath273 note that the virial radius scales as @xmath274 . the mass contained within @xmath275 is @xmath276 \label{nfwmass}\ ] ] and the gravitational potential is @xmath277 the above can be found in @xcite and @xcite . bound orbits in a central force field travel between two radial extremes . these are found by solving . @xmath278}{l^{2}}\ ] ] where @xmath60 and @xmath63 are the energy and angular momentum per unit mass @xcite . the scaled radial extremes , @xmath59 , for an nfw profile are given by @xmath279 \label{turningpoint}\ ] ] where @xmath280 and @xmath92 are defined as @xmath281 @xmath282 the parameter @xmath283 would be the impact parameter for an unbound orbit . conserving angular momentum and using eq . [ turningpoint ] , the orbital speed at the pericenter , @xmath284 , is @xmath285 it is proportional to @xmath274 and depends on the concentration through @xmath286 . at any point in the orbit , the orbital speed , @xmath5 , is @xmath287\ ] ] the virial temperature , @xmath51 , of a gravitational potential is defined by @xmath288 where @xmath289 is the work done by gravity in forming the halo , @xmath47 is the boltzmann constant , @xmath290 is the mass of the proton , and @xmath291 is the mean molecular weight @xcite . combining eqs . [ work ] and [ hydro eq ] , the gas density profile is @xmath299\right)\ ] ] letting the gas mass within @xmath300 equal a fraction @xmath54 of the virial mass results in a central density of @xmath301 the integral , @xmath302 , is @xmath303\right)\ ] ] @xmath6 & eq . [ beta ] & 6.5 & 5.5 & 20 & @xmath307 & 6 & 15 + @xmath11 & eq . [ beta ] & 0.67 & & & @xmath308 & 2 & 3 + & & & 0.57 & & & & + & & & & 0.45 & @xmath308 & 7.4 & 20 + @xmath126 & eq . [ beta ] & 12 & & & 7 - 20 & 6 & 16 + & & & 25 & & & & + & & & & 170 & 75 - 270 & 5.4 & 13 + @xmath162 & sec . [ sec inclination ] & @xmath309 & & & @xmath309 - @xmath310 & 4.5 & 10 + @xmath156 & eq . [ v sat ] & 1.15 & & & @xmath311 & 1.4 & 1.6 + @xmath8 & eq . [ group nfw ] & 5.5 & 7.5 & 10 & 5 - 16 & 1.1 & 1.2 + @xmath17 & eq . full restoring pressure ] & 12 & & 20 & 12 - 20 & 1.4 & 1.7 + @xmath82 & eq . [ sigma not ] & 5.7 & & & @xmath312 & 8 & 20 + @xmath177 & eq . [ stellar disk 2 ] & 6 & & & @xmath313 & 1.2 & 1.4 + @xmath314 & eq . [ stellar disk 2 ] & 0.04 & & & & & + @xmath187 & eq . [ bulge 2 ] & 180 & & na & & & + @xmath81 & eq . [ sigma not ] & 0.008 & & 0.016 & @xmath3151.5 & 1.5 & 1.8 + @xmath192 & eq . [ stellar disk 2 ] & 0.05 & & 0.04 & @xmath316 & 1.3 & 1.4 + @xmath190 & eq . [ bulge 2 ] & 0.017 & & 0 & 0.012 - 0.024 & 1.1 & 1.2 + @xmath54 & eq . [ g halo prof ] & 0.005 & & & @xmath3151.5 & 1.5 & 1.8 + @xmath317 & eq . [ turningpoint ] & 0.35 & & & @xmath318 & 5.4 & 13 +

the evolution of galaxies in groups may have important implications for the evolution of the star formation history of the universe , since many processes which operate in groups may suppress star formation and the fraction of galaxies in bound groups grows rapidly between and the present day . in this paper , we present an investigation of the properties of galaxies in galaxy groups at intermediate redshift ( ) . the groups were selected from the cnoc2 redshift survey as described in , with further spectroscopic follow - up undertaken at the magellan telescope in order to improve the completeness and depth of the sample . we present the data for the individual groups , and find no clear trend in the fraction of passive galaxies with group velocity dispersion and group concentration . we stack the galaxy groups in order to compare the properties of group galaxies with those of field galaxies at the same redshift . the groups contain a larger fraction of passive galaxies than the field , this trend being particularly clear for galaxies brighter than in the higher velocity dispersion groups . in addition , we see evidence for an excess of bright passive galaxies in the groups relative to the field . in contrast , the luminosity functions of the star forming galaxies in the groups and the field are consistent . these trends are qualitatively consitent with the differences between group and field galaxies seen in the local universe . galaxies : fundamental parameters galaxies : evolution galaxies : stellar content catalogues

astro-ph0501182

clusters of galaxies have received intensive observational study over the last decades . this effort has lead to clear results on the way galaxy properties , such as morphology and star formation rate , vary within clusters and how these properties evolve between clusters at different redshifts ( * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * see bower & balogh 2004 for a recent review ) . in contrast , comparably detailed studies of galaxy groups and their evolution have only recently begun in earnest . the group environment is likely to have a significant impact on star formation rates in the member galaxies @xcite and recent work has emphasied that even for galaxies now in rich clusters , much of the transformation of the galaxies properties may have taken place in groups embedded in the filamentary structure @xcite . in addition , while few galaxies in the local universe are located in clusters , up to 50 per cent @xcite may be located in galaxy groups . furthermore , this number is strongly redshift dependent , as a larger and larger fraction of galaxies become members of groups as the large - scale structure of the universe develops . thus the properties of galaxies in groups and the impact of the group environment on the evolution of galaxies can have an important bearing on the decline of the cosmic star formation rate from @xmath0 to the present - day @xcite . studies of nearby groups show that their galaxy populations exhibit properties which vary from cluster - like ( mostly early - type ) to field - like ( mostly late type ) @xcite . however we know that groups span a wide range in local density , upon which the morphological composition @xcite and the mean star forming properties @xcite strongly depend ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? the wide range of galaxy populations found in nearby groups is likely to be a natural consequence of these correlations . conversely , the powerful dependence of galaxy properties on local densities typical of groups suggests that groups provide the ideal environment for galaxy transformations . this may result in the strong dependence of early type fraction and passive dwarf galaxy abundance on group velocity dispersion and x - ray luminosity , observed in nearby groups @xcite . in particular , galaxy interactions are expected to be common in groups , in which the velocity dispersion is typically not much larger than that of the constituent galaxies @xcite . groups have a much lower density contrast against the background galaxy population , so most work has concentrated on groups which are either unusually compact @xcite or x - ray luminous @xcite . more recently , the advent of large field galaxy redshift surveys has made it possible to study galaxy groups selected purely on the basis of their three dimensional galaxy density . in the local universe , early redshift surveys and more recently the extensive 2df galaxy redshift survey ( 2dfgrs ) and sloan digital sky survey ( sdss ) have generated large group catalogues which can be used as the basis for studies of galaxies in the group environment @xcite however , the properties of higher redshift groups have been relatively little explored . this is largely because of the difficulty in finding suitable systems , and the low success rate of spectroscopic follow - up of group members . @xcite used radio galaxies to preselect groups in order to study the evolution of the blue galaxy fraction with redshift , while @xcite have presented a group catalogue based on the cnoc2 galaxy redshift survey . in this paper , we present new data obtained in the region of 26 cnoc2 groups at @xmath3 with ldss2 on magellan . in sections [ sec : cnoc2groupdata ] and [ sec : whyoiiforsf ] we introduce the data and explain why we have chosen to use the [ oii ] emission line to study statistical trends in star formation . in section [ sec : cnoc2dataprocessing ] we begin by describing the data reduction procedure and our method for determination of redshifts and emission line equivalent widths for each galaxy . section [ sec : grpmem ] then goes on to explain our procedure for group membership allocation . in section [ sec : indivgrps ] we present the properties of the 26 individual groups in order to examine any trends of structural group properties with star formation . finally , in section [ sec : stackedgrp ] groups are stacked to enable detailed statistical analaysis , and we investigate the link between the group environment and star formation by contrasting the stacked group with our sample of field galaxies selected from the same redshift range . in a second paper ( * hereafter known as paper ii ) , we present a comparison of the star forming properties of these intermediate redshift groups with local groups selected from the 2pigg catalogue at @xmath4 @xcite . throughout this paper we assume a @xmath5cdm cosmology of @xmath6 , @xmath7 and @xmath8 . the second canadian network for observational cosmology redshift survey ( cnoc2 ) was recently completed with the aim of studying galaxy clustering in the redshift range @xmath9 @xcite . the cnoc2 survey is split into 4 patches , approximately equally spaced in ra and totalling 1.5 square degrees in area . the survey consists of 5 colour @xmath10 photometry with @xmath11 galaxies down to the photometric limit @xmath12 . the mos spectrograph on the cfht 3.6 m telescope was used to obtain spectra for over @xmath13 galaxies in total , 48 per cent complete down to @xmath14 . combination of the imaging and spectroscopy lead to a very well determined selection function for the spectroscopic sample @xcite and comparisons with our deeper spectroscopy suggest that brighter than this limit the cnoc2 survey is not biased towards emission line objects . the transmission efficiency of the band limiting filter and grism combination was above half power in the wavelength range @xmath15 , effectively limiting the redshift range of the survey . the most prominent absorption features ( ca ii h and k ) lie in this wavelength range for galaxies in the redshift range @xmath16 , whereas either the [ oii ] or [ oiii ] emission lines fall in this wavelength range for all galaxies in the range @xmath17 . distant groups have always been difficult to recognize because of their sparse galaxy populations . the presence of high redshift groups has typically been inferred indirectly via the presence of a radio galaxy or x - ray emitting intragroup medium ( igm , * ? ? ? * ; * ? ? ? whilst these surveys provide a useful insight into the evolution of galaxy properties in rich galaxy groups , the selection criteria strongly bias the selection towards these richer elliptical dominated groups , whereas low redshift samples point to a more numerous population of low density , x - ray faint , spiral dominated groups @xcite . the cnoc2 survey provided an powerful opportunity to generate a kinematically selected sample of galaxy groups . a friends - of - friends percolation algorithm was used to detect groups of galaxies in redshift space . this was followed by a trimming step in which a centre , velocity dispersion , @xmath18 and @xmath19 ( @xmath20 virial radius ) were estimated and then members were added or deleted within @xmath21 and @xmath22 in 3 rounds of iteration ( see * ? ? ? a total sample of over 200 galaxy groups was detected . although the spectroscopic sample is incomplete , this group sample represents a kinematically selected sample at intermediate redshifts ( @xmath23 ) , free from the strong biases present in samples selected by other means . we used the multi - object spectrograph ldss2 ( low dispersion survey spectrograph ) on the 6.5 m baade telescope at las companas observatory ( lco ) in chile to obtain a deeper and more complete sample of galaxies in the region of 20 of the cnoc2 groups at @xmath3 located in the 3 out of 4 cnoc2 patches accessible from the latitude of lco . masks were designed using an automated selection algorithm to minimise selection bias and maximise the allocation of targets per mask . due to the proximity on the sky of some of the cnoc2 groups , the masks were chosen to serendipitously sample a further 11 groups from the @xcite sample , although only 6 of these lie at @xmath24 . targets were prioritised using an automated algorithm which favours brighter objects and objects which lie close together along the mask s spectral axis . this second criterion ensures that the spectral coverage does not vary wildly from object to object in each mask . the data were taken in 4 separate observing runs between may 2001 and november 2002 . each ldss2 mask covers approximately 6.5@xmath25 5 and in the cnoc2 fields it is normal to fit between 20 and 30 slits onto a mask . between 1 and 3 masks were observed per target group , depending upon the density of target galaxies in that area of sky . mask exposures varied from 1 to 4 hours depending on the phase of the moon and observing conditions . for maximum throughput , the blue - optimised medium resolution grism was used with a dispersion of 5.3 ( on the fourth run the red optimised grism was used which has the same dispersion and similar efficiency at wavelengths of interest ) . the slits have a width @xmath26which corresponds to @xmath27 of spectral coverage ( compared to @xmath28 in the original cnoc2 masks ) . using the cnoc2 photometric classifier from the cnoc2 photometric catalogue , we targeted 634 objects classified as _ galaxies _ , 102 objects classified as _ possible galaxies_. where there were free spaces on the mask with no galaxies present , we allocated 130 objects classified as _ probable stars_. from this last category , 29 out of 130 objects had galaxy - like spectra which yielded redshifts . the remaining 101 were correctly identified as stars . included in our targets were 35 galaxies which already had redshifts from the initial cnoc2 survey . these were reobserved to form a comparison sample which is used to understand the accuracy of our measurements . ccc targetted & number of & serendipitous group & targetted & @xcite & galaxies & groups22 & 24 & 23,3324 & 45 & 825 & 48 & none28 & 86 & none31 & 38 & none32 & 20 & 1334 & 55 & none37 & 24 & 1,27,3538 & 48 & 4039 & 88 & 23,33134 & 47 & 132,138137 & 52 & 140138 & 54 & 129,132,133,134139 & 28 & none140 & 62 & 137227 & 53 & none228 & 19 & none232 & 24 & 241241 & 14 & 201,232244 & 37 & 243 in table [ tab : obs ] we show the number of galaxies targetted in each field and the serendipitous @xcite groups which lie in these fields . to study the relative levels of star formation in statistical galaxy samples , we use the [ oii]@xmath29 emission line which lies centrally in the visible window in our cnoc2 redshift range of @xmath30 and at wavelengths of low sky emission . at @xmath31 , the emission line disappears entirely from the ldss2 spectrograph window with the instrument sensitivity dropping in the red with the current optics and detector . furthermore , at @xmath32 measurements of are compromised by the increase in sky line density with increasing wavelength . @xcite have recently shown using data from local galaxies in sdss that measurements of sfr using [ oii ] emission are consistent with those from and 1.4ghz luminosities . also the scatter in the -[oii ] relationship is primarily luminosity dependent @xcite and so in a given luminosity range the systematic error in using the [ oii ] measurement to infer star formation rates is significantly reduced . additionally , even at @xmath33 , @xcite place the contribution from dusty starburst galaxies to the total sfr , underestimated by optical emission line measurements at @xmath34 per cent . rather than inferring star formation rates , we deliberately limit our study to direct comparison between [ oii ] measurements to avoid model dependency in our results . finally , we restrict our analysis to ew[oii ] rather than [ oii ] flux and star formation rates . normalisation by the continuum reduces uncertainties related to absorption by dust and aperture bias as well as providing a measure of sfr per unit luminosity . we extracted and calibrated our spectra using mainly iraf tools in the _ onedspec _ and _ twodspec _ packages . wavelength calibration was applied using both iraf tools and new mos reduction software written by dan kelson in carnegie @xcite . we have tested both these methods to ensure that consistent solutions are obtained . the wavelength calibration is based upon arcs taken during daytime and is secure in the wavelength range @xmath35@xmath36 which corresponds to @xmath37 for the [ oii ] emission line and @xmath38 for [ oiii ] which extends far beyond the redshift range of interest . around bright skylines , systematic residuals often remain . we interpolated over these regions . to account for mismatch between the daytime arc and nighttime science observations , we then applied a zero point offset to the wavelength calibration . this is computed by measuring the offset of the 5577skyline in the arc - calibrated frames . most of these offsets were @xmath39 and we did not detect any significant non - linearities . the new wavelength solution appears robust , as evidenced in the comparison of ldss2 and cnoc2 redshifts ( see figure [ fig : zoiicomp ] which is explained in section [ sec : cnoc2cfldss2 ] ) . the iraf tool _ xcsao _ in the _ rvsao _ package @xcite was used to cross - correlate all the spectra with templates of rest - frame early - type ( absorption line ) and late - type ( emission line ) galaxies . the early - type galaxy template was created by coadding a large number of high signal to noise early type galaxy spectra from the sample , each of which had been shifted to zero redshift . the emission line ( sc ) template is a high signal to noise spectrum of ngc4775 from the atlas of @xcite and smoothed to the resolution of ldss2 . @xcite . ] we shall henceforth refer to redshifts obtained using the late - type template as _ emission _ redshifts and those obtained using the early - type template as _ absorption _ redshifts . the peak in the cross - correlation spectrum is selected by _ xcsao _ to give an estimate of the redshift for each object @xcite . to assess redshift measurements , redshifts attained using the early - type ( absorption ) and late - type ( emission ) templates were assessed in each case by manually inspecting the spectra and assigning a redshift quality flag . the quality assessment is based upon the number of believable emission lines in an emission line spectrum . in absorption line spectra , we simply decide whether to accept or reject a solution . for example , a solution with a visible 4000 break would be accepted . spectra containing a single emission line received further checks to ensure the emission is real . finally , redshifts which became apparent upon visual inspection were applied by setting up that solution as the initial guess in _ rvsao_. however , if no peak in the cross - correlation spectrum could be obtained at the correct redshift then the solution was abandoned . the final redshift of a target was selected by comparing the redshift quality flag from emission and absorption redshifts . where the absorption and emission redshifts agree within errors , emission redshifts are preferred due to their greater accuracy . however , where the emission redshift was not secure and a different , higher quality absorption redshift exists , then the absorption redshift is used . objects with no acceptable redshift are discounted from the spectroscopic sample . in total we have 418 redshifts from the ldss2 data of which 240 are in the range @xmath40 which dramatically improves the completeness at this depth . in the whole sample , we have 86 new group members from ldss2 spectroscopy . this increases the total group sample to 295 in 26 groups ( see group membership section for further details of group allocation ) . in the magnitude range @xmath40 and within 150 of targetted group centres , we obtained 179 additional redshifts ( 36 members ) on top of the 115 ( 27 members ) existing from the cnoc2 survey . we consider @xmath41 to be our magnitude limit at which we can obtain redshifts for @xmath42 per cent of objects targetted ; our success rate for all targetted galaxies brighter than @xmath41 is 74 per cent . the selection functions are well understood and targetted galaxies for which we do not manage to obtain a redshift are evenly distributed in colour , suggesting that our sample is unbiased ( see appendix [ sec : completeness ] ) . we measure the [ oii ] equivalent widths using our own purpose written code to compare the flux in two separate continuum regions with that in a feature bandpass . we use the definition from @xcite for [ oii ] , in which the feature bandpass ranges from 3718 to 3738 and the continuua regions have width 65 adjacent to the line region on either side . in figure [ fig : zoiicomp ] we compare the redshift and ew [ ] measurements obtained from the ldss2 spectra and the cnoc2 spectra , for objects in common to both surveys . it can be seen that the redshifts from the two surveys are generally in good agreement . out of 35 galaxies with redshifts from both sources , only four of the redshifts are discrepant . three of these are faint objects ( @xmath43 , @xmath44 ) . we also examine the combined spectral signal ( total counts , _ cts _ ) from the wavelength ranges @xmath45@xmath46 and @xmath47@xmath48 which span the most efficient region of the grism , eliminating the strong night sky lines ( as used to examine targets for which no redshifts were measured in appendix [ sec : zincompl ] ) . these three objects all fall in the low signal range ( @xmath49 ) along with 8 of the remaining 32 galaxies in the combined sample . we exclude the three discrepant redshifts for faint , low signal objects from the figure . the other discrepant redshift is offset by @xmath50 km s@xmath51 ( rest - frame ) . inspection shows that the ldss2 redshift provides a better fit to the emission lines . of the remaining 31 galaxies in this sample , the standard deviation of the rest - frame velocity offsets ( @xmath52 ) , @xmath53 km s@xmath51 with a mean value of only @xmath54 km s@xmath51 . to check the dependence of rest - frame velocity offsets on the spectral counts , we also show the distribution for the 14 galaxies with @xmath55 ( dashed histogram ) . we find no correlation , amongst these remaining objects , between spectral signal strength and velocity offset . using duplicate observations @xcite show that the random rest - frame velocity errors in the cnoc2 redshifts is @xmath56 km s@xmath51 . as the typical velocity difference in our common - object sample is @xmath57 km s@xmath51 , we compute an approximate rest - frame error on ldss2 spectra of @xmath58 km s@xmath51 . we expect that the velocity errors are dominated by the cnoc2 astrometrical errors of @xmath59@xcite . indeed the errors computed using the common - object sample are consistent with an rms astrometrical error of @xmath60 . in the right hand panel of figure [ fig : zoiicomp ] , we show that measurements of ew [ ] in ldss2 spectra ( measured using our code ) are also in excellent agreement with the original cnoc2 survey measurements @xcite within errors , showing that our measurements are consistent with cnoc2 . the code used for both sets of measurements is essentially the same and tests show that they provide identical results . to obtain a consistently selected sample of galaxy groups , we restricted ourselves to examining those groups which were pre - selected from the sample of @xcite and targetted with ldss2 on magellan . this provides a sample of 26 targetted and serendipitous @xcite groups in the range @xmath3 . we derive from the virial theorem @xmath61}^{-1}$ ] where @xmath62 is the velocity dispersion and @xmath63 is the hubble constant at the group redshift z and @xmath64 km s@xmath51 mpc@xmath51 . this means that a group with @xmath65 km s@xmath51 has @xmath66mpc at @xmath67 , with @xmath19 scaling with @xmath62 . the group finding algorithm of @xcite was tuned to identify dense , virialized groups of 3 members or more with the goal of tracing the properties of the underlying dark matter halos . with this in mind , a conservative linking - length ( @xmath68mpc ) and trimming radius of 1.5@xmath19 in the spatial axes of the groups were selected . our objectives differ from those of @xcite in that we wish to understand the global properties of galaxies in groups , not only those galaxies in the virialised core regions . in order to be more representative of the loose group population , whilst retaining the strict selection criteria of @xcite , we choose to relax the projected trimming radius for group members in our redefined group sample . in effect , we find that relaxing this parameter allows the more compact groups to retain the same membership whilst other groups gain extra members . although defining the contents of a group is a subjective problem , there exist some tools which make the task easier . perculiar motion of galaxies moving in a gravitational potential artificially lengthens the group along the line - of - sight direction in redshift space ( the finger - of - god effect ) . from the virial theorem , one can make the assumption that the projected spatial and line - of - sight dimensions of a group in redshift space are approximately in constant proportion ( in rest - frame coordinates ) . @xcite show that an axial ratio of @xmath69 for the length along the line of sight relative to the projected spatial length is most appropriate for a linking volume in a friends - of - friends algorithm . interlopers in redshift space are difficult to identify and eliminate , so we choose a conservative line - of - sight trimming radius of twice the velocity dispersion . we choose a small aspect ratio of @xmath70 , to compute our trimming radius . although this will exceed @xmath19 , it results in a stable membership solution for all 26 groups and allows us to examine how radial trends within the groups extend to the group outskirts and infall regions ( see section [ sec : fpstacked ] ) . however most of our analysis will be conducted within a fixed metric group aperture . we note that these results are insensitive to the choice of a particular value of b , as large as 11 . the algorithm for defining the final membership of each group in our sample works as follows : * the group is initially assumed to be located at the latest position in redshift space determined by @xcite with an initial observed - frame velocity dispersion , @xmath71 , of 500 km s@xmath51 . * the redshift range required for group membership is computed from equation [ eqn : zrange ] , which limits membership in the line - of - sight direction to within twice the velocity dispersion:@xmath72 * this is converted into a spatial distance , @xmath73 , which corresponds to a redshift space distance related to the distance computed in the line - of - sight direction by the aspect ratio , @xmath74 . @xmath73 is computed as : + @xmath75.\left(\frac{d_{\theta}}{h_{75}^{-1}\mbox{mpc}}\right)^{-1 } \label{eqn : phystoang}\end{aligned}\ ] ] + where @xmath76 , the angular diameter distance in physical co - ordinates , is a function of @xmath77 , and + @xmath78 + with @xmath79 . * group members are selected by applying the redshift and positional limits @xmath80 and the angular distance from the group centre , @xmath81 . * we recompute the observed velocity dispersion of the group , @xmath71 , using the gapper algorithm ( equation [ eqn : gappervd ] ) which is insensitive to outliers and thus gives an accurate estimate of the velocity dispersion for small groups @xcite : clipping of a gaussian velocity distribution . ] + @xmath82 + with @xmath83 and @xmath84 . * we also redefine the centre of the group , taking the luminosity weighted centroid in projected spatial coordinates and the mean redshift to be the new group centre . * we then recompute the limiting redshift offset @xmath85 and positional offset @xmath73 using equations [ eqn : zrange ] , [ eqn : phystoang ] and [ eqn : projangrange ] . the galaxies are then reassigned to the group as before . * the whole process is repeated until a stable membership solution is reached . in all 26 groups such a stable solution is found , mostly within 2 iterations although the massive group 138 requires 4 iterations . we finally compute the rest - frame velocity dispersion ( equation [ eqn : convrestvd ] ) and the intrinsic velocity dispersion . the latter is computed by combining the measurement errors of the component galaxies ( equation [ eqn : combmeasvderr ] ) and removing in quadrature from the measured velocity dispersion ( equation [ eqn : remmeasvderr ] ) . + @xmath86 + @xmath87 + where @xmath88 km s@xmath51 ( ldss2 ) and @xmath89 km s@xmath51 ( cnoc2 ) + @xmath90 the combined velocity errors ( @xmath91 ) in group 24 exceed the measured velocity dispersion @xmath92 and so we compute an upper limit to the intrinsic velocity dispersion using monte - carlo simulations . with @xmath93 km s@xmath51 in @xmath94 per cent of iterations a value of @xmath92 less than the measured value is obtained from the simulations , corresponding to the @xmath95 value of a one tailed gaussian . errors on @xmath96 are computed in all other groups using the jackknife technique @xcite . we note that in groups with few known members ( e.g. group 40 ) the true error on @xmath96 may be underestimated using this technique . in figure [ fig : veldistrib ] , we show velocity histograms for two of our groups and the redshift space clustering in the region of those groups . the left - hand panels show the rich group 138 , possibly even a poor cluster , whilst the right - hand panels show the medium - sized group 241 . the overplotted gaussian of width @xmath71 provides a good envelope to the distribution in both cases , which is typical of all 26 groups . we further note that there is no significant trend of velocity dispersion with redshift in our sample indicating that we do not preferentially select higher mass groups at higher redshift ( where the survey does not probe so far down the luminosity function ) . whilst group members lie some distance ( up to @xmath97 ) from the group centre and are included in the calculation of the velocity dispersion , we limit our analysis to galaxies inside a @xmath98mpc projected radius at the group redshift ( @xmath99 for a group with @xmath100 km s@xmath51 ) , where we understand the completeness ( see appendix [ sec : completeness ] ) . rest - frame @xmath101 luminosities are determined to minimise the k - corrections ( the -band well matches the @xmath102 selection band at the redshift of cnoc2 galaxies ) and to allow easy comparison with the properties of the 2dfgrs local galaxy population ( see paper ii for more detail on determination of k - corrections ) . luminosities include a correction for galactic extinction on a patch - to - patch basis , computed by extrapolating from b and v band extinction values obtained from ned ( * ? ? ? * variation within each patch is negligible ) . a luminosity @xmath103 approximates the magnitude limit , @xmath41 for most galaxies at @xmath104 . we compute @xmath105 , an estimate of the number of group members brighter than @xmath103 and within @xmath98mpc projected distance from the group centre . @xmath105 is the sum of the number of known members and candidate members ( the number of galaxies without redshifts which are predicted to lie at the group redshift , and which would meet our luminosity and radial cuts ) . figure [ fig : nmvd ] shows @xmath105 plotted against the intrinsic velocity dispersion @xmath96 of each group . any weak correlation between @xmath105 and @xmath96 is largely masked by a good deal of scatter ( excluding the two largest systems ) . this scatter might be attributed to the variation in group structure and the difficulty of obtaining accurate estimates of velocity dispersion with few members . limiting the membership to galaxies within @xmath106mpc projected radius does not reduce the scatter in this relationship . [ cols="^,^,^,^,^,^,^,^,^,^",options="header " , ] we have used the ldss2 spectrograph on the magellan telescope at lco to obtain 418 new galaxy redshifts in the regions of 26 cnoc2 groups at intermediate redshift ( @xmath3 ) . of these , 86 are group members . with the depth of magellan , we are highly complete and unbiased down to @xmath107 and out to 240(@xmath108 for a 500 km s@xmath51 group at @xmath109 ) . the primary purpose of this paper is to present this data set . our main result of this paper is the significant enhancement of @xmath110 in our group sample relative to the field level , even in the smallest groups . this indicates that star formation is less prevalent among galaxies in the group environment than in the field , even at intermediate redshift . this could be related to either a different formation history in the group environment , intragroup environmental processes accelerating galaxy evolution to the passive state , or a combination of these two effects . we have investigated how this deficit of star forming galaxies depends on the group properties , and find only relatively weak correlations . whilst there is little overall trend of @xmath110 with group velocity dispersion , there is a suggestion that @xmath110 is higher in the more concentrated systems with @xmath111 km s@xmath51 . we examined the radial dependence of @xmath110 , finding a weak trend with radius . comparing @xmath110 with the nearest neighbour distance , the passive fraction declines significantly at large separations , suggesting a relation to the substructure within groups . taking into account the low numbers of galaxies in each system , these trends are qualitatively similar to those reported in local group samples ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? the correlations of passive fraction with environment also mimic the dependence of morphology on environment in local groups ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) . in rich clusters at the epoch of cnoc2 , @xcite find an evolving fraction of elliptical + s0 galaxy types with a fraction @xmath112 in clusters , within a clustercentric radius of @xmath113mpc and brighter than @xmath114 . this is comparible with the mean fraction of passive galaxies in the cnoc2 groups ( down to brighter than @xmath115 , see figure [ fig : fp_vd ] ) but with considerably less scatter in these richer systems . we note that the morphology - density relation is also effective in at least some of these clusters ( see also * ? ? ? * ) , with a much higher fraction of early - type galaxies ( approaching @xmath116 ) in the densest regions . however , the different selection ( morphological , v - band limited ) and our arbitrary division at ew[oii]=5 makes quantitative comparison difficult . the value of @xmath110 in groups is also significantly lower than seen in cnoc1 cluster cores at @xmath117 , with mean values of @xmath118 within the inner @xmath119 mpc and brighter than @xmath120 ( * ? ? ? * note : whilst in that paper the division is at 10 , we use 5 for consistency ) . in mulchaey 2005 we will discuss the morphological properties of our sample in detail , using hst acs imaging ( now mostly complete ) . we have coadded the data to provide a well understood stacked group sample and a stacked field sample over the same redshift range ( @xmath121 ) , and use this sample to investigate the dependence of @xmath110 on galaxy luminosity . we find that @xmath110 is greatly enhanced in group galaxies with respect to the field down to at least @xmath122 . in the brightest galaxies ( @xmath123 ) this trend is strongest in the higher velocity dispersion systems ( @xmath124 km s@xmath51 ) . we also investigate how the luminosity function of galaxies depends on their environment : local studies have shown that luminosity is at least as important as environment in determining whether a galaxy is passive or star forming at the present day ( e.g. * ? ? ? our data suggest that the intermediate redshift group galaxy population is enhanced in bright galaxies and that lower velocity dispersion groups ( @xmath125 km s@xmath51 ) have a deficit of faint galaxies compared to the intermediate redshift field . however , only in the more massive groups is the enhanced bright population primarily passive ( @xmath110 also enhanced ) . it is instructive to see how these trends compare with similar studies at low redshift in the literature . the variation in the shape of the global luminosity function measured from survey to survey is worryingly large @xcite and so it is important to use data from the same survey where possible in studies of how the luminosity function depends upon environment . data from sdss and 2dfgrs suggests that in regions of higher density the local galaxy population is biased towards a brighter characteristic magnitude and steeper faint - end slope @xcite . this density dependence appears to be dominated by the relative abundance of early - type passive galaxies and may not be obvious in each individual cluster ( e.g. the coma cluster , * ? ? ? . in local groups there may also be a correlation between group velocity dispersion and the faint - end slope , driven by the population of faint , early type galaxies @xcite . the large population of faint passive galaxies in our most massive groups ( @xmath126 km s@xmath51 , see figure [ fig : lumfns_bysfandvd ] ) suggests that we might be seeing the same trends at @xmath127 , although there appears to be a deficit of faint galaxies in less massive groups , relative to the field . the generation of this faint passive galaxy population is relavent to downsizing in the red sequence of cluster galaxies , with recent observations indicating that it does not extend faintwards of @xmath128 at @xmath129 @xcite . in paper ii we will compare our cnoc2 groups to a low redshift group sample selected from the 2dfgrs survey . in paper iii we will examine the morphological composition of our groups , extending studies of the morphology - density relation to group environments at @xmath127 . we would like to thank the staff of the magellan telescope for their tremendous support . rgb is supported by a pparc senior research fellowship . djw , mlb and rjw also thank pparc for their support . we are grateful to dan kelson for the use of his spectral reduction software and to david gilbank for his help when learning to use it . we also acknowledge _ noao _ for the wealth of iraf tools used during data reduction . we also acknowledge tom shanks and phil outram for their observations at magellan and the full cnoc2 team for the outstanding dataset . finally , we thank the anonymous referee for some useful feedback which has helped to improve this paper . , j. r. , ellis , r. , zirbel , e. l. , & oemler , a. j. 1993 , apj , 404 , 521 , i. k. , glazebrook , k. , brinkmann , j. , ivezi ' c , . , lupton , r. h. , nichol , r. c. , & szalay , a. s. 2004 , apj , 600 , 681 , m. , eke , v. , miller , c. , lewis , i. , bower , r. , couch , w. , nichol , r. , bland - hawthorn , j. , baldry , i. k. , baugh , c. , bridges , t. , cannon , r. , cole , s. , colless , m. , collins , c. , cross , n. , dalton , g. , de propris , r. , driver , s. p. , efstathiou , g. , ellis , r. s. , frenk , c. s. , glazebrook , k. , gomez , p. , gray , a. , hawkins , e. , jackson , c. , lahav , o. , lumsden , s. , maddox , s. , madgwick , d. , norberg , p. , peacock , j. a. , percival , w. , peterson , b. a. , sutherland , w. , & taylor , k. 2004 , mnras , 348 , 1355 , t. c. , flynn , k. , & gebhardt , k. 1990 , aj , 100 , 32 , m. r. , eisenstein , d. j. , hogg , d. w. , schlegel , d. j. , & brinkmann , j. 2004 , apj , submitted , astro - ph/0310453 , m. r. , hogg , d. w. , bahcall , n. a. , baldry , i. k. , brinkmann , j. , csabai , i. , eisenstein , d. , fukugita , m. , gunn , j. e. , ivezi ' c , . , lamb , d. q. , lupton , r. h. , loveday , j. , munn , j. a. , nichol , r. c. , okamura , s. , schlegel , d. j. , shimasaku , k. , strauss , m. a. , vogeley , m. s. , & weinberg , d. h. 2003 , apj , 594 , 186 , r. g. & balogh , m. l. 2004 , in clusters of galaxies : probes of cosmological structure and galaxy evolution , 326+ , j. , charlot , s. , white , s. d. m. , tremonti , c. , kauffmann , g. , heckman , t. , & brinkmann , j. 2004 , mnras , 351 , 1151 butcher , h. & oemler , a. 1984 , apj , 285 , 426 , r. g. , yee , h. k. c. , morris , s. l. , lin , h. , hall , p. b. , patton , d. r. , sawicki , m. , & shepherd , c. w. 2001 , apj , 552 , 427 , d. 2000 , apj , 545 , 145 , d. j. , farrar , g. r. , norberg , p. , colless , m. , peacock , j. a. , baldry , i. k. , baugh , c. m. , bland - hawthorn , j. , bridges , t. , cannon , r. , cole , s. , collins , c. , couch , w. , dalton , g. , de propris , r. , driver , s. p. , efstathiou , g. , ellis , r. s. , frenk , c. s. , glazebrook , k. , jackson , c. , lahav , o. , lewis , i. , lumsden , s. , maddox , s. , madgwick , d. , peterson , b. a. , sutherland , w. , & taylor , k. 2004 , astro - ph/ , in press : , astro , g. , poggianti , b. m. , arag ' on - salamanca , a. , clowe , d. , halliday , c. , jablonka , p. , milvang - jensen , b. , pell ' o , r. , poirier , s. , rudnick , g. , saglia , r. , simard , l. , & white , s. d. m. 2004 , apjl , 610 , l77 , r. , colless , m. , driver , s. p. , couch , w. , peacock , j. a. , baldry , i. k. , baugh , c. m. , bland - hawthorn , j. , bridges , t. , cannon , r. , cole , s. , collins , c. , cross , n. , dalton , g. b. , efstathiou , g. , ellis , r. s. , frenk , c. s. , glazebrook , k. , hawkins , e. , jackson , c. , lahav , o. , lewis , i. , lumsden , s. , maddox , s. , madgwick , d. s. , norberg , p. , percival , w. , peterson , b. , sutherland , w. , & taylor , k. 2003 , mnras , 342 , 725 dressler , a. 1980 , apj , 236 , 351 , a. , oemler , a. j. , couch , w. j. , smail , i. , ellis , r. s. , barger , a. , butcher , h. , poggianti , b. m. , & sharples , r. m. 1997 , apj , 490 , 577 , a. & shectman , s. a. 1987 , aj , 94 , 899 , s. & de propris , r. 2003 , astrophysics and space science , 285 , 175 , b. 1982 , the jackknife , the bootstrap and other resampling plans ( cbms - nsf regional conference series in applied mathematics , philadelphia : society for industrial and applied mathematics ( siam ) , 1982 ) , v. r. , baugh , c. m. , cole , s. , frenk , c. s. , norberg , p. , peacock , j. a. , baldry , i. k. , bland - hawthorn , j. , bridges , t. , cannon , r. , colless , m. , collins , c. , couch , w. , dalton , g. , de propris , r. , driver , s. p. , efstathiou , g. , ellis , r. s. , glazebrook , k. , jackson , c. , lahav , o. , lewis , i. , lumsden , s. , maddox , s. , madgwick , d. , peterson , b. a. , sutherland , w. , & taylor , k. 2004 , mnras , 348 , 866 , g. , poggianti , b. m. , couch , w. j. , bettoni , d. , kjrgaard , p. , & moles , m. 2000 , apj , 542 , 673 , h. , hammer , f. , elbaz , d. , cesarsky , c. j. , liang , y. c. , fadda , d. , & gruel , n. 2004 , a&a , 415 , 885 , p. l. , nichol , r. c. , miller , c. j. , balogh , m. l. , goto , t. , zabludoff , a. i. , romer , a. k. , bernardi , m. , sheth , r. , hopkins , a. m. , castander , f. j. , connolly , a. j. , schneider , d. p. , brinkmann , j. , lamb , d. q. , subbarao , m. , & york , d. g. 2003 , apj , 584 , 210 , m. , rigoni , e. , mardirossian , f. , & mezzetti , m. 2003 , a&a , 406 , 403 , y. , oemler , a. j. , lin , h. , & tucker , d. l. 1998 , apj , 499 , 589 , p. , kindl , e. , & auman , j. r. 1989 , apjs , 70 , 687 , a. m. 2004 , arxiv astrophysics e - prints , a. m. , miller , c. j. , nichol , r. c. , connolly , a. j. , bernardi , m. , g ' omez , p. l. , goto , t. , tremonti , c. a. , brinkmann , j. , ivezi ' c , . , & lamb , d. q. 2003 , apj , 599 , 971 , j. p. & geller , m. j. 1982 , apj , 257 , 423 , r. a. , franx , m. , & fabricant , d. 2001 , apj , 551 , 825 , l. r. , mchardy , i. , newsam , a. , & mason , k. 2002 , mnras , 334 , 219 , g. , white , s. d. m. , heckman , t. m. , m ' enard , b. , brinchmann , j. , charlot , s. , tremonti , c. , & brinkmann , j. 2004 , mnras , 314 , d. d. 2003 , pasp , 115 , 688 , r. c. 1992 , apjs , 79 , 255 , t. , smail , i. , nakata , f. , okamura , s. , & bower , r. g. 2001 , apjl , 562 , l9 , t. , yamada , t. , akiyama , m. , aoki , k. , doi , m. , furusawa , h. , fuse , t. , imanishi , m. , ishida , c. , iye , m. , kajisawa , m. , karoji , h. , kobayashi , n. , komiyama , y. , kosugi , g. , maeda , y. , miyazaki , s. , mizumoto , y. , morokuma , t. , nakata , f. , noumaru , j. , ogasawara , r. , ouchi , m. , sasaki , t. , sekiguchi , k. , shimasaku , k. , simpson , c. , takata , t. , tanaka , i. , ueda , y. , yasuda , n. , & yoshida , m. 2004 , mnras , 350 , 1005 , m. j. & mink , d. j. 1998 , pasp , 110 , 934 , s. j. , le fevre , o. , hammer , f. , & crampton , d. 1996 , apjl , 460 , l1 + , h. , yee , h. k. c. , carlberg , r. g. , morris , s. l. , sawicki , m. , patton , d. r. , wirth , g. , & shepherd , c. w. 1999 , apj , 518 , 533 madau , p. , pozzetti , l. , & dickinson , m. 1998 , apj , 498 , 106 , h. j. , zandivarez , a. , domnguez , m. , merch ' an , m. e. , & lambas , d. g. 2002 , mnras , 333 , l31 , b. , colless , m. , carter , d. , poggianti , b. m. , bridges , t. j. , kranz , k. , komiyama , y. , kashikawa , n. , yagi , m. , & okamura , s. 2003 , apj , 587 , 605 , j. s. , davis , d. s. , mushotzky , r. f. , & burstein , d. 2003 , apjs , 145 , 39 , f. , g. , b. r. , balogh , m. l. , & j. , w. d. 2004 , mnras , accepted , p. , cole , s. , baugh , c. m. , frenk , c. s. , baldry , i. , bland - hawthorn , j. , bridges , t. , cannon , r. , colless , m. , collins , c. , couch , w. , cross , n. j. g. , dalton , g. , de propris , r. , driver , s. p. , efstathiou , g. , ellis , r. s. , glazebrook , k. , jackson , c. , lahav , o. , lewis , i. , lumsden , s. , maddox , s. , madgwick , d. , peacock , j. a. , peterson , b. a. , sutherland , w. , & taylor , k. 2002 , mnras , 336 , 907 , b. m. , smail , i. , dressler , a. , couch , w. j. , barger , a. j. , butcher , h. , ellis , r. s. , & oemler , a. j. 1999 , apj , 518 , 576 , m. & geller , m. j. 1984 , apj , 281 , 95 , m. , geller , m. j. , & huchra , j. p. 1989 , apj , 344 , 57 , m. , pisani , a. , & geller , m. j. 1997 , aj , 113 , 483 , m. , zamorani , g. , zucca , e. , stirpe , g. m. , vettolani , g. , balkowski , c. , blanchard , a. , cappi , a. , cayatte , v. , chincarini , g. , collins , c. , guzzo , l. , macgillivray , h. , maccagni , d. , maurogordato , s. , merighi , r. , mignoli , m. , pisani , a. , proust , d. , & scaramella , r. 1999 , a&a , 342 , 1 , d. j. , finkbeiner , d. p. , & davis , m. 1998 , apj , 500 , 525 , p. & saracco , p. 2001 , astrophysics and space science , 276 , 749 , c. w. , carlberg , r. g. , yee , h. k. c. , morris , s. l. , lin , h. , sawicki , m. , hall , p. b. , & patton , d. r. 2001 , apj , 560 , 72 , i. , ivezi ' c , . , knapp , g. r. , narayanan , v. k. , strauss , m. a. , gunn , j. e. , lupton , r. h. , schlegel , d. , bahcall , n. a. , brinkmann , j. , brunner , r. j. , budav ' ari , t. , csabai , i. , castander , f. j. , doi , m. , fukugita , m. , gyry , z. , hamabe , m. , hennessy , g. , ichikawa , t. , kunszt , p. z. , lamb , d. q. , mckay , t. a. , okamura , s. , racusin , j. , sekiguchi , m. , schneider , d. p. , shimasaku , k. , & york , d. 2001 , aj , 122 , 1861 , j. & davis , m. 1979 , aj , 84 , 1511 , k. h. , simard , l. , zabludoff , a. i. , & mulchaey , j. s. 2001 , apj , 549 , 172 , d. l. , oemler , a. j. , hashimoto , y. , shectman , s. a. , kirshner , r. p. , lin , h. , landy , s. d. , schechter , p. l. , & allam , s. s. 2000 , apjs , 130 , 237 , r. j. , morris , s. l. , & the cnoc2 team . 2004 , mnras , in preparation , d. j. , balogh , m. l. , bower , r. g. , mulchaey , j. s. , oemler jr , a. , carlberg , r. g. , eke , v. r. , lewis , i. j. , morris , s. l. , & whitaker , r. j. 2004 , mnras , accepted , g. , cowie , l. l. , barger , a. j. , & burke , d. j. 2002 , aj , 124 , 1258 , h. k. c. , ellingson , e. , & carlberg , r. g. 1996 , apjs , 102 , 269 , h. k. c. , morris , s. l. , lin , h. , carlberg , r. g. , hall , p. b. , sawicki , m. , patton , d. r. , wirth , g. d. , ellingson , e. , & shepherd , c. w. 2000 , apjs , 129 , 475 , a. i. & mulchaey , j. s. 1998 , apj , 496 , 39 . 2000 , apj , 539 , 136 despite the greater depth and spectroscopic completeness achieved in the region of the groups with our magellan targetting , it was still necessary to adopt a sparse sampling strategy due to the high density of targets , especially close to our new magnitude limit of @xmath130 . in this section , we investigate the selection functions for the spectroscopic sample , seeking a representative means of stacking the data . the cnoc2 photometric catalogue is complete to @xmath131 and so the probability that we possess a redshift of any galaxy brighter than this limit can be easily understood . in section [ sec : seleffects ] we assume that our magellan data reach sufficient depth to be unbiased at magnitudes @xmath132 . we then describe a simple weighting scheme which accounts for targetting bias in both cnoc2 and magellan spectroscopic samples . this scheme differentiates between galaxies with redshifts from each source which have different selection functions . in section [ sec : zincompl ] we test our assumption of unbiased redshift completeness in the magellan data and show that this assumption holds within the limits of tests using current data . to understand the selection functions for the spectroscopic sample , it must first be split into its component cnoc2 and magellan subsets as selection depends critically on the observing strategies and facilities used . target selection in the cnoc2 spectroscopic survey was primarily dependent upon apparent @xmath102-band magnitude . however , below the nominal spectroscopic limit of @xmath133 in that survey , galaxies with redshifts become biased towards strong ew[oii ] with respect to those obtained using magellan . in contrast target selection with magellan is dependent upon both magnitude and angular distance from the targetted group centre . however there appears to be no strong bias towards emission redshifts brighter than our magnitude limit of @xmath41 ( see section [ sec : zincompl ] ) . therefore we can implement a simple weighting scheme which simulates a @xmath134 per cent complete survey and removes targetting bias and cnoc2 redshift incompleteness bias simultaneously : the weighting scheme requires a radial selection function and a magnitude - dependent selection function . for each galaxy in the spectroscopic sample , a radial weight @xmath135 and a magnitude - dependent weight , @xmath136 are computed in such a way that the total weight of all galaxies with redshifts ( measured - z galaxies ) @xmath137 , the total number of galaxies in the photometric sample brighter than @xmath41 . we denote galaxies with redshifts from the original cnoc2 survey with the suffix @xmath138 and those with magellan ldss2 redshifts with the suffix @xmath139 . we begin by looking at the radial selection function , @xmath140 . the fraction of galaxies with cnoc2 redshifts shows no significant dependence upon angular distance from the centre of the nearest group and so we choose @xmath141 for all these galaxies . however there is a strong dependence on angular radius for the fraction of magellan redshifts ( or magellan targets ) . in figure [ fig : radcompl ] , the fraction of galaxies in the cnoc2 photometric catalogue ( @xmath142 ) targetted by magellan , @xmath143 ( triangles ) , where redshifts were obtained , @xmath144 ( filled circles ) and including the fraction with cnoc2 redshifts , @xmath145 ( diamonds ) is shown as a function of angular radius . most untargetted galaxies lie in the magnitude range @xmath146 . we apply a final cut to stacked data at 240 which is the approximate limiting radius for ldss2 targetting . this corresponds to @xmath98mpc at the low redshift end of our sample , @xmath109 . the linear spline through the filled circles ( @xmath144 ) represents the radial selection function @xmath144 from which the value of @xmath140 is linearly interpolated for any galaxy targetted with magellan . for these galaxies , the radial weight is then simply computed to be @xmath147 , normalised so that galaxies located at the group centre receive a weighting @xmath148 . we note that a galaxy at 240 receives a weighting of @xmath149 . in figure [ fig : magcompl ] , we show the magnitude - dependent selection function . galaxies are limited to within 240 and galaxies with redshifts are weighted by @xmath135 . as in figure [ fig : radcompl ] , we plot the fraction of galaxies with magellan redshifts , @xmath150 ( filled circles ) and the fraction with magellan or cnoc2 redshifts , @xmath151 ( diamonds ) . @xmath136 must weight the galaxies such that @xmath137 in each magnitude bin and the combined weight of all galaxies must accurately emulate the properties of the entire galaxy population . in each bin of magnitude we normalise the total weighted fraction of cnoc2 redshifts to be equal to the fraction of cnoc2 redshifts in the unweighted sample , @xmath152 . we know that in the entire population , that fraction at least will have the properties of cnoc2 spectroscopic galaxies , regardless of cnoc2 redshift incompleteness . magellan targets come from the remaining galaxies without cnoc2 redshifts ( ignoring a tiny comparison sample ) and these have properties typical of that population at any given magnitude ( based upon our assumption of unbiased redshift completeness in the magellan data - see section [ sec : zincompl ] ) . therefore the remainder of the combined weight required to match the total number of galaxies in any magnitude bin ( @xmath153 ) is spread amongst the galaxies with magellan redshifts . thus @xmath136 is computed independently for galaxies with cnoc2 redshifts and magellan redshifts using equations [ eqn : wmagczn ] and [ eqn : wmaglzn ] , below ( which simplify to equations [ eqn : wmagczs ] and [ eqn : wmaglzs ] ) . we note that each quantity is computed at a given @xmath154-band magnitude ( or across each magnitude bin ) . we note from figure [ fig : magcompl ] that in the magnitude bin at @xmath162 the fraction of galaxies with redshifts is down to @xmath163 per cent . however , it is at magnitudes below @xmath164 that the magellan data comes to dominate the statistics and the greater depth of the larger telescope allows us to achieve unbiased and highly successful redshift determination down to @xmath41 . this fainter galaxy population is expected to be more star forming than the brighter population in the local universe ( e.g. * ? ? ? therefore by reaching this depth , we can begin to probe the evolution of group galaxy properties since intermediate redshift significantly below @xmath165 to reach this interesting population of mainly star forming galaxies . to investigate redshift incompleteness in the galaxies targetted with magellan , we examine the distribution of targetted galaxies in colour - magnitude ( @xmath166 ) space ( figure [ fig : colmagzcompl ] ) . if there were a bias associated with preferentially losing absorption - line ( early - type ) galaxies , then one would expect those galaxies for which we could not measure redshifts to be clustered mainly towards the red end of the colour range . in figure [ fig : colmagzcompl ] we compare the colour and magnitude distributions of objects for which we could not measure redshifts ( bottom panel - hereafter referred to as lost objects ) with those for which we have emission redshifts ( top panel ) and absorption redshifts ( centre panel ) . we restrict this analysis to fainter than @xmath167 where not all targets are successful . open symbols represent objects with redshifts outside our range of interest , @xmath3 or below the luminosity range considered in the analysis , @xmath168 . lost objects are not concentrated in the red ( @xmath169 ) , absorption redshift dominated region , as might be expected if we were preferentially losing absorption line ( @xmath20 early type ) galaxies . indeed , the lost galaxies are spread quite evenly across the full colour range , suggesting no colour bias due to redshift - incompleteness ( and thus no bias in star forming properties ) . nonetheless , we are cautious in evaluating any such bias as it will impact directly upon our results . therefore we perform two further checks . the first check involves making a detailed analysis of the lost objects . we begin by making a manual examination of the spectra for the 19 lost objects with magnitudes @xmath170 . six of these objects are stars misclassified as galaxies and are removed from the catalogue . a further 10 of the objects have noisy , low signal spectra , which could represent low surface brightness ( lsb ) galaxies or astrometric offsets from the brightest galaxy regions . most of these objects have extended morphology , as assessed from the cnoc2 imaging . we also characterize the objects by measuring the signal in a significant part of the spectrum . we measure the combined signal ( total counts , _ cts _ ) from the wavelength ranges @xmath45@xmath46 and @xmath47@xmath48 which span the most efficient region of the grism , eliminating the strong night sky lines . from 13 lost objects with @xmath171 , there are 8 stars , 1 quasar , 1 bl lac , 1 lsb galaxy , 1 highly noisy spectrum and 1 unclassifiable object . in the low signal range , most objects with @xmath172 are lost , although there are also some relatively low signal to noise galaxies with clear absorption redshifts . finally , lost objects in the intermediate @xmath173 range tend to have noisy spectra , often lacking emission lines . however , all blue galaxies ( @xmath174 ) with redshifts in this region possess emission lines and so it seems possible that the blue lost objects in this region lie at low redshift where [ oii ] falls out of the spectral window . alternatively , it is possible that those galaxies without redshift do not for some reason possess emission lines despite their blue colours . however , red ( @xmath175 ) lost objects with this signal range simply appear noisier and with less obvious absorption features than their counterparts with redshifts . the second check involves computing a redshift incompleteness weight , @xmath176 which is computed by distributing the weight of the lost object across its five nearest neighbours with redshifts in colour ( @xmath177 ) - magnitude ( @xmath154 ) space . nearest neighbours are computed by equating 1 magnitude in @xmath154 to 0.5 magnitudes in colour . then the combined weight of each galaxy is computed to be @xmath178 . we note that @xcite have performed a similar correction to compensate for galaxies in their sample for which they could not obtain redshifts . in figure [ fig : fpmagnoz ] we show the fraction of passive galaxies in the sample @xmath110 ( at @xmath179 ) as a function of apparent magnitude @xmath154 without applying this additional weight @xmath176 ( solid line ) with statistical errors computed using the jackknife method . we show how this uncorrected version changes when we apply the redshift incompleteness weight ( @xmath176 ) to all galaxies ( dashed line ) and when we apply it to red galaxies only ( @xmath180 , dotted line ) . the latter correction assumes that all lost blue galaxies lie outside our redshift range whilst lost red galaxies are lost simply due to low signal to noise spectra . this is the most biased scenario we can imagine , yet even in the faintest bin , the difference made to @xmath110 is comparable with the statistical error . therefore we consider the ldss2 redshift incompleteness to be unbiased at @xmath181 , to the best of our knowledge .

the gaseous molecular disk that orbits the main sequence a - type star 49 ceti has been known since 1995 , but the stellar age and the origin of the observed carbon monoxide molecules have been unknown . we now identify 49 ceti as a member of the 40 myr old argus association and present a colliding comet model to explain the high co concentrations seen at 49 ceti and the 30 myr old a - type star hd 21997 . the model suggests that massive 400 earth mass analogs of the sun s kuiper belt are in orbit about some a - type stars , that these large masses are composed primarily of comet - like objects , and that these objects are rich in co and perhaps also co . we identify additional early - type members of the argus association and the tucana / horologium and columba associations ; some of these stars display excess mid - infrared emission as measured with the widefield infrared survey explorer ( wise ) .

1207.1747

extensive observations have shown that by the time a typical t tauri star is @xmath15 myr old its protostellar disk will retain insufficient carbon monoxide gas to be detectable with a radio telescope . stars with ages @xmath25 myr are virtually non - existent within @xmath1100 pc of earth ( zuckerman & song 2004 ; torres et al 2008 ) . thus , stars this close to earth with detectable gas - rich circumstellar disks are few and far between . co has been detected from three classical t tauri stars within 100 pc of earth : tw hya , v4046 sgr and mp mus ( kastner et al 2010 ; rodriguez et al 2010 ; and references therein ) . such stars , with likely ages in the range 7 - 12 myr , represent the oldest known examples of remnant gaseous protoplanetary disks around solar - type stars . two dusty a - type stars near earth with gas - rich disks 49 cet and hd 21997 are quite different from classical t tauri stars . the circumstellar gas at 49 cet (= hr 451 & hip 7345 ) was discovered in 1995 and has warranted the attention of single and interferometric radio telescopes ( zuckerman et al 1995 ; dent et al 2005 ; hughes et al 2008 ) . the dust distribution has been investigated with high spatial resolution mid - infrared imaging ( wahhaj et al 2007 ) . the gas around hd 21997 (= hr 1082 & hip 16449 ) was discovered much more recently ( moor et al 2011 ) . 49 cet and hd 21997 are , respectively , 59 and 72 pc from earth ( van leeuwen 2007 ) . hd 21997 is classified , reasonably securely , as a member the 30 myr old columba association ( moor et al 2006 ; torres et al 2008 ) . the age of 49 cet has been much more up in the air . the earliest estimates of this age were typically 8 - 10 myr and 8 myr is the age adopted in the interferometric study by hughes et al ( 2008 ) . with such a young age , no older than that of the three classical t tauri stars mentioned above , it was reasonable to interpret 49 cet as a star in transition between a protoplanetary and a debris disk . however , largely because they did not associate the galactic space motions ( uvw ) of 49 cet with any known young ( 8 - 100 myr old ) stellar moving group , rhee et al ( 2007 ) tentatively assigned an age of `` 20 ? '' myr to the star . based on the discussion that follows below , we believe that all previous age estimates for 49 cet are too young and that the star is actually a member of the 40 myr old argus association ( torres et al 2008 ; zuckerman et al 2011 ) . thus among main sequence stars , 49 cet and hd 21997 contain the longest - lived substantial reservoirs of circumstellar gas currently known in astronomy . moor et al ( 2011 ) outline how difficult it is to explain how so much molecular gas can be present in the circumstellar disks of such old stars . in the present paper we present a model that appears capable of accounting for the properties of such stars . we deduce the age of 49 cet based on its galactic space motions ( uvw ) and its location on an a - type star color - magnitude diagram ( cmd ) . together these demonstrate that 49 cet is a member of the argus association whose age , based on analysis of its g- and k - type members ( torres et al 2008 ) , is @xmath140 myr . table 1 lists a - type stars that we propose as likely or possible members of the argus association ; some of these stars are considered in section 3 . the mean uvw of the association is given in the table notes . as may be seen , the measurement errors in radial velocity for some table 1 stars are substantial ; until more accurate velocities are measured , identification of such stars with argus can be regarded only as possible . however , the listed heliocentric radial velocity ( [email protected] km s@xmath4 ) for 49 cet (= hip 7345 ) is well determined ( gontcharov 2006 ) and agrees well with the systemic velocity ( 12.2 km / s ) determined from the j = 2 - 1 line of co ( hughes et al 2008 ) . while measurement of accurate radial velocities for rapidly rotating early - type stars is challenging , precise velocities can be easily measured for radio emission lines from circumstellar disks . thus the galactic space motion of 49 cet is consistent with that of the argus association . from a list of @xmath11500 early - type main sequence stars within 80 pc of earth very few are found to have all three components of uvw close to those of the argus association ( six such stars are listed in table 4 of zuckerman et al 2011 ) . a star is included in the present tables 1 and 2 or in tables 4 - 6 of zuckerman et al ( 2011 ) if each component of uvw is consistent with that of the association mean . `` consistency '' in this context implies that all 3 stellar components of uvw match those of the association mean within the sum of the quoted error in stellar velocity and the standard deviation of the dispersion among the association velocities . however , because errors in uvw can be underestimated ( see following paragraph ) , because a standard deviation is not an absolute limit , and because these tables list `` proposed '' association members , we have tended to err somewhat on the side of including a potential non - member in preference to losing a true member . in figure 1 we plot the location on a cmd of early a - type stars near earth that have been proposed as members of five young moving groups . table 2 presents early - type stars that we propose as members of the 30 myr old tucana / horologium and columba associations ; membership is based on uvw and location on a cmd . as may be seen from figure 1 and table 3 , for a given b - v there sometimes is a substantial spread in m@xmath5 for stars proposed to be of the same or nearly the same age . if the stars have been correctly placed into the indicated moving groups , then one possibility is that some aspect of the formation and early evolution of early - type stars can lead to a range of stellar sizes at a given color and nominal age . alternatively , some of the stars in table 3 may not actually belong to the indicated moving group . given the errors in measuring radial velocities of early a - type stars and , to a lesser degree , errors in hipparcos parallaxes , some proposed members of a given moving group will ultimately turn out to be non - members . in the present work and that of our 2011 paper ( zuckerman et al 2011 ) we have adopted the gontcharov ( 2006 ) radial velocities and their errors while recognizing that the latter are likely to be too optimistic for some stars . we adopt hipparcos parallaxes from van leeuwen ( 2007 ) while anticipating that some of these may not be correct ( see , for example , the example of hip 30675 considered in section 3 ) . ultimately , high accuracy parallaxes from gaia , a dedicated campaign to confirm the gontcharov radial velocities and reduce their errors , and perhaps a traceback analysis of the moving groups , for example , as demonstrated by ortega et al ( 2002 ) for the @xmath6 pictoris moving group , should distinguish between true members and impostors . notwithstanding caveats expressed in the preceding paragraph , the excellent agreement of the uvw of 49 cet with that of the argus association and the location of 49 cet ( the red square in figure 1 and star a in table 3 ) , indicates that 49 cet is unlikely to be younger than 40 myr . indeed from comparison on figure 1 of the location of the various proposed a - type members of argus with the indicated pleiades stars ( from vigan et al 2012 ) , the argus association might be somewhat older than the 40 myr deduced by torres et al ( 2008 ) . thus , a 40 myr age for 49 cet is at least as secure as the 30 myr age for hd 21997 deduced by moor et al ( 2011 ) . hd 21997 is star `` e '' in figure 1 . various searches for co toward dusty main sequence stars with ages @xmath710 myr have been reported , but only 49 cet and hd 21997 have yielded detections ( e.g. , zuckerman et al 1995 ; dent et al 2005 ; kastner et al 2010 ; moor et al 2011 ) . in appendix a and table 4 we present a revised version of a portion of table 1 in kastner et al ( 2010 ) . a principal motivation for revisiting this table is the absence of co abundance upper limits in the kastner et al table which presents only limits to h@xmath0 abundances based on an assumed h@xmath0/co ratio by number equal to 10@xmath8 . as discussed in section 4.3 below , if the comet model described in section 4.4 can account for the co observed at 49 cet and hd 21997 , then the h@xmath0/co ratio in these debris disks is unconstrained and is unlikely to be 10@xmath8 . in the present section we consider mid ir emission as measured with the widefield infrared survey explorer ( wise ) , uvw velocities , and location on a color - magnitude diagram of some stars listed in tables 1 - 3 along with a few additional stars of interest . we examined the wise catalog to determine if any stars in tables 1 and 2 had excess ir emission above the stellar photosphere . our criteria for deciding whether an excess is present are the same as described in section 2.2 of zuckerman et al ( 2012 ) , so we do not repeat these criteria here . four stars from tables 1 and 2 display excess emission above the stellar photosphere in the wise 22 @xmath9 m filter ; their spectral energy distributions are shown in figure 2 . hip 23585 : this table 1 star is too red to appear in figure 1 , but can be plotted on similar color - magnitude diagrams that appear in zuckerman ( 2001 ) and vigan et al ( 2012 ) . its location is consistent with a 40 myr age . hip 25280 , 59394 , 72408 , 104365 : based on their w component of space velocity , these table 2 stars are members of the tuc / hor association ( rather than columba ) . hip 26309 : this member of the nucleus of the columba association ( zuckerman et al 2011 ) , sits very low on a cmd ( star h in figure 1 ) . hip 26395 and 53771 : based on their w component of space velocity , these table 2 stars are members of the columba association ( rather than the tuc / hor association ) . hip 26395 is a new member of the nucleus of the columba association ; the nuclear region is delineated in section 5.2 of zuckerman et al ( 2011 ) . hip 26395 is star `` i '' in table 3 and figure 1 ; it , along with another columba member , hip 32104 ( star k ) , and a tuc / hor member hip 12413 ( star c ) all sit near each other and very low on the cmd . hip 30675 : this @xmath120 arcsec double star , hr 2328 , is potentially nearby and young . according to simbad and van leeuwen ( 2007 ) , the parallax is 13.26 mas which would put the star @xmath175 pc from earth . however , there are other sources where the listed parallax is much smaller , near 8 mas ; for example in the hipparcos double star catalog and in kharchenko & roeser ( 2009 ) . if 13.26 is to be believed , then the star plots too low in figure 1 to be reasonable , whereas if something like 8 is used , then the location is reasonable . also , the small proper motion is more likely for a more distant star . at a distance of 125 pc , the uvw is -10.0,-19.4,-5.7 ( 1.9,2.0,1.3 ) , consistent with membership in the columba association ( see notes to table 2 ) . the primary is a0-type and the secondary about f5 . even for a young star , the f - star is very bright in x - rays according to figure 4 in zuckerman & song ( 2004 ) ; based on the rosat all - sky survey , its fractional x - ray luminosity ( l@xmath10/l@xmath11 ) is equal to -3.8 . hip 41081 : the u velocity component is that of tuc / hor while the w component is appropriate for columba . similar mixed - messages occur for a few stars in table 3 of zuckerman et al ( 2011 ) . hip 53771 : while too red to appear in figure 1 , the location of this table 2 star on a cmd ( zuckerman 2001 ; vigan et al 2012 ) is plausible for a 30 myr old star . hip 89925 : this double line spectroscopic binary consists of am and a9 stars ( fekel et al 2009 ) . a deconvolution of the light curve , combined with assumed b - v = 0.14 and 0.26 for the primary and secondary respectively ( fekel et al 2009 ) , places both components in the region of a cmd ( zuckerman 2001 ; vigan et al 2012 ) appropriate for 40 myr old stars . what is the origin of the molecular gas at 49 cet and hd 21997 ? some possibilities include ( 1 ) unusually long - lived remnants of the protoplanetary ( herbig ae ) phase of stellar evolution , ( 2 ) collisions of icy comets , and ( 3 ) signature of the aftermath of a recent collision of gas - rich planets or planetary embryos . the lifetime of a given co molecule in a circumstellar disk is set by the intensity of the ultraviolet stellar and interstellar radiation fields modulated by shielding by other co molecules and by h@xmath0 molecules and dust grains . in their model for hd 21997 , moor et al ( 2011 ) derive a co lifetime of @xmath1500 years and , given similarities in the two debris disks , the co lifetime at 49 cet is unlikely to be much different . 500 years is so short compared with estimated ages of 30 - 40 myr that we concur with moor et al that the observed co is not likely to be a remnant of the protoplanetary phase . thus , in what follows , we consider models with rapid production of gas phase co. we assume planetesimal belts that contain volatile - rich comet - like and larger objects , akin to the sun s kuiper belt . moor et al ( 2011 ) and hughes et al ( 2008 ) find that shielding of co by dust grains is unimportant in comparison with shielding by h@xmath0 or co self - shielding . nonetheless , the dust abundance in the debris disk can potentially yield important clues regarding the origin of the co molecules . for example , 49 cet and hd 21997 are among the dustiest a - type stars within 70 pc of earth thus suggesting a direct connection between dust and gas even if the former does not protect the latter . to determine how the amount of dust at 49 cet and hd 21997 compares with that at other a - type stars near earth , we consider stars with excess 60 @xmath9 m measured with the infrared astronomical satellite ( iras ) and listed in table 2 of rhee et al ( 2007 ) . this table should be complete for a - stars within 70 pc of earth with large fractional infrared luminosities , @xmath12 = l@xmath13/l@xmath11 ; all such stars would have been easily detectable with iras . according to rhee et al ( 2007 ) , @xmath12 = 8 x 10@xmath14 and 5 x 10@xmath14 for 49 cet and hd 21997 , respectively . there are @xmath1300 a - type stars within @xmath170 pc of earth ( jura et al 1998 ) . of these only 4 have @xmath12 as large or larger than those of 49 cet and hd 21997 . three of the four hr 4796 , @xmath6 pictoris , hr 7012 have ages @xmath110 myr and may not yet have completely dispersed their primordial dusty disks . in addition , the dust at these three stars is warmer than the dust at 49 cet and hd 21997 , and the closer to a star that dust grains reside the fewer are required to absorb a given fraction of the starlight . the remaining very dusty star is hip 8122 with @xmath12 = 5 x 10@xmath14 and an estimated age of 100 myr ( rhee et al 2007 ) . thus , for stars whose infrared emission is dominated by kuiper belt analog , cool dust , 49 cet and hd 21997 rank among the 1 - 2% dustiest of nearby a - type stars . ( we do not consider here the tiny number of young nearby a - type stars with ir - excess emission dominated by warm , zodiacal - type dust ( melis et al 2012 ) ) . youth and dustiness notwithstanding , co has not yet been detected with a radio telescope at any of hr 4796 , @xmath6 pictoris , or hr 7012 , consistent with the notion that the co gas seen at 49 cet and hd 21997 is not primordial . if the co gas were primordial , then on average one would expect more co in a dusty 10 myr old circumstellar envelope than one of age 30 - 40 myr . while there have been detections via optical absorption lines of trace amounts of gas around some main sequence stars including @xmath6 pictoris ( e.g. , roberge & weinberger 2008 and references therein ) , the derived column densities of gas are far less than those necessary for radio detection of a rotational line in emission . the far infrared emission seen by iras is carried primarily by grains of radius @xmath110 @xmath9 m , just above the radiative blowout size . flux densities at submillimeter wavelengths sample larger grains , with radii up to a few mm . the total mass of these large grains has been measured with radio telescopes at various stars ; results are listed in table 2 in rhee et al ( 2007 ) and for vega , fomalhaut , 49 cet , and hd 21997 are presented in the 4th column of table 5 of the present paper . again , whether one examines table 2 in rhee et al ( 2007 ) or table 5 here , 49 cet and hd 21997 stand out for having large total dust masses in comparison with other nearby stars . if co molecules are sequestered in planetesimals and in the dusty - icy matrices that result from collisions of such objects , then dust and co production rates should be related . as many authors have shown , for debris disks as dusty as those at 49 cet and hd 21997 , collisional cascade followed by radiative blowout of small dust grains is a far faster dust loss mechanism than is poynting - robertson drag . given the plausibility of a direct or indirect connection between dust and co production rates ( as considered above ) , it is important to understand both rates . unfortunately , in two of the best studied debris disks , estimates of dust production rates are wildly discrepant for vega ( su et al 2005 ; mller et al 2010 ) and for fomalhaut ( acke et al 2012 ; krivov 2012 , personal communication and below ) . we therefore consider collisional cascade rates in some detail . 0.2 in one may estimate a minimum dust mass required to block a given fraction of the light of a star ( e.g. , chen & jura 2001 ) ; we denote this fraction as @xmath12 = l@xmath13/l@xmath11 ( see the third column in table 5 ) . for conventional dust grain size distributions , such as in a collisional cascade , the dominant particles contributing to @xmath12 have radii somewhat larger than the blowout radius ( chen & jura 2001 ) . the required dust production rate in a steady - state must balance the most rapid relevant loss rate of these dominant particles ; in a collisional cascade this loss rate is set by destructive collisions of small particles that give rise to even smaller particles that can then be radiatively blown out of the debris disk . @xmath12 is an easily observable quantity and is equivalent to radial optical depth through a spherically symmetric dust shell . in appendix b we derive an expression , as a function of @xmath12 , for collision times of dust particles with sizes slightly above the blowout size . we consider a cylindrical ring of radius r , width in the radial direction of @xmath15r , and full vertical height h. then the collision time is 0.2 in @xmath16 0.2 in where p is orbital period . the dust loss rates at vega and at fomalhaut estimated by su et al ( 2005 ) and acke et al ( 2012 ) , respectively , are much larger than one calculates from equa.1 or in the collisional cascade model proposed by mller et al ( 2010 ) . we consider these discrepancies in detail in appendix c. for reasons given there , and because the su et al and acke et al papers lack a description of a physical mechanism that can generate small dust particles at the rates required in their models , we assume a conventional collisional cascade to produce small particles at a rate given by equa.1 . with this expression and the one that follows , we can calculate the rate of dust grain production / loss for 49 cet and hd 21997 . as shown in detail by melis et al ( 2012 ) , the mass loss rate in a collisional cascade can be written as a function of stellar and dust properties , 0.2 in @xmath17 0.2 in where r@xmath18 , t@xmath18 , and m@xmath18 are in solar units and t@xmath19 is in k. the value of the coefficient c depends on the ratio @xmath15r / r in equa.1 . for @xmath15r / r = 0.5 , appropriate for 49 cet and for hd 21997 ( table 5 ) , c = 1.75 x 10@xmath20 when dm / dt is in g s@xmath4 . we take r@xmath18 , t@xmath18 , @xmath12 , and t@xmath19 from table 2 in rhee et al ( 2007 ) , and assume m@xmath18 = 2.7 and 2.4 for 49 cet and hd 21997 , respectively . the derived loss rate , in g s@xmath4 , for small dust particles is listed in table 5 for these two stars as well as for vega and fomalhaut . the table 5 dust loss rates for 49 cet and hd 21997 are similar to dust production rates derived by kenyon & bromley ( 2010 ) in their model for debris disks around 30 myr old a - type stars ( e.g. , see their figs . 13 & 14 ) . 0.2 in in a steady - state model for the co currently in orbit around hd 21997 , moor et al ( 2011 ) deduce that 140 earth masses of planetesimals in the debris disk would have to have been destroyed in the past 20 myr . this contrasts dramatically with the mller et al ( 2010 ) model for vega where a collisional cascade operating for the past @xmath1340 myr would have destroyed only a few earth masses of solids . moor et al thus favor a situation where the co is currently experiencing a temporary high production rate . the mass loss rate in dust at hd 21997 ( table 5 ) presents another potential problem for any steady state model for generating the observed co. moor et al derive a minimum co production rate of 10@xmath21 g s@xmath4 . this is an order of magnitude faster than the dust production rate ( table 5 ) . suppose the dust is composed of a 50 - 50 mixture by mass of water ice versus silicates plus other solids ( e.g. , section 3.2 in acke et al 2012 ) . optimistically taking the co and co@xmath0 mass fractions in a typical comet to each be 20% that of water ice ( mumma & charnley 2011 ) , and assuming that photodissociation of co@xmath0 produces a co molecule , one then obtains a ratio , co / dust , equal to @xmath10.2 by mass . then the co production rate would seem to be of order a factor of 50 too fast relative to that of the dust , at least in the steady - state . these assumed co and co@xmath0 mass fractions are near the upper end of those measured in solar system comets ( mumma & charnley 2011 ) . however , solar system observations sample only the outer layers of comets that are much older and have been heated to much higher temperatures than young comets at 49 cet and hd 21997 . thus , it would not be surprising if the latter are , on average , more co- and co@xmath0-rich than solar system comets . the discrepancy between the co and dust production rates in the steady state could be reduced somewhat should co and co@xmath0 carry an even larger fraction of the mass of a typical youthful comet than we have assumed . in section 4.4 we present a colliding comet model to account for the co observed at 49 cet and hd 21997 . should outgassing from comets be responsible for the co , then the h@xmath0/co ratio in the debris disks is essentially unconstrained . in particular , there is no reason to expect the large ratios by number , e.g. 10@xmath8 or 10@xmath22 , conventionally assumed in models of co excitation ( e.g , hughes et al 2008 ; moor et al 2011 ) . then excitation of co rotational levels may well be due to something other than collisions with h@xmath0 . one plausible alternative could be collisions of co with electrons . following photodissociation of co , the c will be photoionized and a source of electrons . provided the c+ ions remain in the same volume as the co molecules for a sufficient length of time , then these electrons can excite the co rotational ladder . dickinson et al ( 1977 ) give expressions for calculation of rate coefficients for electron excitation of rotational levels of linear polar molecules in low density clouds . we assume an electron temperature of 70 k and consider the j = 2 - 1 transition of co. then the collision time ( sec ) for excitation of the j = 2 level is 1.4 x 10@xmath23/n@xmath24 , where n@xmath24 is the number of electrons per @xmath25 . for the co 2 - 1 transition 1/a @xmath110@xmath26 sec so , the j = 2 level can be significantly excited in regions with n@xmath24 @xmath1100 @xmath25 . the electron densities are related to the co volume densities and the lifetime of c+ and electrons in the volume where the co resides . given the substantial uncertainties in the volumes occupied by the observed co molecules at 49 cet and hd 21997 , reliable estimation of electron densities will have to await mapping of co with alma . that said , with co volumes based on values for @xmath15r , r and h listed in table 5 , provided that co and c+ abundances are comparable in regions where co resides , then electrons would be a viable collision partner for excitation of co in orbit around 49 cet and hd 21997 . independent of whether or not electrons dominate excitation of co rotational levels , because h@xmath0 helps to shield co against photodissociation , if h@xmath0 abundances are much lower than assumed in conventional models , then this could have substantial implications for the co production rates required to match the observations . two principal constraints on a model that seeks to explain the co abundances at 49 cet and hd 21997 are ( 1 ) the rapid rate of co production necessitated by its short ( @xmath1 500 yr ) lifetime against photodissociation ( e.g. , moor et al 2011 ) , and ( 2 ) the relatively rapid rate of co production compared to dust production ( section 4.2 ) . the model outlined below involves collisions among the great numbers of comets that likely orbit 49 cet and hd 21997 . because these two stars rank among the dustiest a - type stars in the solar vicinity ( section 4 and table 5 ) it is not unreasonable to assume that their debris disks are quite massive relative to those that orbit most other stars . in this sense 49 cet and hd 21997 are very dusty stars ; but in comparison to the amount of co present , there is relatively little dust . schlichting & sari ( 2011 ) present a model of the young sun s kuiper belt ( kb ) . they argue that the total kb mass was dominated during its first 70 myr by comet - size planetesimals and that the kb never possessed many more pluto - size objects than it does now . they take the radius r , radial width @xmath15r , and total mass of the kb to be 40 au , 10 au , and 40 earth masses , respectively . we envision the debris regions at 49 cet and hd 21997 to have r = 100 au and ( @xmath15r)/r = 0.5 ( table 5 ) . neither r nor @xmath15r are particularly well constrained by existing data for these stars ; hopefully alma will substantially improve the situation in the future . for the sun s young kuiper belt region and for 49 cet and hd 21997 we assume a disk surface density proportional to r@xmath27 ( lissauer 1987 ) and a scale height proportional to r. then , scaling from the kb , the total mass of comets that orbit 49 cet and hd 21997 is equal to @xmath1400 earth masses . this estimate assumes that at a given value of r , say 40 au for example , youthful a - type stars will have about the same surface density of comets as did the youthful kb . with co and co@xmath0 cometary mass fractions as assumed in section 4.2 , 400 earth masses of comets corresponds to a co plus co@xmath0 mass of 80 earth masses . with a canonical h@xmath0/co ratio by number of 10@xmath8 , a protoplanetary disk of at least 0.2 m@xmath28 is implied . in a sample of 86 pre - main - sequence stars in the taurus - auriga dark clouds , beckwith et al ( 1990 ) deduced the presence of a few dusty protoplanetary disks with masses as large as a solar mass . the mass of the typical star in the beckwith et al sample is a few times smaller than the masses of 49 cet and hd 21997 , who , as noted above , rank among the dustiest main sequence stars known . we consider each comet to have a radius of 1 km and a density of 2 g @xmath25 , such that a typical comet has a mass of 10@xmath29 g. there are thus 2.4 x 10@xmath21 comets in orbit around 49 cet and hd 21997 . sanford & alamendola ( 1990 ) and tielens et al . ( 1991 ) consider sublimation temperatures for pure co & co@xmath0 ices and also co & co@xmath0 ices trapped in an h@xmath0o ice matrix ; these temperatures are all @xmath3060 k. for typical collision velocities of 100s of m s@xmath4 ( appendix b ) , the bulk of the shattered comets will be heated well above these temperatures . in addition , `` most kilometer - sized asteroids are likely rubble - piles . many comets may also be strengthless or nearly strengthless bodies , their fragility demonstrated when they break up far from perihelion for no obvious reason '' ( see movshovitz et al 2012 and references therein ) . given the high co and co@xmath0 mass fractions assumed in our model , following a cometary collision these molecules should have little trouble escaping as vapors . at the same time , less volatile materials can remain as solids ( e.g. , tielens et al 1994 ; czechowski & mann 2007 ) . we estimate the characteristic time for comet collisions using equa . the number of comets per unit volume ( n ) is equal to 2.4 x 10@xmath21 divided by the volume of the debris ring ( 2@xmath31r@xmath15rh ) . this yields a collision time of @xmath32 where r and @xmath15r are in cm . with a period of 650 yr at 100 au around these a - type stars , t@xmath33 = 5 x 10@xmath34 yr . with 2.4 x 10@xmath21 comets , there are 5 x 10@xmath26 collisions / yr , or one every 6 seconds ! as in section 4.2 , we assume that 50% of the mass of a typical comet is water ice , that co and co@xmath0 each comprise 20% as much mass as water ice , and that photodissociation of co@xmath0 yields a co molecule . when two comets collide , we assume that 50% of the total mass becomes debris consisting of co , co@xmath0 , and solid material composed primarily of equal quantities by mass of water ice and silicates . with these assumptions , co is produced at a rate @xmath110@xmath35 g yr@xmath4 . to reproduce the observed co mass of @xmath13.5 x 10@xmath14 earth masses at hd 21997 , moor et al ( 2011 ) require a co production rate of 4 x 10@xmath36 g yr@xmath4 . hughes et al ( 2008 ) do not give a co production rate appropriate to their model for 49 cet , but the co mass at 49 cet is about 7 times that at hd 21997 or about the mass of pluto ( a.m. hughes , private comm . while this is a lot of co , it is only a tiny fraction of the total mass of comets in the debris disk . as noted in section 4.2 , the co production rate by mass at hd 21997 is a factor of 10 faster than the dust production rate given in table 5 via a collisional cascade . if we assume that the co lifetime against photodissociation at 49 cet is the same as that at hd 21997 ( 500 years ) , then for 49 cet the ratio of co to dust production rate is @xmath120 . in our model the fraction of mass released via comet collisions in the form of co and co@xmath0 is only @xmath120% of the mass released in solid material that will eventually , via a collisional cascade , become the 10 @xmath9 m size dust grains considered in appendix c. the key to why the co - to - dust production rates are skewed so heavily in favor of co is probably related to the much shorter time it takes to release large amounts of co following a disruptive impact compared to the time to whittle a large chunk of solid debris down to 10 @xmath9m - size particles . with equa.1 , the collision time at hd 21997 for dust particles with sizes a bit above the blowout radius is 2.6 x 10@xmath8 yr . for typical solid chunks coming off a comet collision , the ensuing collision times could easily be millions of years . in the model of schlichting & sari ( 2011 ) , dynamical stirring of the young kb increases as a function of time as large bodies grow in the debris belt , they stir up the smaller objects ( the comet - size objects we are considering here ) . a similar picture of the early evolution of debris disks around youthful a - type stars was developed earlier by kenyon & bromley ( 2008 , 2010 ) . thus , 49 cet and hd 21997 may be in a dynamically active phase where comet collisions are frequent , but where the breakdown of large bodies of solid debris into small dust particles has not yet had time to operate to conclusion . the spirit of such a model is not one that relies on an unusual , catastrophic , transient event . rather the two stars are passing through a normal era in the evolution of comet - dominated debris belts . given the discussions in schlichting & sari ( 2011 ) and kenyon & bromley ( 2010 ) , it would not be unreasonable for that era to occur at an age of @xmath130 myr ( see , e.g. , , the 1 km panel in figure 23 in kenyon & bromley 2010 ) . we present a massive ( 400 earth mass ) comet - cloud model to explain the large quantities of carbon monoxide gas seen at the 30 - 40 myr old , a - type , stars hd 21997 and 49 ceti . because co is rapidly photodissociated in the stellar and interstellar radiation fields , it must be produced rapidly . we calculate that this production rate is an order of magnitude faster than the rate of production of dust particles via a model of collisional cascade in the steady state . the implications of this ratio are : ( 1 ) young , pristine comets are likely to be richer in co and perhaps co@xmath0 than typically observed solar system comets , and ( 2 ) the destruction of the youthful comets around hd 21997 and 49 cet are not likely to have been in a steady state situation over periods of time exceeding millions of years . if dynamical activity in debris disks is ramping up at the age of these stars , for example through the build - up of large bodies as in the model of the young kuiper belt proposed by schlichting & sari ( 2011 ) or the model of debris disks around a - type stars by kenyon & bromley ( 2010 ) , then the production of small dust particles may lag well behind the rate of co outgassing . the primary driving force for co outgassing is likely to be collisions among the 100 trillion comets that orbit 49 cet and hd 21997 . such a model for these systems is consistent with origin in a protoplanetary disk of initial mass of order a few tenths of a solar mass and suggests that comets form very early in the history of such disks before there is sufficient time to convert much co to ch@xmath37 . if comet collisions provide the observed co , then the h@xmath0/co ratio is unconstrained and may not be large . if so , then collisional excitation of the co rotational levels may be dominated by something other than h@xmath0 , perhaps electrons . acke et al ( 2012 ) have proposed a rapid comet - destruction model for the dusty debris disk that orbits fomalhaut that , at first glance , might seem similar to our model for hd 21997 and 49 ceti . but , in fact , the two models are very different . as we note in appendix c , the acke et al dust loss rate is 20 times faster than the rate we calculate via a conventional collisional cascade . they do not specify a physical mechanism that can convert comets so rapidly into dust . their rapid dust loss rate at fomalhaut is compelled by their quoted very short lifetime ( 1700 yr ) for dust grains near the blowout size . but a detailed model of the similar debris disk at vega by mller et al ( 2010 and as outlined in our appendix c ) casts doubt on the validity of such a short lifetime . additional observations and modeling of the fomalhaut debris disk are warranted . in particular , a deep search for co at fomalhaut would be worthwhile ; because it is the presence of so much co that compels the rapid comet destruction rate in our model , while combination of observations of co and of dust serve to constrain the model . at the risk of being obvious , dusty young stars will be excellent targets for alma mapping of dust and co gas , should the latter be detectable . it should be possible to map the shape and mass of youthful kuiper belt analogs and to clarify the composition of young comets . 0.2 in we thank meredith hughes , alexander krivov , and hilke schlichting for taking time to clarify some aspects of their papers , david jewitt and michael jura for helpful comments , and the referee for useful suggestions . this research was funded in part by nasa grants to ucla and the university of georgia . table 4 lists a set of upper limits on peak intensities ( @xmath38 ) and integrated intensities in the @xmath39co(21 ) line ( @xmath40 ) as measured with the iram 30 m telescope in 2009 august for a sample of dusty stars within @xmath1100 pc of earth that includes hd 21997 . these upper limits supercede those reported in columns 811 in table 1 of kastner et al . the upper limits on integrated line intensity are computed as in 2.1 of kastner et al . ( 2010 ) for spectrometer channel widths of @xmath41 km s@xmath4 ; but now for an assumed linewidth of @xmath42 km s@xmath4 , whereas they assumed a linewidth of @xmath43 km s@xmath4 . the equivalent line intensities in jy km s@xmath4 are then obtained assuming a conversion factor of 7.85 jy k@xmath4 for the 30 m telescope at 230 ghz . the listed values of @xmath38 were obtained from the rms of the residuals to a fit to the baseline of each spectrum over the range @xmath44 km s@xmath4 to @xmath45 km s@xmath4 ; these values differ only slightly from those reported in kastner et al . ( 2010 ) , which were based on fits over a somewhat longer baseline . the conversion factor employed here differs from that adopted by kastner et al . ( 2010 ) ; this factor was applied erroneously by those authors . note that the @xmath39co(21 ) upper limit for hd 21997 in table 4 , @xmath46 jy km s@xmath4 , is consistent with the integrated line intensity of @xmath47 jy km s@xmath4 reported in moor et al . likewise the upper limit on co mass for hd 21997 in table 4 is consistent with their measured co mass . the easily observable quantity is radial optical depth through a spherically symmetric dust shell , @xmath12 = l@xmath13/l@xmath11 , but collision rates of disk particles are more directly related to what has been called `` face - on fractional surface density @xmath48(r ) '' ( backman & paresce 1993 ) or `` full vertical optical thickness '' (= @xmath12@xmath49 , artymowicz & clampin 1997 ) of disk - like structures . the expressions given by these sets of authors are similar and we begin with equation 8 from artymowicz & clampin ( 1997 ) : 0.2 in @xmath50 0.2 in where p is orbital period . we consider a cylindrical ring of radius r , width in the radial direction of @xmath15r , and full vertical height h. then the ring ( disk ) optical depth in the radial direction , @xmath12@xmath51 , is 0.2 in @xmath52 0.2 in relating @xmath12@xmath51 and @xmath12@xmath49 gives 0.2 in @xmath53 0.2 in or with equa . b1 , 0.2 in @xmath16 0.2 in when p is expressed in terms of orbital radius to the 1.5 power and stellar mass to the -0.5 power ( as per kepler and newton ) , then this expression is in good agreement with equa . ( 25 ) of wyatt et al ( 2007 ) . finally , with @xmath15r / r = 0.3 from table 5 , we obtain a typical dust particle collision time 0.2 in @xmath54 0.2 in with a variation of perhaps a factor of @xmath32 due to differing dust ring , in - plane , extents . one can also employ kinetic theory to independently derive an expression that relates t@xmath33 to @xmath12 , but without use of equa . collision times are equal to the mean free path ( mfp ) between dust particle collisions divided by the relative particle velocities . the relative velocity for typical particles in the ring is the orbital velocity times h/2r . so the collision time is 0.2 in @xmath55 0.2 in the mfp = ( @xmath48n)@xmath4 = ( 4@xmath31a@xmath56n)@xmath4 , where `` a '' is dust grain radius and n is the number of grains per unit volume . then , 0.2 in @xmath57 0.2 in and 0.2 in @xmath58 0.2 in thus , 0.2 in @xmath59 0.2 in in agreement with equa artymowicz & clampin ( 1997 ) use kinetic theory to derive a collisional rate a few times faster than that given in equations b4 and b9 , but then choose to adopt the rate given in equa . b4 ; we do the same . 0.2 in based on spitzer space telescope observations of vega , su et al ( 2005 ) calculated a large dust production rate , @xmath16 x 10@xmath21 g s@xmath4 , and concluded that this rate could not have been sustained over vega s 350 myr lifetime . mller et al ( 2010 ) reinterpreted the spitzer data and derived a dust production / loss rate via collisional cascade two orders of magnitude smaller than that of su et al . in the mller et al model the mass in dust grains with radii @xmath110 @xmath9 m that is , slightly above the blowout size is @xmath17 x 10@xmath14 earth masses . one can calculate a similar dust mass from an estimate of the dust surface area required to block 2 x 10@xmath60 of the light from vega ( e.g. , equa . 4 in chen & jura 2001 ) . in their elaborate vega model , mller et al ( 2010 ) and a. krivov ( private comm . 2012 ) derive a collisional lifetime for these small particles of @xmath13 x 10@xmath61 yr . together , equa . b4 and table 5 yield a collision time of 5 x 10@xmath61 yr , and thus a dust mass loss rate via a collisional cascade of @xmath13 x 10@xmath62 g s@xmath4 . we note in passing , there is a typo in the value for @xmath12@xmath49 = 8.3 x 10@xmath14 given in section 6.8 of mller et al ( 2010 ) . in fact , @xmath12@xmath49 in their model is 8.3 x 10@xmath60 ( a. krivov , private comm . 2012 ) , in good agreement with the value of 10@xmath14 we derive from equa . b3 and table 5 . based on herschel space observatory images , acke et al ( 2012 ) , derived mass loss rates in dust grains for fomalhaut . given that their methods are similar to those used by su et al ( 2005 ) to analyze vega , it is surprising that acke et al cite neither su et al , nor mller et al ( 2010 ) . like su et al , acke et al derive small grain lifetimes of @xmath11000 yr , that is , much shorter than can be achieved via grain production with a conventional collisional cascade . thus , if the acke et al mass loss rate of @xmath17 x 10@xmath63 g s@xmath4 were appropriate for fomalhaut , then dust grain production would likely be due to something other than a collisional cascade ( see paragraph that follows ) ; acke et al ( 2012 ) suggest a quasi - steady state model of colliding comets over a 200 myr lifetime of fomalhaut . should fomalhaut s age instead be @xmath1440 myr , as seems likely ( mamajek 2012 ) , then the acke et al picture would become even more extreme . blockage of 8 x 10@xmath60 of the light from fomalhaut with 10 @xmath9 m radius grains of density 2 g @xmath25 , requires a mass of @xmath110@xmath64 g. this may be compared with the mass of 3 x 10@xmath65 g of comparable size grains with , apparently , somewhat lower density in the model of acke et al ( 2012 ) . with equa . b4 , we find a characteristic dust production timescale of 8 x 10@xmath8 yr and a mass loss rate of 4 x 10@xmath39 g s@xmath4 via collisional cascade . this mass loss rate is @xmath120 times slower than the acke et al mass loss rate . since the mass and luminosity of fomalhaut are similar to that of @xmath66 lep ( rhee et al 2007 ; chen & jura 2001 ) , compact grains of radii only a few microns would be stable against blowout , thus potentially increasing the discrepancy between the quoted acke et al mass loss rate and that estimated from a collisional cascade . it may be possible to match the spectral energy distribution and images of fomalhaut with inclusion of such small , bound , grains ( a. krivov , private comm . should a model similar to that of mller et al ( 2010 ) apply to fomalhaut as well as to vega , then a normal collisional cascade might account for the dust distribution at fomalhaut and invocation of an extreme colliding - comet model might be unnecessary . in particular , use of a single power - law size distribution for both blow - out and bound grains is unphysical ( a. krivov , private comm . 2012 ; mller et al 2010 ) and additional modeling of the data is warranted . as noted in the main text , it is one thing to have comets collide , but converting the resulting debris into small dust particles is something else again . the su et al and acke et al papers lack a description of a physical mechanism that can generate small dust particles at the rates required in their models . therefore , in agreement with mller et al ( 2010 ) , we regard a conventional collisional cascade as the most plausible dust production mechanism . arizona press ) , 1253 @lcccccccccc hip & hd & r.a . & decl . & spec . & v & b - v & dist . & rad . vel . & ( u , v , w ) & uvw error + & & ( h / m ) & ( deg ) & type & ( mag ) & ( mag ) & ( pc ) & ( km s@xmath4 ) & ( km s@xmath4 ) & ( km s@xmath4 ) + 7345 & 9672 & 01 34 & -15 & a1 & 5.6 & 0.07 & 59 & [email protected] & -22.9,-16.4,-5.4 & 0.4,0.3,0.7 + 23585 & 32296 & 05 04 & + 45 & a2 & 6.5 & 0.22 & 61 & [email protected] & -24.4,-11.1,-5.4 & 1.0,0.7,0.3 + 51194 & 90874 & 10 27 & -65 & a2 & 6.0 & 0.09 & 68 & [email protected] & -22.8,-14.3,-8.8 & 0.5,0.6,0.2 + 57013 & 101615 & 11 41 & -43 & a0 & 5.5 & 0.04 & 65 & [email protected] & -20.1,-17.3,-2.9 & 0.9,2.2,0.8 + 76736 & 138965 & 15 40 & -70 & a1 & 6.4 & 0.08 & 78 & [email protected] & -19.4,-15.5,-6.3 & 3.1,2.9,0.9 + 89925 & 168913 & 18 20 & + 29 & a5 & 5.6 & 0.23 & 56 & [email protected] & -23.4,-11.6,-2.4 & 1.5,2.3,0,9 + 98103 & 188728 & 19 56 & + 11 & a1 & 5.3 & 0.01 & 67 & [email protected] & -23.2,-18.5,-3.5 & 2.6,3.2,0.7 + notes @xmath67 input data ( r.a . , decl . , distance , proper motion ) for the uvw calculations in tables 1 and 2 are from the hipparcos catalog ( van leeuwen 2007 ) . radial velocities and their errors are from gontcharov ( 2006 ) , except for hip 57013 where the velocity and its error are from vizier . uvw are defined with respect to the sun , with u positive toward the galactic center , v positive in the direction of galactic rotation , and w positive toward the north galactic pole . additional information regarding the listed stars can be found in sections 2 and 3 . hip 7345 is 49 cet . torres et al ( 2008 ) give a mean uvw for argus stars of [email protected] , [email protected] , [email protected] km s@xmath4 . among the listed stars , hip 98103 appears the most likely to be a non - member ( see table 3 ) . @lcccccccccc hip & hd & r.a . & decl . & spec . & v & b - v & dist . & ( u , v , w ) & uvw error + & & ( h / m ) & ( deg ) & type & ( mag ) & ( mag ) & ( pc ) & ( km s@xmath4 ) & ( km s@xmath4 ) & ( km s@xmath4 ) + 25280 & 35505 & 05 24 & -16 & a0 & 5.6 & 0.0 & 68 & [email protected] & -11.8,-21.2,-1.7 & 0.7,0.6,0.5 + 26395 & 37306 & 05 37 & -11 & a2 & 6.1 & 0.05 & 63 & [email protected] & -13.1,-20.2,-5.8 & 0.5,0.4,0.3 + 41081 & 71043 & 08 22 & -52 & a0 & 5.9 & 0.02 & 70 & [email protected] & -10.3,-21.6,-5.9 & 0.2,1.1,0.2 + 53771 & 95429 & 11 00 & -51 & a3 & 6.2 & 0.19 & 61 & [email protected] & -12.6,-19.0,-5.3 & 0.5,1.0,0.2 + 59394 & 105850 & 12 11 & -23 & a1 & 5.5 & 0.06 & 59 & [email protected] & -10.5,-19.6,-0.3 & 1.3,3.1,2.6 + 72408 & 130556 & 14 48 & + 21 & f0 & 7.9 & 0.41 & 75 & [email protected] & -11.3,-20.6,-0.5 & 0.7,1.0,1.4 + 104365 & 201184 & 21 08 & -21 & a0 & 5.3 & 0.00 & 55 & [email protected] & -7.7,-18.9,0.0 & 2.2,1.2,2.1 + notes @xmath67 characteristic mean uvw for tuc / hor stars are given in zuckerman & song ( 2004 ) as -11 , -21 , 0 km s@xmath4 , and in torres et al . ( 2008 ) as [email protected] , [email protected] , [email protected] km s@xmath4 . torres et al give a mean uvw for columba stars of [email protected] , [email protected] , [email protected] km s@xmath4 . there is a galaxy @xmath150 arcsec from hip 72408 ( mostly toward the south ) that provides the strong 60 & 100 @xmath9 m iras excess emission noted in vizier . hip 72408 is apparently a weak x - ray source in the rosat all - sky survey catalog . hip 30675 (= hr 2328 ) may be a columba star at a distance from earth of @xmath1125 pc ( see section 3 ) . @lcccccc hip & v & b - v & m@xmath5 & moving & age & reference + & ( mag ) & ( mag ) & ( mag ) & group & ( myr ) & for m.g . + a-49 cet & 5.62 & 0.07 & 1.75 & argus & 40 & 1 + b-2578 & 5.07 & 0.04 & 1.78 & tuc / hor & 30 & 2 + c-12413 & 4.74 & 0.06 & 1.98 & tuc / hor & 30 & 3 + d-15353 & 6.03 & 0.13 & 2.33 & ab dor & 70 & 3 + e-16449 & 6.38 & 0.12 & 2.10 & columba & 30 & 4 + f-23179 & 4.93 & 0.04 & 1.34 & columba & 30 * & 3 + g-25280 & 5.65 & 0.00 & 1.48 & tuc / hor & 30 & 1 + h-26309 & 6.26 & 0.15 & 2.65 & columba & 30 & 3 + i-26395 & 6.10 & 0.05 & 2.11 & columba & 30 & 1 + j-26966 & 5.73 & -0.01 & 1.35 & columba & 30 & 4 + k-32104 & 5.20 & 0.06 & 2.00 & columba & 30 & 3 + l-41081 & 5.89 & 0.02 & 1.66 & columba & 30 & 1 + m-50191 & 3.85 & 0.05 & 1.39 & argus & 40 * & 3 + n-51194 & 6.00 & 0.09 & 1.84 & argus & 40 & 1 + o-57013 & 5.54 & 0.04 & 1.46 & argus & 40 & 1 + p-57632 & 2.14 & 0.09 & 1.93 & argus & 40 & 3 + q-59394 & 5.45 & 0.06 & 1.60 & tuc / hor & 30 & 1 + r-76736 & 6.42 & 0.08 & 1.94 & argus & 40 & 1 + s-95261 & 5.03 & 0.02 & 1.61 & @xmath6 pic & 12 & 5 + t-98103 & 5.28 & 0.01 & 1.13 & argus & 40 * & 1 + u-98495 & 3.97 & -0.03 & 1.43 & argus & 40 & 3 + v-99770 & 4.93 & 0.15 & 1.78 & argus & 40 * & 3 + w-104365 & 5.30 & 0.00 & 1.60 & tuc / hor & 30 & 1 + x-115738 & 4.95 & 0.04 & 1.59 & ab dor & 70 & 3 + y-117452 & 4.59 & 0.00 & 1.47 & ab dor & 70 & 3 + z-118121 & 5.00 & 0.06 & 1.62 & tuc / hor & 30 & 2 + notes @xmath67 with the exception of 49 cet , the letters preceding the hip designation are those that are plotted on figure 1 . hd 21997 is star e. additional information regarding some listed stars can be found in section 3 . an asterisk in the right hand column indicates stars that appear high ( bright ) on the figure and thus might be too old to be members of the designated young moving group . some problems with assigning a - type stars to moving groups are considered in sections 2 and 3 . the right hand column gives the earliest reference where one can find a given star placed into a given moving group : ( 1 ) this paper , ( 2 ) zuckerman & webb 2000 , ( 3 ) zuckerman et al ( 2011 ) , ( 4 ) moor et al ( 2006 ) , ( 5 ) zuckerman et al ( 2001 ) . lcccc name & 3@xmath48@xmath68 & 3@xmath48@xmath69 & 5@xmath48@xmath69 & m@xmath70 + & ( mk ) & ( mk km s@xmath4 ) & ( jy km s@xmath4 ) & ( 10@xmath14 m@xmath71 ) + hd 15745 & 72 & 145 & 1.9 & @xmath25.0 + hd 21997 & 119 & 240 & 3.1 & @xmath211.0 + hd 30447 & 139 & 280 & 3.7 & @xmath214.5 + hd 32297 & 89 & 178 & 2.3 & @xmath218.5 + hd 38206 & 146 & 295 & 3.9 & @xmath211.8 + hd 85672 & 64 & 128 & 1.7 & @xmath29.2 + hd 107146 & 55 & 110 & 1.4 & @xmath20.8 + hd 131835 & 193 & 388 & 5.1 & @xmath240.6 + hd 191089 & 124 & 249 & 3.3 & @xmath25.9 + hd 221853 & 54 & 110 & 1.4 & @xmath25.0 + notes @xmath67 this is a revision of columns 8 - 11 in table 1 in kastner et al ( 2010 ) . the listed main beam brightness temperatures ( column 2 ) and the upper limits on integrated line intensity are all only upper limits ; 3 @xmath48 for the second and third columns and 5 @xmath48 for columns 4 and 5 . see appendix a for details . as noted there , the linewidth assumed for calculation of upper limits to integrated line intensity in table 4 is 5 km s@xmath4 , whereas 3 km s@xmath4 was assumed in table 1 in kastner et al ( 2010 ) . m@xmath71 is the mass of earth . @lcccccccc star & age & @xmath12 & m@xmath19 & m@xmath70 & @xmath15r / r & h / r & dm / dt & reference for + & ( myr ) & ( 10@xmath72 & ( m@xmath71 ) & ( 10@xmath14m@xmath71 ) & & & ( 10@xmath39 g s@xmath4 ) & @xmath15r & h + vega & 350 & 0.2 & 0.008 & & 0.4 & 0.2 & 0.3 & muller et al 2010 + fomalhaut & 440 & 0.8 & 0.024 & & 0.13 & & 4 & acke et al 2012 ; + & & & & & & & & boley et al 2012 + 49 cet & 40 & 8 & 0.31 & 25 & @xmath10.5 & 0.04 & 45 & hughes et al 2008 + hd 21997 & 30 & 5 & 0.22 & 3.5 & @xmath10.5 & 0.2 & 13 & moor et al 2011 + notes @xmath67 values of @xmath12 (= l@xmath13/l@xmath11 ) are from rhee et al ( 2007 ) . r is disk radius , @xmath15r is disk width in the radial direction , and h is full vertical height of the disk at radius r. artymowicz & clampin ( 1997 ) give h / r @xmath10.2 for the disk that orbits @xmath6 pictoris . the dust masses , m@xmath19 , obtained directly from submillimeter measurements ( rhee et al 2007 ) , are in units of earth masses . the co masses are in units of 10@xmath14 earth masses . dm / dt is the production / loss rate of dust grains with radii slightly larger than the blow - out radius in a quasi - steady state , collisional cascade , model ( calculated in appendix c for vega and fomalhaut and in section 4.1 for 49 cet and hd 21997 ) . the listed age for fomalhaut is from a recent paper by mamajek ( 2012 ) . color - magnitude diagram of stars listed in table 3 . gold stars are members of the pleiades cluster that have been dereddened and cleaned of binary stars ( from vigan et al 2012 ) . the dashed line is an empirical linear fit to pleiades stars with b - v between 0 and 0.3 magnitudes ( from vigan et al 2012).,width=529 ]

we report the first mid - infrared detection of highly disturbed ionized gas in the ultraluminous infrared galaxy ( ulirg ) irasf001837111 . the gas , traced by the 12.81 m [ ne ii ] and 15.56 m [ ne iii ] lines , spans a velocity range of 3500 to + 3000 km s with respect to systemic velocity . optical and near - infrared spectroscopic studies show no evidence for similarly high velocity gas components in forbidden lines at shorter wavelengths . we interpret this as the result of strong extinction ( a=1050 ) on the high - velocity gas , which identifies the base of the outflow traced in 5007 [ o iii ] as a plausible origin . unusual excitation conditions are implied by a comparison of the mid - infrared low - excitation neon line emission and the polycyclic aromatic hydrocarbon ( pah ) emission for a sample of 56 ulirgs . for irasf001837111 , the neon / pah ratio is 8 times higher than the average ratio . similar mid - infrared kinematic and excitation characteristics are found for only two other ulirgs in our sample : iras121271412ne and iras13451 + 1232 . both sources have an elevated neon / pah ratio and exhibit pronounced blue wings in their 15.56 m [ ne iii ] line profiles . iras13451 + 1232 even shows a strongly blue shifted and broad 14.32 m [ ne v ] line . while for iras13451 + 1232 the observed [ ne iii]/[ne ii ] and [ ne v]/[ne ii ] line ratios indicate exposure of the blue shifted gas to direct radiation from the agn , for irasf001837111 and iras121271412ne the neon line ratios are consistent with an origin in fast shocks in a high - density environment , and with an evolutionary scenario in which strongly blue shifted [ ne ii ] and [ ne iii ] emission trace the ( partial ) disruption of the obscuring medium around a buried agn . the detection of strongly blue shifted [ ne v ] emission in iras13451 + 1232 would then indicate this process to be much further advanced in this ulirg than in irasf001837111 and iras121271412ne , where this line is undetected .

0811.1562

large - scale galactic outflows of gas ( sometimes called `` superwinds '' ) are seen in galaxies both at low and high redshifts ( e.g. * ? ? ? * ; * ? ? ? superwinds are generated when the kinetic energy in the outflows from massive stars and supernovae is thermalized , generating a region of very hot ( t@xmath310@xmath410@xmath5k ) low - density gas in the ism of a starburst galaxy @xcite . as the bubble breaks out of the disk of the galaxy , it will rupture , producing a weakly - collimated bipolar outflow into the galaxy halo . as the bubble expands it sweeps up and shocks ambient material , creating an outflowing wind @xcite . superwinds may also be driven , at least in part , by radiation pressure on dust grains @xcite . they may also be driven directly by the interaction of jets on the dense ism surrounding an agn ( e.g. * ? ? ? there is morphological , physical and kinematic evidence for superwinds in nearby starburst and infrared luminous galaxies ( e.g. * ? ? ? large - scale optical emission - line and associated x - ray nebulae are ubiquitous in starbursts @xcite , and these increase in size and luminosity from dwarf starbursts all the way to ulirgs ( where they can be tens of kpc in size @xcite . in addition , the winds are often seen via blue - shifted absorption features ( e.g. nad ) against the starburst nuclei @xcite indicating that the neutral , cold ism is being swept - up in the outflow . at high redshifts , superwinds have been seen in lyman break galaxies at z=23 @xcite . the ultraluminous infrared galaxy ( ulirg : l(81000@xmath0m)=10@xmath6 l@xmath7 ) irasf001837111 is known from optical emission line spectroscopy to harbour such a galactic - scale outflow @xcite . in this paper , we study the mid - infrared evidence for this outflow based on the _ spitzer_-irs emission line spectroscopy . the emission line data is analyzed and presented along with new 3.6 , 5.8 , 70 and 160@xmath0 m _ spitzer _ photometric observations , obtained to better constrain the infrared spectral energy distribution of this interesting ulirg . with an integrated 81000@xmath0 m luminosity of [email protected]@xmath910@xmath10 l@xmath7 ( this work ) irasf001837111 is one of the most luminous ulirgs discovered by iras . near - infrared imaging by @xcite ( fig.[fig7 ] ) shows a disturbed morphology and what appears to be a single nucleus . evidence for the presence of an accreting central supermassive black hole is provided by the radio luminosity , which is in the range of powerful radio galaxies and a factor ten in excess of what would be expected for starburst galaxies based on the far - infrared luminosity ( q=1.14 ; * ? ? ? * ; * ? ? ? the presence of an agn was recently confirmed by @xcite , who detected a 6.7kev fek@xmath11 line ( fexxv ) with a large equivalent width , indicative of reflected light from a compton thick agn . the most common star formation tracers , the pah emission bands , were not detected in the 512@xmath0 m iso - cam - cvf spectrum @xcite . instead , the spectrum is dominated by a strong silicate absorption band centered at 9.7@xmath0 m , indicative of strong absorption of a background continuum source , most likely a deeply buried agn @xcite . one of the pah emission bands , the c h bending mode at 11.2@xmath0 m , was later identified in the spitzer - irs low - resolution spectrum and was used to constrain the contribution of mid - infrared detectable star formation to the energy budget of irasf001837111 to be @xmath1230% @xcite . the same spectrum may also offer indications for ongoing _ buried _ star formation through the detection of absorption bands of crystalline silicate features at 16 , 19 and 23@xmath0 m @xcite . crystalline silicates are known to be forged by young stars in their circumstellar disks and by evolved stars in their dense stellar winds ( * ? ? ? * and references therein ) . their presence in the general ism of irasf001837111 may indicate fierce ongoing star formation at mid - infrared optical depths significantly larger than one @xcite . , width=321 ] optical long - slit spectroscopy of irasf001837111 has revealed the presence of kinematically highly disturbed ionized gas extending from 10kpc west to @xmath325kpc east of the nucleus . this gas , traced in both the 4959 and 5008lines , is clearly blue shifted between 5 and 20kpc east and shows line widths in excess of 600 km s@xmath1 ( fwhm ) over a range of 25kpc . the double peaked profile of both [ o iii ] lines in this range is in agreement with a superwind model in which the line emission originates from the thin outer shell of shock - heated , swept - up ambient gas . beyond 25kpc east , the gas is much more quiescent , with line widths of @xmath3200 km s@xmath1 centered close to systemic velocity @xcite . recently , @xcite have pointed out that the presence of an outflow may also explain the clear eastward offset by about 22kpc of the soft ( 0.31.0kev ) x - ray radiation detected from irasf001837111 . it also may identify the secondary peak at 25kpc east in the r - band image line @xcite ] by @xcite to be related to this outflow @xcite . a corresponding k - band peak is absent in the 2.2@xmath0 m image ( rest frame 1.65@xmath0 m ) by @xcite ( fig.[fig7 ] ) . in this paper we report the first identification in a ulirg of broad , asymmetric and somewhat blue shifted emission lines of [ ne ii ] at 12.81@xmath0 m and [ ne iii ] at 15.56@xmath0 m in the high - resolution _ spitzer_-irs spectrum of irasf001837111 . we compare our detection to available optical and near - infrared spectroscopy to investigate the processes responsible for this broadening . our paper is organized as follows . in section2 we present our observations of irasf001837111 . in section3 we present the 1120@xmath0 m sed and an analysis of the mid - infrared emission line spectrum . in section4 we analyze line profiles of two other ulirgs which display strong blue shifts in their fine - structure neon line profiles . results are discussed in section5 and conclusions are provided in section6 . additional results for iras f001837111 are presented in the appendix . throughout this paper we assume h@xmath13 = 71 km s@xmath1 mpc@xmath1 , @xmath14 = 0.27 , @xmath15 = 0.73 and @xmath16 = 0 . llc irac-1 & 2.7 & 2.9 @xmath8 0.3 + irac-3 & 4.4 & 22 @xmath8 2 + iras-12 & 9.0 & @xmath1760 + iras-25 & 18.8 & 133 @xmath8 10 + iras-60 & 45 & 1200 @xmath8 837 + mips-70 & 53 & 1500 @xmath8 225 + iras-100 & 75 & 1190 @xmath8 119 + mips-160 & 121 & 540 @xmath8 80 + _ spitzer _ space telescope @xcite irac @xcite 3.6 and 5.8@xmath0 m photometry was obtained on 2006 august 14 ( aor key 17517824 ) . irasf001837111 was imaged at 5 dither positions using 2sec ramp times . the source flux density was measured by integrating over a 12 radius aperture placed upon the ssc pipeline - produced ( s14 ) post - bcd mosaic . we estimate the uncertainty in the photometry due to the background to be @xmath32% . the absolute calibration uncertainties of the irac post - bcd images are 10% . as part of the same program we obtained mips @xcite 70 and 160@xmath0 m photometry on 2006 june 17 ( aor key 17528320 ) . irasf001837111 was imaged for 1 cycle at 70@xmath0 m and for 4 cycles at 160@xmath0 m using 3sec ramps . the source flux densities were measured from the ssc pipeline - produced ( s14 ) unfiltered post - bcd images by integrating the flux within 8 and 20 " radius apertures at 70 and 160@xmath0 m , respectively . aperture corrections were applied to recover the full point - source flux densities . we estimate the uncertainty on the 70 and 160@xmath0 m flux densities due to the background to be around 1 and 2% , respectively . the absolute calibration uncertainties of the mips post - bcd images are 1020% . we present the results of our photometric observations in table[tab2 ] . the calculation of infrared luminosities which utilize this photometry are described in section3.1 . spitzer_-irs low - resolution spectrum @xcite is shown as a continuous line . photometric points are labeled with the following symbols : _ circles _ : eso - ntt j & k @xcite , _ squares _ : irac 3.6 & 5.8@xmath0 m , _ triangles _ : iras 60 & 100@xmath0 m , _ diamonds _ : mips 70 & 160@xmath0 m . error bars on the photometric points are too small to be visible . inset : detailed view of the spectral range around the 9.7@xmath0 m silicate absorption profile . note the emission lines of h@xmath18 s(3 ) at 9.66@xmath0 m and [ ne ii ] at 12.81@xmath0 m . [ fig4],width=321 ] high resolution ( @xmath19@xmath20500 km s@xmath1 ; r@xmath3600 ) 1037@xmath0 m irs @xcite spectra of irasf001837111 were obtained at two epochs : 2005 june 2 ( aor key 6651904 ) and 2008 june 1 ( aor keys 25289472 and 25289984 ) . in both cases the source was acquired using a high - accuracy blue peak - up and observed in staring mode . the 1019.5@xmath0 m portion of the spectrum was obtained in 9 cycles of 120sec integrations in irs short high ( sh ; slit size 4.7@xmath2111@xmath22 ) at pa=91@xmath23 . the 19.337.0@xmath0 m portion of the spectrum was obtained in 2 sets of 35 cycles of 60sec integrations , alternated with sky background observations ( 30 cycles of 60sec ) , in irs long high ( lh ; slit size [email protected]@xmath22 ) at pa=-15@xmath23 . figure[fig7 ] shows the sh slit overplotted on the k - band image of the source @xcite . the lh slit is not shown as it is more than twice as wide and long as the sh slit . data reduction started from 2d droop level ( i.e. pre - flatfield ) images provided by the s17.2 pipeline of the _ spitzer _ science center ( ssc ) . for sh , in a first step we used irsclean to interpolate over the bad and rogue pixels . for lh , this step was preceeded by subtraction of the contemporaneous sky background images . we then used the smart data reduction package @xcite to extract the 1d target spectrum in ` full slit ' mode . the same procedure was followed for a large number of observations of the calibration star @xmath24dra . the final spectrum was obtained by multiplying the spectrum of irasf001837111 by the relative spectral response function created from the ratio of the observed spectrum of @xmath24dra to the stellar reference spectrum ( * ? ? ? sloan priv.comm . ) . note that given the absence of contemporaneous sh background observations , the sh portion of the spectrum is _ not _ background - subtracted . the wavelength calibration of our spectrum is accurate to about 1/5 ( ssc , priv . comm . ) of a resolution element ( 500 km s@xmath1 ) , which amounts to 100 km s@xmath1 . llcc irac-1 & 3.6 & 8.6 & 1% + irac-2 & 4.5 & 29.6 & 1% + irac-3 & 5.8 & 47.3 & 1% + irac-4 & 8.0 & 48.6 & 1% + mips-24 & 24 & 370 & 1% + iras-25 & 25 & 360 & 1% + iras-60 & 60 & 1410 & 1% + mips-70 & 70 & 1190 & 2% + iras-100 & 100 & 890 & 3% + mips-160 & 160 & 490 & 5% + scuba-450 & 450 & 20 & 6% + scuba-850 & 850 & 2.0 & 6% + with the addition of the irac 3.6@xmath0 m and mips 70 and 160@xmath0 m photometry it is possible to fit the infrared sed of irasf001837111 over the entire rest frame wavelength range of 1120@xmath0 m . especially important is the addition of the mips 160@xmath0 m point , which extends the sampling of the rest frame sed from 75@xmath0 m ( pre-_spitzer _ ) to 120@xmath0 m ( see fig.[fig4 ] ) , allowing a more reliable extrapolation of the sed into the submillimetre range . for the latter , we have used the multi - component spectral energy distribution decomposition code of @xcite . based on our fit , we have computed the following integrated rest frame infrared luminosities for irasf00183 - 7111 : l(11000@xmath0m)[email protected]@xmath910@xmath10 l@xmath7 , l(81000@xmath0m)[email protected]@xmath910@xmath10 l@xmath7 , l(40500@xmath0m)[email protected]@xmath910@xmath10 l@xmath7 and l(40120@xmath0m)[email protected]@xmath910@xmath10 l@xmath7 . we have further used the results of the fit to derive rest frame filter flux densities for the irac , mips , iras and scuba bands . these values are tabulated in table[tab4 ] . given the 15% absolute calibration uncertainty of the mips 70 and 160@xmath0 m data , and the fact that the mips 160@xmath0 m data is the longest wavelength point constraining our fit , there is clearly some uncertainty in our extrapolation of the flux density out to 1000@xmath0 m . the spectral shape of the extrapolated flux density beyond 160@xmath0 m is a function of the chosen power - law spectral index of the far - ir dust opacity . the above stated uncertainties in the integrated luminosities incorporate both the systematic uncertainty resulting from our choice of this dust opacity spectral index as well as the statistical uncertainty resulting from the mips calibration uncertainty . the ratio of the observed iras-25 to iras-60 filter fluxes is commonly used to classify ulirgs as `` warm '' ( 25/60@xmath0 m @xmath25 0.2 ) or `` cold '' ( 25/60@xmath0 m @xmath17 0.2 ) , separating ulirgs believed to be agn - dominated from those which are thought to be powered predominantly by star formation @xcite . for the classification to be meaningful also for higher redshift targets like irasf001837111 , rest frame synthetic iras fluxes should be used . for irasf001837111 , this results in an iras-25 to iras-60 flux ratio of 0.26 , which places it among the class of `` warm '' , agn - dominated ulirgs , in agreement with results obtained at other wavelengths ( see section 1 ) . the high - resolution _ spitzer_-irs spectrum of irasf001837111 covers the rest frame wavelength range of 7.527.8@xmath0 m . in table[tab1 ] we list the line fluxes and 3@xmath26 upper limits for 17 emission lines , as well as for the 11.2@xmath0 m pah emission feature . for the latter , the feature flux was measured above a spline continuum defined shortward of 11.1@xmath0 m and longward of 11.6@xmath0 m . the only lines that are detected in our spectrum are the s(1 ) , s(2 ) and s(3 ) pure rotational lines of molecular hydrogen , and the 12.81@xmath0 m [ ne ii ] , 15.56@xmath0 m [ ne iii ] and 18.71@xmath0 m [ s iii ] fine - structure lines . in fig.[fig1 ] we present their continuum - subtracted line profiles as a function of velocity relative to systemic200 km s@xmath1 ( z=0.328 @xmath8 0.001 ) . this redshift is 360 km s@xmath1 higher than the value adopted by @xcite and 180 km s@xmath1 higher than the value used by @xcite ] . llccc @xmath27ne @xmath28 $ ] & 7.65 & @xmath170.25 & ... & ... + h@xmath18 s(4 ) & 8.03 & @xmath170.26 & ... & ... + @xmath27ar @xmath28 $ ] & 8.99 & @xmath170.07 & ... & ... + h@xmath18 s(3 ) & 9.66 & 0.52 @xmath8 0.03 & 650@xmath840 & 410@xmath850 + @xmath27s @xmath28 $ ] & 10.51 & @xmath170.08 & ... & ... + pah & 11.2 & 2.7 @xmath8 0.4 & ... & ... + h@xmath18 s(2 ) & 12.28 & 0.59 @xmath8 0.1 & 950@xmath8130 & 810@xmath8120 + hu@xmath11 & 12.37 & @xmath170.12 & ... & ... + @xmath27ne @xmath28 $ ] & 12.81 & 6.3 @xmath8 0.3 & 1500@xmath8100 & 1410@xmath8100 + @xmath27ne @xmath28 $ ] & 14.32 & @xmath170.13 & ... & ... + @xmath27cl @xmath28 $ ] & 14.37 & @xmath170.14 & ... & ... + @xmath27ne @xmath28 $ ] & 15.56 & 2.1 @xmath8 0.2 & 1300@xmath8120 & 1200@xmath8120 + h@xmath18 s(1 ) & 17.04 & 0.83 @xmath8 0.05 & 660@xmath850 & 430@xmath855 + @xmath27fe ii@xmath28 $ ] & 17.94 & @xmath170.10 & ... & ... + @xmath27s @xmath28 $ ] & 18.71 & 0.23 @xmath8 0.03 & 490@xmath850 & @xmath17200 + @xmath27ne @xmath28 $ ] & 24.30 & @xmath170.20 & ... & ... + @xmath27o iv@xmath28 $ ] & 25.89 & @xmath170.35 & ... & ... + @xmath27fe ii@xmath28 $ ] & 25.99 & @xmath170.31 & ... & ... + @xmath27o @xmath28 $ ] & 0.5007 & 0.45 & ... & 890 + @xmath27o @xmath28 $ ] & 0.5007 & & ... & 6001000 + h@xmath11 & 0.6565 & [email protected] & ... & 720 + @xmath27fe @xmath28 $ ] & 1.257 & 0.228 : & ... & 900 : + pa@xmath29 & 1.282 & 0.115 : & ... & 600 : + @xmath27fe @xmath28 $ ] & 1.644 & 0.376 : & ... & 1000 : of the six detected emission lines , the two neon fine - structure lines appear clearly resolved beyond the instrumental resolution ( 500 km s@xmath1 ; @xmath30=600 ) , most notably the [ neii ] line , at an observed fwhm of 1500@xmath8100 km s@xmath1 ( table[tab1 ] ) . the other emission lines have observed fwhm ranging from 490@xmath850 to 950@xmath8130 km s@xmath1 . in what follows any fwhm values quoted in the text refer to the true fwhm , i.e. the observed fwhm corrected for the instrumental profile . m [ ne ii ] line for four galactic nuclei , at a spectral resolution @xmath30=600 . the spectral features included are the h@xmath18 s(2 ) line at 12.28@xmath0 m , the 12.7@xmath0 m pah emission feature ( extending from -13,500 to + 3,500 km / s ) and the [ ne ii ] line . the latter is clearly broadened in the spectra of irasf001837111 and ngc6240 . note that the 12.7@xmath0 m pah emission feature appears to be absent in the spectra of irasf001837111 and iras08572 + 3915nw [ fig2],width=321 ] m [ ne ii ] ( _ red _ ) and 5007 [ o iii ] ( _ green _ ) at a position 3@xmath22 east of the nucleus of irasf001837111 , where the o@xmath31 gas reaches its highest blue shift . the spectral resolution of the double - peaked [ o iii ] profile ( @xmath30=1900 ) has been degraded to the resolution of the _ spitzer_-irs - sh measurement ( @xmath30=600 ) . overplotted are the 2-component gauss fits to the [ ne ii ] and [ ne iii ] line profiles . the individual components are shown as _ dashed _ lines , their sums as _ dotted _ lines [ fig3],width=321 ] extracting a [ ne ii ] line profile requires careful removal of the underlying spectral structure . as illustrated in fig.[fig2 ] by the starburst galaxy ngc2146 and the ulirg ngc6240 , the [ neii ] line is superimposed on the 12.7@xmath0 m pah emission feature , which arises over a velocity range of -13500 to + 3500kms@xmath1 relative to the [ ne ii ] line . however , unlike ngc2146 and ngc6240 , the spectrum of irasf001837111 shows hardly any pah emission , with the exception of the pah feature at 11.2@xmath0 m , which is detected at very low equivalent width . we conclude that the remarkable width of the 12.81@xmath0 m [ ne ii ] line can not be ascribed to inproper subtraction of the underlying spectral structure . in fact , the resulting profile is consistent with that of the 15.56@xmath0 m [ ne iii ] line . the shape of the [ ne ii ] line profile is asymmetric , with the peak emission arising very close to systemic velocity . to characterize the asymmetry we have decomposed the line profile into two gaussian components ( fig.[fig1 ] ) . the resulting fit is good . we find the dominant component to have a fwhm of 2570@xmath890kms@xmath1 , centered at a blueshift of 400@xmath850 km s@xmath1 w.r.t . the narrow component has a fwhm of 610@xmath860kms@xmath1 and a redshift of 60@xmath820kms@xmath1 w.r.t . the fit components are shown overplotted in red onto the observed [ ne ii ] profile in the top - right panel of fig.[fig1 ] . further fit details are provided in table[tab3 ] . unlike the 12.81@xmath0 m [ ne ii ] line , the 15.56@xmath0 m [ ne iii ] line resides in a relatively uncomplicated part of the spectrum . we determined its profile by defining and subtracting a local third - order polynomial continuum . we then performed a 2-component decomposition in the same way as for the [ ne ii ] line . the narrow component has a fwhm of 460@xmath890 km s@xmath1 and is centered at the systemic velocity , whereas the broad component has a fwhm of 2850@xmath8240 km s@xmath1 and is blue shifted by 450@xmath8100 km s@xmath1 . the fraction of the [ ne iii ] flux emitted in the broad component is 3/4 , slightly lower than the fraction found for the [ ne ii ] line , 4/5 . as a result , the [ ne iii]/[ne ii ] ratios are [email protected] for the broad component and [email protected] for the narrow component . the most detailed spectral line study that we are aware of focusses on the [ o iii ] doublet at 4959 and 5007and was published by @xcite . they observed irasf001837111 in a 2@xmath21100@xmath22 slit along pa 90@xmath23 ( see fig.[fig7 ] ) at a spectral resolution @xmath30=1900 . in fig.[fig3 ] we compare the 5007 [ o iii ] profile obtained at a position 3@xmath22 ( 13kpc ) east of the nucleus to that of our galaxy - integrated spectrum , after degrading the spectral resolution of the [ o iii ] line from 1900 to 600 . according to fig.12 of @xcite , at this position the centroid of the [ o iii ] line profile has its highest blueshift ( 700 km s@xmath1 ; using our preferred redshift of [email protected] ) at a fwhm @xmath3600 km s@xmath1 , typical for the entire region extending from 10kpc west to 20kpc east of the nucleus . clearly , the [ ne ii ] and [ ne iii ] emission extends to much higher velocities than the [ o iii ] line emission does . more recently , @xcite published line fluxes and fwhms for 11 optical lines observed at @xmath30=1000 in a 2@xmath22@xmath99@xmath22 slit oriented along pa=329@xmath23 ( c.l . buchanan , private communication ) . for the h@xmath11 line , which is strongly blended with the @xmath27 n ii@xmath28 $ ] line on either side , they infer a fwhm of 720 km s@xmath1 ( compared to 700 km s@xmath1 reported by * ? ? ? * ) . for the 4959/5007 [ o iii ] doublet , they report fwhms of 800900 km s@xmath1 ( table[tab2 ] ) , which are comparable to those given by @xcite in their figure 12 , even though the slit orientations in the two studies in question differ by 239 degrees ( fig.[fig7 ] ) . we also compared our mid - infrared results to near - infrared h and k band spectroscopy by @xcite obtained using a 1@xmath21290@xmath22 slit oriented at pa=90@xmath23 ( fig.[fig7 ] ) . their data have similar spectral resolution as ours . the two detected [ fe ii ] lines at 1.257 and 1.644@xmath0 m have fwhms in the range 9001000 km s@xmath1 , while the 1.282@xmath0 m pa@xmath29 line appears to be unresolved . again , like in the optical range , none of the lines have fwhms as large as the mid - infrared neon lines ( see table[tab1 ] ) . the rest wavelength range covered by the _ spitzer _ irs high - resolution modules includes three high - ionization lines commonly used as tracers of agn activity : 7.65@xmath0 m [ ne vi ] and 14.32&24.32@xmath0 m [ ne v ] ( e.g. * ? ? ? none of them are detected ( table[tab1 ] ) . the 3@xmath26 limit on the fraction of 81000@xmath0 m infrared power contained in the 14.32@xmath0 m [ ne v ] line is l@xmath32}$]/l@xmath33 @xmath17 1.3@xmath910@xmath34 , which is 12 orders of magnitude smaller than the detections for classical agns , but similar to the limits derived for other ulirgs with deep 9.7@xmath0 m silicate features like iras08572 + 3915 , iras15250 + 3609 and iras201004156 , and for the type-1 ulirg / lobal - quasar mrk231 @xcite . for the limit on 7.65@xmath0 m [ ne vi ] no ulirg comparison data exists . however , the 3@xmath26 upper limit l@xmath35}$]/l@xmath33 @xmath17 2.6@xmath910@xmath34 is at least an order of magnitude below what is found for classical agns @xcite . the quality of our _ spitzer _ irs high - resolution spectrum allows us to measure the flux in the 11.2@xmath0 m pah feature more accurately than has been possible from the low - resolution spectrum @xcite . the latter showed strong disagreements between the spectra obtained in nod positions 1 and 2 at the short wavelength end of irs long - low order 2 . the new feature flux , measured from the irs sh spectrum is tabulated in table[tab1 ] and is 25% smaller than previously reported by @xcite based on the low - resolution spectrum . we also remeasured the upper limit on the 6.2@xmath0 m pah feature from a re - extracted low - resolution spectrum , using the same continuum pivots at 5.65 , 6.07 and 6.50@xmath0 m to define the local spline continuum as in @xcite . given the clear reduction in noise and artefacts between irs pipeline versions 9.1 and 14.2 , the new 3@xmath26 upper limit on the 6.2@xmath0 m pah flux is 1.9@xmath910@xmath36 w @xmath37 , which is a factor of 2 lower than previously reported by @xcite . note that we do not attempt to measure the flux in the 7.7@xmath0 m pah band , as this emission feature if present would be strongly blended with the silicate feature . based on the new pah fluxes we compute an _ limit on the ratio of 6.2-to-11.2@xmath0 m pah of 0.70 . this is at the _ low _ end of the range found for starburst nuclei and ulirgs in our sample ( 2@xmath26 below the average ratio of 1.4 ) , but not untypical for a sample of ir - faint liners ( l@xmath33=10@xmath38 ) studied by @xcite , and for the spiral arms of m51 and the halo of m82 @xcite . the low 6.2-to-11.2@xmath0 m pah ratio may indicate a high ratio of neutral to ionized pahs , which would imply a low ratio of the ionizing radiation field to the electron density , and physical conditions similar to those in the galactic reflection nebula ngc2023 @xcite . in irasf001837111 these conditions could plausibly exist outside the deeply buried nucleus . alternatively , the low observed 6.2-to-11.2@xmath0 m pah ratio for irasf001837111 may indicate destruction of small pahs exposed to hard radiation from an agn . this would selectively lower the fluxes emitted in the pah emission bands at 3.3 , 6.2 and 7.7@xmath0 m relative to those at 11.2@xmath0 m and beyond @xcite . given , however , the absence of detections of high ionization lines , indicative of a hard agn radiation field , this scenario may not hold for irasf001837111 . note that strong extinction on the pah emitting region can be ruled out as a third explanation for the low observed 6.2-to-11.2@xmath0 m pah ratio , as this would have the opposite effect on the pah ratio : the attenuation on the 11.2@xmath0 m pah feature is stronger than the attenuation on the 6.2@xmath0 m pah feature ( e.g. * ? ? ? m [ ne ii ] ( _ red _ ) and 15.56@xmath0 m [ ne iii ] ( _ blue _ ) for iras12127 - 1412ne at a spectral resolution of @xmath30=600 . overplotted are 2-component gauss fits to the line profiles . the individual components are shown as _ dashed _ lines , their sums as _ dotted _ lines [ fig9],width=321 ] the discovery of high velocity [ ne ii ] emission in irasf001837111 raises the question whether there are other luminous infrared galaxies that show evidence for disturbed ( fwhm@xmath39800 km s@xmath1 ) neon gas in their _ spitzer_-irs spectra . if so , do these galaxies have properties similar to irasf001837111 ? to address this question we have inspected the line profiles of the low - ionization emission lines 12.81@xmath0 m [ ne ii ] and 15.56@xmath0 m [ ne iii ] of the 53 ulirgs in the sample of @xcite , as well as the archival spectra of the ulirg iras121271412ne and the hylirgs irasf09104 + 4109 and irasf15307 + 3252 . in our search we distinguish two classes . one consisting of sources with symmetric resolved neon line profiles , and another with sources which show asymmetric resolved line profiles . m [ ne ii ] ( _ red _ ) , 15.56@xmath0 m [ ne iii ] ( _ blue _ ) and 14.32@xmath0 m [ ne v ] ( _ black _ ) for iras13451 + 1232 at a spectral resolution of @xmath30=600 . overplotted are 2-component gauss fits to the neon line profiles . the individual components are shown as _ dashed _ lines , their sums as _ dotted _ lines . the [ ne v ] line has been scaled to match the blue wing of the [ ne iii ] line profile . also shown is the sum of the 3-component gauss fit to the 5007 [ o iii ] line ( _ green _ ) as presented by @xcite , degraded to a resolution of @xmath30=600 , and the pa@xmath11 profile ( _ purple _ ) obtained by @xcite at @xmath30=650 [ fig10],width=321 ] lclcccc + @xmath27ne @xmath28 $ ] & n & 12.816 & + 60 @xmath8 20 & 790 @xmath8 60 & 610 @xmath8 60&1.4 @xmath8 0.16 + @xmath27ne @xmath28 $ ] & b & 12.796 & -400 @xmath8 50 & 2620 @xmath8 90 & 2570 @xmath8 90&4.9 @xmath8 0.20 + @xmath27ne @xmath28 $ ] & n & 15.555 & 0 @xmath8 30 & 680 @xmath8 90 & 460 @xmath8 80&0.55@xmath8 0.09 + @xmath27ne @xmath28 $ ] & b & 15.532 & -450 @xmath8 100 & 2890 @xmath8 240 & 2850 @xmath8 240&1.5 @xmath8 0.12 + @xmath27ne @xmath28 $ ] & - & 14.322 & & & & @xmath170.13 + + @xmath27ne @xmath28 $ ] & n & 12.818 & -80 @xmath8 10 & 530 @xmath8 20 & 175 @xmath8 45 & 1.68 @xmath8 0.08 + @xmath27ne @xmath28 $ ] & b & 12.799 & -530 @xmath8 60 & 1740 @xmath8 100&1670 @xmath8 100 & 1.47 @xmath8 0.10 + @xmath27ne @xmath28 $ ] & n & 15.565 & -7 @xmath8 40 & 430 @xmath8 120 & @xmath17200 & 0.16 @xmath8 0.07 + @xmath27ne @xmath28 $ ] & b & 15.531 & -650 @xmath8 200 & 1580 @xmath8 360&1500 @xmath8 340 & 0.42 @xmath8 0.10 + @xmath27ne @xmath28 $ ] & - & 14.322 & & & & @xmath170.08 + + @xmath27ne @xmath28 $ ] & n & 12.808 & -120 @xmath8 10 & 790 @xmath8 40 & 610 @xmath8 50&4.3 @xmath8 0.30 + @xmath27ne @xmath28 $ ] & b & 12.758 & -1290 @xmath8 215 & 2520 @xmath8 350&2470 @xmath8 340&2.5 @xmath8 0.40 + @xmath27ne @xmath28 $ ] & n & 15.545 & -200 @xmath8 25 & 840 @xmath8 70 & 670 @xmath8 70&1.9 @xmath8 0.25 + @xmath27ne @xmath28 $ ] & b & 15.506 & -950 @xmath8 85 & 2410 @xmath8 110&2360 @xmath8 110&4.1 @xmath8 0.30 + @xmath27ne @xmath28 $ ] & b & 14.263 & -1240 @xmath8 80 & 2360 @xmath8 190&2300 @xmath8 190&1.9 @xmath8 0.15 + the class of asymmetric resolved [ ne ii ] and [ ne iii ] line profiles consists of only two sources besides irasf001837111 . in all three cases the line asymmetry takes the form of a blue wing . one source is iras121271412ne ( fig.[fig9 ] ) , the brighter of the two nuclei in this widely separated liner system ( projected nuclear separation 9@xmath22 , or 21.9kpc ; * ? ? ? in the near - infrared , @xcite classified the ne nucleus as a buried agn based on the absence of 3.3@xmath0 m pah emission and the presence of 3.0@xmath0 m water ice and 3.4@xmath0 m hydrocarbon absorption bands . the deeply buried nature of the nuclear power source(s ) is confirmed by the presence of a deep 10@xmath0 m silicate feature in the _ spitzer_-irs low - resolution spectrum @xcite . on 2008 july 10 , we reobserved the source in staring mode in irs - sh , using an integration time of 12 cycles of 120sec . sky - subtraction was performed using a contemporaneous sky background observation of 6 cycles of 120sec . data reduction proceeded along the same steps as detailed in sect.2.1 . like for irasf001837111 , the asymmetry of the neon emission lines of iras121271412ne may be characterized using a 2-component gaussian fit . for the [ ne ii ] line the narrow component is practically unresolved ( fwhm = 175@xmath845 km s@xmath1 ) and is centered close to systemic velocity s(1 ) and s(3 ) lines and find 39870@xmath8200 km s@xmath1 ( z=0.1330@xmath40 ) , in good agreement with the redshift of 0.133 quoted by @xcite ] ( -80@xmath810 km s@xmath1 ) . the other component has a fwhm of 1670@xmath8100 km s@xmath1 and is centered at a blue shift of 530@xmath860 km s@xmath1 ( fig.[fig9 ] ) . the fraction of the [ ne ii ] line flux emitted in the broad component is slightly smaller than that in the narrow component . this is less than for irasf001837111 , for which about 4/5 of the line flux is associated with the broad component . we also examined the 15.56@xmath0 m [ ne iii ] profile . its shape is not as well constrained as that of the [ ne ii ] line due to a five times lower line flux , but the profile clearly shows the presence of a blue wing . a 2-component gauss fit results in an unresolved narrow component centered at systemic velocity ( -7@xmath840 km s@xmath1 ) and a broad component with fwhm of 1500@xmath8340 km s@xmath1 blueshifted by 650@xmath8200 km s@xmath1 . we find that the contribution of the broad component to the [ ne iii ] line profile is far stronger than to the [ ne ii ] profile , 3/4 as compared to 1/2 ( fig.[fig9 ] ) . this implies a difference in excitation of the narrow and broad components . for the narrow component we find a [ ne iii]/[ne ii ] ratio of [email protected] , while for the broad component we compute a ratio of [email protected] unfortunately , a comparison to the line profile of the optical [ o iii ] doublet is not possible as the equivalent widths of these lines in the spectrum of @xcite are too low for a meaningful analysis of their profiles . the third ulirg with asymmetric neon line profiles is iras13451 + 12320.00001 as the systemic velocity for iras13451 + 1232 . this velocity corresponds to the velocity center of the narrow component in the 3-component fit to the 4959 & 5007 [ o iii ] doublet in the @xmath30=1200 spectrum of @xcite . using this redshift , the line centers of the h@xmath18 s(1 ) and s(3 ) lines occur at 40@xmath41 and 20@xmath835 km s@xmath1 , respectively . note that our adopted redshift is 200 km s@xmath1 lower than the value derived from co observations @xcite ] , also known as pks1345 + 12 ( fig.[fig10 ] ) . the source , which is optically classified as seyfert-2 @xcite , comprises two nuclei separated by 1.8@xmath22 ( 4.3kpc ; unresolved at the spatial resolution of _ spitzer_-irs ) . both the low resolution near - infrared 2.63.65@xmath0 m @xcite and the _ spitzer _ mid - infrared spectrum is virtually featureless , characteristic of agn heated hot dust . at the higher resolving power of irs - sh , emission line profiles of 12.81@xmath0 m [ ne ii ] , 15.56@xmath0 m [ ne iii ] and 14.32@xmath0 m [ ne v ] can be discerned on the continuum , the clearest of which is the is the [ ne iii ] line . we fit its profile with a broad component of fwhm 2360@xmath8110 km s@xmath1 blue shifted by 950@xmath885 km s@xmath1 and a narrow component of fwhm 670@xmath870 km s@xmath1 blueshifted by 200@xmath825 km s@xmath1 . the broad component is responsible for 2/3 of the [ ne iii ] emission , significantly more than for irasf001837111 but somewhat less than for iras121271412ne . for the [ ne ii ] line profile the 2-component decomposition results in a narrow component with a fwhm of 610@xmath850 km s@xmath1 blueshifted by 120@xmath810 km s@xmath1 and a broad component with fwhm 2470@xmath8340 km s@xmath1 blueshifted by 1290@xmath8220 km s@xmath1 . the broad component carries 1/3 of the total [ ne ii ] line flux , significantly less than for irasf001837111 and iras121271412ne . the third line , the [ ne v ] coronal line at 14.32@xmath0 m , _ only _ has a blue shifted component , centered at a -1240@xmath880 km s@xmath1 , which we fit with a gaussian profile of fwhm 2300@xmath8190 km s@xmath1 . the presence of [ ne v ] emission only at blueshifted wavelengths is consistent with the difference in excitation between the narrow and broad components as measured from the [ ne iii]/[ne ii ] ratio . for the narrow component this ratio is [email protected] , while for the broad component it is [email protected] . the [ ne v]/[ne ii ] ratio for the broad component is [email protected] . in fig.[fig10 ] we also show the sum of a 3-component fit to the profile of the 5007 [ o iii ] line , centered on the nucleus @xcite . the spectral resolution was degraded from @xmath30=1200 to 600 for a proper comparison to our irs - sh data . as a result of the degradation , the narrowest ( fwhm=340@xmath823 km s@xmath1 ) and weakest of the three kinematic components , centered at systemic velocity , gets blended with the intermediate width component ( fwhm=1255@xmath812 km s@xmath1 ) centered at -402@xmath89 km s@xmath1 . the broadest , most blueshifted component ( fwhm=1944@xmath865 km s@xmath1 ) , centered at -1980@xmath836 km s@xmath1 , can however still be recognized as a separate component . we also overplot the line profile of pa@xmath11 @xcite with a narrow component of fwhm=773 km s@xmath1 and a broad component of fwhm=2588 km s@xmath1 . the overlay suggests that the 15.56@xmath0 m [ ne iii ] line traces more or less the same blue shifted kinematic components as the 5007 [ o iii ] and pa@xmath11 lines . the comparison further shows that , unlike irasf001837111 , there is no evidence for the 5007 [ o iii ] line to cover a more limited velocity range than the 12.81@xmath0 m [ ne ii ] and 15.56@xmath0 m [ ne iii ] lines ( figs.[fig3]&[fig10 ] ) . the latter is in agreement with the differences in the distribution of the obscuring medium as probed by the 9.7@xmath0 m silicate absorption features ( see section 5.4 ) . observations at various wavelengths suggest that irasf001837111 derives a substantial fraction of its bolometric power from accretion onto a central black hole . especially important in this respect is the detection by @xcite of a 6.7kev fek@xmath11 line ( fe xxv ) with a large equivalent width , indicating the presence of a compton - thick agn in irasf001837111 . another clear indication is a higher than expected radio power based on the radio - far - infrared correlation for starburst galaxies @xcite , as well as unpublished radio interferometric data showing the nucleus to be unresolved on the 200pc scale ( r.p . norris , 2004 , private communication ) . the non - detection of several common mid - infrared agn tracers in our _ spitzer_-irs high resolution spectrum may hence indicate that the agn coronal line region ( clr ) is small or absent or , alternatively , that our line of sight to the agn clr is strongly obscured . the latter would be in agreement with the detection of a deep 9.7@xmath0 m silicate absorption feature ( * ? ? ? * fig[fig4 ] ) , which is unusual for classical seyfert galaxies and quasars @xcite . even stronger obscuration , along all lines of sight , is suggested by @xcite , who concluded from a study of the 4.7@xmath0 m co gas absorption band that the strong detection of this feature in irasf001837111 @xcite and its non - detection in local agns observed with iso indicates that the central power source in irasf001837111 must be fully covered rather than blocked by an edge - on torus . to gain insight into the processes responsible for the presence of high velocity gas in irasf001837111 we compare irasf001837111 to the only other two ulirgs in our sample that show pronounced blue wings in their 12.81@xmath0 m [ ne ii ] and 15.56@xmath0 m [ ne iii ] line profiles : iras12127 - 1412ne ( fig.[fig9 ] ) and iras13451 + 1232 ( fig.[fig10 ] ) . clear evidence for unusual excitation conditions in irasf001837111 emerge if we compare the pah emission and low - ionization neon line emission for agns , ulirgs and starburst galaxies from the various _ spitzer_-irs gto programs , ( u)lirgs from _ spitzer _ program 1096 and iras121271412ne . the pah luminosity should track the low - ionization neon line luminosity to within a factor two , if predominantly excited in star forming regions . this was indeed found to be the case for the sample of ulirgs published by @xcite . in fig.[fig5 ] we plot the ratio of these two quantities ( the combined [ ne ii]+[ne iii ] luminosity and the combined luminosity of the 6.2 and 11.2@xmath0 m pah features ) as a function of increasing apparent depth of the 9.7@xmath0 m silicate feature ( i.e. the negative of the silicate strength , s@xmath42 ; * ? ? ? the symbol size scales linearly with the [ ne v]/[ne ii ] ratio above a threshold value of 0.4 . this distinguishes ulirgs with a strong agn contribution to the [ ne ii]+[ne iii ] luminosity ( e.g. mrk463 ; * ? ? ? * ) from those without . as expected , for the sources lacking a strong agn signature the average value of neon / pah has a reasonably low scatter of a factor 1.4 , in agreement with the scaling relation between neon and pah flux for ulirgs found by @xcite . the notable exception is irasf001837111 , which appears to be at least 8 times overluminous in the neon lines compared to its combined pah luminosity . another exception in the s@[email protected] regime is iras121271412ne , which is overluminous in neon / pah by a factor of 3 . at s@[email protected] , the situation is less straightforward . in this regime there are several sources with large [ ne v]/[ne ii ] ratios for which the agn contributes significantly to the [ ne ii]+[ne iii ] luminosity , thereby raising the neon / pah ratio ( fig.[fig5 ] ) . the only ulirg in this regime which has an elevated neon / pah ratio despite a low [ ne v]/[ne ii ] ratio of 0.28 is iras13451 + 1232 , at @xmath256 times the average neon / pah ratio ( fig.[fig5 ] ) . this brings the number of ulirgs with unusual excitation conditions to three . since these same three ulirgs are also the only ulirgs in our sample that show pronounced blue wings in their 12.81@xmath0 m [ ne ii ] and 15.56@xmath0 m [ ne iii ] line profiles , it seems plausible that the high neon / pah ratios for these sources could be the result of additional neon line emission contributed by these blue wings rather than unusually weak pah emission in these three sources . m [ ne iii]/12.81@xmath0 m [ ne ii ] versus 14.32@xmath0 m [ ne v]/12.81@xmath0 m [ ne ii ] . plotted are predicted line ratios as computed by mappings version iiiq @xcite for combined precursor+shock models at solar metallicity at a gas density n=100@xmath43 ( upper panel ) and n=10@xmath43 ( lower panel ) . different colors ( ranging from _ purple _ to _ red _ ) denote different magnetic field strengths , ranging from 10@xmath44 to 100@xmath0 g . shock velocities range from 200 to 1000 km s@xmath1 and are labeled along the _ red _ curve . overlaid are the positions of irasf001837111 , iras12127 - 1412ne and iras13451 + 1232 for the integrated fluxes ( _ gray disks _ ) and for just the blue shifted component ( _ black disks _ ) . _ gray squares _ mark the positions of the nuclear spectra of the classical agns ngc1068 and ngc4151 @xcite [ fig11],width=321 ] in fig.[fig11 ] we show a mid - infrared diagnostic diagram based on the 12.81@xmath0 m [ ne ii ] , 15.56@xmath0 m [ ne iii ] and 14.32@xmath0 m [ ne v ] lines . in the diagram we plot shock excitation models from the mappings iiiq library @xcite for solar metallicity gas with densities n=100@xmath43 ( upper panel ) and n=10@xmath43 ( lower panel ) . the model tracks plotted include the ionizing effects from uv and soft x - ray photons in the shock cooling zone on the gas ahead of the shock front . overplotted are the positions of irasf001837111 , iras12127 + 1412ne and iras13451 + 1232 from our spitzer observations . black filled circles indicate the results for just the broad blue - shifted emission line components , the gray versions denote the results based on the velocity - integrated line fluxes ( i.e. the narrow and broad components combined ) . for irasf001837111 and iras12127 + 1412ne the low observed [ ne iii]/[ne ii ] line ratios are consistent with high - velocity shocks ( v@xmath25500 km s@xmath1 ) in high - density environments ( n@xmath25100@xmath43 ) with low magnetic field strengths ( @xmath171@xmath0 g ) . at lower gas densities ( n@xmath4510@xmath43 ) , only the highest velocity shocks ( v=1000 km s@xmath1 ) at the lowest magnetic field strengths ( @xmath170.1@xmath0 g ) are consistent with the observed [ ne iii]/[ne ii ] line ratios . below this density the observed line ratios can not have an origin in shock emission . given the absence of density measurements we can not rule out either scenario . however , as we will argue later , if the neon line emission originates at the base of the optical outflow , gas densities of n@xmath25100@xmath43 may not be unreasonable . in iras13451 + 1232 it is the high observed [ ne v]/[ne ii ] ratio which appears to rule out shocks of any velocity and density as the source of excitation for the neon lines . for the latter source , exposure of the blue shifted gas to direct radiation from the agn seems a likely explanation for the presence of [ ne v ] emission at the observed ratio to [ ne ii ] . this explanation is supported by the proximity of the blue shifted component of iras13451 + 1232 to the positions of the classical agns ngc1068 and ngc4151 in fig.[fig11 ] . in sect.3.3 we analyzed the velocity structure of the 12.81@xmath0 m [ ne ii ] and 15.56@xmath0 m [ ne iii ] lines in irasf001837111 and compared the results to those of studies of its optical and near - infrared forbidden lines . this revealed that the mid - infrared neon emission lines have strongly blue shifted components not detected in both the optical 5007 [ o iii ] and near - infrared 1.644@xmath0 m [ fe ii ] line profiles . given the mid - infrared spectral evidence for strong obscuration in the nucleus of irasf001837111 @xcite , the simplest explanation for the absence of strongly blue shifted gas components in the optical and near - infrared emission lines is extinction . we estimate the required extinction by assuming line emission at any wavelength @xmath46 to be effectively suppressed by a specific extinction a(@xmath46)@xmath392 . the level of screen extinction required to obscure the high - velocity components of the 5007 [ o iii ] line then amounts to a@xmath47 = a@xmath2@xmath252 . if the same screen is responsible for obscuring the high - velocity components of the 1.644@xmath0 m [ fe ii ] line , the required extinction increases to a@xmath2@xmath2510 ( a@xmath48@xmath252 ) . then to still be able to see the strongly blue shifted 12.81@xmath0 m [ ne ii ] emission through this screen , the maximum extinction is constrained to a@xmath2@xmath1750 ( a@xmath49@xmath171 ) . this implies a column density of the obscuring screen of n@xmath50 = 210 @xmath9 10@xmath51 @xmath37 , assuming n@xmath50 = 1.9 @xmath9 10@xmath52 @xmath9 a@xmath2 . in the absence of spatial information for the mid - infrared neon line emission , the location of the screen blocking the optical and near - infrared line emission can not be determined with any certainty . however , it is plausible for the screen to be situated at the base of the outflow traced in 5007 [ o iii ] . the observed strong decrease in blueshift of the 5007 [ o iii ] line from 3@xmath22 to 0@xmath22 east of the nucleus ( fig.12 of * ? * ) may hence indicate that the material that obscures the most highly disturbed optical ( but not mid - infrared ) line emission is located within the inner 10kpc . irasf001837111 is a clear outlier in a mid - infrared diagnostic diagram which combines the equivalent width of the 6.2@xmath0 m pah emission feature and the silicate strength in the so - called ` fork ' diagram @xcite . in this diagram most ulirgs are found along one of two branches . one branch connects starburst - dominated mid - infrared spectra ( e.g. ngc7714 and m82 ) to spectra of deeply buried central sources , characterized by very deep silicate features and the ( virtual ) absence of pah emission ( e.g. iras08572 + 3915 ) . the other branch connects starburst - dominated sources with featureless hot dust - dominated sources , which are mostly agn dominated . irasf001837111 is located in the scarcely populated region in between the tips of the two branches and has iras12127 + 1412ne as its nearest neighbor . their location , away from other deeply obscured sources , either implies a more than 10 times lower 6.2@xmath0 m pah equivalent width than other ulirgs with similar silicate strength , or a less pronounced silicate feature than other deeply obscured ulirgs . @xcite have proposed that sources in between the two branches may be agn evolving from a fully covered obscuring geometry to an incomplete or clumpy obscuring geometry , with a decrease in apparent depth of the silicate feature as a result . in this scenario , the detection of strongly blue shifted neon line emission would then trace a unique phase in the ( partial ) disruption of the obscuring medium around the powerful agn . alternatively , the isolated positions of irasf001837111 and iras12127 + 1412ne in the fork diagram may be the result of unusual lines of sight into these systems . the unusual mid - infrared excitation conditions and kinematics may then be associated to that . if sources with mid - infrared kinematic properties and neon / pah ratios similar to irasf001837111 and iras12127 + 1412ne are indeed agn evolving from full to incomplete or clumpy obscuration , there should be observational characteristics linking them . however , the only ulirg in our sample with these same mid - infrared properties , iras13451 + 1232 , has a mid - infrared continuum spectrum which is very different from the other two sources , displaying only a weak silicate absorption feature on an otherwise featureless power law spectrum . this places it firmly on the agn branch in the fork diagram , far off from the other two sources . evidence supporting an unusual evolutionary stage is presented by @xcite . based on the shape of its radio sed and a kinematic study of its optical emission line spectrum , they conclude that iras13451 + 1232 must be a young compact radio source whose radio jets are expanding through a dense nuclear dust shell , entraining gas and dust from this shell . they infer the reddening to be highest toward the broadest ( most disturbed ) most blue shifted kinematic component in their optical line study , reminiscent of our findings for irasf001837111 using optical , near and mid - infrared line profiles . we have discovered high - velocity ionized gas with velocities ranging from -3500 to + 3000 km s@xmath1 as part of a nuclear outflow from the ulirg irasf001837111 . the gas is traced by the mid - infrared 12.81@xmath0 m [ ne ii ] and 15.51@xmath0 m [ ne iii ] lines , which are dominated by blue shifted components . optical and near - infrared spectroscopic studies show no evidence for similar high - velocity gas components in forbidden lines at shorter wavelengths . we interpret this as the result of strong extinction ( a@xmath2=1050 ; a@xmath49@xmath171 ) on the blue shifted gas , identifying the base of the optical outflow ( @xmath173@xmath22 east ) as its most likely origin . unusual excitation conditions in irasf001837111 are further implied by a comparison of the combined [ ne ii ] and [ ne iii ] luminosity and the combined 6.2 and 11.2@xmath0 m pah luminosity for a large sample of agns , ulirgs and starburst galaxies . this reveals the neon line emission in irasf001837111 to be at least 8 times stronger compared to its pah emission than the typical ratio for the sample . two other ulirgs in our sample of 56 display mid - infrared excitation and kinematic properties similar to irasf001837111 : iras121271412ne and iras13451 + 1232 . both sources have elevated neon / pah ratios and exhibit pronounced blue wings in their 15.56@xmath0 m [ ne iii ] line profiles . wings are also present in their 12.81@xmath0 m [ ne ii ] profiles , but are less pronounced , indicating blue shifted gas to have a higher [ ne iii]/[ne ii ] ratio than gas closer to systemic velocity . iras13451 + 1232 even shows strongly blue shifted 14.32@xmath0 m [ ne v ] emission . while for iras13451 + 1232 the observed [ ne iii]/[ne ii ] and [ ne v]/[ne ii ] ratios indicate exposure of the blue shifted gas to direct radiation from the agn , for irasf001837111 and iras121271412ne the observed ratios are consistent with an origin in fast shocks ( v@xmath25500 km s@xmath1 ) in an environment with gas densities n@xmath2510@xmath43 . irasf001837111 and iras121271412ne are clear outliers in the mid - infrared diagnostic diagram of silicate strength and 6.2@xmath0 m pah equivalent width , the so - called ` fork ' diagram @xcite . the sources are found in between the two branches which separate deeply buried nuclei and sources with incomplete or clumpy obscuring geometries . @xcite have proposed that ulirgs in between the tips of these branches may be agn in the process of disrupting their fully covered geometry before settling on the agn branch . iras13451 + 1232 would then already be there , given its nearly featureless mid - infrared power law spectrum . the results from our mid - infrared kinematic and excitation study are consistent with the above evolutionary scenario in that the strongly blue shifted [ ne ii ] and [ ne iii ] emission ( from shocks ) may trace the ( partial ) disruption of the obscuring medium around buried agns in irasf001837111 and iras121271412ne . the detection of strongly blue shifted [ ne v ] emission in iras13451 + 1232 may then indicate the lifting of the obscuration around the agn coronal line region to be further advanced in this ulirg than in irasf001837111 and iras121271412ne , where this line is undetected . our measurements of the s(1 ) , s(2 ) , s(3 ) and s(4 ) pure rotational molecular hydrogen lines allow us to investigate the properties of warm molecular hydrogen in irasf001837111 . unfortunately , a measurement of the important s(0 ) line is lacking as it is outside the irs - lh bandpass . in fig.[fig6 ] we show the excitation diagram for molecular hydrogen for the four measured lines . given the placement of the s(3 ) line at the peak of the silicate absorption feature , we can attempt an extinction correction on the four data points under the assumption that a ) the excitation temperature should be constant or increase monotonically with increasing upper level energy e@xmath53(@xmath54,j ) , b ) that the ortho - to - para ratio is 3 ( i.e. assuming lte ) , and c ) that the cooler gas probed by the s(1 ) and s(2 ) lines is more or less co - spatial with the warmer gas probed by the s(3 ) and s(4 ) lines . adopting the local ism extinction law of @xcite , we then find that the extinction on the h@xmath18 lines ranges between a@xmath2=15 and 25 . this is far lower than the extinction on the continuum source , which is a@xmath2@xmath2554 based on the apparent depth of the 9.7@xmath0 m silicate feature ( @xmath55=3.0 ; * ? ? ? * ) and a@xmath2/@xmath55=18 @xcite . based on the two detected ortho - h@xmath18 lines , it is possible to infer an excitation temperature for the warm molecular hydrogen . the value ranges from 330 to 390k for extinction corrections ranging from a@xmath2=0 to 25 . assuming all h@xmath18 levels to be in lte at 330390k , we derive a total warm molecular hydrogen mass of 820@xmath910@xmath57 m@xmath7 . this value becomes a lower limit if the excitation temperature based on the upper level populations of the s(0 ) and s(1 ) lines is lower than those based on the s(1 ) and s(3 ) lines which is usually the case ( e.g. * ? ? ? unfortunately , we can not determine the warm to total molecular hydrogen fraction and compare this to other ulirgs , as no millimeter co line measurement exists for irasf001837111 from which to infer the cold molecular hydrogen mass . [ [ non - detection - of - absorption - bands - of - c_2h_2-and - hcn - gas ] ] non - detection of absorption bands of c@xmath18h@xmath18 and hcn gas ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ several ulirgs show absorption bands of c@xmath18h@xmath18 and hcn gas in their 13.714.1@xmath0 m _ spitzer _ irs - sh spectra @xcite . detections are especially strong among ulirgs with pronounced silicate absorption features , like iras201004156 and iras15250 + 3609 . far weaker detections are reported for ulirgs with agn - like spectra , like iras051892524 , iras010032238 and mrk231 @xcite . detections at these levels require spectra to have signal - to - noise ratios of at least 100 in their continua . the irs - sh spectrum of irasf001837111 does not show any indication for the presence of either feature , despite a continuum s / n of 70 . this translates into upper limits on the c@xmath18h@xmath18 and hcn gas column densities of n(c@xmath18h@xmath18)@xmath171.2@xmath910@xmath58@xmath37 and n(hcn)@xmath173.0@xmath910@xmath58@xmath37 for an adopted excitation temperature of 300k ( f. lahuis , private communication ) . the authors thank jan cami , helmut dannerbauer , brent groves , lei hao , timothy heckman , lisa kewley , dieter lutz and xander tielens for discussions , catherine buchanan , joanna holt and sylvain veilleux for sharing their published spectra , and mark g. allen for providing mappings shock models ahead of publication . we also would like to thank the slow but dilligent referee for comments which helped to strengthen the paper . support for this work was provided by nasa . this research has made extensive use of the nasa / ipac extragalactic database ( ned ) which is operated by the jet propulsion laboratory , california institute of technology , under contract with nasa . spoon , h.w.w . , keane , j.v . , cami , j. , lahuis , f. , tielens , a.g.g.m . , armus , l. , charmandaris , v. , 2005 , in iau symp . 231 , astrochemistry : recent successes and current challenges , eds . lis , g.a . blake & e. herbst , ( cambridge : cambridge univ .

asteroid ( 99942 ) apophis close approach in 2029 will be one of the most significant small - body encounter events in the near future and offers a good opportunity for exploration to determine the asteroid s surface properties and measure any tidal effects that might alter its regolith configuration . resurfacing mechanics has become a new focus for asteroid researchers due to its important implications for interpreting surface observations , including space weathering effects . this paper provides a prediction for the tidal effects during the 2029 encounter , with an emphasis on whether surface refreshing due to regolith movement will occur . the potential shape modification of the object due to the tidal encounter is first confirmed to be negligibly small with systematic simulations , thus only the external perturbations are taken into account for this work ( despite this , seismic shaking induced by shifting blocks might still play a weak role and we will look into this mechanism in future work ) . a two - stage approach is developed to model the responses of asteroid surface particles ( the regolith ) based on the soft - sphere implementation of the parallel-body gravity tree code ` pkdgrav ` . a full - body model of apophis is sent past the earth on the predicted trajectory to generate the data of all forces acting at a target point on the surface . a sandpile constructed in the local frame is then used to approximate the regolith materials ; all the forces the sandpile feels during the encounter are imposed as external perturbations to mimic the regolith s behavior in the full scenario . the local mechanical environment on the asteroid surface is represented in detail , leading to an estimation of the change in global surface environment due to the encounter . typical patterns of perturbation are presented that depend on the asteroid orientation and sense of rotation at perigee . we find that catastrophic avalanches of regolith materials may not occur during the 2029 encounter due to the small level of tidal perturbation , although slight landslides might still be triggered in positions where a sandpile s structure is weak . simulations are performed at different locations on apophis surface and with different body- and spin - axis orientations ; the results show that the small - scale avalanches are widely distributed and manifest independently of the asteroid orientation and the sandpile location . we also include simulation results of much closer encounters of the apophis with earth than what is predicted to occur in 2029 , showing that much more drastic resurfacing takes place in these cases . * keywords * : asteroids , dynamics ; asteroids , surfaces ; near - earth objects ; regoliths ; tides , solid body

1408.0168

in this paper , we investigate numerically the behavior of granular material at the surface of an asteroid during close approach to the earth . we focus on the specific case of asteroid ( 99942 ) apophis , which will come as close as @xmath3 earth radii on april 13th , 2029 . we provide predictions about possible reshaping and spin - alteration of and surface effects on apophis during this passage , as a function of plausible properties of the constituent granular material . studies of possible future space missions to apophis are underway , including one by the french space agency cnes calling for international partners ( e.g. , michel et al . 2012 ) , with the aim of observing this asteroid during the 2029 close encounter and characterizing whether reshaping , spin - alteration , and/or surface motion occur . the numerical investigations presented here allow for estimation of the surface properties that could lead to any observed motion ( or absence of motion ) during the actual encounter . apophis made a passage to the earth at @xmath4 earth radii in early 2013 . at that time , the herschel space telescope was used to refine the determination of the asteroid s albedo and size ( mller et al . according to these observations , the albedo is estimated to be about 0.23 and the longest dimension about @xmath5 m , which is somewhat larger than previous estimates ( @xmath6 m , according to delbo et al . concurrent radar observations improved the astrometry of the asteroid , ruling out the possibility of a collision with the earth in 2036 to better than 1 part in @xmath7 . however , wlodarczyk et al . ( 2013 ) presented a possible path of risk for 2068 . this finding has put off any crisis by @xmath8 years and makes exploring apophis in 2029 to be more for scientific interest . to date , nothing is known about the asteroid s surface mechanical properties , and this is why its close passage in 2029 offers a great opportunity to visit it with a spacecraft , determine its surface properties , and , for the first time , observe potential modifications of the surface due to tidal effects . and as apophis approaches , it is likely that international interest in a possible mission will increase , since such close approaches of a large object are relatively rare . the case for tidally induced resurfacing was made by binzel et al . ( 2010 ; also see demeo et al . 2013 ) and discussed by nesvorn et al . ( 2010 ) to explain the spectral properties of near - earth asteroids ( neas ) belonging to the q taxonomic type , which appear to have fresh ( unweathered ) surface colors . dynamical studies of these objects found that those bodies had a greater tendency to come close to the earth , within the earth - moon distance , than bodies of other classes in the past @xmath9 kyr . the authors speculated that tidal effects during these passages could be at the origin of surface material disturbance leading to the renewed exposure of unweathered material . we leave a more general and detailed investigation of this issue for future work , but if this result is true for those asteroids , it may also be true for apophis , which will approach earth on friday , april 13 , 2029 no closer than about 29,500 km from the surface ( i.e. , @xmath10 earth radii , or @xmath3 earth radii from the center of the planet ; giorgini et al . it is predicted to go over the mid - atlantic , appearing to the naked eye as a moderately bright point of light moving rapidly across the sky . our aim is to determine whether , depending on assumed mechanical properties , it could experience surface particulate motions , reshaping , or spin - state alteration due to tidal forces caused by earth s gravity field . the classical roche limit for a cohesionless fluid body of bulk density @xmath11 @xmath12 to not be disrupted by tidal forces is @xmath13 earth radii , so we do not expect any violent events to occur during the rocky asteroid s 2029 encounter at @xmath3 earth radii . the presence of granular material ( or regolith ) and boulders at the surface of small bodies has been demonstrated by space missions that visited or flew by asteroids in the last few decades ( e.g. , veverka et al . 2000 ; fujiwara et al . it appears that all encountered asteroids to date , from the largest one , the main belt asteroid ( 4 ) vesta by the dawn mission , to the smallest one , the nea ( 25143 ) itokawa , sampled by the hayabusa mission , are covered with some sort of regolith . in fact , thermal infrared observations support the idea that most asteroids are covered with regolith , given their preferentially low thermal inertia ( delbo et al . there even seems to be a trend as a function of the asteroid s size based on thermal inertia measurements : larger objects are expected to have a surface covered by a layer of fine regolith , while smaller ones are expected to have a surface covered by a layer of coarse regolith ( clark et al . this trend is consistent with observations by the near - shoemaker spacecraft of the larger ( @xmath14 km mean diameter ) eros , whose surface is covered by a deep layer of very fine grains , and by the hayabusa spacecraft of the much smaller ( @xmath15 m mean diameter ) itokawa , whose surface is covered by a thin layer of coarse grains . however , interpretation of thermal inertia measurements must be made with caution , as we do not yet have enough comparisons with actual asteroid surfaces to verify that the suggested trend is systematically correct . thus , we are left with a large parameter space to investigate possible surface motion during an earth close approach of an asteroid with unknown surface mechanical properties . our approach is to consider a range of simple and well - controlled cases that certainly do not cover all possibilities regarding apophis surface mechanical properties , but rather aim at demonstrating whether , even in a simple and possibly favorable case for surface motion , some resurfacing event can be expected to occur during the passage . for instance , instead of considering a flat granular surface , we consider a sandpile consisting of a size distribution of spherical grains ( section [ s : initialcondition ] ) and vary the grain properties in order to include more or less favorable cases for motion ( from a fluid - like case to a case involving rough particles ) . slight disturbances may manifest as very - small - scale avalanches in which grain connections readjust slightly , for example . the forces acting on the sandpile are obtained by measuring all `` external '' perturbations during the encounter , including body spin magnitude and orientation changes , for cases in which the global shape remains nearly fixed , and again assuming simple and favorable configurations of the asteroid . indirectly , the encounter may also lead to internal reconfigurations of the asteroid , which in turn produce seismic vibrations that could propagate to the surface and affect the regolith material . these secondary modifications are not modeled here , although it may be possible in future work to account for this by shaking the surface in a prescribed manner . in any case , for this particular encounter , we demonstrate ( section [ s : reshpeff ] ) that any global reconfiguration will likely be small to negligible in magnitude . in the following , we first present , in section [ s : method ] , the numerical method used to perform our investigation , including the initial conditions of the sandpile adopted to investigate surface motion , the representation of the encounter , and the mechanical environment . results are described in section [ s : results ] , including potential reshaping of the asteroid , tidal disturbances for apophis encounter in 2029 , which is a function of the sandpile properties , spin orientation changes , and the dependency of the location of the sandpile on the asteroid to the outcome of the encounter . we also show the responses of the sandpiles for artificially close approaches ( @xmath16 and @xmath17 earth radii ) to demonstrate that our method does predict significant alteration of the sandpiles when this is certainly expected to happen . the investigation is discussed in section [ s : discuss ] and conclusions are presented in section [ s : concl ] . we use ` pkdgrav ` , a parallel @xmath2-body gravity tree code ( stadel 2001 ) adapted for particle collisions ( richardson et al . 2000 ; 2009 ; 2011 ) . originally collisions in ` pkdgrav ` were treated as idealized single - point - of - contact impacts between rigid spheres . a soft - sphere option was added recently ( schwartz et al . 2012 ) ; with this new functionality , particle contacts last many timesteps , with reaction forces dependent on the degree of overlap ( a proxy for surface deformation ) and contact history this is appropriate for dense and/or near - static granular systems with multiple persistent contacts per particle . the code uses a 2nd - order leapfrog integrator , with accelerations due to gravity and contact forces recomputed each step . various types of user - definable confining walls are available that can be combined to provide complex boundary conditions for the simulations . the code also includes an optional variable gravity field based on a user - specified set of rules . the spring / dash - pot model used in ` pkdgrav ` s soft - sphere implementation is described fully in schwartz et al . briefly , a ( spherical ) particle overlapping with a neighbor or confining wall feels a reaction force in the normal and tangential directions determined by spring constants ( @xmath18 , @xmath19 ) , with optional damping and effects that impose static , rolling , and/or twisting friction . the damping parameters ( @xmath20 , @xmath21 ) are related to the conventional normal and tangential coefficients of restitution used in hard - sphere implementations , @xmath22 and @xmath23 . the static , rolling , and twisting friction components are parameterized by dimensionless coefficients @xmath24 , @xmath25 , and @xmath26 , respectively . plausible values for these parameters are obtained through comparison with laboratory experiments ( also see section [ s : initialcondition ] ) . careful consideration of the soft - sphere parameters is needed to ensure internal consistency , particularly with the choice of @xmath18 , @xmath19 , and timestep a separate code is provided to assist the user with configuring these parameters correctly . the numerical approach has been validated through comparison with laboratory experiments ; e.g. , schwartz et al . ( 2012 ) demonstrated that ` pkdgrav ` correctly reproduces experiments of granular flow through cylindrical hoppers , specifically the flow rate as a function of aperture size , and found the material properties of the grains also affect the flow rate . also simulated successfully with the soft - sphere code in ` pkdgrav ` were laboratory impact experiments into sintered glass beads ( schwartz et al . 2013 ) , and regolith , in support of asteroid sampling mechanism design ( schwartz et al . 2014 ) . we use a two - stage approach to model the effect of a tidal encounter on asteroid surface particles ( regolith ) . first , a rigid ( non - deformable ) object approximating the size , shape , and rotation state of the asteroid is sent past the earth on a fixed trajectory ( in the present study , the trajectory is that expected of ( 99942 ) apophis see section [ s : representme ] ; note that the actual shape of apophis is poorly known beyond an estimate of axis ratios , so we assume an idealized ellipsoid for this study ) . all forces acting at a target point designated on the object surface are recorded ( section [ s : representme ] ) . then , a second simulation is performed in the local frame of the target point , allowing the recorded external forces to affect the motion of particles arranged on the surface ( in the present study we consider equilibrated sandpiles ) . this two - stage approach is necessary due to the large difference in scale between the asteroid as a whole and the tiny regolith particles whose reactive motion we are attempting to observe . we approximate the regolith , which in reality likely consists of a mixture of powders and gravel ( clark et al . 2002 ) , by a size distribution of solid spheres . we mimic the properties of different materials by adjusting the soft - sphere parameters ( section [ s : initialcondition ] ) . the soft - sphere approach permits simulation of the behavior of granular materials in the near - static regime , appropriate for the present case in which the regolith particles generally remain stationary for hours and suffer a rapid disturbance only during the moments of closest approach to the planet . in particular , the model permits a detailed look at the responses of local individual particles even when the tidal effects are too weak to cause any macroscopic surface or shape changes ; this gives insight into the limit of tidal resurfacing effects . in order to assess the effect of tidal encounters on a small surface feature , we carried out numerical simulations in a local frame consisting of a flat horizontal surface ( i.e. , the local plane tangential to the asteroid surface at the target point ) with a `` sandpile '' resting on top . the sandpile consists of @xmath2 = 1683 simulated spherical particles with radii drawn from a power - law distribution of mean @xmath27 cm and @xmath28 cm width truncated ( minimum and maximum particle radii @xmath29 and @xmath30 cm , respectively ) . between the sandpile and the floor is a rigid monolayer of particles drawn from the same distribution and laid down in a close - packed configuration in the shape of a flat disk . this rigid particle layer provides a rough surface for the sandpile to reduce spreading ( fig . [ f : sandpile ] ) . three different sandpiles were constructed using the material parameters described below . particles were dropped though a funnel from a low height onto the rough surface to build up the sandpile . these sandpiles were allowed to settle until the system kinetic energy dropped to near zero . this approach eliminates any bias that might arise from simply arranging the spheres by hand in a geometrical way ; the result should better represent a natural sandpile . figure [ f : sandpile ] . for this study , we compared three different sets of soft - sphere parameters for the sandpile particles ( table [ t : ssdemparams ] ) . our goal was to define a set of parameters that spans a plausible range of material properties given that the actual mechanical properties of asteroid surface material are poorly constrained . in the specific case of apophis , very little is known beyond the spectral type , sq ( binzel et al . there are no measurements of thermal properties that might give an indication of the presence or absence of regolith on apophis . consequently , we chose three sets of material parameters that span a broad range of material properties . the first set , denoted `` smooth '' in the table , consists of idealized frictionless spheres with a small amount of dissipation ( 5% , chosen to match the glass beads case ) . this is about as close to the fluid case that a sandpile can achieve while still exhibiting shear strength arising from the discrete nature of the particles ( and the confining pressure of surface gravity ) . it is assumed this set will respond most readily to tidal effects due to the absence of friction between the particles and the non - uniform size distribution . the second set , `` glass beads , '' is modeled after actual glass beads being used in a set of laboratory experiments to calibrate numerical simulations of granular avalanches ( richardson et al . 2012 ) . in this case @xmath23 was measured directly , which informed our choice for @xmath21 ( schwartz et al . 2012 ) , and @xmath24 and @xmath25 were inferred from matching the simulations to the experiments . the glass beads provide an intermediate case between the near - fluid smooth spheres and the third parameter set , denoted `` gravel . '' table [ t : ssdemparams ] . the gravel parameters were arrived at by carrying out simple avalanche experiments using roughly equal - size rocks collected from a streambed . in these experiments , the rocks ( without sharp edges ) were released near the top of a wooden board held at a 45@xmath31 incline . the dispersal pattern on the stone floor was measured , including the distance from the end of the board to approximately half of the material , the furthest distance traveled by a rock along the direction of the board , and the maximum angle of the dispersal pattern relative to the end of the board ( fig . [ f : rockslide]a c ) . a series of numerical simulations was then performed to reproduce the typical behavior by varying the soft - sphere parameters ( fig . [ f : rockslide]d f ) . the values used in table [ t : ssdemparams ] for `` gravel '' were found in this way . the @xmath24 and @xmath25 values are quite large , reflecting the fact that the actual particles being modeled were not spheres . a correspondingly smaller timestep is needed to adequately sample the resulting forces on the particles in the simulations . the value of @xmath22 was measured by calculating the average first rebound height of sample gravel pieces that were released from a certain height ; @xmath23 was not measured but since the rocks were rough it was decided to simply set @xmath32 . this exercise was not meant to be an exhaustive or precise study ; rather , we sought simply to find representative soft - sphere parameters that can account plausibly for the irregularities in the particle shapes . in any case , we will find that such rough particles are difficult to displace using tidal forces in the parameter range explored here , so they provide a suitable upper limit to the effects . similarly , we do not consider cohesion in this numerical study , which would further resist particle displacement due to tidal forces . figure [ f : rockslide ] . a new adaptation of ` pkdgrav ` developed in this work allows for the simulation of sandpiles located on an asteroid surface , based on the motion equations of granular material in the noninertial frame fixed to the chosen spot . detailed mechanics involved in the local motion of the sandpile during a tidal encounter were considered and represented in the code , including the contact forces ( reaction , damping , and friction between particles and/or walls ) and the inertial forces derived from an analysis of the transport motion of a flyby simulation . here we provide a thorough derivation of the relevant external force expressions . figure [ f : frames ] . figure [ f : frames ] illustrates four frames used in the derivation , the space inertial frame ( spc ) , mass center translating frame ( mct ) , body fixed frame ( bdy ) and local frame ( loc ) . ( [ e : posspc2loc ] ) gives the connection between the spatial position and the local position of an arbitrary particle in the sandpile , for which @xmath33 is the vector from spc s origin to the particle , @xmath34 is the vector from spc to mct / bdy , @xmath35 is the vector from mct / bdy to loc , and @xmath36 is the vector from loc to the particle : @xmath37 equations ( [ e : velspc2loc])([e : accspc2loc ] ) are derived by calculating the first- and second - order time derivatives of eq . ( [ e : posspc2loc ] ) , which denote the connections of velocity and acceleration between the spatial and local representations , respectively . @xmath38 indicates the angular velocity vector of the asteroid in spc or mct . the operator @xmath39 denotes the time derivative with respect to the inertial frame spc , while @xmath40 denotes the time derivative with respect to the body - fixed frame bdy or loc : @xmath41 @xmath42+\frac{\tilde \mathrm{d}}{\mathrm{d } t } { { \mbox{\boldmath{$\omega$ } } } } \times ( \mathbf{l}+\mathbf{r})+2 { { \mbox{\boldmath{$\omega$ } } } } \times \frac{\tilde \mathrm{d}}{\mathrm{d } t } \mathbf{r}+\frac{\tilde \mathrm{d}^2 } { \mathrm{d } t^2 } \mathbf{r}.\ ] ] the dynamical equation for an arbitrary particle in the inertial frame spc can be written as eq . ( [ e : spcdyneqn ] ) , which shows that the forces a particle in the sandpile feels can be grouped into three categories : @xmath43 denotes the local gravity from the asteroid ; @xmath44 denotes the attraction from the planet ; and @xmath45 represents all the contact forces coming from other particles and walls : @xmath46 we get the corresponding dynamical equation in the local frame loc ( eq . ( [ e : locdynequ ] ) ) by substituting eq . ( [ e : accspc2loc ] ) into eq . ( [ e : spcdyneqn ] ) , @xmath47-\frac{\tilde \mathrm{d}}{\mathrm{d } t } { { \mbox{\boldmath{$\omega$ } } } } \times ( \mathbf{l}+\mathbf{r})-2 { { \mbox{\boldmath{$\omega$ } } } } \times \frac{\tilde \mathrm{d}}{\mathrm{d } t } \mathbf{r}.\ ] ] the terms on the right - hand side of eq . ( [ e : locdynequ ] ) represent the mechanical environment of a particle in the local sandpile during a tidal encounter , and can be interpreted as follows . @xmath45 is the reaction force from the walls and surrounding particles , which is directly exported by ` pkdgrav ` . @xmath43 is approximated by a constant value at the loc s origin that acts as the uniform gravity felt throughout the whole sandpile , since the latter s size is negligible compared with the asteroid dimensions . the difference between the planetary attraction and the translational transport inertia force , @xmath48 , is the tidal force that plays the primary role of moving the surface materials . the term @xmath49 $ ] is the centrifugal force due to the rotation of the asteroid , @xmath50 is the librational transport inertia force ( ltif ) that results when the rotational state changes abruptly during the encounter , and @xmath51 is the coriolis effect that only acts when particle movement occurs . figure [ f : rubblepile ] . in the numerical scheme , we derived the component dynamical equations by projecting eq . ( [ e : locdynequ ] ) to loc and treating each term separately . the local gravity @xmath43 and loc origin @xmath35 are constant vectors , which were initialized in the simulations as parameters . the complications arising from the asteroid motion ( tidal force , angular velocity , and acceleration ) were solved within a simulation in advance , using a rigid rubble pile as the asteroid model and exporting the physical quantities required in each step . to specify apophis s encounter with earth in 2029 , we computed the planet trajectory by tracing back from the perigee conditions , @xmath3 @xmath52 distance and @xmath53 km / s encounter speed ( giorgini et al . 2008 ) , where @xmath52 indicates the average earth radius ( @xmath54 km ) . as fig . [ f : rubblepile ] illustrates , a tri - axial ellipsoid model with axis ratio 1.4:1.0:0.8 was employed to approximate the overall shape of apophis ( scheeres et al . 2005 ) , corresponding to an equivalent radius of @xmath55 m ( mller et al . a rubble pile bounded by the tri - axial model was constructed for the flyby simulation . the rubble pile model consisted of @xmath2 = 2855 equal - size spherical particles in a close - packed configuration . the total mass was set to @xmath56 kg ( bulk density @xmath11 @xmath12 ) with an initial rotation around the maximum principle axis of inertia of period @xmath57 h ( tholen et al . 2013 ) . three markers ( non - coplanar with the origin ) were chosen from the surface particles to track the variation of the rubble pile s attitude in spc . a complete list of its motion states during the encounter was exported to a data file , which was accessed by the subsequent simulations with local sandpiles . richardson et al . ( 1998 ) showed the complicated behavior of a rotating rubble pile due to a tidal encounter . generally , a variation in rotational state is usually induced because of the coupling effect between librational motion and orbital motion . especially for elongated bodies like apophis , the terrestrial torques during its flyby may force a strong alteration of the rotational state in a short period ( scheeres et al . the modifications in structure of the rubble pile are also notable . the tidal encounter outcomes show a strong dependence on the progenitor elongation , the perigee distance , and the time spent within the roche sphere . the outcomes also depend on the spin orientation of the progenitor to the extent that retrograde rotators always suffer less catastrophic consequences than prograde ones ( richardson et al . 1998 ) . however , the parameter space considered in that study ( richardson et al . 1998 ) did not include the apophis scenario in 2029 , since the focus was on the conditions for full - scale distortion or disruption . for this study , we first redid the flyby simulations ( section [ s : representme ] ) using a true ( non - rigid ) rubble pile model with the apophis encounter parameters of @xmath3 @xmath52 and @xmath53 km / s . we quantified the tidal reshaping effect by measuring the maximum net change of a normalized shape factor defined by eq . ( [ e : shpfact ] ) ( hu et al . 2004 ) , in which @xmath58 , @xmath59 , @xmath60 are the principal inertia moments , with @xmath61 : @xmath62 the shape factor @xmath63 $ ] roughly describes the mass distribution of the body ( 0=oblate , 1=prolate ) . the initial rubble - pile model ( equilibrated , prior to the encounter ) had @xmath64 , and its relative change , defined by @xmath65 , was measured for several simulations parameterized by bulk density and constituent material type ( table [ t : fullscalechange ] ) . the corresponding roche limits are provided in the table , with the values estimated for the case of a circular orbit . the reshaping effects show consistent results for all three types of materials , varying between a noticeable change in shape ( magnitude @xmath66 ) and a negligible change in shape ( magnitude @xmath67 ) , with the sharpest transition occurring for a critical bulk density @xmath68 @xmath12 with corresponding roche limit of @xmath3 @xmath52 . this agrees with the analysis of apophis disruption limit using a continuum theory for a body made up of solid constituents ( holsapple et al . note we find that even a bulk density as low as @xmath69 @xmath12 did not dislodge any particles enough for them to end up in a new geometrical arrangement , due to the very short duration of the tidal encounter . figure [ f : newpile ] . table [ t : fullscalechange ] . since idealized equal - size spheres can arrange into crystal - like structures that may artificially enhance the shear strength of the body ( tanga et al . 2009 ; walsh et al . 2012 ) , we carried out a second suite of simulations with rubble piles made up of a bimodal distribution of spheres ( fig . [ f : newpile ] ) . the model consisted of @xmath70 big particles of radius @xmath71 m and @xmath72 small particles of radius @xmath73 m , bounded by the same tri - axial ellipsoid as shown in fig . [ f : rubblepile ] , and arranged randomly . as before , the bulk density and material type were varied , with notable differences in outcome compared to the equal - size - sphere cases ( table [ t : fullscalechange ] ) . particles using the smooth " parameter set are not able to hold the overall shape of apophis ( the 1.4:1.0:0.8 tri - axial ellipsoid ) ; instead the rubble pile collapses and approaches a near - spherical shape . for the other two parameter sets , however , the overall shape of the rubble pile is maintained . figure [ f : comp ] shows the relative change of shape factor @xmath74 as a function of bulk density for both types of simulations . as shown , the results from the unimodal and bimodal rubble piles present similar trends : before reaching the roche limit density , the reshaping remains small ( the magnitudes are @xmath75@xmath76 with a certain amount of stochasticity ) ; after that , the reshaping effect sharply increases to around @xmath77 . the bimodal rubble piles show less resistance to the tidal disturbance , since they have less - well - organized crystalline structures compared to the unimodal cases ; they therefore exhibit larger shape modifications ( walsh et al . 2012 ) . figure [ f : comp ] . regardless , catastrophic events were not detected for either type of rubble pile until the bulk density was as small as @xmath69 @xmath12 , for all material parameter sets . therefore , the reshaping effects on apophis in 2029 should be negligibly small for bulk densities in the likely range ( @xmath78@xmath79 @xmath12 ) . even so , minor internal reconfigurations resulting from the tidal encounter may produce seismic waves that could propagate and affect the configuration of surface regolith . since we are only looking at localized areas on the surface , isolated from the rest of the body , vibrations emanating from other regions , in or on the asteroid , are not evaluated in the current work , although again we expect these to be small or non - existent for the specific case of the apophis 2029 encounter . only the external " forces acting directly on the surface particles in the considered localized region are taken into account . we will look into the effects of seismic activity , which may stem from other regions of the asteroid , on an actual high - resolution rubble pile in future work . in this study , we focus on external forces , outside of the rubble pile itself ; the rigid rubble pile model is employed as a reasonable simplification for the purpose of measuring the external forces on a surface sandpile . the right - hand side of eq . ( [ e : locdynequ ] ) provides a description of the constituents of the mechanical environment experienced by a sandpile particle , including local gravity , tidal force , centrifugal force , ltif , and the coriolis effect . note the coriolis force does not play a role before particle motion begins , so it can be ignored when examining the causes of avalanches / collapses . the centrifugal force and ltif show weak dependence on local position @xmath36 , so these terms can be simplified by substituting @xmath80 , since the dimension of the sandpile is much smaller than that of the asteroid . this leads to an approximation of the resultant environmental force ( eq . ( [ e : envirforce ] ) ) , which provides a uniform expression of the field force felt throughout the sandpile : @xmath81 the variation of @xmath82 shows a common pattern for different encounter trajectories and different locations on the asteroid : it stays nearly invariable for hours as the planet is approaching / departing and shows a rapid single perturbation around the rendezvous . this feature enables us to evaluate the mechanical environment by measuring the short - term change of @xmath82 , and moreover , to make a connection between these external stimuli with the responses of sandpiles in the simulations . spherical coordinates @xmath83 are used to represent the resultant force ( fig . [ f : accs ] ) , in which @xmath84 is the magnitude and @xmath85 and @xmath86 denote the effective gravity slope angle and deflection angle , respectively . figure [ f : accs ] . the effects of the environment force must be considered in the context of the sandpile modeling . a rough approximation of the catastrophic slope angle ( eq . ( [ e : critcang ] ) ) can be derived from the conventional theory of the angle of repose for conical piles ( brown et al . 1966 ) , namely that avalanches will occur when the effective gravity slope angle @xmath85 exceeds @xmath87 : @xmath88 here @xmath89 denotes the resting angle of the ( conical ) sandpile , assumed to be less than the repose angle @xmath90 , and @xmath91 denotes the critical slope angle , which , if exceeded by the effective gravity slope angle , results in structural failure of the sandpile , i.e. , an avalanche . essentially , the avalanches of the sandpile depend primarily on the instantaneous change of the slope angle @xmath85 of the resultant force and should be only weakly related to the magnitude @xmath84 and deflection angle @xmath86 . the global distribution of changes of @xmath85 were examined for the duration of the encounter ( see below ) . it would be difficult to make an exhaustive search over all asteroid encounter orientations due to the coupling effects between the asteroid s rotation and the orbital motion . instead , we chose @xmath92 representative trajectories along the symmetry axes of the apophis model ( at perigee ) for study , which serves as a framework for understanding the influence of encounter orientation and the strength of tidal disturbance at different locations . ( the technique used to match the orientation of the asteroid at perigee is to first run a simulation in reverse starting at perigee with the required orientation , thereby providing the correct asteroid state for a starting position far away . ) figure [ f : traject ] shows the @xmath92 trajectories along three mutually perpendicular planes , including both the prograde and retrograde cases . this choice is based on the symmetry of the dynamical system , and all @xmath92 trajectories have a speed of @xmath53 km / s at a perigee distance of @xmath3 @xmath52 . figure [ f : traject ] . figure [ f : traject ] presents these representative trajectories in order , each corresponding to a special possible orientation of the encounter . the effective gravity slope angle changes throughout the surface of the tri - axial ellipsoid model were recorded for each trajectory , with particular attention to the location and time of the maximum change . we found that these maximum changes concentrate at several minutes around perigee . figure [ f : maxcsa ] shows the maximum values of slope angle change along these trajectories , indicating that the largest perturbation on slope angle is less than @xmath93 . we verified that the achievable range of the effective gravity slope angle during the encounter ( section [ s : representme ] ) is within the safe limit predicted by eq . ( [ e : critcang ] ) for all three sandpiles ( section [ s : initialcondition ] ) ; that is , a slope change below @xmath93 is not enough to trigger any massive avalanches throughout the sandpiles . figure [ f : maxcsa ] . figure [ f : maxcsa ] also shows a general dependence of the tidal perturbation on the orientation of the encounter trajectory , namely that the three step levels seen in the figure correspond to three directions of the relative planet motion at perigee . trajectories @xmath94@xmath95 correspond to the direction of the body long axis , which leads to relatively strong tidal effects ; trajectories @xmath96@xmath97 correspond to the direction of the intermediate axis , which leads to moderate effects ; while trajectories @xmath98@xmath92 correspond to the direction of the short axis , which leads to relatively weak tidal effects . figure [ f : patterns ] illustrates the distribution of effective gravity slope angle change at perigee for the three trajectory sets , with a denoting trajectories @xmath94@xmath95 , b denoting trajectories @xmath96@xmath97 , and c denoting trajectories @xmath98@xmath92 . we confirm that the four trajectories in each set present visually the same distribution around perigee , therefore we chose the patterns due to trajectories 1 , 5 , and 9 for demonstration of sets a , b , and c , and generalize these three patterns as representative . several points can be inferred from fig . [ f : patterns ] . first , the direction to the planet at perigee largely determines the global distribution of tidal perturbation , that is , the strongest effects tend to occur along the long axis and the weakest effects along the short axis . second , at perigee , the largest slope change occurs near areas surrounding the pole for the most favorable orientation ( set a ) , while the largest slope change at the pole occurs several minutes before or after perigee ( see fig . [ f : pxpypz ] ) . these maximum slope changes are about equal in magnitude ( the change is only slightly smaller at the pole compared to the area immediately surrounding it , but not enough to make a difference to the avalanches ) , so for simplicity we just use the poles themselves as our testing points . third , the duration of strong tidal effects depends on the eccentricity of the encounter trajectory . we checked the trajectories shown in fig . [ f : traject ] and found the duration of strong perturbation , defined for illustration as the period when the force magnitude stays above @xmath99 of the peak value , is tens of minutes long ; correspondingly , the responses of the surface material are also transitory and weak . figure [ f : patterns ] . the three poles p@xmath100 , p@xmath101 and p@xmath102 ( see figure [ f : traject ] ) were chosen as the test locations for the sample sandpiles since the effective gravity slope angle is near zero at these locations , which enables the sandpiles to hold their initial shapes . figure [ f : pxpypz ] illustrates the time variation of slope angle at the poles during the encounter . the results derived from all @xmath92 trajectories in fig . [ f : traject ] are shown and to the same scale . the most perturbed pole is p@xmath101 on the medium axis , which gains the largest slope angle change for most cases , except in trajectories \{3,4,9,10}. it is notable that trajectories in the same plane always share similar variation at the three poles , such as \{1,2,5,6 } , \{3,4,9,10 } , and \{7,8,11,12}. the curves at p@xmath100 and p@xmath101 show similar doublet shapes since those poles are both located in the rotational plane . the variation at p@xmath102 is more complicated due to the fact that the spin pole is not perpendicular to the orbit plane . figure [ f : pxpypz ] . figure [ f : pxpypz ] can be used to estimate the changes in the mechanical environment at these pole locations . since the slope angle change is the primary mechanism to drive avalanches on the sandpiles , the results derived from different trajectories serve as a framework to determine the magnitude of tidal perturbation at the candidate locations and locations in between , which turns out to be quite small for the apophis encounter ( less than about @xmath103 in slope for these cases ) . as analyzed above , catastrophic avalanches solely due to external forces acting directly on the surface particles may never occur on apophis during the 2029 encounter since the tidal perturbation will be very weak , however it might still have the potential to trigger some local tiny landslides of the surface materials . generally , the regolith experiences more activity than the constituents deep in the asteroid due to the dynamical environment on the surface , including the microgravity , maximum tidal and centrifugal acceleration , and smaller damping forces from the surroundings due to the smaller confining pressure , therefore we conclude that resurfacing due to external perturbations should occur before wholesale reshaping during the encounter ( discussed in section [ s : reshpeff ] ) . in the following sections we detail our numerical examination of the local sandpiles for the predicted apophis encounter scenario to estimate the limit and magnitude of the material responses in 2029 . one problem that must be confronted in the numerical simulations is that the soft - sphere sandpiles exhibit slow outward spreading due to the accumulation of velocity noise , and this slow spreading may eventually lead to a collapse of the whole sandpile . unfortunately , our integration time required is long ( hours ) for a granular system , thus the numerical noise has to be well limited in our method . two techniques are used in this study . first , as described in section [ s : initialcondition ] , a rough pallet made up of closely packed spheres in a disk fixed to the ground is used to reduce spreading of the bottom particles . second , we introduce a critical speed in the code , below which all motions are considered to be noise and are forced to zero . we found a speed threshold @xmath104 m / s by launching hours - long simulations of a static sandpile for different material properties and finding the corresponding minimum value of the critical speed that limits the numerical spreading . simulations of sandpiles without a flyby were carried out first , serving as a reference for subsequent flyby simulations . sandpiles in this situation were confirmed to stay equilibrated for a long period , which suggests the avalanches in our numerical experiments ( if any ) would be attributed entirely to the effects of the tidal encounter . we performed local simulations with the sandpiles generated in section [ s : initialcondition ] to consider the response of different materials to the apophis encounter . the local frames at the three poles p@xmath100 , p@xmath101 and p@xmath102 ( fig . [ f : traject ] ) were used to position the sample sandpiles , and the flyby simulation data derived from the @xmath92 fiducial trajectories were employed as the source of external perturbations for the local simulations . there are @xmath105 possible combinations in total for different materials , locations , and trajectory orientations , covering a wide range of these undetermined parameters . in all the simulations , the start distance of apophis from the center of the planet was set to @xmath106 @xmath52 to ensure the sandpiles were equilibrated fully before it approached the perigee . for our study , we concentrated on the connections between the sandpile s responses and significant variables , e.g. , spin orientation and sandpile locations , by which we will get a better sense of the surface processes due to a tidal encounter . the sandpiles located at pole p@xmath101 for trajectory 2 ( fig . [ f : traject ] ) were examined in detail , which is the case suffering the strongest tidal perturbation ( fig . [ f : pxpypz ] ) . figure [ f : avalanches ] illustrates the disturbances on sandpiles of three different materials for this configuration . each panel includes a snapshot of the sandpile after the tidal encounter with the displaced particles highlighted and a diagram showing the time variations of the total kinetic energy with planetary distance . intensive events can be identified by noting the peaks of the kinetic energy curves , and the particles involved in these events are marked with different colors to show the correspondence between the disturbed site and occurrence time . figure [ f : avalanches ] . the scale of the disturbances proves to be tiny even for the strongest " case in fig . [ f : avalanches ] . as illustrated , the displaced particles are few in number and mostly distributed on the surface of the sandpile . we found that the maximum displacement of a given particle is less than @xmath107 times its radius ; that is , these disturbances only resulted from the collapse of some local weak structures and the small chain reaction among the surrounding particles . moreover , the gravel " sandpile proves to be unaffected by the tidal encounter , because the static friction is quite large ( table [ t : ssdemparams ] ) and all particles are locked in a stable configuration , in which case the total kinetic energy of the sandpile remains very small ( @xmath108 j ) . the glass beads " sandpile suffered small but concentrated disturbances that involved very few particles . accordingly , the displacements of these particles are relatively large . smooth " sandpile presents near - fuild properties with many surface particles experiencing small - amplitude sloshing . the motion eventually damps out and the displacements of the involved particles are tiny ( smaller than that of glass beads " ) . based on the detailed analysis of this representative scenario , we performed simulations at other poles and at other orientations , constructing a database to reveal any connections between disturbances and these parameters . however , the results show little association between these two : for the gravel " sandpile , no disturbances were detected for any location and any orientation due to the large @xmath109 ; for the glass beads " sandpile , some small - scale avalanches can always be triggered , while the occurrence ( site and time ) of these avalanches seems to be widely distributed and independent of location and orientation ; and for the smooth " sandpile , surface particles can feel the tidal perturbation and show slight sloshing , but no visible avalanches eventually resulted , because the initial low slope angle in this case imposes a relatively stable structure that is always able to recover from the external perturbations . to illustrate the effect of stronger perturbations on our model sandpiles , we carried out a few simulations of closer approaches , specifically at @xmath16 and @xmath17 @xmath52 , for the same encounter speed of @xmath53 km / s . these scenarios , though unlikely to occur in 2029 based on current understanding of apophis orbit , are presented here to illustrate the magnitudes of tidal resurfacing effects over a wider range of perturbation strengths . flyby simulations of rubble piles were first carried out to quantify the reshaping effects , using the monodisperse and bidisperse particle rubble - pile models ( see section [ s : reshpeff ] ) . the bulk densities for both models were set to @xmath110 g / cc . table [ t : reshp24 ] presents the relative net changes of shape factor @xmath65 for these rubble piles of different materials at different perigee distances . table [ t : reshp24 ] . table [ t : reshp24 ] shows results consistent with section [ s : reshpeff ] , namely that the bimodal rubble pile shows greater fluidity and larger shape changes during the encounter . to be more specific , the magnitude of reshaping effects at perigee distance @xmath16 @xmath52 ( still larger than the roche limit of @xmath111 @xmath52 for a fluid body ) remains small ( @xmath112 ) , and that at perigee distance @xmath17 @xmath52 ( smaller than the roche limit ) achieves a significant level of @xmath113@xmath77 . it is notable that the rubble piles did not experience any irreversible distortion even for an approach distance as close as @xmath17 @xmath52 , because the duration of strong tidal effects is relatively short ( see section [ s : reshpeff ] ) . anyway , in this section we still adopt the assumption of rigidity to measure the quantities required by the local simulations , which is acceptable since these scenarios are fictitious and only designed to exhibit some massive resurfacing effects . for the same reason , in this section we do not present systematic local simulations for different sandpile locations and orientations as done for the real encounter scenario ; instead , we simply place the sample sandpiles ( see fig . [ f : sandpile ] ) at the pole of the body long axis p@xmath100 , and choose trajectory @xmath94 ( see fig . [ f : traject ] ) as the encounter trajectory to give rise to the maximum tidal effects . figures [ f : refrsh4f ] and [ f : refrsh2f ] illustrate the responses of the three sandpiles for perigee distance @xmath16 @xmath52 and @xmath17 @xmath52 , respectively . each panel includes snapshots of the sandpile before and after the tidal encounter , and the overall shape of the sandpile is traced with a white border for emphasis . diagrams showing the time variation of the total kinetic energy with planetary distance are also included , in which the intensive avalanches can be identified by noting the peaks of the kinetic energy curves ( note the different vertical scales ) . figure [ f : refrsh4f ] . [ f : refrsh2f ] . as illustrated in fig . [ f : refrsh4f ] , the shape changes of the three sandpiles are still small ( but visible ) for the encounter at @xmath16 @xmath52 perigee , and involve many more particles than the encounter at @xmath3 @xmath52 ( fig . [ f : avalanches ] ) . and accordingly , the magnitudes of the total kinetic energy increase consistently for the three sandpiles of different materials , except for the gravel case at @xmath16 @xmath52 , where it appears the frictional `` lock '' established when the pile was first created is disturbed enough to cause a stronger distortion than might otherwise be expected , causing a sharp spike in the kinetic energy plot . essentially , the `` gravel '' sandpile experienced a more significant collapse than the `` glass bead '' sandpile . as stated in section [ s : initialcondition ] , the three sandpiles were constructed in a manner that allowed for some inherent randomness , thus their energy state and strength may differ to some degree , and appearently the `` gravel '' sandpile was closer to its failure limit than the other two . this adds an extra element of stochasticity to the results , an aspect to be explored in future work . figure [ f : refrsh2f ] shows the results from the encounter at @xmath17 @xmath52 perigee , for which the encounter trajectory partly entered the roche sphere ( @xmath111 @xmath52 ) . in this case , the shapes of the three sandpiles are highly distorted during the encounter , with the involved particles slumped towards the direction where the planet is receding . the corresponding changes in total kinetic energy become extremely large when the massive avalanches occur , especially for the `` smooth '' particles . the results of this section suggest that a @xmath16 @xmath52 encounter may alter the regolith on apophis surface slightly , and a @xmath17 @xmath52 encounter may produce a strong resurfacing effect ( of course , we would expect considerable global distortion as well , if the asteroid is a rubble pile ) . the shape changes of the sandpiles depend on the orientation of the encounter trajectory ; i.e. , particles can be dragged away along the direction in which the planet recedes . the argument about whether a terrestrial encounter can reset the regolith of neas has been discussed for a while ( e.g. , binzel et al . 2010 , nesvorn et al . important evidence is provided by measurements of the spectral properties , which suggest that asteroids with orbits conducive to tidal encounters with planets show the freshest surfaces , while quantitative evaluation of this mechanism is still rare and rough . the two - stage approach presented in this paper enables the most detailed simulations to date of regolith migration due to tidal effects , which we have applied to make a numerical prediction for the surface effects during the 2029 approach of ( 99942 ) apophis . we confirm that the shape modification of apophis due to this encounter will likely be negligibly small based on the results of systematic simulations parameterized by bulk density , internal structure , and material properties . the analysis of the global mechanical environment over the surface of apophis during the 2029 encounter reveals that the tidal perturbation is even too small to result in any large - scale avalanches of the regolith materials , based on a plausible range of material parameters and provided that external forces acting directly on the surface dominate any surface effects due to seismic activity emanating from other regions of the body . future work will explore these second - order surface effects and their relevance to this encounter , and to cases of tidal resurfacing of small bodies in general . it is notable that our approach is capable of capturing slight changes in the regolith ( modeled as sandpiles ) ; thus , through numerical simulation , we find that this weak tidal encounter does trigger some local disturbances for appropriate material properties . these possible disturbances , though local and tiny , are essentially related to the nature of the sandpile , which is actually a bunch of rocks and powders in a jammed state . the sandpile is formed by a competition between the constituents to fall under gravity and eventually reach an equilibrium with mutual support . it can hold a stable shape under a constant environment force at low sandpile densities ( compared with close packing ) , which is primarily due to interior collective structures , called bridges , that are formed at the same time as the sandpile and are distributed non - uniformly throughout the sandpile ( mehta 2007 ) . the soft - sphere method provides a detailed approximation to a real sandpile in terms of the granular shape and contact mechanics , which can well reproduce the bridges that dominate the structure of the sandpile . accordingly , we can describe the nuanced responses due to the changes in @xmath83see section [ s : gcme ] and fig . [ f : accs ] . changing the slope angle @xmath85 is the most efficient way to break the equilibrium of substructures in the sandpile , while changing the magnitude @xmath84 and deflection angle @xmath86 may also play a role in the causes of avalanches . changes in @xmath84 can readjust the interparticle overlaps and cause collapse of some bridges , which is a principal mechanism of compaction by filling the voids during small structural modifications ( mehta et al . this dependence was recently demonstrated in experiments of measuring the angle of repose under reduced gravity ( kleinhans et al . similarly , the weak links in bridges of the sandpile may also be disturbed and broken during the sweeping of @xmath86 in the plane . in addition , mehta et al . ( 1991 ) pointed out that even slight vibration caused by the environment force may result in collapse of long - lived bridges , which is also responsible for the sandpile landslides during the encounter . although the magnitudes of the avalanches are different for the three materials we tested , they are nonetheless all very small . we note the occurrence of these local avalanches shows little dependence on the orientation / location of the sandpile , because all the perturbations are small in magnitude and the sandpile behavior , these small - scale avalanches , actually depends on the presence of weak bridges inside the sandpile , which in turn could be sensitive to the changes of the environmental force s magnitude , deflection angle , and slope angle in a somewhat random way . in any case , this numerical study shows that tidal resurfacing may not be particularly effective at moderate encounter distances . we predict that overall resurfacing of apophis , regolith will not occur if the only source of disturbance is external perturbations . mini - landslides on the surface may still be observed by a visiting spacecraft with sufficiently sensitive monitoring equipment . this provides a great motivation for @xmath0 @xmath1 exploration of apophis in 2029 ( michel et al . this paper provides a numerically derived prediction for the surface effects on ( 99942 ) apophis during its 2029 approach , which is likely to be one of the most significant asteroid encounter events in the near future . a two - stage scheme was developed based on the soft - sphere code implementation in ` pkdgrav ` to mimic both a rubble pile s ( rigid and flexible ) responses to a planetary flyby and a sandpile s responses to all forms of perturbations induced by the encounter . the flyby simulations with the rubble pile indicate that reshaping effects due to the tidal force on apophis in 2029 will be negligibly small for bulk densities in the expected range ( @xmath78@xmath79 @xmath12 ) . the resultant environmental force felt by the sandpile on the asteroid surface was approximated with a uniform analytical expression , which led to an estimate of the changes in the global mechanical environment . three typical patterns of perturbation were presented based on the asteroid body and spin orientation at perigee , showing a general dependence of the magnitude of tidal perturbation on the orientation of the trajectory . twelve fiducial trajectories were used to calculate the magnitude of the tidal perturbation at three poles of the tri - axial ellipsoid model , indicating that the strongest tidal perturbation appears where the local slope is originally steep and that the duration of the strong perturbation is short compared with the whole process . the tidal perturbation on surface materials is confirmed to be quite weak for the 2029 encounter , therefore large - scale avalanches may never occur . however , we showed this weak perturbation does trigger some local tiny landslides on the sample sandpiles , though the involved particles are few in number and are distributed on the surface of the sandpile . these small - scale avalanches result from the breaking of weak substructures by slight external perturbations , therefore the occurrence of these local avalanches is widely distributed and presents little dependence on the encounter parameters . the simulations of closer approaches show that an encounter at @xmath17 @xmath52 is capable of triggering some massive avalanches of the sandpiles , i.e. , to alter the regolith on apophis surface significantly ( although the entire body would also undergo significant shape change in this case ) . further research will be performed to generalize our work over a wide range of possible asteroid - planet encounter conditions . and as stated above , we will also investigate whether even slight internal perturbations in the asteroid during tidal encounters may contribute to noticeable surface motions . this material is based on work supported by the us national aeronautics and space administration under grant nos . nnx08am39 g , nnx10aq01 g , and nnx12ag29 g issued through the office of space science and by the national science foundation under grant no . some simulations in this work were performed on the yorp " cluster administered by the center for theory and computation in the department of astronomy at the university of maryland . patrick michel acknowledges support from the french space agency cnes . ray tracing for figs.1 , 2 , 4 , 5 and 12 was performed using pov - ray ( http://www.povray.org/ ) . 1999 catastrophic disruptions revisited icarus 142 520 2009 spectral properties and composition of potentially hazardous asteroid ( 99942 ) apophis icarus 200 480485 2010 earth encounters as the origin of fresh surfaces on near - earth asteroids nature 463 331334 1966 principles of powder mechanics oxford : pergamon 2007 albedo and size determination of potentially hazardous asteroids : ( 99942 ) apophis icarus 188 266269 2013 mars encounters cause fresh surfaces on some near - earth asteroids icarus 227 112122 2006 the rubble - pile asteroid itokawa as observed by hayabusa science 312 13301334 2008 predicting the earth encounters of ( 99942 ) apophis icarus 193 119 2006 tidal disruptions : a continuum theory for solid bodies icarus 183 331348 2004 numerical determination of stability regions for orbital motion in uniformly rotating second degree and order gravity fields planetary and space science 52 685692 2010 fragment properties at the catastrophic disruption threshold : the effect of the parent body s internal structure icarus 207 5465 2011 static and dynamic angles of repose in loose granular materials under reduced gravity journal of geophysical research 116 e11004 2010 do planetary encounters reset surfaces of near earth asteroids ? icarus 209 510519 2007 granular physics cambridge university press 1991 vibrated powders : a microscopic approach physics review letter 67 394397 2004 cooperativity in sandpiles : statistics of bridge geometries journal of statistical mechanics : theory and experiment p10014 2001 collisions and gravitational reaccumulation : forming asteroid families and satellites science 294 16961700 2012 probing the interior of asteroid apophis : a unique opportunity in 2029 special session 7 of the xxviii iau general assembly , 2931 august , 2012 , beijing ( china ) , abstract book 1998 tidal distortion and disruption of earth - crossing asteroids icarus 134 4776 2000 direct large - scale @xmath2-body simulations of planetesimal dynamics icarus 143 4559 2009 numerical simulations of asteroids modeled as gravitational aggregates planetary and space science 57 183192 2011 numerical simulations of granular dynamics : i. hard - sphere discrete element method and tests icarus 212 427437 2012 numerical simulations of landslides calibrated against laboratory experiments for application to asteroid surface processes american astronomical society , dps meeting 44 105.06 2005 abrupt alteration of asteroid 2004 mn4 s spin state during its 2029 earth flyby icarus 178 281283 2012 an implementation of the soft - sphere discrete element method in a high - performance parallel gravity tree - code granular matter 14 363380 2013 numerically simulating impact disruptions of cohesive glass bead agglomerates using the soft - sphere discrete element method icarus 226 6776 2014 low - speed impact simulations into regolith in support of asteroid sampling mechanism design i. : comparison with 1-g experiments planet . space sci . 2001 cosmological @xmath2-body simulations and their analysis university of washington 141 2009 rubble - pile reshaping reproduces overall asteroid shapes the astrophysical journal 706 197202 2013 improved astrometry of ( 99942 ) apophis acta astronautica 90 5671 2000 near at eros : imaging and spectral results science 289 20882097 2012 spin - up of rubble - pile asteroids : disruption , satellite formation , and equilibrium shapes icarus 220 514529 2013 the potentially dangerous asteroid ( 99942 ) apophis monthly notices of the royal astronomical society 434(4 ) 30553060 .soft - sphere parameter sets used to represent three different material properties of sandpiles in the simulations . here @xmath24 is the coefficient of static friction ( @xmath114 , where @xmath115 is the material friction angle ) , @xmath25 is the coefficient of rolling friction , @xmath22 is the normal coefficient of restitution ( 1 = elastic , 0 = plastic ) , @xmath23 is the tangential coefficient of restitution ( 1 = no sliding friction ) , @xmath18 is the normal spring constant ( @xmath116 ) , and @xmath19 is the tangential spring constant ( also @xmath116 ) . [ cols="^,^,^,^",options="header " , ] * figure [ f : sandpile ] : * snapshots of sandpiles constructed using three different materials , which are ( a ) gravel , " ( b ) glass beads , " and ( c ) smooth , " with corresponding soft - sphere parameters listed in table [ t : ssdemparams ] . the same constituent spheres are used in each sandpile both for the free particles ( green ) and the rigid particles ( white ) . the values of average slope and pile height after equilibrium is achieved are indicated in the snapshots . snapshots of the avalanche experiment and corresponding simulations . in the experiments , @xmath117 similar - size gravel pieces were piled up on the slope each time and released all at once by removing the supporting board . snapshots from both the experiment and simulation include frames from the beginning ( a , d ) , middle ( b , e ) , and end ( c , f ) of the avalanche event , respectively . sketch of the frames used for deriving the motion equations of the local sandpile , indicated with different colors . spc ( black ) is an inertial frame with the origin and axes set by ` pkdgrav ` . mct ( blue ) is a noninertial translating frame with the origin at the mass center of the asteroid and @xmath118 , @xmath119 , @xmath120-axes parallel to spc . bdy ( green ) is a noninertial frame that is fixed to the asteroid and initially coincides with mct . loc ( red ) is a frame fixed to the asteroid surface , taking the origin to be where the sandpile is located and choosing the @xmath120-axis to be normal to the surface at that point . diagram of asteroid ( 99942 ) apophis. the tri - axial ellipsoid ( light gray ) denotes the overall shape model , while the arranged particles ( khaki ) denote the rubble - pile model used for numerical simulations . the highlighted particles ( red ) are markers used for determining the attitude of the asteroid in spc . snapshot of the rubble - pile model for apophis with bimodal particles in irregular packing . the tri - axial ellipsoid ( light gray ) denotes the overall shape model , and the arranged particles ( khaki ) show the rubble - pile model used for numerical simulations . the relative change of shape factor @xmath74 as a function of bulk density during the encounter . the solid line ( with triangular points ) shows the results for the original rubble pile with equal - sized particles . the dashed line ( with square points ) shows the results for the bimodal rubble pile with irregular packing . sketch of the environmental force that the sandpile feels in loc . the plane ( gray ) indicates the local tangential plane . the red arrow indicates the environmental force given by eq . ( [ e : envirforce ] ) . label @xmath84 indicates its magnitude , and @xmath85 and @xmath86 indicate the slope angle ( elevation ) and deflection angle ( azimuth ) of the effective gravity , respectively . sketch of the fiducial trajectories in different orientations . these trajectories ( blue numbered lines ) are placed in the perpendicular planes ( yellow ) of the apophis model ( gray ellipsoid ) , matching the three axial directions at perigee . the blue arrows show the prograde and retrograde directions . the red arrow denotes the angular velocity . p@xmath100 , p@xmath101 , p@xmath102 indicate the poles along the three axes @xmath118 , @xmath119 , @xmath120 , respectively . maximum values of slope angle change for the fiducial trajectories . the stars denote the maximum values , and the dashed lines denote the levels of tidal perturbation ( the average of the 4 values in each case ) , which divide the @xmath92 trajectories into @xmath79 basic catagories according to the spin orientation at perigee . see fig . [ f : traject ] for the trajectory orientations . global distribution of slope angle change at perigee for the three trajectory sets a , b and c , each including 4 trajectories of the same spin orientation at perigee ( see text for an explanation of each set ) . a uniform colormap ranging up to @xmath121 is used for the three plots . time variation of the effective gravity slope angle at the poles p@xmath100 ( dashed lines ) , p@xmath101 ( solid lines ) , and p@xmath102 ( dotted lines ) during the encounter . each subgraph corresponds to one fiducial trajectory marked with its number on the upper left ( refer to fig . [ f : traject ] ) . snapshots of sandpiles after the tide - induced avalanches in 2029 , with corresponding time variations of their total kinetic energy . the three panels include the results of sandpiles constructed using the three materials ( a ) gravel , " ( b ) glass beads , " and ( c ) smooth , " respectively . particles that moved more than @xmath69 of a particle radius during the avalanches are highlighted . the peaks in the total kinetic energy curves indicate occurrence of avalanches ; the peaks are labeled , and corresponding particles that moved appreciably are marked in different colors ( red , blue , orange ) in the snapshots . the dotted line in each graph shows the evolution of the planet - asteroid distance during the encounter . snapshots of the sample sandpiles before and after the avalanches induced by the encounter at @xmath16 @xmath52 perigee , with corresponding time variations of their total kinetic energy . the white lines in the snapshots indicate the original and final shapes of the sandpiles . the peaks in the total kinetic energy curves indicate occurrence of avalanches , and the dotted line shows the variation of the planet - asteroid distance during the encounter .

the paper develops a method for discrete computational fourier analysis of functions defined on quasicrystals and other almost periodic sets . a key point is to build the analysis around the emerging theory of quasicrystals and diffraction in the setting on local hulls and dynamical systems . numerically computed approximations arising in this way are built out of the fourier module of the quasicrystal in question , and approximate their target functions uniformly on the entire infinite space . the methods are entirely group theoretical , being based on finite groups and their duals , and they are practical and computable . examples of functions based on the standard fibonacci quasicrystal serve to illustrate the method ( which is applicable to all quasicrystals modeled on the cut and project formalism ) .

0808.1814

in this paper we consider the problem of discrete methods for dealing with functions that are intrinsically almost periodic , but not actually periodic . quasicrystals , quasicrystalline photonic crystals , faraday wave experiments , and other physical phenomena arising from the interaction of incommensurate frequencies , all display the features of almost periodicity . as a typical example one may think of a potential field of a physical quasicrystal . the salient features of quasicrystals are highly structured long - range order ( represented by pure point or near pure point diffraction ) but no periodic order . thus the potential is not a periodic function , but rather belongs to the domain of almost periodic functions . here we put forth a method for finite discrete analysis of almost periodic functions that has the following main features : * it is entirely based on group theoretical methods , primarily finite groups and their duals . * the discretely computed fourier approximants are themselves almost periodic and uniformly approximate their target functions over their entire domains . * the fourier frequencies involved in the approximation lie in the module of fourier frequencies of the target function . a standard approach to modeling such structures is to take a finite part of it , impose periodic boundary conditions , rationalize and reduce the object to a periodic approximant , and then apply usual crystallography . although this type of periodization is used routinely and successfully for many modelling problems in the theory of quasicrystals , it is not entirely satisfactory . almost periodic order goes beyond periodic order in fundamental ways , its essence appearing as a underlying incommensurability which pervades every part of the theory . for instance , a key feature of quasicrystals appears in the fourier module which parameterizes the bragg spectrum and always has rank higher than ( typically double ) the dimension of the ambient space of the quasicrystal . periodization destroys this , and by its nature can only produce results that can fit data over the finite range specified by the imposed periodization boundaries , whereas the essence of the material is that its order is long - range . the present paper does not involve any periodization and avoids these issues . the theory of almost periodic functions was initiated by h. bohr @xcite , on the basis of earlier work on uniform approximation of functions by trigonometric polynomials by p. bohl . it was greatly extended by the work of a. besicovitch @xcite , s. bochner , j. von neumann , @xcite , n. wiener @xcite , h. weyl @xcite , and others @xcite . the advent of quasicrystals and aperiodic tilings instigated a revival of the field , and led to extensive study of the cut and project formalism and the theory of pure point diffraction @xcite which have become the mainstays of experimentalists and theorists alike . an important component of this is the use of dynamical systems and dynamical hulls . these are ideally suited to the phenomenon of almost periodicity , which appears in the dynamics as recurrence , and provide a natural setting for the fourier analysis that is used to study it . although our study is of almost periodic functions , their importance in the subject of quasicrystals is that they arise from functions whose behaviour is dominated by the local environments of the quasicrystal in question . the way in which this happens is mostly taken for granted , but in fact there are some interesting assumptions involved , and for this reason we begin by formally defining local functions with respect to a given point set @xmath0 , and showing how it is that they are connected with almost periodic functions . we can create a _ dynamical hull _ @xmath1 from @xmath0 . this is a compact space that arises from @xmath0 and its translations , and it lies at the heart of the fourier analysis of local functions on @xmath0 . one assumes ( in many important cases it is forced ) a probability measure @xmath2 on which @xmath3 acts in a measure preserving way . a local function @xmath4 lifts to some new function @xmath5 on @xmath1 , and this is an @xmath6 function . now the analysis of @xmath4 can be related directly to the analysis of @xmath5 , and for this we have a powerful tool in the form of the action of @xmath3 on @xmath7 , which is unitary . all of this material is explained in 2 . to go further , we next place ourselves in the situation of the cut and project formalism , which is the standard method of modeling used in the study of quasicrystals . the set @xmath0 is now assumed to be a model set ( cut and project set ) . in this setting the hull @xmath1 , and more particularly @xmath7 , can be described explicitly in terms of a higher dimensional torus(higher dimensional periodicity ! ) , and it is a straightforward matter to carry out fourier analysis of @xmath5 . it is the restriction of this fourier analysis back to @xmath4 that provides the required fourier analysis of @xmath4 . by its very nature this is almost periodic and captures the full aperiodic nature of @xmath0 , including the correct fourier module in which physical information actually appears . this is the content of 3 . readers familiar with the cut and project method who do not wish to go through the theory of local hulls and local functions may read [ cpsmstp ] and [ fouriercoefficients ] , and then continue starting at [ discretization ] . this theoretical analysis is based on higher dimensional structures that are not explicitly computatable , as well as the usual array of countably many fourier coefficients , each of which is the outcome of integration . to be a practical tool , the analysis has to be reduced to finitely many objects that are explicitly computable entirely in the context of the given function @xmath4 . the resulting approximants are trigonometric polynomials ( quasi - periodic functions ) whose frequencies come from the fourier module of the original function . in 4 we outline the method of discretization , which depends primarily on the construction of a refinement lattice of the lattice of the cut and project scheme and , along with it , its dual lattice . together these produce two finite groups which are in @xmath8-duality to each other . the selection of data points and appropriate fourier frequencies is governed by these two groups . this is best illustrated by examples and for this we have chosen two local functions based on one of the famous fibonacci point sets . this has the advantage of being straightforward to construct and easy to visualize , while at the same time containing all the essential features of more general model sets . 5 prepares the mathematics of the fibonacci cut and project scheme , and 6 shows some explicit computations for the particular local functions we have chosen . as is evident from the examples , the main effort required is in the creation of the data points . once this is done , the same set of data points and fourier frequencies will work for any almost periodic function arising from the same cut and project scheme . the results are striking in two ways . first of all , the approximating functions are remarkably good , given the amount of data from which they are produced . second , the approximating functions are not just local approximations they are also _ global _ approximations , in the sense that they provide finite fourier series that approximate @xmath4 throughout its entire domain ( namely @xmath9 ) . of course this was to be expected , but it is impressive to see it in action . the numerical methods we introduce here are designed to be efficient and , of course , to utilize the inherent almost periodicity . in the development of the theory we provide error estimates that can be worked out specifically in the cases of interest . we note that the primary weakness in the error estimates is not one that is due to the aperiodic nature of the problem , but one that always appears in fourier analysis , namely how well can one approximate a function if one uses only finitely many of its fourier coefficients . [ hulls ] we work in @xmath3 , a real euclidean space of finite dimension @xmath10 . for @xmath11 , let @xmath12 , be all point sets @xmath13 for which the distance @xmath14 for all @xmath15 . this means that @xmath16 consists of all discrete point sets with minimal separation @xmath17 between their points . the _ local topology _ on @xmath12 can be intuitively introduced as follows . two sets @xmath18 and @xmath19 of @xmath12 are ` close ' if , for some large @xmath20 and some small @xmath21 , one has @xmath22 where @xmath23 ( resp . @xmath24 ) is the ball of radius @xmath20 ( resp . @xmath21 ) around @xmath25 . thus for each point of @xmath18 within the ball @xmath23 , there is a point of @xmath19 within the distance @xmath21 of that point ; and vice versa . pairs @xmath26 satisfying are called @xmath27-close . and @xmath19 of @xmath12 that are close . ] the local topology is actually a metric topology , although we make no use of this fact here . for @xmath28 the _ local hull _ of @xmath0 is @xmath29 i.e. take all translates of @xmath0 and take their closure in the local topology . @xcite the translation action of @xmath3 on @xmath0 lifts to a translation action on @xmath30 . the local hull @xmath30 is compact and the @xmath3-action on it is continuous . let @xmath0 be the lattice @xmath31 in @xmath3 . then @xmath32 is the @xmath10-torus ( with its usual topology ) . intuitively we translate the lattice around with @xmath3 . but translation by an element of @xmath31 leaves @xmath0 invariant . so @xmath33 parameterizes all distinct positions of @xmath0 under translation . let @xmath0 be any penrose tiling . then @xmath30 is the set of all penrose tilings that are locally indistinguishable from some translate of @xmath0 . @xmath30 contains considerably more than just the translations of @xmath0 . in fact @xmath30 consists of all penrose tilings based on the same pair of penrose rhombs and the same orientations as appearing in @xmath0 . generally one may think of @xmath30 as some sort of local indistinguishability class of @xmath0 . we assume that @xmath34 and @xmath30 are as in [ hulls ] . consider a function @xmath35 we can define from it a function @xmath36 by @xmath37 if @xmath5 is continuous then we note that for all @xmath38 , @xmath39 thus continuity of @xmath5 implies continuity of @xmath4 , and we see that @xmath4 is local , or almost periodic , with respect to @xmath0 in the following sense : a function @xmath40 is called _ local with respect to a set @xmath28 _ , or @xmath0-_local _ , if for all @xmath41 there exist @xmath20 and @xmath21 so that whenever @xmath38 satisfy that @xmath42 and @xmath43 are @xmath27-close then latexmath:[\[\begin{gathered } the intuitive meaning of this is that @xmath4 has the very natural property , at least from the perspective of physical systems , that it looks very much the same at places where the local environment looks the same . local functions are easily seen to be continuous on @xmath3 . using locality , we can go in the opposite direction . let @xmath28 and let @xmath45 be local with respect to @xmath0 . define @xmath46 ( so that @xmath5 is a function on a part of @xmath12 ) by @xmath47 then @xmath5 is continuous on @xmath48 with respect to the local topology . in fact it is _ uniformly _ continuous . the reason for this is that the continuity condition which defines the local - ness of @xmath4 is based on the uniformity ( i.e. the notion of @xmath27-closeness ) defining the local topology of @xmath48 . it follows that @xmath5 lifts uniquely to a continuous function on the local hull @xmath30 . for each local function @xmath4 with respect to @xmath0 there is a unique continuous function on the local hull , whose restriction to the orbit of @xmath0 is @xmath4 . every continuous function on the local hull of @xmath0 arises in this way . thus we see that a locality with respect to @xmath0 and the existence and continuity of an extension function on @xmath30 amount to the same thing . in the situation that @xmath30 is equipped with an @xmath3-invariant probability measure @xmath2 ( i.e. a positive borel measure @xmath2 with @xmath49 for all borel sets and with @xmath50 ) , the action of @xmath3 on @xmath30 leads to unitary action @xmath51 of @xmath3 on @xmath52 . namely for all @xmath53 and for all @xmath54 , @xmath55 is the function defined by @xmath56 and with @xmath57 we have @xmath58 in principle the spectral theory of @xmath52 should allow one analyze @xmath0-local functions @xmath59 by analyzing their corresponding functions @xmath5 on @xmath52 . for one very important class of subsets @xmath0 this can actually be carried out in detail namely the class of model sets , which we now introduce . an important class of point sets @xmath60 for which we know a considerable amount about the corresponding hulls @xmath30 is the class of cut and project sets , or the model sets as they are often called @xcite . consider the cut and project scheme @xmath61 and a window @xmath62 . here @xmath63 is a lattice in @xmath64 which is oriented so that the projections into @xmath3 are @xmath65 and dense respectively . in ( [ cps ] ) the left - hand @xmath3 is _ physical space _ , the space in which @xmath0 is going to lie . the right - hand @xmath3 is _ internal space _ , the one that will be used to control the projection of the lattice @xmath63 into physical space . the image of @xmath63 under projection into physical space is denoted by @xmath66 . since this projection is one - one , @xmath67 as groups , so @xmath66 is a free abelian group of rank @xmath68 , i.e. it has a @xmath8-basis of @xmath68 elements . however , it necessarily has accumulation points , and the typical situation is that @xmath66 is dense in physical space . it is convenient to use notation like @xmath69 for the elements of @xmath70 and their respective left and right projections . then @xmath71 where @xmath72 runs through @xmath66 . this implies existence of the mapping @xmath73 defined by @xmath74 , which passes from physical to internal space . note that this mapping @xmath75 , as given , is only defined on @xmath66 . it can not be extended in any canonical way to a mapping @xmath76 . however it does extend canonically to the rational spans of the objects in question , and we shall make use of this later . we choose a subset @xmath62 in internal space . window _ is assumed to be compact , equal to the closure of its interior , and to have boundary of measure 0 . using it we define @xmath77 sets of the form @xmath78 , @xmath54 , are called _ cut and project sets _ or _ model sets _ . in particular , for each @xmath79 we may define @xmath80 if @xmath81 , then @xmath82 , as can be verified directly from the definitions . thus these model sets are parameterized by the torus of dimension @xmath68 , @xmath83 we note , though , that the parametrization need not necessarily be @xmath65 , i.e. in general , @xmath84 . in the sequel , for simplicity , we shall usually write @xmath85 for the congruence class @xmath86 there is a natural measure , the haar measure , @xmath87 on @xmath88 . this measure is the obvious ` area ' measure in the case of @xmath89 , and ` length ' measure for @xmath90 . it is invariant under the @xmath3-action : @xmath3 acts on @xmath91 by @xmath92 of particular importance to us is natural embedding ( see fig . 2 . ) @xmath93 which lies behind the connection between almost periodicity in physical space and periodicity in some higher dimensional setting . the image of this mapping is easily established to be dense in @xmath88 . now start with @xmath94 , and translate it by elements @xmath54 : @xmath95 form the local hull @xmath30 , the closure of the set of all translates @xmath96 of @xmath0 under the local topology . @xcite there is a continuous mapping , @xmath97 called the torus parametrization , such that * @xmath98 is onto ; * @xmath98 is 1 - 1 almost everywhere , seen from the perspective of haar measure on @xmath88 ; * for all @xmath54 , for all @xmath99 , one has @xmath100 . * @xmath101 for all @xmath54 . @xmath30 and @xmath88 are both compact spaces with natural @xmath3 action , and @xmath102 is an onto @xmath3 mapping . but @xmath30 and @xmath88 are subtly different . although @xmath103 and @xmath104 for all @xmath105 , not every element of @xmath30 is a @xmath106 for some @xmath107 . rather , when @xmath108 , then for all @xmath109 there are always @xmath110 of elements of @xmath30 that are mapped by @xmath98 to the same @xmath85 . when @xmath111 with @xmath112 then there will always be @xmath113 with @xmath114 , yet @xmath115 @xmath116 . in other words , there are ambiguities regarding the lattice points that project onto the boundary @xmath117 of @xmath62 . an example of this is shown in the footnote appearing in [ ssec_fibonacci_model_set ] . the fact that @xmath98 is @xmath65 almost everywhere is due to our assumption that the boundary of @xmath62 has measure 0 . when we say that @xmath98 is 1 - 1 almost everywhere , we mean that the set @xmath118 of points @xmath119 in @xmath88 , for which there is more than one point set over @xmath119 , satisfies @xmath120 . there is a _ unique @xmath3-invariant ergodic measure _ @xmath2 on @xmath30 with @xmath50 . in fact @xmath98 relates @xmath2 and @xmath121 : @xmath122 or more specifically , @xmath123 for all measurable subsets @xmath118 of @xmath88 . with @xmath2 in hand , we can introduce the space @xmath52 of square integrable functions on @xmath30 . as already pointed out , the natural action of @xmath3 on this is unitary . square integrable functions on @xmath30 and square integrable functions on @xmath88 can be identified , @xmath124 the isomorphism is easy to understand : x ( ) ^ _ g & ^g + & gives us a map @xmath125 since @xmath98 is almost everywhere 1 - 1 , the map is a bijection . this allows us to analyze functions on @xmath30 by treating them as functions on @xmath88 . the advantage of this is that functions in @xmath126 have fourier expansions @xmath127 here @xmath128 is the natural dot product on @xmath129 and @xmath130 is the lattice which is @xmath8-dual to @xmath131 : @xmath132 assuming that the dot product is rational valued on @xmath133 , this new lattice is in the @xmath134 span of @xmath131 and its elements @xmath135 have the same type of decompositions @xmath136 as elements of @xmath131 . in particular , for each @xmath135 there is a unique @xmath137 . we prepare for the ultimate reduction of everything to the physical space by already using the symbols @xmath138 rather than @xmath139 for the coefficients of the fourier expansion . we shall write @xmath140 for the projection of @xmath141 on the physical side , so we can just as well write @xmath142 as @xmath143 . we will call @xmath140 the dual ( or @xmath144$]-dual ) of @xmath66 . the mapping @xmath145 here is compatible with the one on @xmath66 , indeed @xmath146 . the corresponding functions on @xmath30 have similar expansions . this works as follows : if @xmath147 corresponds to @xmath148 then , for @xmath149 with @xmath150 , we have as @xmath6-functions @xmath151 let @xmath0 be a model set arising from the cut and project scheme of . let @xmath152 be the torus with torus parametrization @xmath153 then we have the identification of the corresponding @xmath6 spaces . each element @xmath99 maps by @xmath98 to a point @xmath154 in @xmath88 . we also know @xmath155 , and @xmath156 . so we know how @xmath98 works on @xmath157 . suppose @xmath4 is a local function with respect to the model set @xmath0 . from the local function @xmath4 we have its extension @xmath53 which is continuous . then we obtain @xmath158 , where @xmath159 and we can write @xmath160 where @xmath140 is the dual of @xmath66 . the _ fourier - bohr expansion _ of the local function @xmath4 is @xmath161 as seen in the torus parametrization . ] our study of almost periodic functions on @xmath0 becomes the study of functions on @xmath88 and their restrictions to the ` spiral ' orbit @xmath157 in @xmath88 given by the embedding ( [ torusembedding ] ) , i.e. restriction to @xmath162 . let us continue with the situation in [ modelsets ] . for all @xmath163 , @xmath164 unfortunately , we do nt have total control over @xmath165 . we know it only on @xmath162 . to compute @xmath138 out of @xmath4 alone , we use the birkhoff ergodic theorem or @xmath9 . in @xcite there is a proof over @xmath31 which is easy to adapt to @xmath3 . ] : for all continuous functions @xmath166 on @xmath88 , @xmath167 thus @xmath168 here we use @xmath169 and @xmath170 , so @xmath171 the averaging sequence @xmath172 that we have employed here can be replaced by any unbounded ascending sequence @xmath173 , where @xmath174 is compact , @xmath175 , and the boundary of @xmath176 has measure 0 for all @xmath177 . in actual practice one should adapt the averaging sequence to the problem at hand . our objective is to devise a discrete method by which to estimate the coefficients @xmath138 of of a local function @xmath4 with respect to a model set @xmath178 . this means replacing the integral by a finite sum of values of the integrand . the problem is to do this in such a way that it respects the cut and project scheme in which the model set lives , can be guaranteed to converge in the limit to the required integral , and can be carried out efficiently from a computational point of view . there are three components to this : * deciding on a suitable domain of integration ( what should we use for @xmath23 ? ) ; * creating the points of evaluation of the integrand , including how many there should be ; * deciding to which set of fourier coefficients ( which values of @xmath179 ) we should restrict our attention . a key feature of discrete methods involving periodic functions is the use of finite groups arising from refinements of the period lattice and quotients of its dual lattice , e.g. @xcite . we need to translate this concept into the context of cut and project schemes . the set of points on which the integrand is evaluated is created out of the same cut and project process that creates the original model set . the ingredients are a choice of a suitable lattice @xmath180 , which then gives rise to the finite group @xmath181 . the data points in @xmath3 at which computations of our functions will be made come by projection into physical space of a suitable set of coset representatives of @xmath182 modulo @xmath131 . the corresponding frequencies ( wave vectors ) @xmath179 are chosen from the dual lattice @xmath140 . the choice of values of @xmath179 at which we should evaluate the fourier coefficients @xmath138 come by selecting suitable representatives of @xmath141 modulo @xmath183 . the key point is the duality @xmath184 in this section we give more precise details as to how the goals of [ tdm ] can be achieved . the process necessarily involves a number of decisions , which can only be made in the context of the situation at hand . in [ fibonacci ] and [ twoee ] we shall see how this looks in particular examples . it should be noted that although getting the computational details set up is somewhat involved , these details depend only on the cut and project scheme and the degree of accuracy required from the computation . once this is established the data points and choices of frequencies are already determined , and they suffice for the fourier analysis of _ all _ functions that arise out of the same almost periodic family , and are algorithmically easy to compute . we begin with the cut and project scheme with torus @xmath88 and note the natural extension of the mapping @xmath185 to the rational span of the module @xmath66 : @xmath186 we assume that @xmath187 is supplied with the standard dot product ( denoted @xmath188 ) , and then define the dual lattice : @xmath189 then @xmath130 is a @xmath8-module of the same rank as @xmath133 , namely @xmath68 . there is a cut and project scheme dual to of which @xmath130 is the lattice @xcite : @xmath190 we shall use the same type of notation as in for this scheme too . it arises by taking the pontryagin duals of all the groups in , whereupon @xmath130 appears as the dual of the torus @xmath88 . for more on dual cut and project schemes see @xcite . dualizing can be considerably simplified if the inner product on @xmath191 is rational valued on the lattice @xmath63 . this quite often happens in actual practice . for instance , it happens in the fibonacci example below , where the inner product arises from the trace form on @xmath192 $ ] . when this happens one can identify @xmath130 as a subset of @xmath193 and thereby avoid having to find the new @xmath185 mapping . however , in the general situation such simplifications need not exist , and we do not assume them here . in reading what follows it is good to keep in mind the equation @xmath194 which lies at the bottom of the approximation . the discrete fourier analysis is accomplished by a dual pair of finite groups from which the @xmath195 and the @xmath196 will come . the actual values of @xmath195 and @xmath196 are important only modulo the lattices @xmath197 and @xmath131 respectively , and this freedom lies at the heart of the process . the values of @xmath198 should be in the range @xmath118 of our integration , and ideally we would have the corresponding @xmath199 since the @xmath196 are supposed to be representing points of the physical space @xmath3 . however , the latter is not possible , so we attempt to choose the @xmath200 along with @xmath201 as small as possible . the approximation then works by throwing away the term which involves @xmath202 in . the set of values of @xmath195 is constrained primarily by the requirement that the values of @xmath203 should be small , see the discussion after step 5 below . * step 1 * : choose a finite subgroup of @xmath88 of order @xmath204 . this appears in the form @xmath205 where @xmath206 is a lattice refining @xmath63 . the number @xmath204 will determine the number of points of evaluation in approximating the integrals by sums . we have @xmath207 and it has a @xmath8-dual @xmath208 which is of index @xmath204 in @xmath130 . this affords the natural pairing @xmath209 induced by @xmath210 ( and still denoted by @xmath210 ) . using the @xmath185 mappings , we obtain , in the obvious notation , the @xmath8-modules @xmath211 , @xmath212 , @xmath213 , @xmath214 , and an induced pairing @xmath215 replacing @xmath131 by @xmath216 , we have the refined cut and project scheme @xmath217 and similarly its dual . * step 2 * : choose a fundamental domain @xmath218 for @xmath131 . a canonical choice would be the voronoi cell of @xmath63 at 0 , but any other choice is allowable . in the examples below we use the parallelogram defined by a pair of basis vectors of @xmath131 . cover @xmath219 with a set of translates @xmath220 of @xmath218 by elements of @xmath131 . the projection of these cells into the physical space @xmath3 covers it , though in general the projected @xmath220 have many overlaps . * step 3 * : form @xmath221 . this provides a complete set of representatives in @xmath222 for the group @xmath223 . * step 4 * : choose a region @xmath118 which will be the delimiting the range over which the fourier coefficients of @xmath4 will be estimated in the form @xmath224 these integrals have to be computed for values of @xmath179 that come from @xmath225 . take as @xmath118 the image of a finite set of the translates of @xmath218 appearing in step 2 , i.e. @xmath226 for some finite subset of @xmath227 elements of @xmath131 . the choice of @xmath118 is again determined by the problem at hand . we need next to determine a set of data points in @xmath3 which will serve to replace the integrals of step 4 by finite sums . this is the purpose of the next step . as we pointed out above , we are free to translate the elements of @xmath228 by @xmath131 as we please , and we wish to do this so that the projected images are in our region of integration @xmath229 . we also keep in mind that we wish to do this so as to minimize the size of the corresponding @xmath201 . * step 5 * : for each @xmath230 find a @xmath231 for which @xmath232 is minimal . then the set of data points is @xmath233 at this point , for each @xmath142 we have @xmath234 and the resulting approximation of @xmath4 is @xmath235 the set @xmath236 is to run over a complete set of representatives of @xmath237 . the choice seems free , but one may assume that in most cases the lower frequencies ( smaller @xmath203 ) are most essential in approximating @xmath4 by using only finitely many of its frequencies . for this reason we have : * step 6 : * choose @xmath179 for each class of @xmath237 with @xmath203 taken as small as possible . this concludes the broad description of the algorithm . to make all this more concrete we work through the details of a one dimensional example , the well - known fibonacci sequence , where the cut and project scheme lives in two dimensions and the geometry of the data set @xmath238 and the set of translates @xmath239 via a refinement lattice and new windows are easily visualized . this involves first setting up the cut and project scheme in detail , [ fcps ] and [ duallattice ] , and then describing a fibonacci point set arising from the standard fibonacci substitution in terms of it , [ ssec_fibonacci_model_set ] . we then follow steps 1 through 6 of [ 6steps ] , which provide the data points and corresponding frequencies that will work for the analysis of any local function that we may choose . in [ twoee ] we apply this information to two simple examples of local functions to see how well the methods actually work . one of the great virtues of the cut and project method is that it is primarily an algebraic tool and does not require great geometric insight to use it . given that for aperiodic structures in dimension greater than @xmath240 we almost always in the situation of lattices of rank greater than @xmath241 , and hence forced into spaces in dimensions greater than @xmath241 , this type of algebraic formalism is of enormous value . however , in the fibonacci example , in which everything can be done in @xmath242-dimensions , it is useful to see the underlying geometry explicitly . thus we have gone to some effort to show the geometric meaning of the central feature of the method , that is , the creation of the data points , and to show what the approximations look like and how good they are in comparison with exact computation . in actual practice all this is unnecessary . the only part of the algorithm that requires any serious insight into the geometry ( and it is actually trivial in the fibonacci example ) is the selection of the fundamental domain and the translates of it that are to be used . it is their projection that makes the domain in physical space in which the data will lie . let @xmath243 and let @xmath244=\z+\z\tau$ ] . then @xmath244 $ ] is the ring of integers of the field @xmath245=\q[\sqrt5]$ ] and @xmath246 . let @xmath185 on @xmath244 $ ] and @xmath245 $ ] be the conjugation that interchanges @xmath247 and @xmath248 . we define @xmath249}:= \{(x , x')\ : \ x\in\z[\tau]\}\subset\r\times\r\,.\end{gathered}\ ] ] @xmath250}$ ] is a lattice in @xmath251 and its natural projections @xmath252 into @xmath9 provide the set - up for the fibonacci cut - and - project scheme : @xmath253&&\quad\longleftarrow\quad&&\wt{\z[\tau ] } & & \quad\longrightarrow\quad&&\z[\tau ] \\ & x & & \quad\tomaps\quad&\wt x&=(x , x ' ) & & \quad\mapsto\quad & & x'\ , . \end{alignedat}\ ] ] a natural basis of @xmath250}$ ] is @xmath254 , @xmath255 and the standard inner product on @xmath250}$ ] is defined by @xmath256 here the notation @xmath257 indicates taking the rational component @xmath258 of @xmath259 = : a + b\tau$ ] , and @xmath260 in particular , @xmath261 the geometry of the fundamental cell for the lattice @xmath262}$ ] is illustrated in fig . details of the material here can be found in @xcite . the inner product @xmath263 allows us to identify @xmath250}^\circ$ ] inside the rational span of @xmath250}$ ] and to express @xmath264 in terms of @xmath263 . the basis dual to @xmath265 is given by @xmath266 and the dual lattice is @xmath267})^\circ}=\z\wt\omega_1+\z\wt\omega_2 = \tfrac12\wt{\z[\tau]}\,.\ ] ] the elements of @xmath268})^\circ}$ ] are always of the form @xmath170 and we can write @xmath269^\circ:=\{k\ : \ \wt k=(k , k')\in({\wt{\z[\tau]})^\circ}\}\ ] ] then @xmath270})^\circ= \wt{\z[\tau]^\circ}= \{\wt k\ : \ k=(k , k')\in\z[\tau]^\circ\}\,,\ ] ] and @xmath271 becomes @xmath272 , which is more useful notation for the sequel . the 2-torus @xmath88 of the cut - and - project scheme is then @xmath273}\,.\end{gathered}\ ] ] fourier series on @xmath88 are expressed in terms of the characters @xmath274 , namely @xmath275 here @xmath276 can be arbitrary in @xmath251 , but we shall need to compute only with @xmath277 for which @xmath278 is well - defined . for these elements @xmath279 . the product @xmath263 extends to @xmath251 , but care has to be taken . for a typical element @xmath280 , @xmath281 of our superspace @xmath251 the inner product is calculated as @xmath282 where @xmath283 consider @xmath284}^\circ\,,\end{gathered}\ ] ] where @xmath285 . suppose we want to compute @xmath286 , @xmath287 . from @xmath288 where @xmath289 are from , we have @xmath290 and so , @xmath291 we are interested in model sets @xmath0 coming from the cut and project scheme . in the context of , the fourier - bohr expansion assumes the simpler form @xmath292}})\\ = \sum_{k\in\z[\tau]^\circ } a_k \ , e^{2\pi i2\delta kt}\,.\ ] ] we shall use the @xmath138 computed in the form @xmath293 for future use note that @xmath294 the standard @xmath242-letter fibonacci sequence is the fixed point of the substitution @xmath295 , @xmath296 : namely , @xmath297 with tile lengths @xmath298 for @xmath258 symbols and @xmath240 for @xmath299 symbols , and starting at @xmath25 , we get the sequence of tiles that cover the non - negative part of the real line . the left - hand ends of these tiles , @xmath300 form an infinite sequence of points on the non - negative real line . this set appears explicitly as the non - negative part of the model set @xmath301 arising from the cut and project scheme .)$ ] , which differs from @xmath0 in the two points coming from the ends of the interval @xmath302 . this is an example of a pair of sets that map to the same place in @xmath88 . this ambiguity shows up in an interesting way later on , see [ sos ] . ] [ torusconfusion ] @xmath0 decomposes as @xmath303 and @xmath304 , which give the left - hand end points of the @xmath258 and @xmath299 points respectively . of the lattice @xmath250}=\z(1,1)+\z(\tau,\tau')$ ] is shown in terms of the standard coordinate system for @xmath251 . when @xmath218 is imagined as a torus by identifying opposite sides and the physical space is wrapped onto it using these identifications , then the @xmath258 and @xmath299 regions shown here indicate which parts of the physical space are lying in @xmath258 or @xmath299 tiles respectively . ] a useful and commonly used way to visualize the distribution of @xmath258 and @xmath299 points is to view them on the torus @xmath88 after the embedding of @xmath9 in @xmath88 . the fundamental cell , see fig . 3 , with opposite edges identified is @xmath88 . to see @xmath0 itself being formed , we start at @xmath305 and trace out @xmath306 , @xmath307 , with the usual rules for exiting and re - entering the fundamental cell . we move continuously . our moving point has three ` whiskers ' attached to it . @xmath308 : a whisker of length @xmath309 facing vertically upwards ; @xmath310 : a whisker of length @xmath311 facing down ; @xmath312 : a whisker of length @xmath309 that faces down , but has an initial gap of size @xmath311 . the rule is this . as our point moves along the line on the fundamental cell , if the @xmath308 whisker hits @xmath313 ( i.e. @xmath313 is close enough to get cut by this whisker ) , we get an @xmath258-point of @xmath0 . if the @xmath310 whisker hits the point @xmath314 , we also get an @xmath258-point of @xmath0 . if the @xmath312 whisker hits the point @xmath314 , we get a @xmath299-point . essentially the window , represented by the whiskers , is carried along with the moving point , and exits and re - enters the fundamental cell with the moving point . the advantage of this point of view is that it allows us to divide the fundamental cell into regions which deliver the @xmath258- and @xmath299-points , and thus to see what our function @xmath315 looks like on the fundamental cell . in fig . 3 the vertical dotted lines are the key thing . our moving point moves to the right and every time it crosses a dotted vertical line , we get a new point of @xmath0 , thus starting a new interval . we stay on that interval until the next crossing . the @xmath258 and @xmath299 regions are indicated in the fig . we now go into the details of how to deal with the fourier analysis using discrete methods . here we follow the six steps outlined in [ 6steps ] . write @xmath131 for @xmath250}$ ] , so @xmath316 . let @xmath317 be a fundamental region for @xmath131 . its volume is equal to @xmath247 . fix any @xmath318 . we want to create a lattice @xmath319 that is a refinement of @xmath131 . in the [ 6steps ] we took @xmath320 with @xmath321=n$ ] . in our present situation we use the most obvious lattices , namely @xmath322 . these actually have index @xmath323 , so there are some slight notational differences between this section and [ 6steps ] . since @xmath324 , so @xmath325 . let @xmath326 , so @xmath327 is a complete set of representatives of @xmath328 . together this completes steps 1 , 2 , 3 of [ generalsetting ] . for understanding the approximations better we also introduce a fundamental region @xmath329 for @xmath330 , chosen so that @xmath331 for step 4 we choose as our delimiting range in @xmath9 an interval @xmath332 $ ] where @xmath20 is a positive real number taken so that @xmath333 taken modulo @xmath131 is on the boundary of @xmath218 . following along the path @xmath306 , @xmath334 , and wrapping around @xmath218 as indicated in fig . 4 , this amounts to stopping at some point @xmath20 where the path is just exiting @xmath218 , so that we have an exact number of passes of @xmath218 . thus the path @xmath306 , @xmath335 involves an explicit set of translates @xmath336 , @xmath337 of @xmath218 . let @xmath338 . let @xmath339 be any continuous local function with respect to the model set @xmath340 , and let @xmath341 be its extension to a continuous function on @xmath342 . generally we are interested in fourier decomposition of @xmath165 and along with it the corresponding decomposition of @xmath4 . thus we wish to compute things like @xmath343 here @xmath344 is just some other continuous function on @xmath88 that is local with respect to @xmath0 . so it suffices to deal with some general continuous function @xmath166 on @xmath88 and its restriction @xmath345 to the line @xmath346 . let @xmath347 our first estimate is @xmath348 since @xmath349 , the error in this is estimated by @xmath350 i.e. the error in our approximation is bounded by @xmath351 . in practice @xmath330 and @xmath204 would be chosen so as to provide a suitably _ a priori _ assigned value of @xmath352 . in order to be useful , the computation must be restricted to values of the function @xmath4 , since this is all that one is given in practice . we wish to use an approximation of the form @xmath353 in accordance with . in this section we indicate the geometry behind step 5 . again , it should be pointed out that in the final analysis much of the detail that appears here need not appear at all in the actual algorithm . the path @xmath354 , wrapped around @xmath218 , is divided by @xmath218 into segments @xmath355 . by the path @xmath354 are shown in the torus . ( opposite sides of the parallelogram coincide . ) ] let @xmath356 be the projections onto internal space of the boundary cutting points of @xmath357 . define @xmath358 the strips @xmath359 formed by passing the interval @xmath360 in internal space through @xmath218 in the direction of the physical axis , form a partition of @xmath218 . through each strip in @xmath218 runs a part @xmath361 of our line @xmath362 . the idea is to use the intervals @xmath360 as windows and the line segments @xmath361 as ( part of ) the physical space for a model set construction based on the lattice @xmath182 . this will then produce the points on @xmath361 which will be our data points for the evaluation of the functions @xmath166 , and then @xmath363 . in other words , we are implicitly using the partial model sets @xmath364 . let @xmath365 and @xmath366 into internal space of the boundary cutting points of the path . the points @xmath367 mark the boundaries of the smaller windows into which the original window @xmath368 is partitioned . ] each @xmath369 lies in exactly one strip @xmath359 . let @xmath370 be its projection onto the line segment @xmath361 . then @xmath371 for some @xmath372 $ ] . thus we obtain @xmath373 on our path @xmath374 . lying in the strip @xmath359 formed by the window @xmath375 are projected onto the segment @xmath361 of the path @xmath376 producing data points @xmath377 , shown here as small crosses . ] we have the estimate . moreover , @xmath378 since @xmath379 thus @xmath380 since @xmath381 and @xmath382 both have second component in the same interval @xmath360 . combining and , we have @xmath383 this provides a method of estimating the integral @xmath384 using only @xmath363 on @xmath332 $ ] along with well chosen points in the interval . the two parameters @xmath204 and @xmath20 control the estimates . in spite of the apparent complexity of strips , what is going on is easy to implement . for each @xmath385 there is a translation vector @xmath386 for which @xmath387 is minimal . the corresonding data point is then @xmath388 . this is step 5 of [ generalsetting ] . in this section we apply the methods outlined above to two almost periodic functions @xmath389 , both local with respect to the model set @xmath340 of [ ssec_fibonacci_model_set ] . the first is the distance - to - the - nearest neighbour function @xmath390 this is a continuous and piecewise linear function that is local with respect to @xmath0 . the second is the function @xmath391 which local , but only piecewise continuous with breaks wherever @xmath72 switches from a long to a short interval . graphs of these functions are shown ( solid lines ) in fig . 8 and fig . 13 respectively . our objective is to see how well the approximations we have discussed compare with the actual functions when the calculations are done with specific choices of data points . we approximate each of the two functions by a finite number of terms of its fourier - bohr expansion , which in our setting here reads : @xmath392^{\circ}}}a_k e^{2\pi i(k|x)}\ , , \qquad x\in\r\,.\end{gathered}\ ] ] for the two functions @xmath4 chosen here , it is easy to determine the corresponding functions @xmath393 on the torus @xmath394 and hence to compute the fourier - bohr coefficients @xmath138 exactly by , see [ exact ] . for the nearest neighbour function these are given explicitly in . on the other hand , the approximation depends on the choice of the refinement lattice @xmath182 , which shall here always be of the form @xmath395}$ ] ( so the index @xmath396 $ ] is @xmath323 ) . this determines both the points that will eventually be projected into our data points and also the values of @xmath179 which shall be included in approximating the sum , namely a set @xmath397 of @xmath398}^\circ $ ] chosen as representatives of the group @xmath250}^\circ / n\wt{\z[\tau]}^\circ$ ] dual to the group @xmath399})/\wt l$ ] . our choice is take the @xmath195 with @xmath203 as small as possible , so @xmath236 consists of elements @xmath400 $ ] , one for each congruence class of @xmath144 $ ] modulo @xmath401 $ ] , chosen so that @xmath203 is minimal in its class . we denote by @xmath402 the finite series , taken from , approximating @xmath403 , @xmath404 : @xmath405 next we replace the coefficients @xmath138 in @xmath402 by their approximations @xmath406 following from ( [ coeff2 ] ) , and denote the resulting function by @xmath407 . we shall use various values of @xmath20 , all of which correspond to a set of complete passes across the fundamental region , as illustrated in fig . the integrals are to be estimated by reducing them to finite sums where the integrand is computed on the finite set of data arising as projections @xmath408 of the @xmath323 points in @xmath218 , as explained in [ 6steps ] : @xmath409 the resulting approximation to @xmath402 is denoted @xmath410 . this is the approximation that we have been working towards . written out in full it reads : @xmath411 it is computed out of data points in @xmath9 , is a finite sum of exponential functions whose frequencies come from the fourier module of @xmath4 , and utilizes discrete groups arising out of the periodic setting of the underlying cut and project scheme . by its form , @xmath412 is almost periodic and approximates the function @xmath393 everywhere on the real line as it lies wrapped around the torus . thus we have four functions : @xmath413 , all defined for all @xmath404 . the three approximation functions depend on the number @xmath414 of passes of the real line going through the fundamental region @xmath218 ( see [ ssec_discr_of_analysis ] and [ ssec_restric_to_f ] for details ) , and on the number @xmath323 of lattice points of @xmath222 found in @xmath218 . each of them restricts the full summation of the fourier - bohr expansion to the same finite set @xmath236 of frequencies . they differ due to the ways in which the fourier - bohr coefficients are obtained : they are exact in the first case ( or at least as exact as real computation on computers can be ) , are derived from the integral approximation in the second , and come from the finite sum approximation to the integral in the third . the calculations and graphs shown below allow one to compare these functions . the exact coefficients and their approximations by integrals are sufficiently close as to make little difference to the graphs , so the figures are restricted to comparing @xmath415 . by way of comparison , there is one further fourier approximant , of the type that we are obtained by periodically extending @xmath4 from its values on a fixed finite interval @xmath416 of @xmath9 to the entire space @xmath9 . thus we introduce @xmath417 needless to say , @xmath418 can not be expected to have much relationship to @xmath4 outside @xmath416 , and it does nt . in fact , @xmath418 does not appear to be a good choice even on the limited domain @xmath416 . it may be of interest to indicate how we computed the exact fourier - bohr coefficients . we can reorganize fig . 3 , translating the @xmath258 and @xmath299 regions into single blocks ( see fig . 7 ) , a procedure that is familiar from the klotz construction see @xmath4197 often used in studying cut and project sets , @xcite . it is then straightforward to understand the corresponding function @xmath393 on the 2-dimensional torus , i.e. a genuinely periodic function for which @xmath420}}$ ] . and @xmath299 regions of fig . 3 are reorganized here into two rectangles ( heavy lines ) . the function @xmath165 on @xmath88 arising from the local function is supported on these two rectangles . it is constant on vertical lines . along the horizontal axis , on the @xmath299 rectangle the function values start at @xmath25 , increase linearly to @xmath421 at the mid point of the rectangle and then linearly decrease to @xmath25 again . similarly on the @xmath258 rectangle they increase to @xmath422 and then return to @xmath25 . ] furthermore , it is easy to compute the ` exact ' fourier coefficients of @xmath423 . let @xmath424}^\circ$ ] , where @xmath285 . then @xmath425 where @xmath426 the fourier coefficient for @xmath135 is @xmath427 writing @xmath428 we have @xmath429 making the change from the variables @xmath430 and @xmath431 to @xmath72 and @xmath432 , and using the definition of @xmath165 , we obtain @xmath433 note that the factor @xmath434 comes from @xmath435 , @xmath436 . @xmath437 inspection of the aperiodic approximations shows that they are remarkably faithful to the originals . but there is apparently a strange fuzziness in the approximation in the interval @xmath438 $ ] . the explanation for this is quite interesting and shows that the approximation method we use here is sensitive to quite subtle qualities of the aperiodicity . the particular fibonacci set that we have used in our example is _ singular _ , that is to say , it is one at which the torus parametrization is not one - one . as pointed out in footnote of [ torusconfusion ] , @xmath0 is but one of two distinct sets in @xmath30 that map to the same point on the torus @xmath88 under the torus map . the other one differs from @xmath0 only in that it contains @xmath439 but not @xmath440 . the fourier analysis takes place on @xmath88 and treats these two sets equally . however the nearest neighbour functions of the two sets are different . this is a difference that it immaterial to @xmath393 , which our function @xmath412 is approximating , but which the original function @xmath5 can see . what we see is the fourier analysis hedging between the two scenarios . further refinements of the lattice will never improve the situation on this interval . however , if we had used an element of @xmath30 at which the torus map were @xmath65 ( and this is the case with probabilistic certainty if one chooses randomly from @xmath30 ) then this phenomenon would not occur and the approximation would be uniformly valid over the entire space . all one need do is to shift the window so that its end points do not lie in the set @xmath144 $ ] , e.g. @xmath441 $ ] . we should point out that there is considerable scope for adapting the scenario sketched out here . it is possible to arrange things so that they are related to tilings . one elegant tiling method is the klotz construction @xcite . one begins with the decomposition into voronoi cells of the lattice @xmath131 in @xmath191 , and along with it the corresponding dual cell decomposition into delone cells . for each pair of @xmath442 consisting of intersecting @xmath10-dimensional faces @xmath443 and @xmath444 from a voronoi cell and a delone cell respectively , form the _ klotz _ @xmath445 in @xmath191 . the set of these kltze form a tiling of @xmath191 . furthermore the intersection of @xmath446 with this tiling produces a tiling of physical space . ( if one chooses instead to do the projections the other way around , one gets a different tiling ) . several of these kltze can be combined to form a fundamental region for @xmath191 , as we see in fig . 7 , and this type of choice should lead to computational methods that are adapted to the these tilings . for instance , the region @xmath118 of integration might well be chosen as the union of a finite number of tiles . for more on determining voronoi and delone cells in the context of high symmetry , see @xcite . the method advocated here is based on the idea of local functions , the extending of them into the context of compact abelian groups , and the discretization of the resulting fourier analysis by the use of finite groups . in the 1d setting that was explored in detail here , the only symmetries involved arise from translational symmetry ( which is at the base of the almost periodicity ) . in higher dimensions , especially those of interest to the quasicrystal community , decagonal , icosahedral , or other symmetries appear . in these cases there are a number of ways of utilizing the symmetry to greatly improve the efficiency of the computation , in the same spirit as @xcite . as we have pointed out , the preparations required for this depend largely on the cut and project scheme and not so much on the actual model set involved . fortunately the cut and project schemes for these settings are essentially canonical and their strong algebraic nature makes this program quite feasible . finally , although we have not spelled it out here , the way in which finite groups and their duals are used here makes the process amenable to the technique of the fast fourier transform . details will appear later . work supported in part by the natural sciences and engineering research council of canada , the mind research institute of santa ana , calif . , the aspen center for physics , and mitacs . the authors are grateful to the referees for their constructive comments . 14 l. amerio and g. prouse _ almost - periodic functions and functional equations , _ litton educational publishing , new york 1971 . m. baake , d. lenz , and r.v . moody , characterisation of models sets by dynamical systems , _ ergodic th . & dynam . * 27 * ( 2007 ) 341382 ; ` arxiv : math.ds/0511648 ` . a. s. besicovitch , _ almost periodic functions , _ dover publications , cambrige 1954 . s. bochner , _ properties of fourier series of almost periodic functions , _ proc . london math . * 26 * ( 1927 ) 433 - 452 . s. bochner , _ a new approach to almost periodicity , _ proceedings of the nat . usa * 48 * ( 1962 ) 195 - 205 . s. bochner , j. von neumann , _ almost periodic functions of groups ii , _ trans . am . * 37 * ( 1935 ) 21 - 50 . h. bohr , _ almost periodic functions , _ chelsea publishing company , 1947 . p. bohl , _ ber die darstellung von funktionen einer variabeln durch trigonometrische reihen mit mehreren einer variabeln proportionalen argumenten _ , dorpat ( 1893 ) ( thesis ) . burckel , _ weakly almost periodic functions on semigroups , _ gordon & breach , science publishers , new york 1970 . l. chen , r. v. moody , and j. patera , _ non - crystallographic root systems , _ in _ quasicrystals and discrete geometry , _ fields institute monograph series * 10 * ( 1998 ) 135178 , ed.j . patera , amer . a. hof , _ on diffraction by aperiodic structures _ , commun . * 169 * ( 1995 ) 2543 . gerhard keller , _ equilibrium states in ergodic theory _ , lms student texts , 42 , cambridge u. press , cambridge 1998 . p. kramer , _ atomic order in quasicrystals is supported by several unit cells _ , . lett . * b1*(1987 ) 7 - 18 . kramer and m. schlottmann , _ dualization of voronoi domains and klotz construction : a general method for the generation of proper space filling _ , j. phys . * a22*(1989 ) l1097-l1102 . b.m . levitan and v.v . zhikov _ almost periodic functions and differential equations , _ cambridge university press , cambridge 1982 . yves meyer , _ algebraic numbers and harmonic analysis _ , north - holland publ . , amsterdam - london , 1972 . r. v. moody , _ model sets and their duals , _ in _ the mathematics of long - range order , _ ed . moody , nato asi , series * c489 * , kluwer , dordrecht ( 1997 ) 239 - 268 . r. v. moody and j. patera , _ characters of elements of finite order in lie groups _ , siam journal for algebraic and discrete methods , 5 ( 1984 ) 359 - 383 . r. v. moody and j. patera , _ voronoi domains and dual cells in the generalized kaleidoscope with applications to root and weight lattices , _ can . j. math . , * 47 * ( 1995 ) 573605 . moody and j. patera , _ orthogonality within the families of @xmath218- , @xmath447- , and @xmath448-functions of any compact semisimple lie group , _ sigma ( symmetry , integrability and geometry : methods and applications ) * 2 * ( 2006 ) 076 , 14 pages , math - ph/0611020 . c. radin and m. wolf , _ space tilings and local isomorphism _ , geometriae dedicata * 42 * ( 1992 ) 355 - 360 . m. schlottmann , _ generalized model sets and dynamical systems , _ in _ directions in mathematical quasicrystals _ eds . m. baake and r. moody , crm monograph series vol . * 13 * ams providence ri ( 2000 ) pp . 143 - 159 . h. weyl , _ integralgleichungen und fastperiodische funktionen , _ math . ann . * 97 * ( 1926 - 7 ) 473 - 498 . n. wiener , _ on the representation of functions by trigonometrical integrals , _ math . zeitschr . * 24 * ( 1926 ) 575 - 616 . moody : department of mathematics , university of victoria , victoria , british columbia , canada ; [email protected] . m. nesterenko : institute of mathematics , nas of ukraine , 3 tereshchenkivska street , kyiv-4 , 01601 , ukraine ; [email protected] . j. patera : centre de recherches mathmatiques , universit de montral , c.p.6128-centre ville , montral , h3c3j7 , qubec , canada ; [email protected] . .comparison of the calculated coefficients @xmath138 in the approximations @xmath449 , @xmath450 , and @xmath412 based on the parameters of fig [ cols="^ , < , < , < " , ] of . the dotted and dashed curves correspond to the @xmath412 and @xmath450 approximations . both approximations were calculated using a total of @xmath451 lattice points in @xmath218 , @xmath452 passes of @xmath218 , and corresponding interval of integration given by @xmath453 . ] of is drawn for parameters : @xmath454 , @xmath455 , @xmath456 . also the periodic approximation @xmath418 is shown as thin line . the number of cosine terms used is @xmath457 and coefficients were calculated by . here the dotted curve corresponds to the @xmath412 approximation and the dashed curve corresponds to @xmath449 approximation . ]

the question of existence and properties of stationary solutions to langevin equations driven by noise processes with stationary increments is discussed , with particular focus on noise processes of pseudo - moving - average type . on account of the wold karhunen decomposition theorem , such solutions are , in principle , representable as a moving average ( plus a drift - like term ) but the kernel in the moving average is generally not available in explicit form . a class of cases is determined where an explicit expression of the kernel can be given , and this is used to obtain information on the asymptotic behavior of the associated autocorrelation functions , both for small and large lags . applications to gaussian- and lvy - driven fractional ornstein uhlenbeck processes are presented . a fubini theorem for lvy bases is established as an element in the derivations .

0912.3091

this paper studies the existence and properties of stationary solutions to langevin equations driven by a noise process @xmath0 with , in general , stationary dependent increments . we shall refer to such solutions as quasi ornstein uhlenbeck ( qou ) processes . of particular interest are the cases where the noise process is of the pseudo - moving - average ( pma ) type . in wide generality , the stationary solutions can , in principle , be written in the form of a wold karhunen - type representation , but it is relatively rare that an explicit expression for the kernel of such a representation can be given . when this is possible it often provides a more direct and simpler access to the character and properties of the process ; for instance , concerning the autocovariance function . this will be demonstrated in applications to the case where the noise process @xmath0 is of the pseudo - moving - average kind , including fractional brownian motion and , more generally , fractional lvy motions . of some particular interest for turbulence theory is the large and small lags limit behavior of the autocovariance function of the ornstein uhlenbeck - type process driven by fractional brownian motion , which has been proposed as a representation of homogeneous eulerian turbulent velocities ; see shao @xcite . the fractional brownian and lvy motions are not of the semimartingale type . another non - semimartingale case covered is @xmath1 where the processes @xmath2 are brownian motions in different filtrations and @xmath3 is a measure on some space @xmath4 . in recent applications of stochastics , particularly in finance and in turbulence , modifications of classic noise processes by time change or by volatility modifications are of central importance ; see barndorff - nielsen and shephard @xcite and barndorff - nielsen and shiryaev @xcite and references given therein . prominent examples of such processes are @xmath5 , where @xmath6 is brownian motion and @xmath7 is a predictable stationary process for instance , the square root of a superposition of inverse gaussian ornstein uhlenbeck processes ( cf . barndorff - nielsen and shephard @xcite and barndorff - nielsen and stelzer @xcite ) and @xmath8 , where @xmath9 is a lvy process and @xmath10 is a time change process with stationary increments ( cf . et al_. @xcite ) . the theory discussed in the present paper applies also to processes of this type . the structure of the paper is as follows . section [ exi ] defines the concept of qou processes and provides conditions for existence and uniqueness of stationary solutions to the langevin equation . the form of the autocovariance function of the solutions is given and its asymptotic behavior for @xmath11 is discussed . as an intermediate step , a fubini theorem for lvy bases is established in section [ fubini ] . in section [ ma ] explicit forms of wold karhunen representations are derived and used to analyze the asymptotics , under more specialized assumptions , of the autocovariance functions , both for @xmath11 and for @xmath12 . the establishes an auxiliary continuity result of a technical nature . let @xmath13 be a measurable process with stationary increments and let @xmath14 be a positive number . by a qou process @xmath15 driven by @xmath0 and with parameter @xmath16 , we mean a stationary solution to the langevin equation @xmath17 , that is , @xmath18 is a stationary process that satisfies @xmath19 where the integral is a pathwise lebesgue integral . for all @xmath20 we use the notation @xmath21 . recall that a process @xmath22 is measurable if @xmath23 is @xmath24-measurable , and that @xmath25 has stationary increments if , for all @xmath26 , @xmath27 has the same finite distributions as @xmath28 . for @xmath29 we will say that a process @xmath25 has finite @xmath30 moments if @xmath31<\infty$ ] for all @xmath32 . moreover , for @xmath33 or @xmath34 , we will write @xmath35 , @xmath36 or @xmath37 provided that @xmath38 , @xmath39 or @xmath40 , respectively . for each process @xmath25 with finite second moments , let @xmath41 denote its variance function . when @xmath25 , in addition , is stationary , let @xmath42 denote its autocovariance function and @xmath43 $ ] its complementary autocovariance function . before discussing the general setting further , we recall some well - known cases . the stationary solution @xmath15 to ( [ eq:5 ] ) when @xmath44 ( with @xmath6 the brownian motion ) is the gaussian ornstein uhlenbeck process , @xmath45 is the mean level , @xmath16 is the speed of reversion and @xmath46 is the volatility . when @xmath0 is a lvy process , the corresponding qou process , @xmath15 , exists if and only if @xmath47<\infty$ ] or , equivalently , if and only if @xmath48 where @xmath49 is the lvy measure of @xmath0 ; see sato and yamazato @xcite or wolfe @xcite . in this case @xmath15 is called an ornstein uhlenbeck - type process ; for applications of such processes in financial economics , see barndorff - nielsen and shephard @xcite . the first result below shows the existence and uniqueness for the stationary solution @xmath15 to the langevin equation @xmath17 in the case where the noise @xmath0 is integrable . that is , we show existence and uniqueness of qou processes @xmath15 . moreover , we provide an explicit form of the solution that is used to calculate the mean and variance of @xmath15 . [ lemma_masani ] let @xmath0 be a measurable process with stationary increments and finite first moments , and let @xmath14 be a positive real number . then , @xmath18 , given by @xmath50 is a qou process driven by @xmath0 with parameter @xmath16 ( the integral is a pathwise lebesgue integral ) . furthermore , any other qou process driven by @xmath0 and with parameter @xmath16 equals @xmath15 in law . finally , if @xmath0 has finite @xmath30 moments , @xmath51 , then @xmath15 also has finite @xmath30 moments and is continuous in @xmath52 . it is an open problem to relax the integrability of @xmath0 in theorem [ lemma_masani ] ; that is , is it enough that @xmath0 has finite @xmath53 moments ? recall that when @xmath0 is a lvy process , finite @xmath53 moments is a necessary and sufficient condition for the existence of the corresponding ornstein uhlenbeck - type process . proof of theorem [ lemma_masani ] existence : let @xmath51 and assume that @xmath0 has finite @xmath30 moments . choose @xmath54 , according to corollary [ p_cont ] , such that @xmath55 for all @xmath32 . by jensen s inequality , @xmath56 & \leq & ( { \mathrm{e}}^{\lambda t}/\lambda)^{p-1}\int_{-\infty}^t { \mathrm{e}}^{\lambda s}\mathrm{e}[|n_s|^p]\,{\mathrm{d}}{s}\\ & \leq & ( { \mathrm{e}}^{\lambda t}/\lambda)^{p-1}\int_{-\infty}^t { \mathrm{e}}^{\lambda s}(\alpha+\beta|s|)^p\,{\mathrm{d}}{s}<\infty,\end{aligned}\ ] ] which shows that the integral in ( [ eq:88 ] ) exists almost surely as a lebesgue integral and that @xmath57 , given by ( [ eq:88 ] ) , is @xmath30-integrable . using substitution we obtain from ( [ eq:88 ] ) , @xmath58 by corollary [ p_cont ] , @xmath0 is @xmath52-continuous and , therefore , it follows that the right - hand side of ( [ eq:23 ] ) exists as a limit of riemann sums in @xmath52 . hence the stationarity of the increments of @xmath0 implies that @xmath15 is stationary . moreover , using integration by parts on @xmath59 , we get @xmath60 which shows that @xmath15 satisfies ( [ eq:5 ] ) , and hence @xmath15 is a qou process driven by @xmath0 with parameter @xmath16 . since @xmath15 is a measurable process with stationary increments and finite @xmath30 moments , proposition [ p_cont ] shows that it is continuous in @xmath52 . to show uniqueness in law , let @xmath61 be a qou process driven by @xmath0 with parameter @xmath14 , that is , @xmath61 is a stationary process that satisfies ( [ eq:5 ] ) . for all @xmath62 we have , with @xmath63 , that @xmath64 solving ( [ eq:48 ] ) pathwise , it follows that for all @xmath65 , @xmath66 note that @xmath67 a.s . , thus for all @xmath68 and @xmath69 , the stationarity of @xmath61 implies that for @xmath70 , @xmath71 ( for all random vectors , @xmath72 denotes convergence in distribution ) . therefore , as @xmath70 , @xmath73 this shows that the distribution of @xmath61 only depends on @xmath0 and @xmath16 , and completes the proof . proposition 2.1 in surgailis _ et al_. @xcite and proposition 2.1 in maejima and yamamoto @xcite provide existence results for stationary solutions to langevin equations . however , these results do not cover theorem [ lemma_masani ] . the first result considers only bochner - type integrals and the second result requires , in particular , that the sample paths of @xmath0 are riemann integrable . let @xmath74 denote an @xmath75-brownian motion indexed by @xmath76 and @xmath77 be a predictable process ; that is , @xmath46 is measurable with respect to @xmath78\times a\dvt s , t\in{\mathbb{r}},s < t , a\in{\mathcal f}_s\bigr).\ ] ] assume that , for all @xmath79 , @xmath80 has the same finite - dimensional distributions as @xmath81 and that @xmath82 . then @xmath0 , given by @xmath83 is a well - defined continuous process with stationary increments and finite second moments . ( recall that for @xmath84 , @xmath85 . ) [ cor_sigma ] let @xmath0 be given by ( [ eq:125 ] ) . then , there exists a unique - in - law qou process @xmath15 driven by @xmath0 with parameter @xmath14 , and @xmath15 is given by @xmath86 since @xmath0 is a measurable process with finite second moments , it follows by theorem [ lemma_masani ] that there exists a unique - in - law qou process @xmath15 , and it is given by @xmath87\\[-8pt ] & = & \lambda\int_{-\infty}^0 \biggl(\int_{\mathbb{r}}1_{(t+s , t]}(u ) { \mathrm{e}}^{\lambda s}\sigma_u\,{\mathrm{d}}{b}_u\biggr)\,{\mathrm{d}}{s}.\nonumber\end{aligned}\ ] ] by an extension of protter @xcite , chapter iv , theorem 65 , from finite intervals to infinite intervals we may switch the order of integration in ( [ eq:128 ] ) and hence we obtain ( [ eq:127 ] ) . let us conclude this section with formulas for the mean and variance of a qou process @xmath15 . in the rest of this section , let @xmath0 be a measurable process with stationary increments and finite first moments and let @xmath15 be a qou process driven by @xmath0 with parameter @xmath14 ( which exists by theorem [ lemma_masani ] ) . since @xmath15 is unique in law , it makes sense to consider the mean and variance function of @xmath15 . let us assume for simplicity that @xmath88 a.s . the following proposition gives the mean and variance of @xmath15 . [ m_and_v ] let @xmath0 and @xmath15 be given as above . then , @xmath89= \frac{\mathrm{e}[n_1]}{\lambda } \quad\mbox{and}\quad \operatorname{var}(x_0)=\frac{\lambda}{2}\int_0^\infty { \mathrm{e}}^{-\lambda s}{\mathrm{v}_{n}}(s)\,{\mathrm{d}}{s}.\ ] ] in the part concerning the variance of @xmath90 , we assume that @xmath0 has finite second moments . note that proposition [ m_and_v ] shows that the variance of @xmath90 is @xmath91 times the laplace transform of @xmath92 . in particular , if @xmath93 , where @xmath94 is a fractional brownian motion ( fbm ) of index @xmath95 ( see @xcite or @xcite for properties of the fbm ) , then @xmath96=\mu$ ] and @xmath97 and hence , by proposition [ m_and_v ] , we have that @xmath98=\frac{\mu}{\lambda } \quad\mbox{and}\quad \operatorname{var}(x_0)=\frac{\sigma^2\gamma(1 + 2h)}{2\lambda^{2h}}.\ ] ] for @xmath99 , ( [ eq:119 ] ) is well known , and in this case @xmath100 . before proving proposition [ m_and_v ] , let us note that @xmath101=\mathrm{e}[n_1]t$ ] for all @xmath32 . indeed , this follows by the continuity of @xmath102 $ ] ( see corollary [ p_cont ] ) and the stationarity of the increments of @xmath0 . proof of proposition [ m_and_v ] recall that , by corollary [ p_cont ] , we have that @xmath103\leq\alpha+\beta{\vert t\vert}$ ] for some @xmath54 . hence , by ( [ eq:88 ] ) and fubini s theorem , we have that @xmath104 & = & \mathrm{e}\biggl[-\lambda\int_{-\infty}^0 { \mathrm{e}}^{\lambda s } n_s\,{\mathrm{d}}{s}\biggr]=-\lambda\int_{-\infty}^0 { \mathrm{e}}^{\lambda s}\mathrm{e}[n_s]\,{\mathrm{d}}{s } \\ & = & -\lambda\mathrm{e}[n_1]\int_{-\infty}^0 { \mathrm{e}}^{\lambda s}s\,{\mathrm{d}}{s}=\mathrm{e}[n_1]/\lambda.\end{aligned}\ ] ] this shows the part concerning the mean of @xmath90 . to show the last part , assume that @xmath0 has finite second moments . by using @xmath105=\mathrm{e}[n_1]/\lambda$ ] , ( [ eq:88 ] ) shows that , with @xmath106t$ ] , we have @xmath107)^2\bigr]= \mathrm{e}\biggl[\biggl(\lambda\int_{-\infty}^0 { \mathrm{e}}^{\lambda s } \tilde n_s\,{\mathrm{d}}{s}\biggr)^2 \biggr].\ ] ] since @xmath108 for some @xmath54 by corollary [ p_cont ] , fubini s theorem shows @xmath109)\,{\mathrm{d}}{s } \,{\mathrm{d}}{u},\ ] ] and since @xmath110=\frac{1}{2}[{\mathrm{v}_{n}}(s)+{\mathrm{v}_{n}}(u)-{\mathrm{v}_{n}}(s - u)],$ ] we have @xmath111\\[-8pt ] & = & \lambda\int_{-\infty}^0 { \mathrm{e}}^{\lambda s}{\mathrm{v}_{n}}(s)\,{\mathrm{d}}{s}-\frac{\lambda^2}{2}\int_{-\infty}^0 { \mathrm{e}}^{\lambda u}\biggl(\int_{-\infty}^{-u } { \mathrm{e}}^{\lambda(s+u ) } { \mathrm{v}_{n}}(s)\,{\mathrm{d}}{s}\biggr)\,{\mathrm{d}}{u}.\nonumber\end{aligned}\ ] ] moreover , @xmath112 which , by ( [ eq:121 ] ) , gives the expression for the variance of @xmath90 . the next result shows that the autocovariance function of a qou process @xmath15 driven by @xmath0 with parameter @xmath16 has the same asymptotic behavior at infinity as the second derivative of the variance function of @xmath0 divided by @xmath113 . [ var ] let @xmath0 be a measurable process with stationary increments , @xmath88 a.s . , and finite second moments . let @xmath15 be a qou process driven by @xmath0 with parameter @xmath14 . assume that @xmath92 is three times continuous differentiable in a neighborhood of @xmath34 , and for @xmath114 we have that @xmath115 , @xmath116 and @xmath117 . then , for @xmath11 , we have @xmath118 . assume for @xmath33 that @xmath119 , then , for @xmath120 we have @xmath121 . more generally , let @xmath51 and assume that @xmath0 has finite @xmath30 moments and @xmath122 as @xmath33 . then , for @xmath12 , we have @xmath123 . note that by proposition [ var ] the short - term asymptotic behavior of @xmath124 is not influenced by @xmath16 . proof of proposition [ var ] let @xmath125 and assume that @xmath92 is three times continuous differentiable on @xmath126 ; that is , @xmath127 . let @xmath128 , and let us show that for @xmath65 and @xmath129 , @xmath130 if we have shown ( [ eq:100 ] ) , then , by using that @xmath116 , @xmath117 and lhpital s rule , follows . similar to the proof of proposition [ m_and_v ] , let @xmath131t$ ] . to show ( [ eq:100 ] ) , recall that by corollary [ p_cont ] we have @xmath132 for some @xmath54 . hence , by ( [ eq:88 ] ) and fubini s theorem , we find that @xmath133)(x_0-\mathrm{e}[x_0])\bigr ] = g(t)-\lambda { \mathrm{e}}^{-\lambda t}\int_{-\infty}^t { \mathrm{e}}^{\lambda s}g(s)\,{\mathrm{d}}{s},\ ] ] where @xmath134\,{\mathrm{d}}{s},\qquad t\in{\mathbb{r}}.\ ] ] since @xmath135=\frac{1}{2}[{\mathrm{v}_{n}}(t)+{\mathrm{v}_{n}}(s)-{\mathrm{v}_{n}}(s - t)],$ ] we have that @xmath136\,{\mathrm{d}}{s}\nonumber \\[-8pt]\\[-8pt ] & = & -\frac{1}{2}\biggl({\mathrm{v}_{n}}(t)-\lambda { \mathrm{e}}^{\lambda t}\int_{t}^\infty { \mathrm{e}}^{-\lambda s}{\mathrm{v}_{n}}(s)\,{\mathrm{d}}{s}\biggr)\nonumber -\frac{\lambda}{2}\int_{-\infty}^0 { \mathrm{e}}^{\lambda s}{\mathrm{v}_{n}}(s)\,{\mathrm{d}}{s}.\end{aligned}\ ] ] from ( [ eq:68 ] ) it follows that @xmath137 and hence , using partial integration on ( [ eq:66 ] ) , we have for @xmath65 , @xmath138 moreover , from ( [ eq:68 ] ) and for @xmath65 , we find @xmath139 for @xmath140 we have , by assumption , that @xmath115 , and hence @xmath141 . thus , from ( [ eq:57 ] ) and a double use of partial integration , we obtain that @xmath142 using ( [ eq:70 ] ) , fubini s theorem and that @xmath143 we have for @xmath65 , @xmath144 combining this with ( [ eq:109 ] ) we obtain ( [ eq:100 ] ) , and the proof of is complete . using ( [ eq:5 ] ) we have for all @xmath145 that @xmath146 on the other hand , @xmath147 which shows that @xmath148 a similar inequality is available when @xmath84 , and hence for @xmath33 we have that @xmath123 if @xmath149 . when @xmath0 is an fbm of index @xmath150 then @xmath151 , and hence @xmath152 the conditions in proposition [ var ] are clearly fulfilled and thus we have the following corollary . [ cor_var ] let @xmath0 be an fbm of index @xmath95 , and let @xmath15 be a qou process driven by @xmath0 with parameter @xmath14 . for @xmath153 and @xmath11 , we have @xmath154 . for @xmath95 and @xmath12 , we have @xmath155 . the above result concerning the behavior of @xmath156 for @xmath157 when @xmath0 is an fbm has been obtained previously via a different approach by cheridito _ et al . _ @xcite , theorem 2.3 . a square - integrable stationary process @xmath158 is said to have _ long - range dependence _ of order @xmath159 if @xmath160 is regularly varying at @xmath34 of index @xmath161 . recall that a function @xmath162 is regularly varying at @xmath34 of index @xmath163 if , for @xmath140 , @xmath164 where @xmath165 is slowly varying , which means that for all @xmath166 , @xmath167 . many empirical observations have shown evidence for long - range dependence in various fields , such as finance , telecommunication and hydrology ; see doukhan _ et al_. @xcite . let @xmath15 be a qou process driven by @xmath0 ; then proposition [ var ] shows that @xmath15 has long - range dependence of order @xmath159 if and only if @xmath168 is regularly varying at @xmath34 of order @xmath161 . furthermore , proposition [ cor_ma_noise ] below shows how to construct qous with long - range dependence . more precisely , if @xmath15 is a qou driven by @xmath169 where @xmath0 is given by ( [ pma_noise ] ) , and for some @xmath170 and @xmath140 , @xmath171 , then @xmath15 has long - range dependence of order @xmath172 . the example @xmath173 , with @xmath174 and @xmath175 is considered in corollary [ exp_fbm ] and it follows that the qou process @xmath15 has long - range dependence of order @xmath176 . here @xmath15 is a fractional ornstein uhlenbeck process if @xmath177 , and a semimartingale if and only if @xmath178 . a quite different type of semimartingale with long - range dependence is obtained for @xmath179 with @xmath46 and @xmath6 independent and @xmath180 being a supou process with long - range dependence , cf . barndorff - nielsen @xcite , barndorff - nielsen and stelzer @xcite and barndorff - nielsen and shephard @xcite . hence , by considering more general processes than the fractional type , we can easily construct stationary processes with long - range dependence within the semimartingale framework . let @xmath181 denote a centered lvy basis on a non - empty space @xmath182 equipped with a @xmath183-ring @xmath184 , see rajput and rosiski @xcite . ( a lvy basis is an infinitely divisible , independently scattered random measure . recall also that a @xmath183-ring on @xmath182 is a family of subsets of @xmath182 that is closed under union , countable intersection and set difference ) . as usual , we assume that @xmath184 is @xmath46-finite , meaning that there exists @xmath185 such that @xmath186 . all integrals @xmath187 will be defined in the sense of rajput and rosiski @xcite . we can now find a measurable parametrization of lvy measures @xmath188 on @xmath76 , a @xmath46-finite measure @xmath3 on @xmath182 and a positive measurable function @xmath189 , such that for all @xmath190 , @xmath191=\exp\biggl(\int_a\biggl[-\sigma^2 ( s)y^2/2+\int_{\mathbb{r } } ( { \mathrm{e}}^{\mathrm{i}yu}-1-\mathrm{i}yu)\nu({\mathrm{d}}{u},s)\biggr ] m({\mathrm{d}}{s})\biggr),\qquad y\in{\mathbb{r}},\ ] ] see @xcite . let @xmath192 be given by @xmath193\nu({\mathrm{d}}{u},s),\ ] ] and for all measurable functions @xmath194 define @xmath195.\ ] ] moreover , let @xmath196 denote the _ musielak orlicz space _ of measurable functions @xmath197 with @xmath198 . note that @xmath199 if and only if @xmath200 , since @xmath201 for some @xmath202 and all @xmath203 . we refer to musielak @xcite for the basic properties of musielak orlicz spaces . when @xmath204 and @xmath199 , theorem 2.1 in marcus and rosiski @xcite shows that @xmath205 is well defined , integrable and centered and @xmath206\leq c_2 \|g\|_\phi,\ ] ] and we may choose @xmath207 and @xmath208 . hence for general @xmath209 it is easily seen that for all @xmath210 , @xmath205 is well defined , integrable and centered and @xmath211\leq2c_2 \|g\|_\phi.\ ] ] let @xmath10 denote a complete separable metric space , and @xmath212 be given by @xmath213 for some measurable function @xmath214 for which the integrals are well defined . then we can choose a measurable modification of @xmath61 . indeed , the existence of a measurable modification of @xmath61 is equivalent to measurability of @xmath215 according to theorem 3 and the remark in cohn @xcite . hence , since @xmath216 is measurable , the maps @xmath217 for all @xmath199 are measurable . this shows that @xmath218 is measurable since @xmath219 is a separable banach space . hence by continuity of @xmath220 , see rajput and rosiski @xcite , it follows that @xmath215 is measurable . assume that @xmath221 is a @xmath46-finite measure on a complete and separable metric space @xmath10 . then we have the following stochastic fubini result extending rosiski @xcite , lemma 7.1 ; prez - abreu and rocha - arteaga @xcite , lemma 5 ; and basse and pedersen @xcite , lemma 4.9 . stochastic fubini - type results for semimartingales can be founded in protter @xcite and ikeda and watanabe @xcite ; however , the assumptions in these results are too strong for our purpose . [ stoc_fubini ] let @xmath222 be an @xmath223-measurable function such that @xmath224 then @xmath225 for @xmath3-a.a . @xmath226 and @xmath227 belongs to @xmath219 , all of the below integrals exist and @xmath228 if @xmath221 is a finite measure , then the last condition in ( [ eq:21 ] ) is equivalent to @xmath229m({\mathrm{d}}{s})\mu({\mathrm{d}}{x})<\infty.\ ] ] we will need theorem [ stoc_fubini ] to be able to prove proposition [ moving_average_rep ] . that proposition yields , in particular , examples for which the conditions of theorem 3.1 are fulfilled . but before proving theorem [ stoc_fubini ] , we will need the following observation . [ remark_int ] for all measurable functions @xmath230 we have @xmath231 moreover , if @xmath230 is a measurable function such that @xmath232 , then for @xmath3-a.a . @xmath226 , @xmath225 and @xmath233 is a well - defined function that belongs to @xmath219 . let us sketch the proof of ( [ eq:101 ] ) . for @xmath216 of the form @xmath234 where @xmath235 , @xmath236 and @xmath237 are disjoint measurable subsets of @xmath10 of finite , ( [ eq:101 ] ) easily follows . hence , by a monotone class lemma argument , it is possible to show ( [ eq:101 ] ) for all measurable @xmath216 . the second statement is a consequence of ( [ eq:101 ] ) . recall that if @xmath238 is a separable banach space , @xmath221 is a measure on @xmath10 and @xmath239 is a measurable map such that @xmath240 , then the bochner integral @xmath241 exists in @xmath242 and @xmath243 . even though @xmath244 is a banach space , this result does not cover lemma [ remark_int ] . proof of theorem [ stoc_fubini ] for @xmath216 of the form @xmath245 where @xmath68 , @xmath246 are measurable subsets of @xmath10 of finite @xmath221-measure , @xmath247 and @xmath248 , the theorem is trivially true . thus , for a general @xmath216 as in the theorem , choose @xmath249 for @xmath68 of the form ( [ eq:16 ] ) such that @xmath250 . indeed , the existence of such a sequence follows by an application of the monotone class lemma . let @xmath251 and let us show that @xmath15 is well defined and @xmath252 in @xmath253 . this follows since @xmath254\leq 2c_2 \int_e \|f(x,\cdot)\|_\phi\mu({\mathrm{d}}{x})<\infty\ ] ] and @xmath255\leq 2 c_2 \int_e \|f_n(x,\cdot)-f(x,\cdot)\|_\phi \mu({\mathrm{d}}{x}).\ ] ] similarly , let @xmath256 and let us show that @xmath61 is well defined and @xmath257 in @xmath253 . by remark [ remark_int ] , @xmath258 is a well - defined function that belongs to @xmath219 , which shows that @xmath61 is well defined . by ( [ eq:25 ] ) and ( [ eq:101 ] ) we have @xmath259\leq2 c_2 \int_e \|f_n(x,\cdot)-f(x,\cdot)\|_\phi\mu({\mathrm{d}}{x}),\ ] ] which shows that @xmath257 in @xmath253 . we have , therefore , proved ( [ eq:15 ] ) , since @xmath260 a.s . , @xmath252 and @xmath257 in @xmath253 . let @xmath22 denote an integrable and centered lvy process with lvy measure @xmath49 and gaussian component @xmath209 . then @xmath25 induces a lvy basis @xmath261 on @xmath262 and @xmath263 , the bounded borel sets , which is uniquely determined by @xmath264)=z_b - z_a$ ] for all @xmath265 with @xmath20 . in this case @xmath3 is the lebesgue measure on @xmath76 and @xmath266 we will write @xmath267 instead of @xmath268 . note that , @xmath269 exists and is integrable if and only if @xmath270 , that is , @xmath271 moreover , if @xmath25 is a symmetric @xmath172-stable lvy process , @xmath272 $ ] , then @xmath273 , where @xmath274 is the space of @xmath172-integrable functions with respect to the lebesgue measure @xmath16 . in wide generality , if @xmath15 is a continuous - time stationary process , then it is representable , in principle , as a moving average ( ma ) , that is , @xmath275 where @xmath276 is a deterministic function and @xmath277 has stationary and orthogonal increments , at least in the second - order sense . ( for a precise statement , see the beginning of section [ w_k ] . ) however , an explicit expression for @xmath278 is seldom available . we show in section [ e_ma_rep ] that an expression can be found in cases where the process @xmath15 is the stationary solution to a langevin equation for which the driving noise process @xmath0 is a pma , that is , @xmath279 where @xmath22 is a suitable process specified later on and @xmath162 is a deterministic function for which the integrals exist . in section [ a_beha ] , continuing the discussion from section [ asy_auto ] , we use the ma representation to study the asymptotic behavior of the associated autocovariance functions . section [ cancel ] comments on a notable cancellation effect . but first , in section 4.1 we summarize known results concerning wold karhunen - type representations of stationary continuous - time processes . let @xmath18 be a second - order stationary process of mean zero and continuous in quadratic mean . let @xmath280 denote the spectral measure of @xmath15 , that is , @xmath280 is a finite and symmetric measure on @xmath76 satisfying @xmath281=\int_{\mathbb{r}}{\mathrm{e}}^{\mathrm{i}(t - u)x}f_x({\mathrm{d}}{x}),\qquad t , u\in{\mathbb{r}},\ ] ] and let @xmath282 denote the density of the absolutely continuous part of @xmath280 . for each @xmath32 let @xmath283 , @xmath284 and @xmath285 ( @xmath286 denotes the @xmath287-closure of the linear span ) . then @xmath15 is called deterministic if @xmath288 and purely non - deterministic if @xmath289 . the following result , which is due to satz 56 in karhunen @xcite ( cf . also doob @xcite , chapter xii , theorem 5.3 ) , provides a decomposition of stationary processes as a sum of a deterministic process and a purely non - deterministic process . let @xmath15 and @xmath280 be given as above . if @xmath290 then there exists a unique decomposition of @xmath15 as @xmath291 where @xmath292 is a lebesgue square - integrable deterministic function and @xmath277 is a process with second - order stationary and orthogonal increments , @xmath293={\vert u - s\vert}$ ] . furthermore , for all @xmath32 , @xmath294 , and @xmath295 is a deterministic second - order stationary process . moreover , if @xmath280 is absolutely continuous and ( [ eq:28 ] ) is satisfied , then @xmath296 and hence @xmath15 is a backward ma . finally , the integral in ( [ eq:28 ] ) is infinite if and only if @xmath15 is deterministic . the results in karhunen @xcite are formulated for complex - valued processes ; however , if @xmath15 is real - valued ( as it is in our case ) , then one can show that all the above processes and functions are real - valued as well . note also that if @xmath15 is gaussian , then the process @xmath277 in ( [ eq:13 ] ) is a standard brownian motion . if @xmath46 is a stationary process with @xmath297=1 $ ] and @xmath6 is a brownian motion , then @xmath298 is of the above type . a generalization of the classical wold karhunen result to a broad range of non - gaussian , infinitely divisible processes was given in rosiski @xcite . assume initially that @xmath25 is an integrable and centered lvy process , and recall that @xmath219 is the space of all measurable functions @xmath162 satisfying ( [ eq:52 ] ) . let @xmath299 be a measurable function such that @xmath300 for all @xmath32 , and let @xmath0 be given by @xmath301 [ moving_average_rep ] let @xmath0 be given as above . then there exists a unique - in - law qou process @xmath15 driven by @xmath0 with parameter @xmath14 , and @xmath15 is an ma of the form @xmath302 where @xmath303 belongs to @xmath219 and is given by @xmath304 since @xmath305 is measurable we may choose a measurable modification of @xmath0 see section [ fubini ] and hence , by theorem [ lemma_masani ] , there exists a unique - in - law qou process @xmath15 driven by @xmath0 with parameter @xmath16 . for fixed @xmath32 , we have by ( [ eq:88 ] ) and with @xmath306 for all @xmath307 and @xmath308 that @xmath309 by theorem [ thm_cont ] there exist @xmath54 such that @xmath310 for all @xmath79 , implying that @xmath311 . by theorem [ stoc_fubini ] , @xmath312 for lebesgue almost all @xmath26 , which implies that @xmath313 for all @xmath145 , and hence @xmath314 , defined in ( [ def_of_psi ] ) , is a well - defined function . moreover , by theorem [ stoc_fubini ] , @xmath315 and @xmath316 which completes the proof . note that for @xmath317 , we have @xmath318 and @xmath319 . thus , in this case we recover the well - known result that the qou process @xmath15 driven by @xmath25 with parameter @xmath14 is an ma of the form @xmath320 . let us use the notation @xmath321 , and let @xmath322 be given by @xmath323 a pma @xmath0 of the form ( [ eq:9 ] ) , where @xmath25 is an @xmath172-stable lvy process with @xmath324 $ ] and @xmath216 is given by @xmath325 , is called a _ linear fractional @xmath172-stable motion _ of index @xmath326 ; see samorodnitsky and taqqu @xcite . moreover , pmas with @xmath327 for @xmath328 and where @xmath25 is a square - integrable and centered lvy process are called _ fractional lvy processes _ in marquardt @xcite ; these processes provide examples of @xmath216 and @xmath25 for which proposition 4.2 applies . moreover , @xcite , theorems 6.2 and 6.3 , studies mas driven by fractional lvy processes , which in some cases also have a representation of the form ( [ eq:34 ] ) . [ ma_frac ] let @xmath329 $ ] and @xmath0 be a linear fractional @xmath172-stable motion of index @xmath95 . then there exists a unique - in - law qou process @xmath15 driven by @xmath0 with parameter @xmath14 , and @xmath15 is an ma of the form @xmath330 where @xmath331 is given by @xmath332 for @xmath140 , we have @xmath333 , and for @xmath33 , @xmath334 . a qou process driven by a linear fractional @xmath172-stable motion is called a fractional ornstein uhlenbeck process . in maejima and yamamoto @xcite , the existence of the fractional ornstein uhlenbeck process is shown in the case where @xmath335 and @xmath336 . ( the case @xmath337 is trivial since @xmath338 . ) the existence in the case @xmath339 ( see corollary [ ma_frac ] ) is somewhat unexpected due to the fact that the sample paths of the linear fractional @xmath172-stable motion are unbounded on each compact interval ; cf . page 4 in maejima and yamamoto @xcite , where non - existence is surmised . in the case @xmath340 ( i.e. , @xmath0 is a fractional brownian motion ) , cheridito _ et al . _ @xcite show the existence of the fractional ornstein uhlenbeck process . in the next lemma we will show a special property of @xmath314 , given by ( [ def_of_psi ] ) ; namely that @xmath341 whenever this integral is well defined and @xmath216 tends to zero at @xmath34 . this property has a great impact on the behavior of the autocovariance function of qou processes . we will return to this point in section [ cancel ] . [ remark_f ] let @xmath162 be a locally integrable function that is zero on @xmath342 and @xmath343 . then , @xmath344 . for @xmath145 , @xmath345 and hence , by using that @xmath346 we obtain that @xmath347 proposition [ moving_average_rep ] carries over to a much more general setting . for example , if @xmath0 is of the form @xmath348 \lambda ( \mathrm{d}s,\mathrm{d}x),\qquad t\in{\mathbb{r}},\ ] ] where @xmath261 is a centered lvy basis on @xmath349 ( @xmath295 is a non - empty space ) with control measure @xmath350 ; @xmath351 and @xmath352 , from ( [ eq:36 ] ) , do not depend on @xmath26 ; and @xmath353 for all @xmath32 , then , using theorems [ thm_cont ] , [ lemma_masani ] and [ stoc_fubini ] , the arguments from proposition [ moving_average_rep ] show that there exists a unique - in - law qou process @xmath15 driven by @xmath0 with parameter @xmath14 , and @xmath15 is given by @xmath354 where @xmath355 we recover proposition [ moving_average_rep ] when @xmath356 and @xmath357 is the dirac delta measure at 0 . the representation , from the previous section , of qou processes as mas enables us to handle the autocovariance function analytically . in section [ gen_ma ] we discuss how the tail behavior of the kernel @xmath276 of a general ma process determines that of the covariance function . by use of those results , section [ cov_ma ] relates the asymptotic behavior of the kernel of the noise @xmath0 to the asymptotic behavior of the autocovariance function of the qou process @xmath15 driven by @xmath0 , both for @xmath33 and @xmath140 . let @xmath276 be a lebesgue square - integrable function and @xmath25 be a centered process with stationary and orthogonal increments . assume for simplicity that @xmath358 a.s . and let @xmath360 be a backward ma ; @xmath156 be its autocovariance function , that is @xmath361=\int_0^\infty\psi(t+s)\psi(s)\,{\mathrm{d}}{s},\qquad t\in{\mathbb{r}};\ ] ] and @xmath43 $ ] . the behavior of @xmath156 at @xmath362 or @xmath34 corresponds in large extent to the behavior of the kernel @xmath276 at @xmath362 or @xmath34 , respectively . indeed , we have the following result , in which @xmath363 and @xmath364 are constants given by @xmath365 [ pro_cov_ma ] let the setting be as described above . for @xmath11 and @xmath366 , @xmath367 implies @xmath368 , provided @xmath369 for all @xmath145 and some @xmath370 . for @xmath157 and @xmath371 , @xmath372 implies @xmath373 , and hence @xmath374 , provided @xmath375 . for @xmath12 and @xmath376 , @xmath367 implies @xmath377 , provided @xmath276 is absolutely continuous on @xmath378 with density @xmath379 satisfying @xmath380 for all @xmath145 and some @xmath381 . let @xmath382 and assume that @xmath367 as @xmath11 and @xmath369 for @xmath145 . then @xmath383\\[-8pt ] & = & t^{2\alpha+1 } \int_0^\infty\frac{\psi(t(1+s ) ) \psi(ts ) } { ( t(1+s))^{\alpha}(ts)^{\alpha}}(1+s)^{\alpha } s^{\alpha } \,{\mathrm{d}}{s}\nonumber \\ & \sim & t^{2\alpha+1 } c^2 \int_0^\infty(1+s)^{\alpha } s^{\alpha } \,{\mathrm{d}}{s } \qquad\mbox{as } t\rightarrow\infty.\nonumber\end{aligned}\ ] ] since @xmath384 ( [ eq:130 ] ) shows that @xmath385 for @xmath11 . let @xmath371 and assume that @xmath372 for @xmath11 . note that @xmath386 and for some @xmath387 we have for all @xmath388 and @xmath389 that @xmath390 . hence , by applying lebesgue s dominated convergence theorem , we obtain @xmath391 by letting @xmath392 we have @xmath393= t\int\bigl[\psi\bigl(t(s+1)\bigr)-\psi(ts)\bigr]^2\,{\mathrm{d}}{s}=t^{2\alpha+1 } \int \vert f_t(s)\vert^2\,{\mathrm{d}}{s}.\ ] ] as @xmath12 , we find @xmath394 choose @xmath178 such that @xmath395 for @xmath396 . by our assumptions we have for all @xmath397 that @xmath398}{\vert \psi'(u)\vert } \\ & \leq & c_2 t^{-\alpha+1}\sup_{u\in[st , t(s+1 ) ] } { \vert u\vert}^{\alpha-1}= c_2 t^{-\alpha+1}{\vert ts\vert}^{\alpha-1 } = c_2s^{\alpha-1},\end{aligned}\ ] ] and for @xmath399 , @xmath400 $ ] . this shows that there exists a function @xmath401 such that @xmath402 for all @xmath145 , and thus , by lebesgue s dominated convergence theorem , we have @xmath403 together with ( [ eq:1 ] ) , ( [ eq:41 ] ) shows that @xmath404 for @xmath12 . it would be of interest to obtain a general result covering proposition [ pro_cov_ma ] in the case @xmath405 . recall that @xmath314 , given by ( [ def_of_psi ] ) , often satisfies that @xmath406 , according to lemma [ remark_f ] . consider the case where @xmath407 for @xmath408 and @xmath14 . for @xmath409 , @xmath410 , and hence @xmath411 for @xmath12 and @xmath412 , by proposition [ pro_cov_ma ] ( compare with barndorff - nielsen _ et al_. @xcite ) . note that if @xmath413 is a moving average , as above , then by proposition [ pro_cov_ma ] and for @xmath140 , @xmath414 with @xmath159 , provided that @xmath415 and @xmath416 . this shows that @xmath15 has long - range dependence of order @xmath172 . let us conclude this section with a short discussion of when an ma @xmath413 is a semimartingale . it is often very important that the process of interest is a semimartingale , especially in finance , where the semimartingale property of the asset price is equivalent to the property that the capital process depends continuously on the chosen strategy ; see section 8.1.1 in cont and tankov @xcite . in the case where @xmath25 is a brownian motion , theorem 6.5 , in knight @xcite shows that @xmath15 is an @xmath417 semimartingale if and only if @xmath276 is absolutely continuous on @xmath418 with a square - integrable density . ( here @xmath419)$ ] . ) for a further study of the semimartingale property of pma and more general processes , see @xcite in the gaussian case , and basse and pedersen @xcite for the infinitely divisible case . let us return to the case of a qou process driven by a pma . let @xmath25 be a centered lvy process , @xmath162 be a measurable function that is 0 on @xmath342 and satisfies @xmath420 for all @xmath32 and @xmath0 be given by @xmath421\,{\mathrm{d}}{z } _ s,\qquad t \in{\mathbb{r}}.\ ] ] first , we will consider the relationship between the behavior of the kernel of the noise @xmath0 and that of the kernel @xmath314 of the corresponding moving average @xmath15 . [ cor_ma_noise ] let @xmath0 be given by ( [ pma_noise ] ) , and @xmath15 be a qou process driven by @xmath0 with parameter @xmath14 . let @xmath422 and assume that , for some @xmath423 , @xmath216 is continuous differentiable in a neighborhood of @xmath34 with @xmath424 for @xmath140 . then , for @xmath140 , we have @xmath425 , provided @xmath426 for all @xmath145 and some @xmath427 . let @xmath412 and @xmath428 for @xmath33 . then , for @xmath33 , we have @xmath429 , provided @xmath216 is two times continuous differentiable in a neighborhood of @xmath34 with @xmath430 for @xmath140 , and that @xmath216 is absolutely continuous on @xmath378 with a density @xmath431 satisfying @xmath432 for all @xmath433 . choose @xmath125 such that @xmath216 is continuous differentiable on @xmath434 . by partial integration , we have for @xmath435 , @xmath436 showing that @xmath437 for @xmath140 . choose @xmath438 such that @xmath439 for all @xmath440 . by ( [ def_of_psi ] ) we have that @xmath441}{\vert \psi_f(t)t^{-\alpha}\vert}<\infty,$ ] since @xmath442}{\vert f(t)t^{-\alpha}\vert}<\infty$ ] , and hence there exists a constant @xmath370 such that @xmath443 for all @xmath145 . therefore , ( i ) follows by proposition [ pro_cov_ma ] . choose @xmath125 such that @xmath216 is two times continuous differentiable on @xmath434 . by ( [ eq:107 ] ) and partial integration we have for @xmath444 and @xmath140 , @xmath445 where we in the last equality have used that @xmath430 for @xmath140 . using that @xmath446 and @xmath447 for all @xmath433 , it follows that there exists a @xmath381 such that @xmath448 for all @xmath145 . moreover , for @xmath33 , we have that @xmath449 . hence , ( ii ) follows by proposition [ pro_cov_ma ] . now consider the following set - up : let @xmath22 be a centered and square - integrable lvy process , and for @xmath95 , @xmath450 , @xmath174 , let @xmath451\,{\mathrm{d}}{z}_s.\ ] ] note that when @xmath177 and @xmath25 is a brownian motion , @xmath452 is a constant times the fbm of index @xmath453 , and when @xmath178 , @xmath452 is a semimartingale . we have the following corollary to proposition [ cor_ma_noise ] . [ exp_fbm ] let @xmath452 be given by ( [ eq:108 ] ) , and let @xmath454 be a qou process driven by @xmath452 with parameter @xmath14 . then , for @xmath175 and @xmath140 , @xmath455 and for @xmath95 and @xmath33 , @xmath456 for @xmath175 , let @xmath457 . then @xmath458 and , for @xmath444 , @xmath459 , where @xmath460 and @xmath461 . moreover , @xmath462 . thus , proposition [ cor_ma_noise ] shows that @xmath463 . to show ( [ eq:43 ] ) assume that @xmath95 . for @xmath33 , we have @xmath464 , where @xmath465 and @xmath466 when @xmath177 , and @xmath467 and @xmath468 when @xmath178 . for @xmath457 , @xmath469 with @xmath470 , showing that @xmath430 for @xmath140 ( both for @xmath178 and @xmath177 ) . moreover , @xmath216 is absolutely continuous on @xmath378 with density @xmath471 . this shows that @xmath472 for all @xmath433 ( both for @xmath178 and @xmath177 ) . hence ( [ eq:43 ] ) follows by proposition [ cor_ma_noise ] . let @xmath0 be a pma of the form ( [ eq:9 ] ) , where @xmath25 is a centered square - integrable lvy process , and @xmath473 , where @xmath95 . ( recall that if @xmath25 is a brownian motion , then @xmath0 is an fbm of index @xmath453 . ) let @xmath15 be a qou process driven by @xmath0 with parameter @xmath14 , and recall that by proposition [ moving_average_rep ] , @xmath15 is an ma of the form @xmath474 where @xmath475 below we will discus some stability properties for the autocovariance function under minor modification of the kernel function . for all bounded measurable functions @xmath476 with compact support , let @xmath477 . we will think of @xmath478 as an ma where we have made a minor change of @xmath15 s kernel . note that if we let @xmath479 , then the autocovariance function @xmath480 , of @xmath481 , is zero whenever @xmath482 is large enough due to the fact that @xmath216 has compact support . [ cor_stab ] we have the following two situations in which @xmath483 are non - zero constants . for @xmath484 and @xmath485 , we have for @xmath140 , @xmath486 for @xmath175 , we have for @xmath140 , @xmath487 thus for @xmath484 , the above shows that the behavior of the autocovariance function at infinity is changed dramatically by making a minor change of the kernel . in particular , if @xmath216 is a positive function , not the zero function , then @xmath488 behaves as @xmath489 at infinity . on the other hand , when @xmath490 the behavior of the autocovariance function at infinity does not change if we make a minor change to the kernel . that is , in this case the autocovariance function has a stability property , contrary to the case where @xmath484 . note that the dramatic effect appearing from corollary [ cor_stab ] is associated with the fact that @xmath491 , as shown in lemma [ remark_f ] . proof of corollary [ cor_stab ] by corollary [ ma_frac ] we have for @xmath140 that @xmath492 , where @xmath493 and @xmath494 . to show , assume that @xmath495 and hence @xmath371 . according to lemma [ remark_f ] , we have that @xmath491 and hence @xmath496\,{\mathrm{d}}{s}\neq0,$ ] since @xmath497 by assumption . from proposition [ pro_cov_ma ] and for @xmath498 we have that @xmath499 , where @xmath500\,{\mathrm{d}}{s}$ ] . on the other hand , by corollary [ cor_var ] we have that @xmath501 for @xmath140 , and hence we have shown with @xmath502 . for @xmath175 we have that @xmath503 , and hence follows by proposition [ pro_cov_ma ] . in this appendix we will show an auxiliary continuity result used several times in the paper . the main result in this appendix is theorem [ thm_cont ] ; corollary [ p_cont ] is used in theorem [ lemma_masani ] , while the general modular setting is needed to prove proposition [ moving_average_rep ] . for the basic definitions and properties of linear metric spaces , modulars and @xmath242-norms , we refer to rolewicz @xcite . let @xmath504 be a @xmath46-finite measure space , and @xmath505 an even and continuous function that is non - decreasing on @xmath506 , with @xmath507 . assume there exists a constant @xmath202 such that @xmath508 for all @xmath509 ( that is , @xmath278 satisfies the @xmath510 condition ) . let @xmath511 denote the space of all measurable functions from @xmath512 into @xmath76 ; @xmath513 denote the modular on @xmath514 given by @xmath515 and @xmath516 denote the corresponding modular space . furthermore , for @xmath517 , define @xmath518 then @xmath519 is an @xmath242-norm on @xmath219 and , in particular , @xmath520 is an invariant metric on @xmath219 . moreover , when @xmath278 is convex , the _ luxemburg norm _ @xmath521 is a norm on @xmath219 ; see khamsi @xcite . [ thm_cont ] let @xmath522 denote a measurable function satisfying @xmath523 for all @xmath32 , and @xmath524 then , @xmath525 is continuous . moreover , if @xmath278 is convex , then there exist @xmath54 such that @xmath526 for all @xmath32 . proof of lemma [ borel_m ] we will use a monotone class lemma argument to prove this result , so let @xmath530 be the set of all functions @xmath216 for which lemma [ borel_m ] holds and @xmath531 be the set of all functions @xmath216 of the form @xmath532 where , for @xmath68 , @xmath246 are measurable subsets of @xmath76 , @xmath533 are measurable subsets of @xmath512 of finite @xmath221 measure and @xmath248 . let us show that @xmath534 is measurable . since , for all @xmath199 , @xmath535 is measurable , we get that for all @xmath199 and @xmath427 , @xmath536 is measurable ( we use the notation , @xmath537 ) . therefore , since @xmath538 has separable range , it follows that @xmath538 is measurable ( recall that the borel @xmath46-field in a separable metric space is generated by the open balls ) . this shows that @xmath539 . note that the set @xmath540 of bounded elements from @xmath541 is a vector space with @xmath542 , and that @xmath543 with @xmath544 implies that @xmath545 . moreover , since @xmath531 is stable under pointwise multiplication , the monotone class lemma ( see @xcite , chapter ii , theorem 3.2 ) shows that @xmath546 ( for a family of functions @xmath547 , @xmath548 denotes the least @xmath46-algebra for which all the functions are measurable , and for each @xmath46-algebra @xmath549 , @xmath550 denotes the space of all bounded @xmath549-measurable functions . ) for a general function @xmath216 , define @xmath551 by @xmath552 . for all @xmath68 , @xmath551 is a bounded measurable function and hence @xmath553 is a measurable map with a separable range . moreover , @xmath554 pointwise in @xmath219 , showing that @xmath538 is measurable and has a separable range . proof of theorem [ thm_cont ] let @xmath538 denote the map @xmath525 , and for fixed @xmath555 and arbitrary @xmath32 , consider the ball @xmath556 . by lemma [ borel_m ] , @xmath538 is measurable , and hence @xmath557 is a measurable subset of @xmath76 for all @xmath32 . according to lemma [ borel_m ] , @xmath538 has a separable range and , therefore , there exists a countable set @xmath558 such that the range of @xmath538 is included in @xmath559 , implying that @xmath560 . in particular , there exists an @xmath561 such that @xmath562 has a strictly positive lebesgue measure . by the steinhaus lemma , see @xcite , theorem 1.1.1 , there exists a @xmath178 such that @xmath563 . note that by ( [ eq:8 ] ) it is enough to show continuity of @xmath538 at @xmath564 . for @xmath565 there exists , by definition , @xmath566 such that @xmath567 for @xmath568 , showing that @xmath569 which completes the proof of the continuity part . to show the last part of the theorem , assume that @xmath278 is convex . for each @xmath145 choose @xmath570 such that @xmath571 . then , @xmath572\\[-8pt ] & \leq & n { \vert f_1-f_{0}\vert}_\phi+{\vert f_{t - n}-f_0\vert}_\phi\leq t \beta+a,\nonumber\end{aligned}\ ] ] where @xmath573 and @xmath574}{\vert f_s - f_0\vert}_\phi$ ] . we have already shown that @xmath575 is continuous , and hence @xmath576 . since @xmath577 for all @xmath32 , ( [ eq:37 ] ) shows that @xmath578 for all @xmath32 , implying that @xmath579 where @xmath580 . [ p_cont ] let @xmath585 and @xmath18 be a measurable process with stationary increments and finite @xmath30 moments . then @xmath15 is continuous in @xmath52 . moreover , if @xmath51 , then there exist @xmath54 such that @xmath586 for all @xmath32 . note that in corollary [ p_cont ] the reversed implication is also true ; in fact , all stochastic processes @xmath18 that are continuous in @xmath514 have a measurable modification according to theorem 2 in cohn @xcite . the idea of using the steinhaus lemma to prove theorem [ thm_cont ] is borrowed from surgailis _ et al_. @xcite , where corollary [ p_cont ] is shown for @xmath584 . furthermore , when @xmath221 is a probability measure and @xmath583 , lemma [ borel_m ] is known from cohn @xcite .

we have explored the interplay of star formation and agn activity in soft x - rays ( 0.5 - 2 kev ) in two samples of seyfert 2 galaxies ( sy2s ) . using a combination of low resolution ccd spectra from _ chandra _ and _ xmm - newton _ , we modeled the soft emission of 34 sy2s using power law and thermal models . for the 11 sources with high signal - to - noise _ chandra _ imaging of the diffuse host galaxy emission , we estimate the luminosity due to star formation by removing the agn , fitting the residual emission . the agn and star formation contributions to the soft x - ray luminosity ( i.e. l and l ) for the remaining 24 sy2s were estimated from the power law and thermal luminosities derived from spectral fitting . these luminosities were scaled based on a template derived from xsings analysis of normal star forming galaxies . to account for errors in the luminosities derived from spectral fitting and the spread in the scaling factor , we estimated l and l from monte carlo simulations . these simulated luminosities agree with l and l derived from _ chandra _ imaging analysis within a 3 confidence level . using the infrared [ neii]12.8 m and [ oiv]26 m lines as a proxy of star formation and agn activity , respectively , we independently disentangle the contributions of these two processes to the total soft x - ray emission . this decomposition generally agrees with l and l at the 3 level . in the absence of resolvable nuclear emission , our decomposition method provides a reasonable estimate of emission due to star formation in galaxies hosting type 2 agn .

1208.6233

supermassive black holes ( smbhs ) and their parent galaxies co - evolve @xcite . in particular , observational and theoretical work has established a link between accreting smbhs ( active galactic nuclei , or agn ) and host galaxy star formation . common mechanisms have been proposed for triggering star formation while fueling smbh accretion , including galaxy mergers ( e.g. * ? ? ? * ; * ? ? ? * ) and secular processes involving gravitational instabilities induced by spiral arms or galactic sized bars ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? disentangling star formation from agn activity becomes a necessary endeavor to investigate smbh and host galaxy co - evolution . this separation has been explored extensively and has included identifiying optical and ultraviolet ( uv ) spectral signatures of starburst activity in agn @xcite , performing principal component analysis on infrared spectra of agn ( e.g. * ? ? ? * ) , and analyzing optical and infrared ( ir ) diagnostics that parameterize the relative contribution of agn to star formation , such as ratios of infrared and optical emission lines that indicate ionization field hardness @xcite and strength of the polycyclic aromatic hydrocarbon features @xcite . here we extend the study of the interplay between agn activity and star formation to the x - ray regime . quiescent galaxies emit thermal and non - thermal x - ray emission . hot gas energized by stellar winds and supernova explosions radiates in soft x - rays ( 0.5 - 2 kev ) whereas x - ray binares and supernova remnants dominate the non - thermal emission above 2 kev ( see * ? ? ? * for a review ) . due to the relatively short life time of high mass x - ray binaries ( hmxbs , @xmath4 yr ) and the short delay between starburst and generation of soft x - rays from hot gas , hard and soft x - ray emission can trace the instantaneous star formation rate ( sfr ) . indeed , x - ray emission from quiescent galaxies is well correlated with radio , infrared , optical and ultraviolet sfr indicators ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? various x - ray sfrs have thus been calibrated in the literature based on the correlation of x - ray emission with far - infrared and radio luminosities @xcite , x - ray emission from hmxbs and cumulative galactic x - ray point sources with far - infrared luminosities @xcite and x - ray luminosity with infrared plus ultraviolet emission @xcite . however , low mass x - ray binaries , which trace stellar mass ( m@xmath5 ) rather than star formation , also contributes to the x - ray emission and becomes the primary source of x - ray radiation in galaxies with low sfr / m@xmath5 @xcite . an active nucleus complicates using x - ray emission as a sfr indicator : accretion onto the smbh dominates emission above 2 kev . isolating the hard x - ray emission due to star formation then becomes prohibitively difficult when the nucleus is not resolved . however , soft x - ray emission can have comparable contributions from scattered / reflected agn emission and thermal emission from gas associated with starburst activity when the agn is obscured or has very low luminosity . determining the relative contribution of agn and star formation to the soft x - ray emission can be achieved through high resolution spectroscopy which resolves narrow emission lines . various diagnostics can be used to distinguish between collisionally and photoionized plasma , such as the width of the radiative recombination continua @xcite and ratios of forbidden , intercombination and resonance lines in the ovii triplet @xcite . however , with current x - ray missions ( i.e. _ chandra _ and _ xmm - newton _ ) high resolution spectra are only obtainable through grating observations , necessitating long exposure times on x - ray bright sources to achieve adequate signal to noise . how well can less time - intensive low resolution ccd spectroscopy achieve the goal of disentangling starburst from agn activity in soft x - rays ? local obscured agn , seyfert 2 galaxies ( sy2s ) , provide an ideal laboratory to answer this question . as these sources are relatively nearby and the accretion disk is hidden behind an obscuring `` torus , '' circumnuclear starbursts are visible . hence , both agn and host galaxy star formation can be modeled . levenson et al . ( 2004 , 2005 ) analyzed _ chandra _ observations of two sy2/starburst composites ( ngc 5135 and ngc 7130 ) and demonstrated that a thermal component due to stellar processes was necessary to adequately model the soft x - ray emission . we expand this methodology to sy2s in general to investigate the efficacy of modeling soft x - ray emission with a thermal component to describe host galaxy star formation . our analysis focuses on two samples of sy2s : an optically selected [ oiii]5007 sample @xcite and a mid - infrared selected 12@xmath3 m sample @xcite . using _ chandra _ and _ xmm - newton _ observations , we decomposed the soft x - ray emission into a star - forming and an agn component . for sy2s observed with _ chandra _ that have significant host galaxy emission , we are able to resolve the point source associated with the agn and can therefore remove it and analyze emission soley from the host galaxy , which we ascribe to star formation ( l@xmath1 ) . for the sy2s observed with _ xmm - newton _ and the remaining _ chandra _ sources , we fit the x - ray spectra with a combination of a thermal and power - law model . to account for the presence of x - ray binaries , we scale the soft thermal and power law fluxes using star - forming galaxies from the xsings sample as a template ( ptak et al . 2012 ) , obtaining soft x - ray luminosity values associated with star formation ( l@xmath1 ) and agn activity ( l@xmath0 ) . we compare these estimates of l@xmath1 and l@xmath0 with infrared ( ir ) spectral signatures that accurately describe star formation ( [ neii]12.8@xmath3 m luminosity , l@xmath6}$ ] , @xcite and agn activity ( [ oiv]26@xmath3 m luminosity , l@xmath7}$ ] , @xcite to test the accuracy of this decomposition . the selection details for the [ oiii ] and 12@xmath3 m samples are presented in @xcite and @xcite and are therefore only briefly mentioned here . the [ oiii ] sample was selected from sdss data release 4 @xcite . these sources live in the seyfert 2 locus of the bpt diagram @xcite according to the @xcite demarcation , and the sample is complete to a flux limit of 4@xmath8 erg @xmath9 s@xmath10 , totaling 20 sources . seventeen of these sy2s were observed with _ xmm - newton_. the 12@xmath3 m sample is comprised of sy2s from the _ iras _ point source catalog , is complete to a flux density limity of 0.3 jy at 12@xmath3 m and totals 31 galaxies . twenty - eight of these sources have archival _ chandra _ and/or _ xmm - newton _ data . in @xcite and @xcite we discussed in detail the x - ray data reduction , using xassist @xcite , and broad - band x - ray ( 0.5 - 10 kev ) spectral fitting of these sources with xspec . each galaxy was fit with an absorbed power law or when necessary , double absorbed power law , with the spectral indices initially tied together , simulating a partial covering geometry where a certain percentage of the transmitted x - ray emission is absorbed and the remainder scattered into the line of sight ; fe k@xmath11 emission , when present , was modeled with a gaussian component . to accomodate the possible presence of starburst emission in soft x - rays ( 0.5 - 2 kev ) , we included a thermal model ( apec ) in these spectral fits which describes emission from a collisionally ionized plasma . here we revisit more detailed modeling of the soft emission . since our aim is to study host galaxy star formation in tandem with agn activity , we limit our analysis to the sy2s where addition of a thermal model component results in a detection of thermal emission , rather than an upper limit . as a result , ngc 291 , cgcg 064 - 017 , cgcg 218 - 007 and 2masx j13463217 + 6423247 were dropped from the [ oiii ] sample . we also exclude the three 12@xmath3 m sources that were not detected in 2 - 10 kev x - rays : f08572 + 3915 , ngc 5953 and ngc 7590 . ngc 1068 , part of the original 12@xmath3 m sample , has soft x - ray emission dominated by agn photoionization ( e.g. * ? ? ? * ) and has therefore been omitted from this analysis . in total , 38 seyfert 2 galaxies remain , 14 from the [ oiii ] sample and 24 from the 12@xmath3 m sample . as noted in @xcite , several 12@xmath3 m sy2s had both _ chandra _ and _ xmm - newton _ archival observations . we tested if differences in the spectra for an individual source between the two observatories were due to aperture effects , since _ chandra _ has a higher spatial resolution than _ xmm - newton _ , or variability . our aim in that analysis was to constrain the properties of the agn , so we focused on the _ chandra _ data to isolate the point source , using _ xmm - newton _ information to help constrain the spectral fit for non - variable sources . in this analysis , we favor larger aperture extraction areas as they encompass extended x - ray emission due to host galaxy star formation . for the sy2s in the 12@xmath3 m sample with both _ xmm - newton _ and _ chandra _ observations , the _ chandra _ extraction area was adjusted to match the area from _ xmm - newton _ or to encompass all visible extended x - ray emission . for the sources with data , we increased the extraction area from the analysis presented in @xcite to ensure all extended x - ray emission is covered . in tables [ o3_ap ] and [ 12m_ap ] , we list the aperture sizes used to extract spectra for the [ oiii ] and 12@xmath3 m sy2s , respectively . the spectral fitting results for the [ oiii ] and 12@xmath3 m samples are summarized in tables [ o3_apec ] and [ 12m_apec ] . the broad - band x - ray spectra ( 0.5 - 10 kev ) were fitted to most fully constrain the agn contribution to the soft x - ray emission . multiple spectra of the same source ( from the three _ xmm - newton _ detectors and/or multiple observations ) , were initially fit simultaneously with a multiplicative constant allowed to vary to account for differences in detector sensitivity . for variable sources , individual model parameters were fit independently . we imposed a lower column density ( n@xmath12 ) limit of that due to our galaxy along the line of sight to the source . in some cases , the best fit absorption pegged at this lower limit and n@xmath12 was subsequently frozen to this value . for the sources better described by a double power law model , which as noted above represents a partial covering geometry , the photon indices were tied together , with the normalizations left free to vary and an independent absorption attenuating the second power law component ( n@xmath13 ) ; the sy2s best accommodated with a single power law can be identified by the blank entry in the n@xmath13 column of tables [ o3_apec ] and [ 12m_apec ] . the lower temperature limit ( kt ) was set at 0.1 kev . ten sy2 galaxies were well fit by a second apec component , significant at greater than the 3@xmath2 level according to the f - test : 2masx j08244333 + 2959238 , ngc 424 , ngc 1320 , ngc 1386 , ngc 4388 , m-3 - 34 - 64 , ngc 5194 , mrk 463 , ngc 7582 and ngc 7674 . this result indicates the presence of a two temperature gas , similar to the results of x - ray observations of starburst galaxies @xcite : one phase at kt between 0.6 - 0.75 kev and a lower temperature gas at @xmath140.15 kev , with the exception of m-3 - 34 - 64 which has two hot gas components at 0.77@xmath15 kev and 2.85@xmath16 . we list the observed soft fluxes in tables [ o3_flux ] and [ 12m_flux ] : total 0.5 - 2 kev flux ( f@xmath17 ) , power law component of soft x - ray flux ( f@xmath18 ) and apec component of soft x - ray flux ( f@xmath19 ) . in the subsequent analysis , we omit the sy2s with a poorly costrained apec flux , i.e. the sources where the error is an order of magnitude or more higher or lower than the measured value ( 2masx j11110693 + 0228477 and cgcg 242 - 028 ) , leaving us with 35 sy2s . before assigning `` thermal '' ( f@xmath19 ) flux to starburst activity and power law emission ( f@xmath18 ) to agn , we need to account for the effects of x - ray binaries which contribute to the non - thermal emission . as noted above , high mass x - ray binaries indicate instantaneous star formation in the host galaxy and this effect is incorporated into x - ray sfr calibrations . since x - ray binaries boost the non - thermal flux , f@xmath18 will overestimate the agn emission while f@xmath19 under - represents host galaxy star formation . to assess the contribution of x - ray binaries to the observable emission , we model just the extended emission when the agn is resolvable . for the remaining sources , we apply a correction to f@xmath19 and f@xmath18 using results of x - ray analysis of star - forming galaxies from the xsings sample ( ptak et al . 2012 ) . spatially resolved emission from the narrow line region ( nlr ) can also potentially affect the results , over - predicting the luminosity attributable to host galaxy star formation . for a handful of 12@xmath3 m sources , the major radius ( @xmath20 ) of the nlr was measured from @xcite , while we used the relations from @xcite to estimate nlr @xmath20 values for the remaining 12@xmath3 m sources and the [ oiii ] sample . for the [ oiii ] sy2s , the nlr radius ranges from 280 - 900 pc , with a median size of @xmath140.5 kpc . similarly , the 12@xmath3 m selected sy2s have nlr sizes ranging from 80 - 1320 pc , with a median @xmath20 value of @xmath140.39 kpc . for all [ oiii ] and a majority of 12@xmath3 m sources , the nlr is an order of magnitude or more smaller than the x - ray extraction radius . the nlr is thus likely not resolvable at these scales , so we only focus on quantifying the effects of x - ray binaries on the observed emission . one notable exception is ic 5063 , where the radius of the nlr ( @xmath141.3 kpc ) comprises a signficant fraction of the x - ray extraction radius ( 3 kpc ) . we will return to this sy2 in section [ results ] . another potential bias could be due to photoionized gas from the nlr being indistinguishable from collisionally ionized gas at low ccd resolution . however as we show in section [ results ] , such an effect does not systematically affect our results . due to _ chandra _ s arc second resolution and the close proximity of the 12@xmath3 m sy2s , the central point source ( i.e. agn ) is resolvable for a subset of our sample . to accurately constrain the agn position , we ran wavedetect on 2 - 10 kev _ chandra _ images of the 12@xmath3 m sample . spectra were then extracted from the remaining emission , omitting the agn identified by wavedetect . similar to above , the broad band x - ray emission was fitted with ( an ) absorbed power law(s ) and an apec component . we only include those sources where the spectra from the residual emission have a high enough signal - to - noise to be fit with the @xmath21 statistic ( minimum grouping of 10 counts per bin ) , therefore dropping the sources where the x - ray emission is concentrated in the nucleus : ngc 424 , tololo 1238 - 364 , f05189 - 2524 and ngc 5347 . the summary of the spectral fitting for these 11 sources is listed in table [ pt_src_rem ] , and the associated total soft x - ray flux from the unresoloved emission ( i.e. host galaxy emission after agn removal ) is listed in the second column of table [ 12m_flux ] . to first order , the total soft x - ray flux from the unresolved emission can be considered an estimate of host galaxy star formation , l@xmath1 . l@xmath0 is then estimated by taking the difference between l@xmath1 and l@xmath17 for these sy2s , with the associated error derived by scaling the error of total soft x - ray luminosity by the ratio of the agn luminosity to the total luminosity . we have used the results of xsings to estimate the contribution of x - ray binaries ( xrbs ) to the soft x - ray emission in host galaxies . the xsings project entails the analysis of 96 chandra observations of 56 spitzer infrared nearby galaxy survey ( sings ) galaxies ( kennicutt et al , 2003 ) . sings is designed to cover a wide range of galaxy properties such as morphological type , star formation rate ( sfr ) , stellar mass and metallicity , and the main aim of xsings is to study the relationships between x - ray emission ( x - ray binaries , nuclear activity and hot ism ) with these properties . in ptak et al . ( 2012 , in prep . ) the spatially - averaged x - ray emission of xsings galaxies is discussed and compared to the summed x - ray binary luminosities for galaxies with significant detections ( x - ray binary catalogs for all xsings galaxies will be presented in jenkins et al . xrbs emit as a power law , but provide a measure of the instantaneous sfr along with thermal emission in soft x - rays . we therefore have : @xmath22 @xmath23 we further assume that a certain fraction of x - ray emission from xrbs contributes to host galaxy star formation : @xmath24 using the 22 pure star - forming galaxies in xsings ( i.e. non sy2s and non liners ) , we have calculated the mean ratio of x - ray emission from resolved non - nuclear point sources ( i.e. x - ray binaries ) to total x - ray emission in the 0.5 - 2 kev band ( @xmath25 ) , deriving @xmath26 . solving the above equations , we obtain : @xmath27 @xmath28 to account for uncertainties in the luminosities derived from spectral fitting ( i.e. l@xmath19 and l@xmath18 ) and the wide spread in @xmath25 values , we have performed monte carlo simulations to derive l@xmath1 and l@xmath0 . we have drawn 1000 random values for l@xmath19 , l@xmath18 and @xmath25 from a gaussian distribution centered on the best - fit values of l@xmath19 and l@xmath18 ( and mean value of @xmath25 ) , with @xmath2 corresponding to the errors derived from spectral fitting ( and standard deviation on @xmath25 ) . using these simulated parameters and eqs . 4 and 5 , we calculated 1000 simulated values for l@xmath1 and l@xmath0 . the mode of these distributions is then taken as l@xmath1 and l@xmath0 , with the 68% confidence interval as the associated errors . for the subset of sy2s with high signal - to - noise _ chandra _ spectra of the soft unresolved emission , we compare the results of l@xmath1 ( l@xmath0 ) derived via _ chandra _ imaging analysis and the simulation approach described above . here , l@xmath29 ( l@xmath30 ) represents the soft x - ray luminosity due to star formation ( agn ) based on spectral fitting of just the unresolved emission , while l@xmath31 ( l@xmath32 ) is the estimate of the star formation ( agn ) component of the soft x - ray flux calculated via simulations . to account for errors in both parameters , we use a bayesian approach to linear regression @xcite . both here and in section [ results ] , the high and low error bars on the x - ray flux were averaged to provide symmetric errors for the linear regression routine . we plot l@xmath31 as a function of l@xmath29 and l@xmath32 against l@xmath30 in figure [ chandra_v_sings ] . the grey shaded regions enclosed by the dashed lines delineate the 3@xmath2 confidence level on the regression line , based on the median of the posterior distribution of the slope , intercept and mean distribution of the independent variable , with the dotted - dashed line denoting the line of equality . though the l@xmath31 values are somewhat systematically higher than l@xmath29 , they do agree within the 3@xmath2 contours . the agreeement among these independent methods suggests that using xsings galaxies as a template to correct the l@xmath19 and l@xmath18 values for the presence of x - ray binaries provides a reasonable estimate to decompose the soft x - ray emission in the absence of resolveable nuclear emission . in the subsequent analysis ( i.e. section [ results ] ) , we use l@xmath29 and l@xmath30 as l@xmath1 and l@xmath0 for the 11 _ chandra _ sources with high signal - to - noise . l@xmath1 and l@xmath0 for the remaining 24 sy2s are estimated via the monte carlo simulations outlined above . these values are listed in tables [ o3_lum ] and [ 12m_lum ] for the [ oiii ] and 12@xmath3 m samples , respectively . we compare our x - ray decomposition with information gleaned from high resolution infrared spectroscopy , using diagnostics that cleanly trace the separate processes of agn and star formation activity . the flux of the [ oiv]26@xmath3 m line has been shown to accurately describe intrinsic agn flux @xcite while the [ neii]12.8@xmath3 m emission reliably tracks star formation in both quiescent and active systems ( lamassa et al . , 2012 ) . we therefore use these fluxes to decompose the soft x - ray emission into a star forming and an agn component by solving l@xmath17 = @xmath33l@xmath6}$ ] + @xmath34l@xmath7}$ ] using ordinary least squares multiple linear regression on both the 12@xmath3 m and [ oiii ] samples , obtaining @xmath11=1.07 and @xmath35=0.16 ( see * ? ? ? * ; * ? ? ? * for calculated values of l@xmath7}$ ] and l@xmath6}$ ] for these samples ) . here l@xmath17 represents the total soft x - ray emission ( i.e. column 1 in tables [ o3_flux ] and [ 12m_flux ] ) . figure [ softx_ne2_o4 ] illustrates the results of this decomposition . we consider the x - ray variable sy2s , ngc 5506 and ngc 7172 as independent data points , but we do average the soft x - ray flux for the latter two ngc 5506 observations , which are consistent . though we have only one ir measurement for each of these sources , we do not expect the infrared emission to vary as much as the x - ray emission . we color code the sources in figure [ softx_ne2_o4 ] according to sample : cyan triangles represent the 12@xmath3 m sy2s ( with the x - ray variable sources connected by a vertical line ) , blue diamonds denote the subset of 12@xmath3 m sources with high signal - to - noise _ chandra _ imaging of the unresolved emission and red squares mark the [ oiii ] selected sources . the overplotted line indicates where the two quantities are equal . with the exception of a handful of outliers , a relatively good agreement is present between the observed soft x - ray emission and the ir decomposition . we note that this consistency holds for both the 12@xmath3 m and [ oiii ] samples , with no apparent systematic offset introduced from different sample selection techniques . in the following analysis , we therefore use @xmath33l@xmath6}$ ] and @xmath34l@xmath7}$ ] as an independent estimate of the starburst and agn contribution , respectively , to the soft x - ray emission . we plot l@xmath1 as a function of @xmath33 l@xmath6}$ ] and l@xmath0 as a function of @xmath34 l@xmath7}$ ] in figure [ lx_ir ] . we note that the abscissa errors are on the order of the symbol size , due to the 5@xmath2 detection limit we imposed on ir detections ( see lamassa et al . 2012 ) , and therefore are not plotted in figure [ lx_ir ] , though they are included in the errors for the bayesian linear regression routine . due to the large error on l@xmath0 for ngc 7674 , this source was dropped from the l@xmath0 vs. @xmath34 l@xmath7}$ ] analysis . in figure [ lx_ir ] , a dotted - dashed line is plotted to indicate where the two quantities are equal while the grey shaded region illustrates the 3@xmath2 confidence interval on the regression line . l@xmath1 and l@xmath0 are approximately equivalent to @xmath33 l@xmath6}$ ] and @xmath34 l@xmath7}$ ] , respectively , albeit with wide scatter for a handful of individual sources . deviations of the regression fit from equality at greater than the 3@xmath2 level are slight , and appear at higher luminosities for the star formation decomposition ( @xmath36 erg / s/@xmath37 ) and moderate luminosities for the agn parameterization ( @xmath38 l@xmath39 erg / s/@xmath37 ) . as mentioned previously , the nlr for ic 5063 takes up a signficant fraction of the the x - ray extraction region , where contributions from resolved nlr emission could potentially bias the x - ray estimate of host galaxy star formation . however , as seen in figure [ lx_ir ] , where ic 5068 is the blue diamond at @xmath33 l@xmath6}$ ] = 40.4 dex and l@xmath1 = 40.1 dex , this source lies well within the shaded region . hence nlr contamination appears to be negligible , possibly due to the removal of the agn from the _ chandra _ imaging analysis . we note that aperture bias can contribute to these slight disagreements : [ neii ] ( [ oiv ] ) is measured through the short - high ( long - high ) module on the irs spectrograph on _ spitzer _ , corresponding to sizes [email protected] ( [email protected] ) , whereas the x - ray apertures capture the full emission from the host galaxy ( see tables [ o3_ap ] and [ 12m_ap ] ) . hence , biases can be introduced from comparing parameters that sample different physical scales of the host galaxy , introducing scatter into the relation . in the agn parameterization , there are 4 obvious outliers from the relation : ic 0486 and mrk 463 ( members of the [ oiii ] sample with over - predicted l@xmath0 , red squares in figure [ lx_ir ] ) , ngc 4388 ( from the 12@xmath3 m sample with under - predicted l@xmath0 , blue diamond in figure [ lx_ir ] ) and ngc 5506 ( from the 12@xmath3 m sample with over - predicted l@xmath0 , an x - ray variable source with a vertical line connecting the cyan triangles in [ lx_ir ] ) . interestingly , there are no correlations between outliers in the l@xmath0 parameterization and physical size of x - ray emitting region ( where mis - match beween aperture sizes between _ chandra _ or _ xmm - newton _ and _ spitzer _ could introduce disagreements ) , quality of soft x - ray spectra , sample selection , source variability or even method for estimating l@xmath0 . in general , however , deviations of the best - fit trends between l@xmath1 versus @xmath33 l@xmath6}$ ] and l@xmath0 versus @xmath34 l@xmath7}$ ] from equality at the 3@xmath2 level are slight . such an agreement demonstrates that our method of using low resolution ccd spectroscopy to decompose the soft x - ray emission into a star - forming and an agn component is moderately successful , with greater consistency in the star formation estimation . l@xmath1 can then be used to estimate the host galaxy sfr in sy2s using the l@xmath17 calibrations from @xcite and @xcite . for the 12@xmath3 m sample , we have tested the efficacy of our decomposition by deriving a soft x - ray sfr using the @xcite calibration and comparing this with the far - infrared ( fir ) sfr using the @xcite calibration , using iras flux densities reported in @xcite . the [ oiii ] sample , however , lacks a reliable independent host - galaxy wide sfr . these sources were not detected by _ iras _ and the sdss sfrs derived through the 3 fiber ( see * ? ? ? * for a discussion ) only covers the innermost regions of the galaxy , while the x - ray aperture captures the full emission . we therefore use only the 12@xmath3 m sample to test the x - ray derived sfrs . to estimate sfr@xmath41 , we used the sum of 60@xmath3 m and 100@xmath3 m _ iras _ flux densities . we note that the @xcite calibration is based on mir through fir luminosities for starburst galaxies , and in such systems , including the iras 12@xmath3 m and 25@xmath3 m luminosities as part of the fir luminosity is important . however , these mir luminosities have significant contributions from dust heated by the agn , so we only consider the 60@xmath3 m and 100@xmath3 m luminosities when deriving sfr@xmath41 , which are less contaminated by the agn . as shown in figure [ firsfr_xsfr ] , a good agreement exists between sfr@xmath17 and sfr@xmath41 , despite the latter sfr parameterization being most appropriate for starburst galaxies . this agreement suggests that any unresolved photoionized emission from the nlr does not introduce any systematic offsets into our results , but rather operates at a level less than the observed scatter . the 2 sy2s that are the greatest outliers ( i.e. , largest sfr@xmath41 values ) are arp 220 and f05189 - 2524 , which are ultra luminous infrared galaxies ( ulirgs ) . we have modeled the 0.5 - 10 kev spectra of two homogeneous samples of seyfert 2 galaxies to disentangle the starburst and agn emission in soft x - rays ( 0.5 - 2 kev ) . eleven of these sy2s , observed with _ , had high signal - to - noise unresolved emission after removing the agn . we derive l@xmath1 from spectrally fitting this unresolved emission and assign l@xmath0 to the difference between the total soft x - ray emission and l@xmath1 . the remaining 24 sources were decomposed into an agn and star formation component by modeling the soft x - ray emission with a power law and thermal model . the luminosities of these sources were converted to l@xmath1 and l@xmath0 values based on a scaling factor derived from xsings analysis on normal star forming galaxies . to account for errors on the luminosities derived from spectral fits and the scaling factor , we executed monte carlo simulations , assuming a gaussian distribution of random variables centered on the best - fit values of l@xmath19 , l@xmath18 and @xmath25 , with the spread corresponding to the errors on these parameters . using eqs . 4 and 5 , we calculate l@xmath1 and l@xmath0 from the 1000 simulated parameters and use the mode of this distribution as our estimate of the x - ray luminosities due to star formation and agn activity . * the soft x - ray spectra of 10 sy2s were well fitted by two thermal model components , indicating the presence of a two temperature gas . this result is similar to what has been observed in starbust galaxies @xcite . * for the subset of 11 sy2s with high signal - to - noise _ chandra _ imaging of unresolved host galaxy emission , estimates of the soft x - ray emission due to star formation and agn from both _ imaging analysis and estimates from the monte carlo simulations agree at the 3@xmath2 level . in the absence of resolved nuclear emission , scaling l@xmath18 and l@xmath42 by the factors derived from xsings analysis is thus a reasonable method to estimate the agn and star formation contributions to the soft x - ray emission . * the independent decompositions of the soft x - ray luminosity into a star formation and agn component using ir data as a proxy and scaling l@xmath19 and l@xmath18 based on xsings galaxies largely agree within a 3@xmath2 confidence interval . deviations of the best fit regression line from this agreement are slight and appear at higher luminosities ( @xmath43 erg / s/@xmath37 ) in the star formation decomposition and at moderate luminosities ( @xmath38 l@xmath44 erg / s/@xmath37 ) when describing agn emission . though more scatter in individual sources is evident in the agn decomposition , the star formation relationship is more consistent among both methods . * comparison of our calculated x - ray sfr using the @xcite calibration with an fir derived sfr from @xcite for the 12@xmath3 m sample shows general agreement . we have demonstrated that analysis of low resolution ccd x - ray spectra can effectively disentangle emission from agn activity and star formation in 0.5 - 2kev x - rays . using the decomposition we have presented , l@xmath1 can be used to cleanly estimate the sfr in sy2s using existing calibrations ( * ? ? ? * ; * ? ? ? * e.g. ) , complementing previous studies in the optical and infrared and providing a more panchromatic view of the interplay between smbh accretion and star formation . mrk 0609 & 0.034 & 65/34/35 & 44/23/24 + ic 0486 & 0.027 & 34/33/35 & 19/18/19 + 2masx j08035923 + 2345201 & 0.029 & 17/20/18 & 10/12/11 + 2masx j08244333 + 2959238 & 0.025 & 40/40/40 & 21/21/21 + 2masx j10181928 + 3722419 & 0.049 & 20/22/22 & 20/22/22 + 2masx j11110693 + 0228477 & 0.035 & 18/-/- & 13 + cgcg 242 - 028 & 0.026 & 19/22/13 & 10/12/7 + sbs 1133 + 572 & 0.050 & 20/16/17 & 20/16/17 + mrk 1457 & 0.049 & 19/13/13 & 18/13/13 + 2masx j11570483 + 5249036 & 0.036 & 35/35/35 & 26/26/26 + 2masx j12183945 + 4706275 & 0.094 & -/30/31 & 53/55 + 2masx j12384342 + 0927362 & 0.083 & 33/30/30 & 52/47/47 + ngc 5695 & 0.014 & 31/22/17 & 9/7/5 + ngc 0424 & 0.012 & _ xmm - newton _ & 00029242301 & 34/33/33 & 8/7/7 + & & _ chandra _ & 03146 & 15 & 3 + ngc 1144 & 0.028 & _ xmm - newton _ & 0312190401 & 36/33/33 & 20/18/18 + ngc 1320 & 0.009 & _ xmm - newton _ & 0405240201 & 37/35/35 & 6/6/6 + ngc 1386 & 0.003 & _ xmm - newton _ & 0140950201 & 45/40/40 & 2/2/2 + & & _ chandra _ & 04076 & 19 & 1 + ngc 1667 & 0.015 & _ xmm - newton _ & 0200660401 & 32/22/34 & 10/7/10 + f05189 - 2524 & 0.043 & _ xmm - newton _ & 0085640101 & 41/31/33 & 35/26/28 + & & _ chandra _ & 02034 & 40 & 34 + & & _ chandra _ & 03432 & 40 & 34 + ngc 3982 & 0.004 & _ xmm - newton _ & 0204651201 & 34/37/44 & 3/3/4 + & & _ chandra _ & 04845 & 34 & 3 + ngc 4388 & 0.008 & _ xmm - newton _ & 0110930701 & 36/35/43 & 7/7/8 + & & _ xmm - newton _ & 0110930301 & 37/32/33 & 7/6/6 + & & _ chandra _ & 01619 & 36 & 7 + ngc 4501 & 0.008 & _ xmm - newton _ & 0112550801 & 18/22/33 & 3/4/6 + & & _ chandra _ & 02922 & 20 & 4 + tololo 1238 - 364 & 0.011 & _ chandra _ & 04844 & 20 & 5 + ngc 4968 & 0.010 & _ xmm - newton _ & 0002940101 & 17/18/14 & 4/4/3 + & & _ xmm - newton _ & 0200660201 & 21/18/18 & 5/4/4 + m-3 - 34 - 64 & 0.017 & _ xmm - newton _ & 0206580101 & 59/34/38 & 21/12/14 + ngc 5135 & 0.014 & _ chandra _ & 02187 & 20 & 6 + ngc 5194 & 0.002 & _ xmm - newton _ & 0112840201 & 65/42/49 & 3/2/2 + & & _ xmm - newton _ & 0303420101 & -/56/49 & -/2/2 + & & _ xmm - newton _ & 0303420201 & -/61/46 & -/3/2 + & & _ chandra _ & 00354 & 40 & 2 + & & _ chandra _ & 01622 & 40 & 2 + & & _ chandra _ & 03932 & 40 & 2 + ngc 5347 & 0.008 & _ chandra _ & 04867 & 6 & 1 + mrk 463 & 0.050 & _ xmm - newton _ & 0094401201 & 49/45/45 & 49/45/45 + & & _ chandra _ & 04913 & 40 & 40 + ngc 5506 & 0.006 & _ xmm - newton _ & 0013140101 & 60/-/- & 9/-/- + & & _ xmm - newton _ & 0013140201 & 54/-/- & 8/-/- + & & _ xmm - newton _ & 0201830201 & 56/-/53 & 8/-/8 + & & _ xmm - newton _ & 0201830301 & 52/-/64 & 8/-/9 + & & _ xmm - newton _ & 0201830401 & 55/-/55 & 8/-/8 + & & _ xmm - newton _ & 0201830501 & 68/-/98 & 10/-/14 + & & _ xmm - newton _ & 0554170201 & 64/-/- & 9/-/- + & & _ xmm - newton _ & 0554170101 & 66/-/- & 10/-/- + arp 220 & 0.018 & _ xmm - newton _ & 0101640801 & 20/21/25 & 8/8/9 + & & _ xmm - newton _ & 0101640901 & 28/25/23 & 11/9/9 + & & _ xmm - newton _ & 0205510201 & 34/28/22 & 13/11/8 + & & _ chandra _ & 00869 & 40 & 15 + ngc 6890 & 0.008 & _ xmm - newton _ & 0301151001 & 17/13/15 & 3/2/2 + ic 5063 & 0.011 & _ chandra _ & 07878 & 12 & 3 + ngc 7130 & 0.016 & _ chandra _ & 02188 & 17 & 5 + ngc 7172 & 0.009 & _ xmm - newton _ & 0147920601 & 52/41/34 & 8/7/5 + & & _ xmm - newton _ & 0202860101 & 61/59/60 & 10/9/10 + & & _ xmm - newton _ & 0414580101 & 53/59/55 & 8/9/9 + ngc 7582 & 0.005 & _ xmm - newton _ & 0112310201 & 44/41/41 & 4/4/4 + & & _ xmm - newton _ & 0204610101 & 60/43/47 & 6/4/4 + & & _ chandra _ & 00436 & 43 & 4 + & & _ chandra _ & 02319 & 43 & 4 + ngc 7674 & 0.029 & _ xmm - newton _ & 0200660101 & 19/18/16 & 11/10/9 + mrk 0609 & 0.04 & 0.27@xmath45 & ... & 1.77@xmath45 & ... & ... & 160 ( 203 ) + ic 0486 & @xmath460.06 & @xmath460.16 & ... & 1.23@xmath47 & 1.00@xmath48 & ... & 636 ( 629 ) + 2masx j08035923 + 2345201 & 0.61@xmath49 & @xmath460.12 & ... & 2.84@xmath50 & 46.7@xmath51 & ... & 120 ( 95 ) + 2masx j08244333 + 2959238 & 0.03 & 0.68@xmath52 & @xmath460.12 & 1.54@xmath53 & 15.5@xmath54 & 229 ( 186 ) & 263 ( 188 ) + 2masx j10181928 + 3722419 & 0.01 & 0.18@xmath55 & ... & 2.63@xmath56 & ... & ... & 52.1 ( 61 ) + 2masx j11110693 + 0228477 & @xmath460.62 & 0.23@xmath57 & ... & 2.04@xmath58 & ... & ... & 37.5 ( 36 ) + cgcg 242 - 028 & 0.69@xmath59 & @xmath460.15 & ... & 0.31@xmath60 & ... & ... & 87.0 ( 90 ) + sbs 1133 + 572 & @xmath460.10 & 0.824@xmath61 & ... & 3.08@xmath62 & 57.6@xmath63 & ... & 38.1 ( 48 ) + mrk 1457 & 0.66@xmath64 & 0.14@xmath65 & ... & 1.29@xmath66 & 27.5@xmath67 & ... & 35.2 ( 35 ) + 2masx j11570483 + 5249036 & @xmath460.1 & 0.17@xmath68 & ... & 2.85@xmath69 & 83.9@xmath70 & ... & 123 ( 76 ) + 2masx j12183945 + 4706275 & 0.02 & @xmath460.24 & ... & 1.95@xmath71 & 87.2@xmath72 & ... & 15 ( 19 ) + 2masx j12384342 + 0927362 & @xmath460.07 & @xmath460.23 & ... & 2.17@xmath73 & 29.3@xmath74 & ... & 195 ( 164 ) + ngc 5695 & @xmath460.59 & 0.24@xmath75 & ... & 2.55@xmath76 & ... & ... & 75.7 ( 62 ) + ngc 0424 & 0.04@xmath65 & 0.69@xmath77 & @xmath460.13 & 2.21@xmath78 & 32.6@xmath79 & 246 ( 178 ) & 267 ( 180 ) + ngc 1144 & 0.06 & 0.37@xmath80 & ... & 1.91@xmath81 & 47.0@xmath82 & .. & 175 ( 149 ) + ngc 1320 & @xmath460.08 & 0.15@xmath65 & 0.73@xmath83 & 2.97@xmath84 & 48.0@xmath85 & 233 ( 188 ) & 279 ( 190 ) + ngc 1386 & 0.04@xmath86 & 0.67@xmath65 & 0.13@xmath86 & 2.72@xmath87 & 31.4@xmath88 & 398 ( 332 ) & 435 ( 334 ) + ngc 1667 & 0.05 & 0.33@xmath89 & ... & 2.18@xmath90 & ... & ... & 49.8 ( 38 ) + f05189 - 2524 & 0.02 & 0.1 & ... & 2.05@xmath91 & 6.77@xmath92 & ... & 501 ( 374 ) + ngc 3982 & 0.01 & 0.29@xmath65 & ... & 2.39@xmath93 & 40.3@xmath94 & ... & 174 ( 160 ) + ngc 4388 ( xmm ) & 0.03 & 0.60@xmath95 & 0.15@xmath96 & 1.25@xmath64 & 24.3@xmath97 & 510 ( 495 ) & 580 ( 498 ) + ngc 4388 ( chandra ) & 0.03 & 0.60@xmath98 & 0.15@xmath99 & 0.38@xmath100 & 25.6@xmath101 & 210 ( 166 ) & 269 ( 168 ) + ngc 4501 & 0.03 & 0.36@xmath102 & ... & 1.52@xmath103 & ... & ... & 94.2 ( 102 ) + tololo 1238 - 364 & @xmath460.19 & 0.32@xmath104 & ... & 2.45@xmath105 & ... & ... & 91.6 ( 105 ) + ngc 4968 & 0.08 & 0.68@xmath64 & ... & 1.72@xmath106 & ... & ... & 325 ( 268 ) + m-3 - 34 - 64 & 0.05 & 2.85@xmath107 & 0.77@xmath15 & 3.24@xmath108 & 53.6@xmath109 & 780 ( 492 ) & 848 ( 493 ) + ngc 5135 & 0.05 & 0.73@xmath65 & ... & 2.34@xmath110 & ... & .. & 166 ( 114 ) + ngc 5194 & 0.04@xmath111 & 0.18@xmath111 & 0.62@xmath111 & 3.18@xmath91 & 10.2@xmath112 & 1560 ( 1291 ) & 2021 ( 1293 ) + ngc 5347 & 0.02 & @xmath460.21 & ... & 1.53@xmath103 & 32.6@xmath113 & ... & 69.9 ( 82 ) + mrk 463 & 0.02 & 0.74@xmath99 & 0.20@xmath114 & 1.76@xmath115 & 28.6@xmath116 & 365 ( 316 ) & 392 ( 318 ) + ngc 5506 & 0.11@xmath111 & 0.74@xmath117 & ... & 1.71@xmath86 & 2.69@xmath96 & ... & 2646 ( 2380 ) + ngc 5506 & & 0.94@xmath118 & & & & & + ngc 5506 & 0.17@xmath111 & ... & ... & 1.83@xmath15 & 2.80@xmath119 & ... & 3982 ( 3137 ) + ngc 5506 & & & & 1.76@xmath120 & & & + ngc 5506 & 0.12@xmath111 & 0.83@xmath65 & & & & & + ngc 5506 & & 0.96@xmath117 & & & & & + arp 220 & 0.04 & 0.79@xmath98 & ... & 0.95@xmath121 & ... & ... & 309(302 ) + ngc 6890 & @xmath460.22 & 0.78@xmath122 & ... & 3.28@xmath123 & 27.4@xmath124 & ... & 164 ( 148 ) + ic 5063 & 0.06 & 0.60@xmath125 & ... & 1.85@xmath126 & 21.0@xmath127 & ... & 198 ( 141 ) + ngc 7130 & 0.06@xmath128 & 0.63@xmath65 & ... & 2.43@xmath129 & 75.4@xmath130 & ... & 334 ( 240 ) + ngc 7172 & 0.02 & 0.28@xmath45 & ... & 1.56@xmath96 & 7.36@xmath131 & ... & 2074 ( 1746 ) + ngc 7172 & & 0.26@xmath15 & ... & 1.54@xmath96 & 8.13@xmath132 & & + ngc 7582 ( chandra ) & 0.01 & 0.72@xmath117 & ... & 1.94@xmath133 & 24.6@xmath134 & ... & 355 ( 301 ) + ngc 7582 ( xmm ) & 0.01 & 0.18@xmath135 & 0.72@xmath111 & 1.75@xmath102 & 27.0@xmath136 & 1438 ( 886 ) & 1586 ( 888 ) + ngc 7674 & 0.04 & @xmath460.11 & 0.67@xmath137 & 0.62@xmath138 & 89.0@xmath139 & 66.6 ( 70 ) & 113 ( 72 ) + mrk 0609 & 79.1@xmath140 & 4.82@xmath141 & 74.3@xmath142 + ic 0486 & 35.0@xmath143 & 0.45@xmath144 & 34.6@xmath145 + 2masx j08035923 + 2345201 & 1.24@xmath146 & 0.83@xmath147 & 0.41@xmath148 + 2masx j08244333 + 2959238 & 5.25@xmath149 & 3.04@xmath150 & 2.21@xmath151 + 2masx j10181928 + 3722419 & 2.27@xmath152 & 1.04@xmath153 & 1.23@xmath154 + 2masx j11110693 + 0228477 & 1.44@xmath155 & 0.42@xmath156 & 1.02@xmath157 + cgcg 242 - 028 & 1.17@xmath158 & 0.80@xmath159 & 0.37@xmath160 + sbs 1133 + 572 & 3.46@xmath161 & 0.52@xmath162 & 2.94@xmath163 + mrk 1457 & 2.90@xmath164 & 2.36@xmath165 & 0.54@xmath166 + 2masx j11570483 + 5249036 & 3.83@xmath167 & 0.91@xmath168 & 2.92@xmath169 + 2masx j12183945 + 4706275 & 1.54@xmath170 & 0.55@xmath171 & 0.99@xmath172 + 2masx j12384342 + 0927362 & 6.23@xmath173 & 0.86@xmath174 & 5.37@xmath175 + ngc 5695 & 2.57@xmath176 & 0.64@xmath177 & 1.93@xmath178 + ngc 0424 & 31.5@xmath179 & ... & 10.6@xmath180 & 20.9@xmath181 + ngc 1144 & 8.39@xmath182 & ... & 2.87@xmath183 & 5.52@xmath184 + ngc 1320 & 28.4@xmath185 & ... & 10.5@xmath186 & 17.9@xmath187 + ngc 1386 & 25.2@xmath188 & 8.48@xmath189 & 11.7@xmath190 & 13.5@xmath191 + ngc 1667 & 7.31@xmath192 & ... & 3.22@xmath193 & 4.09@xmath194 + f05189 - 2524 & 12.1@xmath195 & ... & 2.22@xmath196 & 9.88@xmath197 + ngc 3982 & 9.14@xmath198 & 7.47@xmath199 & 3.39@xmath200 & 5.75@xmath201 + ngc 4388 & 28.5@xmath202 & 23.3@xmath203 & 17.8@xmath204 & 10.7@xmath205 & _ chandra _ & _ xmm - newton _ observations averaged + ngc 4501 & 7.24@xmath206 & 5.47@xmath207 & 3.83@xmath208 & 3.41@xmath209 + tololo 1238 - 364 & 14.2@xmath210 & ... & 5.43@xmath211 & 8.77@xmath212 + ngc 4968 & 5.85@xmath213 & ... & 1.27@xmath214 & 4.58@xmath215 + m-3 - 34 - 64 & 50.8@xmath216 & ... & 33.4@xmath217 & 17.4@xmath218 + ngc 5135 & 34.5@xmath219 & 16.4@xmath220 & 17.0@xmath221 & 17.5@xmath222 + ngc 5194 & 69.6@xmath223 & 56.1@xmath224 & 52.7@xmath225 & 16.9@xmath226 + ngc 5347 & 2.94@xmath227 & ... & 0.50@xmath228 & 2.44@xmath229 + mrk 463 & 14.3@xmath230 & 6.04@xmath231 & 7.50@xmath232 & 6.80@xmath233 + ngc 5506 & 308@xmath234 & ... & 6.07@xmath235 & 302@xmath234 & obsids 0013140101 , 0201830201 , 0201830301 , 0201830401 + ngc 5506 & 444@xmath236 & ... & 0 & 444@xmath236 & obsids 0013140201 & 0201830501 + ngc 5506 & 457@xmath237 & ... & 10.5@xmath238 & 446@xmath237 & obsids 0554170201 & 0554170101 + arp 220 & 4.90@xmath239 & 4.38@xmath240 & 3.05@xmath241 & 1.85@xmath242 + ngc 6890 & 11.5@xmath243 & ... & 2.62@xmath244 & 8.88@xmath245 + ic 5063 & 22.9@xmath246 & 4.79@xmath247 & 4.57@xmath248 & 18.3@xmath249 + ngc7130 & 23.6@xmath250 & 8.55@xmath251 & 10.3@xmath252 & 13.3@xmath253 + ngc7172 & 20.3@xmath254 & ... & 2.05@xmath255 & 18.3@xmath256 & obsid 0414580101 + ngc7172 & 11.8@xmath257 & ... & 2.06@xmath258 & 9.74@xmath259 & obsids 0147920601 & 0202860101 + ngc7582 & 40.3@xmath260 & 30.8@xmath261 & 15.2@xmath262 & 25.1@xmath263 & _ chandra _ & _ xmm - newton _ observations averaged + ngc7674 & 17.5@xmath264 & ... & 13.6@xmath265 & 3.90@xmath266 + ngc 1386 & @xmath460.05 & 0.58@xmath267 & ... & 3.23@xmath268 & 48.9@xmath269 & ... & 79.0 ( 54 ) + ngc 3982 & 0.01 & 0.20@xmath52 & ... & 2.04@xmath270 & ... & ... & 46.5 ( 25 ) + ngc 4388 & 0.03 & 0.60@xmath98 & 0.16@xmath99 & 0.98@xmath271 & ... & 85.4 ( 83 ) & 123 ( 84 ) + ngc 4501 & 0.03 & 0.37@xmath272 & ... & 2.26@xmath273 & ... & ... & 31.2(38 ) + ngc 5135 & @xmath460.13 & 0.84@xmath274 & 0.39@xmath275 & 2.24@xmath276 & ... & 62.8 ( 60 ) & 77.9 ( 62 ) + ngc 5194 & 0.02 & 0.60@xmath111 & 0.18@xmath128 & 3.13@xmath48 & 7.92@xmath277 & 525 ( 352 ) & 616 ( 353 ) + mrk 463 & @xmath460.05 & 0.62@xmath278 & ... & 2.50@xmath279 & 34.6@xmath280 & ... & 86 ( 71 ) + arp 220 & @xmath460.04 & 0.73@xmath281 & ... & 2.10@xmath282 & ... & ... & 69.3 ( 57 ) + ic 5063 & 0.06 & 0.34@xmath283 & ... & 1.51@xmath284 & 28.5@xmath285 & ... & 71.5 ( 57 ) + ngc 7130 & 0.05@xmath286 & 0.47@xmath287 & ... & 2.24@xmath288 & ... & ... & 58.2 ( 50 ) + ngc 7582 & 0.03@xmath15 & 0.73@xmath55 & ... & 2.37@xmath289 & 41.8@xmath290 & ... & 142 ( 128 ) + mrk 0609 & 41.38@xmath291 & 42.28@xmath52 + ic 0486 & 40.39@xmath292 & 41.83@xmath293 + 2masx j08035923 + 2345201 & 40.45@xmath294 & 39.64@xmath295 + 2masx j08244333 + 2959238 & 40.81@xmath296 & 40.18@xmath297 + 2masx j10181928 + 3722419 & 40.99@xmath298 & 40.54@xmath293 + sbs 1133 + 572 & 40.76@xmath299 & 41.17@xmath300 + mrk 1457 & 41.74@xmath301 & 40.15@xmath302 + 2masx j11570483 + 5249036 & 40.67@xmath303 & 40.81@xmath64 + 2masx j12183945 + 4706275 & 41.68@xmath304 & 41.20@xmath295 + 2masx j12384342 + 0927362 & 41.69@xmath292 & 41.90@xmath305 + ngc 5695 & 40.78@xmath306 & 39.79@xmath307 + ngc 0424 & 40.99@xmath308 & 40.72@xmath309 + ngc 1144 & 40.78@xmath310 & 40.87@xmath311 + ngc 1320 & 40.33@xmath312 & 40.42@xmath311 + ngc 1386 & 39.23@xmath110 & 39.52@xmath313 + ngc 1667 & 40.30@xmath314 & 40.06@xmath315 + f05189 - 2524 & 41.20@xmath316 & 41.56@xmath305 + ngc 3982 & 39.42@xmath317 & 38.77@xmath55 + ngc 4388 & 40.52@xmath318 & 39.87@xmath319 + ngc 4501 & 39.89@xmath49 & 39.40@xmath274 + tololo 1238 - 364 & 40.63@xmath320 & 40.31@xmath321 + ngc 4968 & 39.67@xmath322 & 39.88@xmath64 + m-3 - 34 - 64 & 41.47@xmath323 & 40.84@xmath324 + ngc 5135 & 40.86@xmath325 & 40.90@xmath15 + ngc 5194 & 39.69@xmath326 & 39.08@xmath98 + ngc 5347 & 39.34@xmath327 & 39.47@xmath305 + mrk 463 & 41.55@xmath328 & 41.69@xmath329 + ngc 5506 & 39.88@xmath330 & 41.38@xmath331 & obsids 0013140101 , 0201830201 , 0201830301 , 0201830401 + ngc 5506 & 0 & 41.53@xmath332 & obsids 0013140201 & 0201830501 + ngc 5506 & 40.06@xmath323 & 41.53@xmath52 & obsids 0554170201 & 0554170101 + arp 220 & 40.51@xmath333 & 39.58@xmath334 + ngc 6890 & 39.83@xmath335 & 39.94@xmath315 + ic 5063 & 40.11@xmath267 & 40.69@xmath52 + ngc 7130 & 40.69@xmath336 & 40.94@xmath45 + ngc 7172 & 39.85@xmath337 & 40.45@xmath52 & obsid 0414580101 + ngc 7172 & 39.67@xmath338 & 40.18@xmath64 & obsids 0147920601 & 0202860101 + ngc 7582 & 40.23@xmath98 & 39.72@xmath339 + ngc 7674 & 41.53@xmath340 & 40.03@xmath341 +

we present a detailed , two dimensional numerical study of the microphysical conditions and dynamical evolution of accretion disks around black holes when neutrino emission is the main source of cooling . such structures are likely to form after the gravitational collapse of massive rotating stellar cores , or the coalescence of two compact objects in a binary ( e.g. , the hulse taylor system ) . the physical composition is determined self consistently by considering two regimes : neutrino opaque and neutrino transparent , with a detailed equation of state which takes into account neutronization , nuclear statistical equilibrium of a gas of free nucleons and alpha particles , blackbody radiation and a relativistic fermi gas of arbitrary degeneracy . various neutrino emission processes are considered , with capture onto free nucleons providing the dominant contribution to the cooling rate . we find that important temporal and spatial scales , related to the optically thin / optically thick transition are present in the disk , and manifest themselves clearly in the energy output in neutrinos . this transition produces an inversion of the lepton gradient in the innermost regions of the flow which drives convective motions , and affects the density and disk scale height radial profiles . the electron fraction remains low in the region close to the black hole , and if preserved in an outflow , could give rise to heavy element nucleosynthesis . our specific initial conditions arise from the binary merger context , and so we explore the implications of our results for the production of gamma ray bursts .

astro-ph0506121

gas accretion by a concentration of mass is an efficient way to transform gravitational binding energy into radiation @xcite , and is responsible for observable phenomena at all scales in astrophysics . the more compact the object , the greater the efficiency . in many cases , systems are observed in steady or quasi steady state during this process ( e.g , cvs , lmxbs and agns ) . some external agent ( e.g. , the interstellar medium , a companion star ) provides a continuous supply of mass , energy and angular momentum over a timescale that is much longer than the accretion time . the energy dissipated by the flow before it is accreted , and more importantly , what happens to this energy , determines many of the properties of the flow itself . in most cases cooling is present through some radiative electromagnetic process , which acts as a sink , removing the dissipated energy . in the standard thin disk theory developed by @xcite , it is extremely efficient , maintaining a `` cool '' ( in the sense that @xmath1 , where @xmath2 is the mass of the central object ) and thin ( with a scale height @xmath3 ) , nearly keplerian disk . the requirements of steady state , hydrostatic equilibrium in the vertical direction , energy and angular momentum balance , and an equation of state yield a solution for all relevant variables as a function of the disk radius , @xmath4 . the key question of angular momentum transport was addressed by shakura & sunyaev with their famous @xmath5 prescription , which allows a parametrization of the viscous stresses and energy dissipation rates . magnetic fields could be at the physical origin this effect , through the magneto rotational instability ( mri , * ? ? ? * ; * ? ? ? in the case of inefficient cooling , a solution was found in the 1970s for so called slim disks @xcite , and later for a class that became known as adafs ( advection dominated accretion flows , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) . in this case , the cooling is negligible , either because the optical depth is large , or because the density is so low that the radiative efficiency is extremely small . the dissipated energy is advected with the flow and may be absorbed by the central object . such flows are geometrically thick , with scale heights @xmath6 . what little radiation does emerge from them arises , as in the thin disks , through various electromagnetic processes . in either of the above scenarios , accretion ( and the accompanying radiation ) is usually thought to be limited by the eddington rate , a self regulatory balance imposed by newtonian gravity and radiation pressure . the standard argument gives a maximum luminosity @xmath7 erg s@xmath8 . although this may not be strictly the case in reality as in the current argument concerning the nature of ulxs observationally @xcite , and also quite general theoretical considerations @xcite it does exhibit the qualitative nature of the effect of radiation pressure on accreting plasma , in the limit of large optical depth . the problem is circumvented ( or at least deferred by nearly sixteen orders of magnitude in luminosity ) if the main cooling agent is emission of neutrinos , instead of photons . this regime requires correspondingly large accretion rates , of the order of one solar mass per second , and is termed hypercritical accretion @xcite . it is important , for example , in the context of post supernova fallback accretion onto a proto neutron star . in such a situation , the densities and temperatures are so large ( @xmath9 g @xmath10 , @xmath11 k ) that photons are completely trapped , and energetic neutrinos are emitted in large amounts , cooling the gas and allowing accretion to proceed . the reader may wish to consult the very clear introduction in this context given by @xcite . when the stellar envelope experiences fallback onto the central object , the system may be considered to be nearly in a steady state , because of the timescales involved . there are other , more violent situations , relevant to the study of gamma - ray bursts ( grbs ) in which the assumption of steady state is not justified because of a time varying mass and energy supply , and which therefore require a dynamical analysis for their proper description . in general , whether the system cools via electromagnetic radiation or neutrinos , analytic steady state solutions are found by assuming that the cooling is either efficient ( thin disk ) or inefficient ( thick disk ) . over the last several years , many groups have considered these cases in the neutrino cooled regime @xcite and established the general features of the solutions . the real solution may be very different , in particular ( and in what concerns us for the following ) because of the previous history of the gas that constituted the accretion disk and how the disk formed . it turns out this may be quite important in the case of grbs . numerical analysis of the evolution of the disk with no assumptions concerning the steady state has been carried out by @xcite . they have reported a three dimensional calculation with detailed microphysics . unfortunately , computational limitations do not allow one to continue such calculations for more than approximately 50 ms , whereas the typical short grb lasts 0.2 s. the sources of grbs are now established ( at least based on observations of x ray , optical and radio counterparts ) to lie at cosmological distances , with redshifts @xmath12 ( * ? ? ? * and references therein ) . at such scales , the absolute energetics of each event is approximately @xmath13 erg , assuming isotropic emission . apparent collimation inferred from achromatic breaks in the afterglow light curve may reduce this to @xmath14 erg , depending on the particular event @xcite . although one generally considers two classes of grbs based on duration ( shorter and longer than about 2 s , * ? ? ? * ) , all afterglows to date ( and the corresponding inferences ) come from long bursts ( although see * ? ? ? strong evidence in favor of a sn / grb association @xcite now comes from several events ( albeit all relatively nearby ) , hinting at an underlying physical connection between the two phenomena . this is the central basis of the collapsar model @xcite , in which rotation of the pre supernova star allows for the creation of a massive accretion disk , fed by the infalling stellar envelope . the grb is then powered for a fallback time , which can be long enough to account for the observed durations . in the case of short bursts there is less evidence to go on , but one possibility is that they arise from compact binary mergers , with various combinations of black holes and neutron stars in the binary @xcite such as in the first binary pulsar to be discovered @xcite . the tidal disruption of one star by the other gives rise to an accretion disk which powers the grb . a black hole is either present from the start ( as a variation of the hulse taylor system ) or is produced by the collapse of a supramassive neutron star shortly after the merger itself . in this case there is no external agent feeding the accretion disk , and thus the event is over roughly on an accretion timescale ( which would be on the order of one second ) . clearly , investigating either of these scenarios requires time dependent , multidimensional calculations with detailed microphysics @xcite . the energy released by accretion is then transformed into a relativistic outflow which produces the gamma ray burst ( see * ? ? ? * ; * ? ? ? * ; * ? ? ? * for reviews ) . we note that either scenario would produce a distinct gravitational wave signal , which could in principle be detected by interferometric systems such as ligo in the future , thus establishing the nature of the progenitor system without a doubt . the regime in which the gas lies is unlike any other commonly encountered in astrophysics . the high densities and temperatures lead to photodisintegration of the nuclei and the establishment of nuclear statistical equilibrium ( nse ) . furthermore , neutronization becomes important in the innermost regions of the flow and weak interactions determine the composition , with the electron fraction falling substantially below 1/2 , and the gas correspondingly becoming neutron rich . if this composition is somehow frozen and transported out of the gas and into an outflow , interesting nucleosynthesis of heavy elements could occur @xcite . realistic physics input of this kind allow us to obtain more reliable estimates of the actual energy released from the disk , and potentially available to power a grb . an added complication , which affects the composition @xcite is that even if the main cooling mechanism is neutrino emission , these are not entirely free to leave the system , as scattering ( mainly off free nucleons ) is important enough to suppress the emission in the dense inner disk . we find that in fact the opaqueness of the material may lead to convection through the establishment of a composition gradient that does not satisfy the classical requirements for stability . to our knowledge , this is the first time that this has been addressed in this context . in previous work we initially studied the merger process for black hole neutron star binaries in three dimensions , paying particular attention to the structure of the accretion disks that formed as a result of the tidal disruption @xcite . follow up work used the results of these simulations , mapped to two dimensions in azimuthal symmetry , as initial conditions with simple ( ideal gas ) input physics , and no realistic cooling included @xcite . a first approximation at a realistic equation of state and the effects of neutrino opacities was reported more recently @xcite . in this paper we improve upon our earlier results in three important ways , that make them more realistic , and particularly relevant to the study of grbs . first , we use a much more detailed equation of state , appropriate for the actual physical conditions found in post merger accretion disks . this includes a relativistic electron gas of arbitrary degeneracy , radiation pressure and an ideal gas of free nucleons and @xmath5 particles . the composition is determined self consistently by considering weak interactions in two different regimes : neutrino opaque and neutrino transparent . second , we include neutrino emission as the main source of cooling , by considering the relevant reaction rates ( electron and positron capture onto free nucleons , bremsstrahlung , pair annihilation and plasmon decays ) , taken from tables and fitting formulae valid over wide ranges of temperature and density whenever possible . third , we compute an approximate optical depth for the fluid to neutrinos , using scattering off free nucleons and @xmath5 particles as a source of opacity . the emission rates and pressure due to neutrinos are then suppressed and enhanced , respectively , by an appropriate factor . we begin this paper by a presentation of the physical conditions likely to occur in post merger accretion disks in [ pmdisks ] , immediately following the coalescence of the two compact objects . our results follow in [ results ] , and a discussion of these , and the implications they have for grbs is presented in [ ccl ] . the accurate study of the dynamical evolution of the accretion disks requires knowledge of the physical conditions within them , and the use of an appropriate equation of state . below we give a general overview of the physical conditions in the disk , followed by a detailed presentation of the equation of state and the effects of neutrinos . the accretion structures that are formed as a result of the merger of two neutron stars or the tidal disruption of a neutron star by the black hole become azimuthally symmetric fairly quickly , within a few tens of milliseconds @xcite . these disks typically contain a few tenths of a solar mass , and are small , with the bulk of the mass being contained within 200 km of the black hole ( which harbors about 35 solar masses ) . the disks are dense ( @xmath15}\leq 10^{12}$ ] ) , with high internal energies ( @xmath16}\leq10^{11}$ ] ) ( see e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? in fact , the temperature is high enough that nuclei become photodisintegrated and there is a mixture of @xmath5 particles , free neutrons and protons , electrons and positrons . the timescale for @xmath17equilibrium is given by @xmath18 , where @xmath19 is the cross section for @xmath0 capture onto free nucleons and @xmath20 is the number density of free nucleons ( see e.g. , * ? ? ? * ) . under these conditions , @xmath21 s , which is much shorter than the accretion timescale @xmath22 of a fluid element , so that weak interactions determine the composition . photons are trapped and are advected with the flow . for neutrinos the situation is more complicated , since the optical depth is of order unity . in the outer regions of the disk , they escape freely , whereas for densities larger than @xmath23 g @xmath10 they undergo diffusion on a timescale @xmath24 ms . for simplicity of presentation , we will first assume that all nucleons are free ( no @xmath5 particles are present ) , and include the necessary corrections subsequently . under the conditions described above , the temperature is high enough for electron positron pair creation . the number of pairs thus produced is very sensitive to the degeneracy of the electrons . in fact the number density of pairs is suppressed exponentially with increasing degeneracy , because of fermi blocking . in previous work @xcite the assumption of full degeneracy has been made for simplicity ( note also that in some cases the presence of electron positron pairs has been assumed while at the same time retaining a degeneracy pressure term , which is inconsistent ) . here we take a different approach , using an exact expression for the pressure as a function of the temperature and the chemical potential , valid for arbitrary degeneracy in the limit of relativistic electrons ( i.e. , @xmath25 g @xmath10 ) , due to @xcite , namely : @xmath26 . \label{pe}\ ] ] the number densities of electrons and positrons are related by : @xmath27 \label{ye}\ ] ] where @xmath28 is the electron fraction . the chemical potential of species @xmath29 is denoted by @xmath30 throughout . this expression reduces to the well known limits when the temperature is low ( @xmath31 , which gives @xmath32 , appropriate for a cold relativistic fermi gas ) and when it is high ( @xmath33 , which gives @xmath34 , when the pressure comes from relativistic electron positron pairs ) . the full equation of state then reads : @xmath35 where @xmath36 @xmath37 and @xmath38 is the pressure due to neutrinos ( discussed below ) . here @xmath39 is the radiation constant , @xmath40 is boltzmann s constant and @xmath41 g is the atomic mass unit . since the presence of pairs is automatically taken into account in the expression for @xmath42 , there is no alteration to the numerical factor @xmath43 in the expression for @xmath44 . for the conditions encountered in the accretion disks presented below , gas pressure dominates at the 80% level over the other terms . regarding the photons , since the temperature is @xmath45 mev , the peak in the blackbody spectrum is at @xmath46 mev . on the other hand , the plasma frequency , @xmath47 , corresponds to @xmath48 mev . thus the standard expression for radiation pressure may be considered accurate , and plasma effects negligible as far as the photons are concerned . the computation of the electron fraction and the chemical potential of electrons follows from the assumption of @xmath17equilibrium between neutrons , protons and electrons , and the condition of charge neutrality . as noted by @xcite , a distinction needs to be made to determine the equilibrium composition depending on the optical depth of the material ( the determination of the opacities is presented in [ nuopacities ] ) . if it is transparent to its own neutrino emission , we fix equilibrium by equating the capture rates of electrons and positrons onto protons and neutrons respectively . for mild degeneracy , as is the case here , this leads to the following expression for the electron fraction as a function of the temperature and the electron chemical potential @xcite : @xmath49 where @xmath50 mev . if , however , the material is opaque , and the neutrinos are allowed to diffuse out on a timescale shorter than the accretion timescale , then we may write @xmath51 for the equilibrium composition , and is strictly valid only when there are equal numbers of neutrinos and anti neutrinos . otherwise the electron pressure , bulk viscosity and joule like heating due to the induced lepton current will be affected @xcite . we neglect all these corrections in the present treatment . ] . since the nucleons are not degenerate , we may use maxwell boltzmann statistics to describe their distribution function , and obtain @xmath52 \label{pnfrac}\ ] ] for the ratio of proton to neutron number densities . further , with @xmath53 we arrive at @xmath54 . \label{yethick}\ ] ] we now have the three equations ( [ ye ] ) , ( [ eos ] ) and ( [ yethin ] ) or ( [ yethick ] ) for the three functions @xmath55 and so the system is closed . to allow for a transition from the optically thin to optically thick regime , in fact we solve these in a combined form , weighted by factors @xmath56 $ ] or @xmath57)$ ] . as a practical matter , the internal energy per unit mass , @xmath58 , is used instead of the pressure to solve equation ( [ eos ] ) , since its variation in time is what is determined in the code , using the first law of thermodynamics . we finally include the effects of incomplete photodisintegration of @xmath5 particles by using nuclear statistical equilibrium for the three species @xmath59 to fix the mass fraction of free nucleons as @xcite : @xmath60}\right)^{9/8 } \ , \left(\frac{\rho}{10^{10}\mbox{g~cm$^{-3}$}}\right)^{-3/4 } \ , \exp(-8.2 [ 10^{10}\mbox{k}]/t).\label{nuc}\ ] ] whenever this expression results in @xmath61 we set @xmath62 . the corresponding alterations in the previous derivation are simple , and lead to the full set of equations which we write for each gas element and at each time . note that these do not allow for an explicit solution , so an iterative scheme is used in each instance . the internal energy per unit mass is @xmath63 @xmath17equilibrium gives @xmath64 f(\tau_{\nu})+[1+\exp(\{\eta_{e}-q\})/kt)]g(\tau_{\nu } ) \right ) , \label{betaeq}\ ] ] and charge neutrality implies @xmath65 , \label{neutrality}\ ] ] as given by equation ( [ ye ] ) . the modifications to the equation for @xmath17 equilibrium simply reflect the fact that if the fluid is composed primarily of @xmath5 particles , there will be one neutron per proton , and hence @xmath66 . several processes contribute to the emission of neutrinos . first , we take into account electron and positron capture onto free nucleons using the tables of @xcite . this is an improvement over calculations done previously @xcite , where assumptions concerning the degeneracy were made . the rates are thus accurate over the entire disk , whether the degeneracy is significant or not . the tables cover the following ranges : @xmath67 } ) \leq 11 $ ] ; @xmath68}\leq 11 $ ] . a bilinear interpolation in the @xmath69 plane is performed using the table to obtain the cooling rate for a given mass element . the result is then multiplied by the mass fraction of free nucleons , @xmath70 , since we are not considering capture of electrons and protons by @xmath5 particles . second , the annihilation of electron positron pairs is a source of thermal neutrinos , and the corresponding cooling rate is computed using the fitting functions of @xcite . these cover the range @xmath71 } ) \leq 12 $ ] in density and @xmath72}\leq 11 $ ] in temperature . finally , nucleon nucleon bremsstrahlung , @xmath73 , and plasmon decay , @xmath74 , are considered , with rates given by : @xmath75 @xcite and @xmath76 where @xmath77^{2})/3}$ ] @xcite . for the conditions encountered in the disk , capture by nucleons completely dominates all other processes , and is the main source of cooling . finally , the creation and disintegration of @xmath5 particles leads to a cooling term in the energy equation given by : @xmath78 the material in the disk is dense enough that photons are completely trapped , and advected with the flow . for neutrinos however , the situation is more complicated , since the outer regions are transparent to them , while the inner portions are opaque . the main source of opacity is scattering off free nucleons and @xmath5 particles , with cross sections given by @xcite @xmath79 and @xmath80^{2},\ ] ] where @xmath81 @xmath82 and @xmath83 is the weinberg angle . since capture processes dominate the neutrino luminosity , we assume that the mean energy of the neutrinos is roughly equal to the fermi energy of the partially degenerate electrons @xmath84 mev , where @xmath85 g @xmath10 . so the mean free path is @xmath86 , @xmath20 and @xmath87 being the number densities of free nucleons and @xmath5 particles respectively . to compute an optical depth , we take @xmath88 , with @xmath89 a typical scale height of the disk . inspection of the disk shape ( through the density contours in the inner regions ) and an estimation of the scale height with @xmath90 suggests that @xmath91 , with @xmath92 a constant of order unity , so that our final , simplified expression for the optical depth reads @xmath93 . with the above expressions for the cross section as a function of density and electron fraction , and defining @xmath94 we arrive at @xmath95 . \label{eq : nudepth}\ ] ] the term which depends on @xmath70 reflects the two species composition we have assumed . its influence is limited , since when @xmath70 varies from 0 to 1 , the optical depth is altered by a factor 1.2 , all other values being equal . in practice , and for the compositions found in the disks , this expression corresponds to having an optical depth of unity at approximately @xmath96 g @xmath10 , which can be thought of as a neutrino surface , where the last scattering occurs and neutrinos leave the system . accordingly , we modify the expressions for the neutrino emission rates , suppressing it in the opaque regions , and enhancing the pressure through @xmath97 and @xmath98,\ ] ] where @xmath99 is the unmodified energy loss rate ( in erg g@xmath8 s@xmath8 ) calculated from the rates given in [ nuemission ] . the total neutrino luminosity ( in erg s@xmath8 ) is then computed according to @xmath100 \ , \exp(-\tau_{\nu } ) \ , dm . \label{eq : nulum}\ ] ] for the actual hydrodynamical evolution calculations , we use the same code as in previous work @xcite , and refer the reader to that paper for the details . this is a two dimensional ( cylindrical coordinates @xmath101 in azimuthal symmetry ) smooth particle hydrodynamics code @xcite , modified from our own 3d version used for compact binary mergers @xcite . the accretion disk sits in the potential well of a black hole of mass @xmath102 , and which produces a newtonian potential @xmath103 . accretion is modeled by an absorbing boundary at the schwarzschild radius @xmath104 , and the mass of the hole is updated continuously . the transport of angular momentum is modeled with an @xmath5 viscosity prescription , including _ all _ components of the viscous stress tensor ( not only @xmath105 ) . gravity of the disk is neglected . the main modifications from the previous version are in the implementation of a new equation of state [ eoscomp ] , neutrino emission , [ nuemission ] and the treatment of neutrino optical depths , [ nuopacities ] . as done previously @xcite , the initial conditions are taken from the final configuration of 3d calculations of black hole neutron star mergers , after the accretion disk that forms through tidal disruption of the neutron star has become fairly symmetric with respect to the azimuthal coordinate . we show in table [ table : ics ] the parameters used in each of the dynamical runs included in this paper . as the dynamical evolution calculations begins , the disks experience an initial transient , which is essentially numerical in origin , and is due to the fact that the configuration is not in strict hydrostatic equilibrium ( recall that it is obtained from azimuthally averaging the results of 3d calculations ) . this transient , and its effects , are negligible , and it is essentially over in a hydrodynamical timescale ( @xmath106 ms ) . after that , the disk proceeds to evolve on a much longer timescale , determined by accretion onto the central black hole . we first present the details of the spatial structure of the disk due to fundamental physical effects , and then proceed to show the temporal evolution and associated transitions . the fundamental variable affecting the instantaneous spatial structure of the disk is the optical depth to neutrinos , @xmath107 , since it determines whether the fluid cools efficiently or not . all quantities show a radical change in behavior as the threshold @xmath108 is passed . figure [ compa0.01t100 ] shows color coded contours in a meridional slice of the thermodynamical variables , while figure [ compa0.01t100r ] displays the run of density and entropy per baryon along the equator , @xmath109 , where the density and temperature are highest ( we will refer here mainly to run a2 m , unless noted otherwise ) . the density increases as one approaches the black hole , varying as @xmath110 in the outer regions , where @xmath111 . the critical density for opaqueness is reached at @xmath112 cm , and for @xmath113 , @xmath114 . since the energy that would otherwise be lost via neutrino emission remains in the disk if the cooling is suppressed by scattering , the density does not rise as fast . in fact , one can note this change also by inspecting the scale height @xmath115 , which scales as @xmath116 for @xmath117 and @xmath118 const for @xmath113 ( and thus in terms of surface density the change is from @xmath119 to @xmath120 const at @xmath121 ) . the opaque region of the disk remains inflated to a certain extent , trapping the internal energy it holds and releasing it only on a diffusion timescale . at densities @xmath122 g @xmath10 the optical depth can reach @xmath123 , and thus the suppression of cooling is dramatic , as can be seen in figure [ compa0.01t100 ] , where the contours of @xmath107 and @xmath124 are shown . the flattening of the entropy profile is related to the change in composition in the optically thick region and the occurrence of convection ( see [ nuconvection ] below ) . as the density rises in the inner regions of the disk , the electron fraction @xmath28 initially drops as neutronization becomes more important , with the equilibrium composition being determined by the equality of electron and positron captures onto free neutrons and protons ( see equation ( [ yethin ] ) and the discussion preceding it ) . the lowest value is reached at @xmath125 , where @xmath126 . thereafter it rises again , reaching @xmath127 close to the horizon . thus flows that are optically thin everywhere will reach a higher degree of neutronization close to the black hole than those which experience a transition to the opaque regime . the numerical values for the electron fraction at the transition radius and at the inner boundary are largely insensitive to @xmath5 , as long as the transition does occur . the baryons in the disk are essentially in the form of free neutrons and protons , except at very large radii and low densities ( @xmath128 cm and @xmath129 g @xmath10 ) where @xmath5 particles form . figure [ rtye ] shows the region in the density temperature plane where the fluid lies , for runs a2 m and a1 m at two different times ( and color coded according to the electron fraction ) . most of the gas lies close to the line determining the formation of helium nuclei ( see equation [ nuc ] ) , but does not cross over to lower temperatures . the reason for this is that the energy that would be released by the creation of one helium nucleus ( 28.3 mev ) would not leave the disk ( recall that we consider neutrino emission as the only source of cooling ) and thus immediately lead to the photodisintegration of another @xmath5 particle . an equilibrium is thus maintained in which the gas is close to helium synthesis , but this does not occur . in the opaque regime , the gas moves substantially farther from the transition line to helium , since it can not cool efficiently and , as mentioned before , the density does not rise as quickly . figure [ rtye ] also shows clearly why one must make use of an equation of state which takes into account properly the effects of arbitrary degeneracy of the electrons and positrons . the solid straight line marks the degeneracy temperature as a function of density , given by @xmath130 mev . the disk straddles this line , with a degeneracy parameter @xmath131 in the inner regions , and @xmath132 at lower densities . thus making the approximation that the electrons in the flow are fully degenerate is not accurate . modeling the accretion flow in two dimensions @xmath101 , without the assumption of equatorial symmetry allows one to solve clearly for the vertical motions in the disk , something which is not possible when considering vertically integrated flows . it was pointed out by @xcite that a standard @xmath5 viscosity could lead to meridional flows in which @xmath133 changed sign as a function of height above the mid plane , leading to inflows as well as outflows . @xcite , @xcite and @xcite later considered a similar situation , and found that for a range of values in @xmath5 , the gas flowed inward along the surface of the disk , and outward in the equatorial regions . in previous work @xcite , we found this solution in disk flow , with large scale circulations exhibiting inflows and outflows for high viscosities ( @xmath134 ) , and small scale eddies at lower values ( @xmath135 ) . how vertical motions affect the stability of accretion disks and may lead to the transport of angular momentum is a question that has become of relevance in this context @xcite . here we show in figure [ velcont ] the velocity field and magnitude of meridional velocity for run a2 m . the small scale eddies are clearly visible , with their strength usually diminishing as the disk is drained of matter and the density drops . the effect of a different value of @xmath5 can be seen in the top row of figure [ compa0.01a00.1t100t200 ] , where the comparison between @xmath136 and @xmath137 is made . the large , coherent lines of flow aimed directly at the origin for @xmath136 correspond to inflow , while the lighter shades along the equator at larger radii show outflowing gas . the instantaneous structure of the disk concerning the density , temperature and composition profiles is largely independent of the viscosity , as long as there is enough mass to produce the optically thin / optically thick transition . even in the optically thin regime , the structure can not be determined analytically following the standard arguments applied to thin , cool disks of the shakura sunyaev type . the reason for this is the following : a central assumption in the standard solution is that the disk is cool , i.e. , @xmath138 . this comes from the requirement that all the energy dissipated by viscosity , @xmath139 be radiated away efficiently , and produce the observed flux , @xmath140 . in our case , the disk is most certainly not cool , as it originates from the tidal disruption of a neutron star by a black hole ( or possibly the coalescence of two neutron stars , and subsequent collapse of the central mass to a black hole ) . the gas that constitutes the disk is dynamically hot because it was in hydrostatic equilibrium in a self gravitating configuration , where @xmath141 and has not been able to release this internal energy ( the merger process itself may lead to additional heating ) . a further deviation from the standard solution is that the pressure support in the disk leads to a rotation curve that is slightly sub keplerian . the dissipation by viscosity is in fact smaller than the cooling rate over much of the disk , and the released luminosity comes from a combination of viscous dissipation and the store of internal energy given to the disk at its conception . the classical requirement for convective instability in the presence of entropy and composition gradients , as well as rotation , is the solberg hoiland criterion @xcite : @xmath142 where @xmath143 is the radial epicyclic frequency ( @xmath144 being the angular velocity ) , and @xmath145 \label{bv}\ ] ] is the brunt visla frequency ( see * ? ? ? * ; * ? ? ? the adiabatic index @xmath146 in our case , since @xmath147 is the most important contribution to the total pressure . thus a region may be convectively unstable because of a composition gradient , or an entropy gradient , or both . strictly speaking , here one should consider the total lepton fraction @xmath148 . we do not consider the transport of neutrinos in detail , as already mentioned ( see [ nuopacities ] ) , and do not calculate explicitly the contribution of neutrinos to this @xmath148 . for what follows , we will thus simply take @xmath28 to represent the behavior of the full @xmath148 . the origin of convection in the lepton inversion zone can be understood as follows @xcite . consider a fluid element in the lepton inversion zone that is displaced in the outward direction and then comes to pressure equilibrium with its surroundings . the displaced element , which is lepton rich relative to its new surroundings , attains the ambient pressure at a lower density than the surrounding fluid , because the pressure depends directly on the lepton number , and thus tends to drift outwards . by the same token , an inwardly displaced fluid element in the lepton inversion zone is depleted in leptons relative to its new surroundings and thus tends to sink . when temperature gradients are allowed for , the displaced fluids , in general , have temperatures which differ from those of their surroundings . these variations tend to promote stability or instability depending on whether the existing temperature gradient is less than or greater than the adiabatic gradient , respectively . in the present scenario then , the transition to the optically thick regime leads to instability , because initially @xmath149 , and also @xmath150 ( see also figure [ compa0.01t100r ] ) . the entropy profile is then flattened by efficient convective mixing , and the lepton gradient inversion is due to the different way in which the composition is determined through weak interactions once the neutrinos become trapped . in a dynamical situation such as the one treated here , convection tends to erase the gradients which give rise to it . accordingly , the sum of the corresponding terms in equation ( [ sh ] ) tends to zero once the simulation has progressed and convection has become established in the inner disk ( note that in our case the rotation curve there is sub keplerian , with @xmath151 , so that @xmath144 and @xmath152 are not equal ) . this is analogous to what occurs in proto neutron stars following core collapse @xcite and has actually been confirmed in numerical simulations of such systems @xcite , where the convection leads to strong mixing . for a neutrino driven convective luminosity @xmath153 at radius @xmath154 and density @xmath155 , the convective velocity may be estimated by standard mixing length theory as ( see e.g. , * ? ? ? * ) @xmath156 where we have used the fact that gas pressure dominates in the fluid . the assumption of spherical symmetry implicit in this expression is not strictly met in our case , but it may provide us nevertheless with a useful guide . the overturn time of a convective cell is then given by @xmath157 ms , where @xmath158 km is the mixing length , usually set to a pressure scale height . in our calculations , we see that the magnitude of the meridional velocity @xmath159 decreases as @xmath4 decreases , then rises again as the opaque region is reached , reaching @xmath160 cm s@xmath8 , at @xmath161 km . the associated turnover times are thus @xmath162 ms , in good agreement with the estimate made above . to isolate this effect from that due to viscosity ( which generates the meridional circulations mentioned previously ) we have performed a simulation in which @xmath163 ( run aim in table [ table : ics ] ) . this calculation shows essentially no accretion onto the black hole except for a small amount of gas transferred at early times ( because the initial condition is not in strict equilibrium ) . the result concerning the profile of @xmath28 as a function of radius is as we have described above , i.e. , increasing neutronization as @xmath4 decreases , until the opaque region is reached , followed by an increase in @xmath28 as the black hole boundary is approached . the revealing difference lies in the time evolution of this profile . as mentioned above , convection generates motions which tend to eliminate the composition gradient which drives it . in the presence of viscosity , matter is continuously transported radially , and the gradient is not erased entirely . height integrated profiles of @xmath164 at 50 ms and 200 ms show a similar behavior at small radii . when viscosity is removed , the initial composition gradient is gradually softened until it disappears in the innermost regions ( see figure [ convection ] ) . the disk then has a nearly constant electron fraction along the equator for @xmath113 , with a sharp transition region leading to an increase with radius in the transparent regime . this may resemble the convection dominated accretion solution found analytically by @xcite and numerically by several groups @xcite , where convection transports angular momentum inward , energy outward , and gives a radial profile in density @xmath165 . we find a different power law , with @xmath114 . there are several factors which may account for this difference : the disk vs. spherical geometry ; the initial condition with a large amount of internal energy ; and the limited , but present cooling rate on the boundaries of the flow . since the neutrinos are diffusing out of the fluid ( the mean free path is small compared to the size of the system in the optically thick regime ) , one would expect a corresponding viscosity through @xmath166 ( see , e.g. * ? ? ? this needs to be compared to the corresponding viscosity generated by the assumed @xmath5 prescription for consistency . we may assume @xmath167 , where @xmath168 is the energy density associated with neutrinos , and @xmath169 is their mean free path ( see [ nuopacities ] ) . thus @xmath170 for conditions near the equatorial plane , @xmath109 . this value will increase in the outer regions , since the mean free path becomes larger as the density drops . at the neutrino surface , where @xmath171 , the mean free path is @xmath172 cm , and so an averaged value of the viscosity over the entire neutrino opaque region will result in an increase of about one order of magnitude over the estimate given in equation ( [ nuvisc ] ) ( note however , that strictly speaking the diffusion approximation is no longer valid in the outer regions , so this must be interpreted with care ) . for the @xmath5 prescription , @xmath173 . the rotation curve is not too far from keplerian , and scaling this expression to typical values we find @xmath174 the effects on angular momentum transport are thus a full two orders of magnitude below those arising from our viscosity prescription ( for @xmath175 ) . to put it another way , a lower limit for the viscosity under these conditions ( and in the optically thick portion of the disk ) would be @xmath176 . an alternative analysis would be to consider the corresponding timescales induced by this viscosity . since @xmath177 , smaller viscosities imply longer timescales , as expected . aside from convection , several general criteria for stability can be analyzed in our case . evidently , since we are performing dynamical calculations , any instability that arises will quickly lead to a change in structure . it is nevertheless instructive to consider the corresponding conditions within the disk . we first consider the toomre criterion , comparing gravitational and internal energies , with @xmath178 where @xmath152 is the local epicyclic frequency ( essentially equal in this case to the angular frequency @xmath144 ) , @xmath179 is the local sound speed and @xmath180 is the surface density . we find @xmath181 in all cases throughout the calculations , and thus that the disks are stable in this respect ( figure [ qthermalstability]a shows a typical profile of @xmath182 $ ] ) . previous studies @xcite have shown that neutrino cooled disks are thermally unstable if radiation pressure dominates in the flow , and stable otherwise , although the effects of neutrino opacities were not considered . the criterion for stability can be written in this case as @xmath183 where @xmath184 and @xmath185 are the volume heating and cooling rates . essentially , in order to be stable the disk must be able to get rid of any excess internal energy generated by an increase in the heating rate . we show in figure [ qthermalstability]b typical values for @xmath184 and @xmath185 as functions of the central ( equatorial ) temperature , for run a2 m at @xmath186 ms . in the optically thin region the disks are thermally stable , because the pressure is dominated by the contribution from free nucleons and cooling is efficient . in the optically thick region the cooling is greatly suppressed and the criterion would indicate that the disk becomes thermally unstable . this is reasonable , since with optical depths @xmath187 , essentially no direct cooling takes place , and dissipation is not suppressed . this is why the disk is geometrically thick in the inner regions , as already described above ( [ structure ] ) . the balance that gives a quasi steady state structure is achieved mainly through the diffusion of neutrinos , since the diffusion timescale is shorter than the accretion timescale . the evolution of the disk on long timescales is determined by the balance between two competing effects : on one hand , viscosity transports angular momentum outwards , matter accretes and the disk drains into the black hole on an accretion timescale @xmath188 . the trend in this respect is towards lower densities and temperatures . on the other hand , cooling reduces pressure support and leads to vertical compression , increasing the density . the internal energy of the fluid is released on a cooling timescale , @xmath189 . the presence of an optically thick region in the center of the disk limits the luminosity ( dominated by electron and positron capture onto free nucleons ) to @xmath190erg s@xmath8 , and the initial internal energy is @xmath191 erg , so @xmath192 s. if @xmath193 , the disk will cool before its mass or internal energy reservoir is significantly affected by accretion onto the black hole . the maximum density ( shown in figure [ rhomax ] for runs a1 m , a2 m and a3 m ) actually increases slightly due to vertical compression , and subsequently drops once mass loss through accretion dominates ( the @xmath194 ms delay at the start , during which it is approximately constant , is simply the sound crossing time across the optically thick region of the disk ) . the accretion rate onto the black hole , @xmath195 , the accretion timescale @xmath196 and the total neutrino luminosity @xmath197 are shown in figures [ mdot ] , [ tacc ] and [ lum ] . they all show the same qualitative behavior , remaining fairly constant ( or changing slowly ) for an accretion timescale , and abruptly switching thereafter . the accretion timescales are approximately 0.5 s and 5 s for @xmath198 respectively . for high viscosity , @xmath136 , the transport of angular momentum is vigorous , and the black hole quickly accretes a substantial amount of mass ( @xmath199 within the first 100 ms ) . the accretion timescale is @xmath200 ms . the circulation pattern consists of large scale eddies , with @xmath6 . in fact there is essentially one large eddy on each side of the equatorial plane , with mass inflow along the surface of the disk , and an equatorial outflow . part of the outflowing gas moves away from the equator and reverses direction close to the surface of the disk , contributing to the inflow . for an intermediate viscosity , @xmath137 , the intensity of the circulations is smaller , but also , the eddies are smaller , with several of them clearly occurring in the disk at once . the transport of angular momentum being less vigorous , the accretion rate is substantially smaller than in the previous case . finally , for a yet lower viscosity , @xmath201 , the trend continues , and the eddies become smaller still . in these last two cases , angular momentum transport is so low that very high densities are maintained in the central regions of the disk for a large number of dynamical times , contrary to what is seen in the high viscosity case . a simple way to quantify how much of the mass flow is actually making it to the central black hole is to measure the fraction of mass at any given radius that is moving inwards , @xmath202 , where we have divided the mass flow rate into two height integrated parts , one with @xmath203 and another with @xmath204 . for example , for run a2 m at @xmath205 ms and @xmath206 it is approximately 1/3 . the ratio tends to unity only in the innermost regions of the disk , and shows the true black hole mass accretion rate , plotted in figure [ mdot ] for the same cases . we now turn our attention to the neutrinos , which are the main source of cooling . the results in this case are markedly different from what we initially found @xcite , simply because of the new equation of state , the more realistic cooling rates and the approximate computation of opacities . they are in general agreement with the preliminary results we presented before @xcite , which used a less detailed equation of state than the one shown here . the optically thick region is present in every disk at the start of the calculation . as already mentioned , this has two important effects : enhancing the pressure and suppressing the neutrino emission . this reflects upon the total luminosity , since the suppression occurs precisely in the hottest regions , where most of the energy would otherwise be released . the most important qualitative difference between the runs presented here is that for a high viscosity ( @xmath136 , runs a1 m and a1 m ) , the disk is drained of mass so fast that it has no chance to cool ( @xmath207 ) and release most of its internal energy . it is in fact advected into the black hole . moreover , the optically thick region disappears entirely from the disk ( see figure [ compa0.01a00.1t100t200 ] ) by @xmath208 ms . the emission is then no longer suppressed ( see the contours of @xmath124 in the same figure ) , and the disk radiates at the maximum possible cooling rate . the thermal energy content of the disk is so large , that as it thins , the luminosity actually increases briefly around @xmath209 ms before dropping again . the drop at late times follows an approximate power law , with @xmath210 . in this interval the energy release comes from a combination of residual internal energy and viscous dissipation within the disk . for lower values of the viscosity , @xmath211 , the central regions of the disk remain optically thick throughout the calculations . note that reducing the disk mass by a factor of five ( as was done for runs a1 m , a2 m , a3 m ) does not affect these overall conclusions . the fluid is simply compressed into a smaller volume , and thus the densities and temperatures that are reached are similar than for the high mass runs . for run a3 m the neutrino luminosity is practically constant at @xmath212 erg s@xmath8 for @xmath213 ms . table [ table : evol ] summarizes the typical disk mass , energy density , accretion rate , luminosity , and duration and energetics of neutrino emission for all runs . we have performed two - dimensional hydrodynamical simulations of accretion disks in the regime of hypercritical accretion , where neutrino emission is the main cooling agent . the disks are assumed to be present around a stellar mass black hole , and are evolved for a few hundred dynamical timescales . we have paid particular attention to the relevant microphysical processes under the conditions at hand , and used a detailed equation of state which includes an ideal gas of @xmath5 particles and free baryons in nuclear statistical equilibrium and a relativistic fermi gas of arbitrary degeneracy . the composition of the fluid is determined by weak interactions . the density and temperature are such that the inner regions of the disk become opaque to neutrinos , and this is taken into account in a simple approximation . * once the fluid becomes photodisintegrated into free nucleons , neutronization becomes important and lowers the electron fraction substantially below 1/2 , with the electron fraction @xmath28 reaching @xmath214 at its minimum . this value , however , does not occur in the immediate vicinity of the black hole , but rather at the transition radius where the fluid becomes optically thick to its own neutrino emission , @xmath215 cm . at smaller radii , the electron fraction rises again , reaching @xmath216 close to the horizon . * neutrino trapping produces a change in composition and an inversion in the electron fraction . the associated negative gradient in @xmath164 induces convective motions in the optically thick region of the disk . this is analogous to what presumably occurs following core collapse , in a proto neutron star and its surrounding envelope . due to the radial flows induced by viscosity , convection is unable to suppress this composition gradient . it would appear , however , that the entropy per baryon in the optically thick region is very close to being constant , with @xmath217 . * the spatial structure of the disk is characterized by the transition radius @xmath121 where the material becomes optically thick . for @xmath117 the disk cools efficiently , whereas for @xmath113 the emission is suppressed and the fluid is unable to cool directly ( although it does so on a neutrino diffusion timescale ) . this leads to larger pressures and a more moderate rise in density . * the temporal evolution of the disk is determined by the balance between accretion and neutrino emission . for low viscosities ( @xmath218 ) , the disk is able to cool in a quasi steady state and radiate its internal energy reservoir . this lasts for approximately 0.1 - 0.4 s , with @xmath219 erg s@xmath8 . thereafter the typical luminosity and density quickly decay . for large viscosities ( @xmath220 ) the disk is drained of mass on an accretion timescale which is shorter than the cooling timescale , and the internal energy of the disk is essentially advected into the black hole . an interesting result in this case is that as the disk becomes transparent before being engulfed by the hole , it undergoes a re - brightening , as some of the stored energy escapes . * the total energy output in neutrinos is @xmath221 erg , over a timescale of @xmath222 ms . the typical accretion rates are @xmath223 s@xmath8 , and neutrino energies are @xmath224 mev at @xmath121 . energy densities in the inner regions of the disk are @xmath225erg @xmath10 . there are two main ingredients in the results presented here that contrast with those available previously in the literature concerning the steady state structure of neutrino cooled accretion disks . the first is the ability to dynamically model the evolution of the system for hundreds of dynamical timescales , taking the previous history of the fluid into consideration through the choice of initial conditions . this allows us to consider the energetics on more relevant timescales ( cooling , viscous ) than the dynamical one accessible in three dimensional studies . the second is the realization that the structure of the disks is affected qualitatively by the presence of an optically thick region at high densities . in this sense the situation is similar to that encountered following massive core collapse , where convection occurs ( recent work assessing the relative importance of the mri , neutrino and convection driven viscosity in rotating collapsing cores has been reported by * ? ? ? * ; * ? ? ? * we comment further on this issue in [ magnetic ] ) . clearly there is room for improvement in the results presented here . to begin with , our expressions for the effect of a finite optical depth on cooling and pressure are too simple , and attempt only to capture the essential physical behavior of the system . there is an important region of the disk where the optical depth is not large enough to consider the diffusion approximation , and where more detailed transport effects ought to be considered . we have not separated the neutrino variables into three species , which would be more rigorous ( e.g. , * ? ? ? * have noted that in such flows , the electron neutrinos @xmath226 might be preferentially absorbed with respect to electron anti neutrinos @xmath227 , thus affecting the neutrino annihilation luminosity above the surface of the disk in an important way ) . we have only considered the effects of coherent scattering off nucleons and @xmath5 particles for opacity purposes . this specifically ignores absorptive scattering , which would : ( i ) lead to a modification of the neutrino emergent spectrum ; ( ii ) produce heating of the fluid . the latter may be particularly important concerning the driving of powerful winds off the surface of the disk , and needs to be addressed more carefully in the context of grbs . as in previous work , we have chosen to maintain the newtonian expression for the potential well of the central black hole for ease of comparison to previous results ( this will be addressed in future work ) . the sudden release of gravitational binding energy of a neutron star is easily sufficient to power a grb . the minimum energy requirement is @xmath228 erg if the burst is beamed into a solid angle @xmath229 . as discussed in [ intro ] , a familiar possibility , especially for bursts belonging to the short duration category , is the merger of a neutron star - black hole , or double neutron star binary by emission of gravity waves , which , as illustrated here , is likely to generate a black hole surrounded by a lower mass accretion disk . how is the available rotational and gravitational energy converted into an outflowing relativistic plasma ? a straightforward way is that some of the energy released as thermal neutrinos is reconverted , via collisions outside the disk , into electron - positron pairs or photons . the neutrino luminosity emitted when disk material accretes via viscous ( or magnetic ) torques on a timescale @xmath230 s is roughly @xmath231 for a canonical radiation efficiency of 0.1 . during this time , the rate of mass supply to the central black hole is of course much greater than the eddington rate . although the gas photon opacities are large , the disk becomes sufficiently dense and hot to cool via neutrino emission . there is in principle no difficulty in dissipating the disk internal energy , but the problem is in allowing these neutrinos to escape from the inflowing gas . at sufficiently low accretion rates , @xmath232 , we find that the energy released by viscous dissipation is almost completely radiated away on a timescale given by @xmath233 s. in contrast , for a higher mass supply , @xmath234 , energy advection remains important until the entire disk becomes optically thin . the restriction on the cooling rate imposed by high optical depths is key because it allows the energy loss to be spread over an extended period of time during which the neutrino luminosity stays roughly constant . this gives a characteristic timescale for energy extraction and may be essential for determining the duration of neutrino - driven short grbs @xcite . neutrinos could give rise to a relativistic pair - dominated wind if they converted into pairs in a region of low baryon density ( e.g. along the rotation axis , away from the equatorial plane of the disk ) . the @xmath235 process can tap the thermal energy of the torus produced by viscous dissipation . for this mechanism to be efficient , the neutrinos must escape before being advected into the hole ; on the other hand , the efficiency of conversion into pairs ( which scales with the square of the neutrino density ) is low if the neutrino production is too gradual . typical estimates suggest a lower bound of @xmath236 ( e.g * ? ? ? if the pair - dominated plasma were collimated into a solid angle @xmath229 then of course the apparent isotropized energy would be larger by a factor @xmath237 , but unless @xmath229 is @xmath238 this may fail to satisfy the apparent isotropized energy of @xmath13 ergs implied by a redshift @xmath239 for short grbs . one attractive mechanism for extracting energy that could circumvent the above efficiency problem is a relativistic magneto hydrodynamic ( mhd ) wind @xcite . such a wind carries both bulk kinetic energy and ordered poynting flux , and it possible that gamma - ray production occurs mainly at large distances from the source @xcite . a rapidly rotating neutron star ( or disk ) releases energy via magnetic torques at the rate @xmath240 , where @xmath241 s is the spin period , and @xmath242 g is the strength of the poloidal field at a radius @xmath243 cm . the last stable orbit for a schwarzschild hole lies at a coordinate distance @xmath244 km , to be compared with @xmath245 km for an extremal kerr hole . thus the massive neutron disk surrounding a schwarzschild black hole of approximately @xmath246 should emit a spin - down luminosity comparable to that emitted by a millisecond neutron star . a similar mhd outflow would result if angular momentum were extracted from a central kerr hole via electromagnetic torques @xcite . the field required to produce @xmath247 is colossal , and may be provided by a helical dynamo operating in hot , convective nuclear matter with a millisecond period @xcite . a dipole field of the order of @xmath248 g appears weak compared to the strongest field that can in principle be generated by differential rotation ( @xmath249^{-1}\;{\rm g}$ ] ) , or by convection ( @xmath250 ) , although how this may come about in detail is not resolved . we examine in more detail the possible generation of strong magnetic fields below in [ magnetic ] . computer simulations of compact object mergers and black hole formation can address the fate of the bulk of the matter , but there are some key questions that they can not yet tackle . in particular , high resolution of the outer layers is required because even a tiny mass fraction of baryons loading down the outflow severely limits the attainable lorentz factor - for instance a poynting flux of @xmath251 erg could not accelerate an outflow to @xmath252 if it had to drag more than @xmath253 solar masses of baryons with it . one further effect renders the computational task of simulating jet formation even more challenging . this stems from the likelihood that the high neutrino fluxes ablate baryonic material from the surface of the disk at a rate @xcite @xmath254 a rest mass flux @xmath255 limits the bulk lorentz factor of the wind to @xmath256 if one assumes that the external poloidal field strength is limited by the vigor of the convective motions , then the spin - down luminosity scales with neutrino flux as @xmath257 , where @xmath258 is the convective velocity . the ablation rate given in equation ( [ ablation ] ) then indicates that the limiting bulk lorentz factor @xmath259 of the wind decreases as @xmath260 . thus the burst luminosity emitted by a magnetized neutrino cooled disk may be self - limiting . the mass loss would , however , be suppressed if the relativistic wind were collimated into a jet . this suggests that centrifugally driven mass loss will be heaviest in the outer parts of the disk , and that a detectable burst may be emitted only within a certain solid angle centered on the rotation axis ( see e.g. , * ? ? ? there is also the question of magnetic fields , which we have not included , but should obviously be considered . the field in a standard disk is probably responsible for viscous stresses and dissipation , through the mri . in this respect the current scenario should exhibit these characteristics . the mri operates on an orbital timescale , and so the field would grow in a few tens of milliseconds . it may also be amplified by the convective motions described above , [ nuconvection ] . the saturation value for the field can be naively estimated as that at which its energy density is in equipartition with the gas , @xmath261 , or when the alfven speed , @xmath262 is comparable with the azimuthal velocity , @xmath263 . this gives @xmath264 g. it is not clear at all , however , that the field amplitude will reach such high levels , because the magnetic reynolds number is far beyond its critical value ( where diffusion balances dynamo driven growth ) , and amplification can lead to field expulsion from the convective region , thereby destroying the dynamo @xcite . precious little is known about the growth of magnetic fields at such overcritical levels @xcite , and a definitive answer will require the self consistent inclusion of full mhd into the problem at hand ( but with a level of resolution which may be well above present computational capabilities ) . it is not clear either that the magnetic shearing instability can generate a mean poloidal field as strong as @xmath265 g , since to first order it does not amplify the total magnetic flux . the non - linear evolution of the instability depends sensitively on details of magnetic reconnection , and it has indeed been suggested that this can smooth reversals in the field on very small scales , pushing the dominant growing mode to much larger scales @xcite . it is certainly possible , as shown here , that compact binary mergers do form a neutron disk that is hot enough to be optically thick to neutrinos , and convective instability is a direct consequence of the hot nuclear equation of state . a neutron disk is likely to be convective if the accretion luminosity exceeds @xmath266 . note that even if the accretion luminosity is lower , a hot , massive disk ( such as those forming in collapsars , * ? ? ? * ) would undergo a brief period of convection as a result of secular cooling ( notice that convection is driven by secular neutrino cooling , whereas the mri is powered by a release of shear kinetic energy ) . if the dense matter rotates roughly at the local keplerian angular velocity , @xmath267 , then @xmath268 is approximately independent of radius , and the required poloidal field for a given luminosity is @xmath269 . if a period of convection is a necessary step in the formation of a strong , large - scale poloidal field , an acceptable model thus requires that the surrounding torus should not completely drain into the hole on too short of a timescale . whether a torus of given mass survives clearly depends on its thickness and stratification , which in turn depends on internal viscous dissipation and neutrino cooling . a large amount of differential rotation ( as may occur in newborn neutron stars or those in x ray binaries , and is definitely the case in toroidal structures supported mainly by centrifugal forces ) , combined with short periods , may produce substantial magnetic field amplification @xcite . the energy transferred to the magnetic field is released in episodic outbursts when the buoyancy force allows the field to rise to the surface of the star or disk . the amplification of a magnetic field to such strong values would clearly have important consequences on the evolution and time variability of the disk and its energy output . it would probably lead to strong flaring and reconnection events accompanied by the release of large amounts of energy , if the growth time for field amplification , @xmath270 , is shorter than the accretion timescale , @xmath188 ( otherwise the disk would drain of matter before the field had the chance to reach large values ; in this case , the survival of a massive , rapidly rotating neutron star as the end - point of binary ns merger might be preferred over the prompt formation of a bh ) . an effective helical dynamo of the @xmath271 type should be favored by a low effective viscosity , @xmath5 , because , as stated in [ nuconvection ] , the overturn time ms . ] is @xmath272ms ( this applies only if the disk is not fed with matter externally for a time longer than @xmath188 , otherwise convection would also be able to amplify the magnetic field ) . core collapse sne and compact object mergers are natural astrophysical sites for the production of heavy elements @xcite . in particular , nucleosynthesis in neutrino driven winds is an issue that may be relevant for iron group elements , as well as for heavier nuclei through the r process @xcite . initial investigations into this matter @xcite determined that the entropy in the outflow arising from a newborn neutron star was probably too low to give rise to the r process efficiently . more recently , this problem has been addressed again in the specific case of collapsar or post merger accretion disks , based on the results of analytical calculations of neutrino cooled disks in one dimension @xcite . if the wind consists of a uniform outflow driven from the surface of the disk , the entropy is too low , and essentially only iron group elements are synthesized , in agreement with the earlier results . however , their results also indicate that in a bubble - type outflow ( against a background steady wind ) , where episodic expulsion of material from the inner regions takes place , material may be ejected from the disk and preserve its low electron fraction , thus allowing the r - process to occur . the convective motions reported here , occurring in the optically thick portion of the disk , would represent one way such cells could be transported to the disk surface if they can move fast enough to preserve the neutron excess . since the existence of a convection region is dependent upon the densities reached in the inner disk ( so that it becomes opaque ) , the synthesized nuclei ( iron group vs. heavier , r process elements ) could be a reflection of its absence / presence . we have benefited from many useful discussions and correspondence with u. geppert , n. itoh , k. kohri , p. kumar , a. macfadyen , p. mszros , m. prakash , m. rees , t. thompson , s. woosley and a. socrates . financial support for this work was provided in part by conacyt 36632e ( whl , dp ) and by nasa through a chandra postdoctoral fellowship award pf3 - 40028 ( err ) . part of this work was done during visits to the institute for advanced study ( whl ) and instituto de astronoma , unam ( err ) , whose hospitality is gratefully acknowledged . we thank the anonymous referee for helpful comments and suggestions on the initial manuscript . abramowicz , m.a . , in growing black holes : accretion in a cosmological context `` , in press : eds . a. merloni , s. nayakshin and r. sunyaev , ' ' eso astrophysics symposia series " , springer - verlag , berlin , 2004 ( astro - ph/0411185 ) clccc aim & 0.0 & 3.85 & 0.308 & 19,772 + a1 m & 0.1 & 3.85 & 0.308 & 19,772 + a2 m & 0.01 & 3.85 & 0.308 & 19,772 + a3 m & 0.001 & 3.85 & 0.308 & 19,772 + a1 m & 0.1 & 3.85 & 0.062 & 19,772 + a2 m & 0.01 & 3.85 & 0.062 & 19,772 + a3 m & 0.001 & 3.85 & 0.062 & 19,772 + cccccccc aim & 0 & 6@xmath273 & 2@xmath274 & 1.5@xmath274 & 8 & @xmath275140 & 0.308 + a1 m & 7 & 2@xmath276 & 2@xmath277 & 2@xmath274 & 8 & 60 & 0.098 + a2 m & 0.7 & 2@xmath276 & 6@xmath274 & 2@xmath274 & 8 & 150 & 0.2 + a3 m & 0.05 & 1.6@xmath276 & 2.@xmath274 & 1.55@xmath274 & 8 & @xmath275140 & 0.287 + a1 m & 0.7 & 4@xmath273 & 2@xmath274 & 5.4@xmath278 & 8 & 25 & 0.022 + a2 m & 0.05 & 3@xmath273 & 1@xmath274 & 5.5@xmath278 & 8 & 75 & 0.045 + a3 m & 0.005 & 3@xmath273 & 1.5@xmath274 & 7@xmath278 & 8 & @xmath275160 & 0.059 +

let g be an arbitrary finite weighted digraph with weights in the set of complex rational functions . a general procedure is proposed which allows for the reduction of g to a smaller graph with a less complicated structure having the same spectrum as of g ( up to some set known in advance ) . the proposed procedure has a lot of flexibility and could be used e.g. for design of networks with prescribed spectral and dynamical properties .

0909.0053

the structure of a given graph can range in complexity from being quite simple , having some regular features or small edge and vertex sets to being extremely complicated where basic characteristics of the graph are hard to obtain or estimate . such complicated structure is typical if for instance the graph represents some real network @xcite . an important problem therefore is whether it is possible to simplify or reduce a graph while maintaining its basic graph structure as well as some characteristic(s ) of the original graph . a related key question then is which characteristic(s ) to conserve while reducing a graph . studies of dynamical networks ( i.e. networks of interacting dynamical systems which could be cells , power stations , etc . ) reveal that an important characteristic of a network s structure is the spectrum of the network s adjacency matrix @xcite . with this in mind we present an approach which allows for the reduction of a general weighted digraph in such a way that the spectrum of the graph s ( weighted ) adjacency matrix is maintained up to some well defined and known set . we denote the class of graphs for which reductions are possible by @xmath0 which consists of the set of all finite weighted digraphs without parallel edges but possibly with loops having weights in the set @xmath1 of complex rational functions . a graph @xmath2 can therefore be written as the triple @xmath3 where @xmath4 and @xmath5 are the vertex and edge sets of @xmath6 respectively and @xmath7 . each such graph has an adjacency matrix with a well defined spectrum which we denote by @xmath8 . with this in place a graph reduction of @xmath6 can be described as follows . given a specific subset @xmath9 which we call a _ structural sets _ of @xmath6 , an _ isospectral reduction _ of @xmath6 over the vertex set @xmath10 is a weighted digraph @xmath11 ( see section 3 for the exact definitions ) . the main result of the paper is the following theorem ( see theorem [ theorem1 ] ) . + * theorem : * let @xmath2 and @xmath10 be a structural set of @xmath6 . then @xmath12 and @xmath13 differ at most by @xmath14 . + the set @xmath14 is a finite set of complex numbers which is known and is the largest set by which @xmath12 and @xmath13 can differ . as a typical graph has many different structural sets it is possible to consider different isospectral reductions of the same graph . moreover , since a reduced graph is again a weighted digraph it is possible to consider sequences of such reductions . the flexibility of this process is reflected in the fact that for a typical graph @xmath2 it is possible to reduce @xmath6 to a graph on any nonempty subset of its original vertex set . that is , we may simplify the structure of @xmath6 to whatever degree we desire . from this it follows that if @xmath15 is any nonempty subset of the vertices of @xmath6 then there are typically multiple ways to sequentially reduce @xmath6 to a graph on @xmath15 . as it turns out each such reduction results in the same graph independent of the particular sequence . this uniqueness result can be interpreted as the property that sequential reductions on @xmath0 are commutative . furthermore , the class of graphs which can be reduced via this method is very general . specifically , we may reduce those graphs in @xmath0 which consist of weighted digraphs without parallel edges . since undirected and unweighted graphs or graphs with parallel edges can be considered as weighted digraphs via some standard conventions then such graphs are automatically included as special cases of graphs that may be reduced by this procedure . because of the flexibility in reducing a graph the relation of having the same branch reduction is not an equivalence relation on the set @xmath0 as this relation is not transitive . however , it is possible to construct specific types of structural sets as well as rules for sequential reductions which do induce equivalence relations on the graphs in @xmath0 which we give examples of . also we note that generally , the tradeoff in reducing a graph is that although the graph structure becomes simpler the weights of edges become more complicated . therefore , we also consider graph reductions over fixed weight sets in which the weight set of the graph is maintained under this reduction while the vertex set is reduced . the structure of the paper is as follows . in sect . 2 we present notation and some general definitions . sect . 3 contains the description of the reduction procedure as well as some results on sequences of such reductions . some of the main results and techniques of the paper are then stated in sect . proofs of these statements are given in sect.5 . in sect . 6 we study the relations between the strongly connected components of the graph and its reductions . sect . 7 considers reductions over fixed weight sets and sect . 8 contains some concluding remarks . in what follows we formally consider the class of digraphs consisting of all finite weighted digraphs with or without loops having edge weights in the set @xmath1 of complex rational functions described below . we denote this class of graphs by @xmath0 . as previously mentioned , graphs or which are either undirected , unweighted , or have parallel edges can be considered as graphs in @xmath0 . this is done by making an undirected graph @xmath6 into a directed graph by orienting each of its edges in both directions . similarly , if @xmath6 is unweighted then it can be made weighted by giving each edge unit weight . also multiple edges between two vertices of @xmath6 may be considered as a single edge by adding the weights of the multiple edges and setting this to be the weight of this single equivalent edge . we will typically assume that the graph @xmath2 or use these conventions to make it so . by way of notation we let the digraph @xmath6 , possibly with loops , be the pair @xmath16 where @xmath4 and @xmath5 are the finite sets denoting the _ vertices _ and _ edges _ of @xmath6 respectively , the edges corresponding to ordered pairs @xmath17 for @xmath18 . furthermore , if @xmath6 is a _ weighted digraph _ with weights in @xmath1 then @xmath19 together with a function @xmath7 where @xmath20 is the _ weight _ of the edge @xmath21 for @xmath22 . we use the convention that @xmath23 if and only if @xmath24 . importantly , if @xmath2 then similar to digraphs we will denote this by writing @xmath3 . in order to describe the set of weights @xmath1 let @xmath25 $ ] denote the set of polynomials in the single complex variable @xmath26 with complex coefficients . we define the set @xmath1 to be the set of rational functions of the form @xmath27 where @xmath28 $ ] such that @xmath29 and @xmath30 have no common factors and @xmath30 is nonzero . the set @xmath1 is then a field under addition and multiplication with the convention that common factors are removed when two elements are combined . that is , if @xmath31 then @xmath32 where the common factors of @xmath33 and @xmath34 are removed . similarly , in the product @xmath35 of @xmath27 and @xmath36 the common factors of @xmath37 and @xmath34 are removed . however , we may at times leave sums and the products of sums of elements in @xmath1 uncombined and therefore possibly unreduced but this is purely cosmetic since there is one reduced form for any element in @xmath1 . to introduce the spectrum associated to a graph having weights in @xmath1 we will use the following notation . if the vertex set of the graph @xmath3 is labeled @xmath38 then we denote the edge @xmath39 by @xmath40 . the matrix @xmath41 defined entrywise by @xmath42 is the _ weighted adjacency matrix _ of @xmath6 . we let the _ spectrum _ of a matrix @xmath43 with entries in @xmath1 be the solutions including multiplicities of the equation @xmath44 and for the graph @xmath6 we let @xmath12 denote the spectrum of @xmath45 . the spectrum of a matrix with entries in @xmath1 is therefore a generalization of the spectrum of a matrix with complex entries . moreover , the spectrum is a _ list _ of numbers . that is , @xmath46 where @xmath47 is the multiplicity of the solutions @xmath48 to equation ( [ eq1 ] ) , @xmath29 the number of distinct solutions , and @xmath49 the elements in the list . in what follows we may write a list as a set with multiplicities if this is more convenient . in this section we describe the main results of the paper . that is , we present a method which allows for the reduction a graph while maintaining the graph s spectrum up to some known set . we also give specific examples of this process notably using this method to reduce graphs associated with the laplacian matrix of a graph . some natural consequences and extensions of this process are also mentioned . here we first introduce some definitions as well as some terminology that allow us to be precise in our formulation of an isospectral reduction . in the following if @xmath9 where @xmath4 is the vertex set of a digraph let @xmath50 denote the complement of @xmath10 in @xmath4 . also , as is standard , a _ path _ @xmath51 in a digraph @xmath19 is a sequence of distinct vertices @xmath52 such that @xmath53 for @xmath54 and in the case that the vetices @xmath55 are distinct , except that @xmath56 , @xmath51 is a _ cycle_. moreover , let the vertices @xmath57 of @xmath51 be the _ interior _ vertices of @xmath51 . [ def1 ] for @xmath19 let @xmath58 be the digraph @xmath6 with all loops removed . we say the nonempty vertex set @xmath9 is a _ structural set _ of @xmath6 if @xmath50 induces no cycles in @xmath58 and for each @xmath59 , @xmath60 . we denote by @xmath61 the set of all structural sets of @xmath6 . eigen3.eps ( 25.5,-4)@xmath6 ( 4,10.5)@xmath62 ( 14,10.5)@xmath63 ( 75,10.5)@xmath63 ( 28.5,11.5)@xmath64 ( 28.5,2.5)@xmath65 ( 42,10.5)@xmath66 ( 93,10.5)@xmath66 ( 53,10.5)@xmath67 ( 83,12)@xmath68 ( 85,1.5)@xmath69 ( 65,7)@xmath68 ( 100.5,7)@xmath70 ( 82,-4)@xmath71 for @xmath19 with @xmath72 let @xmath73 be the set of paths or cycles from @xmath74 to @xmath75 in @xmath6 having no interior vertices in @xmath10 . furthermore , let @xmath76 we call the set @xmath77 the set of all _ branches _ of @xmath6 with respect to @xmath10 . [ branchprod ] let @xmath3 and @xmath78 for some @xmath79 . if @xmath80 for @xmath81 we define @xmath82 as the _ branch product _ of @xmath83 . if @xmath84 we define @xmath85 . [ reductiondef ] let @xmath3 with structural set @xmath86 . define @xmath11 to be the digraph such that @xmath87 if @xmath88 and @xmath89 we call @xmath71 the _ isospectral reduction _ of @xmath6 over @xmath10 . the graph @xmath90 since @xmath91 implying @xmath92 is as well . figure 1 gives an example of a reduction of the graph @xmath6 . we note here that all figures in this paper follow the aforementioned conventions that undirected and unweighted edges are assumed to be oriented in both directions and have unit weight . as another example of a graph reduction consider the complete undirected unweighted graph without loops @xmath93 on @xmath94 vertices . if @xmath38 then @xmath95 has @xmath94 structural sets @xmath96 each given by @xmath97 , @xmath98 . for each @xmath99 the graph @xmath100 has an @xmath101 adjacency matrix @xmath102 where @xmath103 for all @xmath104 . for the complete bipartite graph @xmath105 where @xmath4 is partitioned into the sets @xmath106 and @xmath107 having @xmath108 and @xmath94 vertices respectively it follows that both @xmath109 . moreover , @xmath110 is the digraph with all possible edges including loops on @xmath108 vertices each having weight @xmath111 ( see figure 2 ) . in order to understand the extent to which the spectrum of a graph is maintained under different reductions we introduce the following . if @xmath10 is a structural set of the graph @xmath3 where @xmath38 let @xmath112 that is , @xmath14 is the set of @xmath113 for which there is some vertex @xmath74 off the structural set @xmath10 where @xmath114 or , as @xmath115 , the values of @xmath26 at which @xmath116 . as an example for the graph @xmath6 and structural set @xmath10 in figure 1 @xmath117 . bipartite.eps ( 12,-5.5)@xmath118 ( 1.5,4)@xmath106 ( 28,-4)@xmath107 ( 82,-5.5)@xmath119 ( 80,17)@xmath120 ( 101,17)@xmath120 ( 88.5,0)@xmath121 ( 88.5,35)@xmath121 if @xmath122 and @xmath123 then let @xmath124 be the list given by @xmath125 moreover , if it happens that @xmath126 then we say @xmath12 and @xmath127 differ at most by @xmath107 . the main result of this paper can then be phrased as follows . [ theorem1 ] let @xmath2 with @xmath79 . then @xmath12 and @xmath13 differ at most by @xmath14 . that is , the spectrum of @xmath6 and the spectrum of its reduction @xmath71 differ at most by elements of @xmath14 which justifies our use of the terminology isospectral reduction as the spectrum is preserved up to some known set . we note that two weighted digraphs @xmath128 , and @xmath129 are _ isomorphic _ if there is a bijection @xmath130 such that there is an edge @xmath40 in @xmath131 from @xmath74 to @xmath75 if and only if there is an edge @xmath132 between @xmath133 and @xmath134 in @xmath135 with @xmath136 . if the map @xmath137 exists it is called an _ isomorphism _ and we write @xmath138 . let @xmath122 . we say @xmath6 and @xmath139 have a reduction in common via the structural sets @xmath10 and @xmath140 respectively if @xmath79 , @xmath141 and @xmath142 . the fact that isomorphic graphs have the same spectrum with theorem [ theorem1 ] together imply the following . if @xmath122 have a reduction in common via the structural sets @xmath10 and @xmath140 respectively then @xmath12 and @xmath127 differ at most by @xmath143 . figure 3 gives an example of a reduction of @xmath139 in which @xmath144 where @xmath6 and @xmath10 are the graph and structural set respectively in figure 1 . therefore , the adjacency matrices of @xmath6 and @xmath139 in figures 1 and 3 respectively have the same spectrum up to @xmath145 . in this case one can compute @xmath146 , @xmath147 , and @xmath148 fig10.eps ( 14,-5)@xmath139 ( 0,-.5)@xmath149 ( 0,27)@xmath150 ( 26,27)@xmath151 ( 26,-0.5)@xmath152 ( 73,-5)@xmath153 ( 51.5,13)@xmath68 ( 66,18.5)@xmath150 ( 76,22)@xmath68 ( 78,3)@xmath69 ( 90,18.5)@xmath152 ( 101,13)@xmath70 we note that the representation of matrices with nonnegative entries by smaller matrices with polynomial entries has been used before ( see e.g. @xcite ) . however , the reason for doing so is different from our motivation in this paper . an alternate view of the graph reductions presented in section 3.1 is to consider the reduction process one in which the matrix @xmath45 is reduced to the matrix @xmath154 and view theorem [ theorem1 ] as a theorem about matrix reductions . viewed this way , an important application of theorem [ theorem1 ] is that one may reduce not only the graph @xmath6 but also the graphs associated with both the combinatorial laplacian matrix and the normalized laplacian matrix of @xmath6 . to make this precise , let @xmath19 be an unweighted undirected graph without loops , i.e. a _ simple graph_. if @xmath6 has vertex set @xmath38 and @xmath155 is the degree of vertex @xmath74 then its _ combinatorial laplacian matrix _ @xmath156 is given by @xmath157 on the other hand the _ normalized laplacian matrix _ @xmath158 of @xmath6 is defined as @xmath159 the interest in the eigenvalues of @xmath156 is that @xmath160 gives structural information about @xmath6 ( see @xcite ) . on the other hand knowing @xmath161 is useful in determining the behavior of algorithms on the graph @xmath6 among other things ( see @xcite ) . as every @xmath162 matrix with weights in @xmath1 has a unique weighted digraph associated to it then let @xmath163 be the graph with adjacency matrix @xmath156 and similarly let @xmath164 be the graph with adjacency matrix @xmath158 . since both @xmath163 and @xmath164 can be considered in @xmath0 via our conventions then either may be reduced . we summarize this as the following theorem which is a corollary to theorem [ theorem1 ] . suppose @xmath6 is a simple graph with vertex set @xmath4 . if @xmath9 is such that @xmath50 induces no cycles in @xmath6 then @xmath165 and @xmath166 and @xmath167 differ at most by @xmath168 . similarly , @xmath169 and @xmath170 and @xmath171 differ at most by @xmath172 for example if @xmath173 is the complete graph on 3 vertices then the graph @xmath163 , shown in figure 4 , has the structural set @xmath174 since @xmath175 induces no cycles in @xmath6 . reducing over this set yields @xmath176 where , as can be verified , @xmath177 implying @xmath178 . laplacian.eps ( 8,-2.5)@xmath6 ( 40,-2.5)@xmath163 ( 77,-2.5)@xmath179 ( -1,4)@xmath150 ( 17,4)@xmath149 ( 5,16)@xmath151 ( 33.5,5)@xmath150 ( 50.5,5)@xmath149 ( 38,16)@xmath151 ( 34,10)@xmath180 ( 48.5,10)@xmath180 ( 41,3)@xmath180 ( 28,1)@xmath181 ( 57,1)@xmath181 ( 46.5,17.5)@xmath181 ( 78,4.5)@xmath150 ( 93,4.5)@xmath149 ( 82,10)@xmath182 ( 66.5,7)@xmath183 ( 100,7)@xmath183 one could generalize @xmath156 to any @xmath2 where @xmath6 has no loops and @xmath94 vertices by setting @xmath184 for @xmath185 and @xmath186 . this generalizes and is consistent with what is done for weighted digraphs in @xcite for example . as any reduction @xmath71 of a graph @xmath2 over the structural set @xmath10 is again a graph in @xmath0 it is natural to consider sequences of reductions on a graph as well as to what degree a graph can be reduced via such reductions . in order to address this we need to first extend our notation to an arbitrary sequence of reductions . for @xmath3 suppose @xmath187 such that @xmath188 , @xmath189 and @xmath190 if this is the case then we say @xmath191 _ induces a sequence of reductions _ on @xmath6 and we write @xmath192 for @xmath193 . moreover , we let @xmath194 the following is an immediate corollary of theorem [ theorem1 ] . suppose @xmath191 induces a sequence of reductions on the graph @xmath2 . then @xmath12 and @xmath195 differ at most by @xmath196 . for @xmath197 $ ] let @xmath198 be the degree of @xmath29 and for @xmath199 let @xmath200 . let @xmath201 be the set of graphs with the property that for any @xmath202 with vertex set @xmath203 , @xmath204 for all @xmath205 . note that any graph @xmath6 where @xmath206 is in the set @xmath207 . with this in place we give the following theorem on sequential reductions . [ theorem3]*(commutativity of reductions ) * for @xmath202 suppose the sequences @xmath191 and @xmath208 both induce a sequence of reductions on @xmath6 . if @xmath209 then @xmath210 . that is , the final vertex set in a sequence of reductions completely specifies the reduced graph irrespective of the specific sequence . however , if @xmath191 and @xmath208 both induce a sequence of reductions on @xmath202 it is possible that @xmath211 even if @xmath209 . a natural goal then in designing a sequence of reductions is to somehow minimize @xmath212 by carefully choosing the structural set @xmath213 at the @xmath214 step in a particular sequence . to address the extent to which a graph may be reduced via some sequence of reductions as well as indicate the uniqueness and flexibility of such reductions we give the following theorem . [ theorem-1]*(existence and uniqueness ) * let @xmath3 and @xmath15 be any nonempty subset of @xmath4 . if @xmath202 then there exist a sequence @xmath215 inducing a sequence of reductions on @xmath6 . moreover , for any such sequence @xmath216 there is a unique graph @xmath217=\mathcal{r}(g;t_1,\dots , t_{n-1},\mathcal{v})$ ] independent of the particular sets @xmath218 . it is important to mention that @xmath15 in theorem [ theorem-1 ] is any subset of vertices of the graph @xmath6 . in particular , @xmath15 may not be a structural set of @xmath6 . in this section we introduce a related but alternate way of describing the branch structure of a graph where this structure is again related to the graph s spectrum . we then use this structure along with a small number of graph transformations to give a constructive method for reducing any graph @xmath2 over any @xmath79 . for a given branch @xmath80 in the graph @xmath3 let the _ weight sequence _ of @xmath83 be the sequence given by @xmath219 for a structural set @xmath220 of a digraph @xmath6 let @xmath221 be the _ branch decomposition _ of @xmath6 with respect to @xmath10 where this set includes multiplicities . note that @xmath222 can be written as lists but the formulation above is more convenient . [ def3 ] let @xmath122 . we say @xmath6 and @xmath139 have a common branch decomposition via the structural sets @xmath10 and @xmath140 respectively if @xmath79 , @xmath141 and there is a one - to - one map @xmath223 such that @xmath224 for example consider the unweighted digraphs @xmath6 and @xmath139 in figures 1 and 3 . for @xmath225 and @xmath226 @xmath227 @xmath228 these graphs have the same branch decomposition via the map @xmath223 given by @xmath229 and @xmath230 note that if two graphs @xmath6 and @xmath139 have a common branch decomposition with respect to the structural sets @xmath10 and @xmath140 respectively then there is some @xmath223 such that @xmath231 for all @xmath232 . this implies the following corollary of theorem [ theorem1 ] . let @xmath122 having a common branch decomposition via @xmath10 and @xmath140 respectively . then @xmath142 and @xmath12 and @xmath127 differ at most by @xmath143 . that is , common branch decompositions imply common reductions . one of the useful distinctions between branch reductions and decompositions is that the branch decomposition of a graph contains the lengths of the individual branches whereas the reduction does not . moreover , we note that in some ways branch decompositions are a more natural graph theoretic object since graphs with a common branch decomposition share the same weight set . it will be the fact that the branch decomposition of a graph retains the weight set and branch lengths that will allow us to prove theorem [ theorem1 ] . if @xmath2 and @xmath79 then two branches @xmath233 are said to be _ independent _ if they have no interior vertices in common . let @xmath234 be graphs with a common branch decomposition both with respect to the same set @xmath10 . then @xmath235 is a _ branch expansion _ of @xmath6 with respect to @xmath10 if any two @xmath236 are independent and each vertex of @xmath235 is on a branch of @xmath237 . note that @xmath6 and its branch expansion @xmath235 may have all vertices in common or share only the vertices in @xmath10 . * ( branch expansions)*[lemma0 ] let @xmath238 for @xmath2 . then an expansion @xmath235 of @xmath6 with respect to @xmath10 exists and @xmath12 and @xmath239 differ at most by @xmath14 . an example of a branch expansion is seen in figures 1 and 3 , @xmath6 being an expansion of @xmath139 over the set @xmath225 if the vertices of @xmath6 are relabeled via @xmath240 and @xmath241 . let @xmath3 be a weighted digraph where @xmath38 and @xmath242 . suppose @xmath243 for @xmath244 where @xmath245 . if we replace @xmath246 in @xmath6 by the two edges and loop with associated weights as in figure 5 then we call the resulting graph the graph @xmath6 with _ loop bisected edge _ @xmath246 with intermediate vertex @xmath75 . graphic2.eps ( 10,4)@xmath247 ( 0,-3)@xmath74 ( 54,-3)@xmath74 ( 43,-3)@xmath248 ( 98,-3)@xmath248 ( 76,-3)@xmath75 ( 75.5,7)@xmath249 ( 64,2)@xmath250 ( 85,2)@xmath251 [ lemma2]*(loop bisection ) * let @xmath3 where @xmath242 such that @xmath252 . if @xmath253 is the graph @xmath6 with loop bisected edge @xmath246 and @xmath79 then @xmath254 and the spectra @xmath12 , @xmath255 differ at most by @xmath14 . the goal now is to combine lemmas [ lemma0 ] and [ lemma2 ] to construct the reduction @xmath71 which can be done as follows . the first step in the reduction of @xmath6 over the structural set @xmath10 is to in fact do the opposite . that is , we first would like to find a branch expansion of @xmath6 . note that we can explicitly construct an expansion @xmath235 of @xmath6 via @xmath222 by taking each pair of vertices in @xmath10 and connecting them by independent branches with weight sets and multiplicities as specified in this decomposition . since each vertex of @xmath235 is by construction on some branch of @xmath235 then this is in fact an expansion of @xmath6 giving a proof to the existence claim in lemma [ lemma0 ] . for the next step in this reduction we use the lemma [ lemma2 ] to shorten the lengths of the independent branches in the expanded graph @xmath235 specifically , if @xmath256 has weight set @xmath257 , @xmath258 then by lemma [ lemma2 ] we may modify this weight set to @xmath259 without effecting the spectrum of the graph by more than @xmath14 . if this is continued until @xmath83 is reduced to a single edge @xmath260 from @xmath74 to @xmath75 then @xmath260 has weight @xmath261 . if every branch of @xmath235 is contracted to a single edge in this way then , after the multiple edges are made single via our convention , the resulting graph is the graph @xmath71 defined in definition [ reductiondef ] . moreover , as each step in this process does not change the spectrum of the graph by more than @xmath14 then the proof of theorem [ theorem1 ] follows once lemmas [ lemma0 ] and [ lemma2 ] are known to hold . before we prove the main results of this paper we note the following ( see @xcite for details ) . first , a directed graph is _ strongly connected _ if there is a path ( possibly of length zero ) from each vertex of the graph to every other vertex . the _ strongly connected components _ of @xmath19 are its maximal strongly connected subgraphs . moreover , its vertex set @xmath38 can always be labeled in such a way that @xmath45 has the following triangular block structure @xmath262\ ] ] where @xmath263 is a strongly connected component of @xmath6 and @xmath264 are block matrices with possibly nonzero entries . as @xmath265 then , since edges between strongly connected components correspond to the entries in the block matrices below the diagonal blocks , these edges may be removed or their weights changed without effecting @xmath12 . moreover , as an edge of @xmath6 is in a strongly connected component of @xmath6 if and only if it is on some cycle then all edges belonging to no strongly connected components of @xmath6 can be removed without effecting @xmath12 . with this in mind , for the graph @xmath3 where @xmath38 and @xmath79 , we say an edge @xmath266 is not on any branch of @xmath77 if @xmath267 do not both belong to some @xmath83 for all @xmath268 . if @xmath40 is not on any branch of @xmath77 then the claim is that it can not be on a cycle unless the cycle is a loop . to see this note that every cycle which is not a loop must contain a vertex of @xmath10 for @xmath50 not to induce a cycle in @xmath58 hence , every cycle is either a single branch or the union of several branches in @xmath77 . this implies that all edges except for loops off the branch set @xmath77 may be removed without effecting the graph s spectrum . on the other hand , suppose @xmath269 is a loop , having weight possibly equal to 0 , on the vertex @xmath75 where @xmath75 is not a vertex on any branch of @xmath77 . then this vertex may also be removed from @xmath6 without effecting the spectrum of the graph by more than @xmath14 . this follows from the fact in this case that the vertex @xmath75 is itself a strongly connected component of @xmath6 since , by the discussion above , it lies on no other cycle of @xmath6 . hence , if @xmath270 are the strongly connected components of @xmath6 where @xmath271 then @xmath272 therefore , removing @xmath75 from @xmath6 ( which removes @xmath269 ) changes @xmath12 at most by solutions to the equation @xmath273 or the @xmath26 such that @xmath274 is undefined , all of which are in @xmath14 . we record this as the following proposition . [ prop1 ] let @xmath3 and @xmath79 where @xmath275 are the set of vertices and edges respectively not on any branch of @xmath77 . if @xmath276 then the spectra @xmath12 and @xmath277 differ at most by @xmath14 . for ease of notation we adopt the following . for a square matrix @xmath106 denote by @xmath278_{ij}$ ] the _ minor _ of @xmath106 given by the determinant of the reduced matrix formed by omitting the @xmath279th row and @xmath280th column of @xmath106 . continuing in this manner we let @xmath281_{ij}]_{kl}=[m]_{ij , kl}$ ] be the determinant when row @xmath279 , column @xmath280 are omitted then row @xmath99 , column @xmath280 and so on . we now give a proof of lemma [ lemma0 ] . let @xmath282 where @xmath38 , @xmath283 and @xmath284 for @xmath285 . also let @xmath286 , and @xmath287 . from @xmath139 we construct the graph @xmath288 first by switching the edge @xmath289 to @xmath290 while maintaining its edge weight . second , if @xmath291 make @xmath292 the same as @xmath293 such that @xmath294 for all @xmath295 . let the resulting graph be the graph @xmath288 ( see figure 6 ) . brnchsep.eps ( 14,-5)@xmath139 ( 75,-5)@xmath288 ( 30,16)@xmath150 ( 87,16)@xmath150 ( 18,16)@xmath149 ( 75,16)@xmath149 ( 30,0)@xmath151 ( 91,0)@xmath151 ( -5,0)@xmath152 ( 53,0)@xmath152 ( 0,25)@xmath296 ( 24,25)@xmath297 ( 58,25)@xmath296 ( 92,25)@xmath297 if @xmath298 and @xmath299 the claim is that @xmath300 . to see this note that @xmath301_{31,22}=[\tilde{m}]_{11,32}$ ] and as well that @xmath301_{31,i2}=0 $ ] for all @xmath302 since the first two rows in the associated matrices are identical . if @xmath303 for all @xmath266 then @xmath304_{11}+\omega_{32}[\tilde{m}]_{31}=\\ & ( \omega_{22}-\lambda)\big((\omega_{22}-\lambda)[\tilde{m}]_{11,22}+\sum_{i=4}^n(-1)^i \omega_{i2}[\tilde{m}]_{11,i2}\big)-\omega_{32}(\omega_{22}-\lambda)[\tilde{m}]_{31,22}=\\ & ( \omega_{22}-\lambda)\big((\omega_{22}-\lambda)[\tilde{m}]_{11,22}+\sum_{i=3}^n(-1)^i \omega_{i2}[\tilde{m}]_{11,i2}\big).\end{aligned}\ ] ] since @xmath301_{11,i2}=[m]_{11,i2}$ ] for @xmath305 then @xmath306_{11}=\\ & ( \omega_{22}-\lambda)\big((\omega_{22}-\lambda)[m]_{11,22}+\sum_{i=3}^n(-1)^i \omega_{i2}[m]_{11,i2}\big)=\det(\tilde{m}).\end{aligned}\ ] ] hence , @xmath307 . moreover , it follows that if @xmath308 where @xmath309 then @xmath310 and both @xmath139 and @xmath288 have the same branch decomposition with respect to @xmath10 . to use this construction to our advantage let @xmath311 such that @xmath312 , @xmath283 . if @xmath313 , @xmath314 , and @xmath315 then for any @xmath316 where @xmath317 it follows that @xmath79 . suppose then that @xmath318 are distinct but not independent where their first interior point of intersection is the vertex @xmath149 . if it is the case that the vertex preceding @xmath149 on @xmath297 is @xmath151 and the vertex preceding @xmath149 on @xmath296 is @xmath152 where @xmath319 then @xmath6 has the structure of @xmath139 in the figure 6 . applying the same procedure to @xmath6 as we did to @xmath139 we obtain the graph @xmath253 having the same spectrum as @xmath6 where @xmath254 such that both graphs have the same branch decomposition with respect to @xmath10 . moreover , the branches @xmath297 and @xmath296 in @xmath253 no longer meet at @xmath149 but at some vertex following @xmath149 , if at all . brnchsep2.eps ( 14,-5)@xmath6 ( 75,-5)@xmath253 ( 28,19)@xmath150 ( 89,19)@xmath150 ( 18,19)@xmath149 ( 82,19)@xmath149 ( 17,0)@xmath74 ( 82,0)@xmath74 ( 3,25)@xmath296 ( 24,25)@xmath297 ( 67,25)@xmath296 ( 97,25)@xmath297 if the vertex @xmath320 , then under the assumption that @xmath321 , by a relabeling of the vertices in @xmath6 , with the exception of @xmath150 , let @xmath149 be the first vertex on @xmath297 and @xmath296 where the following vertex on these branches differ . if @xmath322 is the graph with adjacency matrix @xmath323 ( where @xmath140 denotes transpose ) then @xmath324 and @xmath325 . moreover , the branches @xmath326 corresponding to @xmath318 can be separated at @xmath149 again using the method above . if this modified graph is denoted @xmath327 then the graph @xmath253 with adjacency matrix @xmath328 ( see figure 7 ) and @xmath6 again have the same spectrum as well as having the same branch decomposition with respect to @xmath10 . as before , @xmath297 and @xmath296 in @xmath253 meet at some point following @xmath149 , if at all . since the subgraph @xmath329 of @xmath6 is a strongly connected component with single vertex @xmath330 then proposition [ prop1 ] implies @xmath277 and @xmath12 differ at most by @xmath14 which inturn implies the same for @xmath277 and @xmath255 . let @xmath331 denote the graph given by repeating this process over all branches in @xmath77 until each is independent of all other branches . then @xmath332 and @xmath277 and @xmath333 differ at most by @xmath14 . again proposition [ prop1 ] can be used to remove any vertex or edge not belonging to a branch of @xmath334 without effecting the spectrum of @xmath331 by more than @xmath14 . the removal of these edges and vertices produces the graph @xmath235 where @xmath235 is a branch expansion of @xmath335 with the desired properties . we now give a proof of lemma [ lemma2 ] . we proceed as in the previous proof . let @xmath336 have vertex set @xmath337 , @xmath283 where @xmath338 is an edge with weight @xmath339 . let @xmath253 be the graph @xmath335 with loop bisected edge @xmath338 with intermediate vertex @xmath150 . that is , @xmath253 has vertex set @xmath203 where @xmath340 , @xmath341 , and @xmath342 . the claim is that @xmath343 . to see this let @xmath315 where @xmath344 as well as @xmath303 and consider the two matrices @xmath345 and @xmath346 . assuming @xmath347 @xmath348_{11}-\omega_{21}[\tilde{m}]_{21}\big)=\\ & -\lambda\big((\omega_{11}-\lambda)\big((\omega_{22}-\lambda)[\tilde{m}]_{11,22}+\sum_{i=4}^n(-1)^i\omega_{2i}[\tilde{m}]_{11,2i}\big)-\omega_{21}(-\omega_{13}[\tilde{m}]_{21,13})\big).\end{aligned}\ ] ] on the other hand , @xmath349_{11}=\\ & -\lambda(\omega_{11}-\lambda)\big((\omega_{22}-\lambda)[m]_{11,22}-\frac{\omega_{21}\omega_{13}}{\omega_{11}-\lambda}[m]_{11,23}+\sum_{i=4}^n(-1)^i\omega_{2i}[m]_{11,2i}\big)=\\ & -\lambda\big((\omega_{11}-\lambda)(\omega_{22}-\lambda)[m]_{11,22}+\omega_{21}\omega_{13}[m]_{11,23}+(\omega_{11}-\lambda)\sum_{i=4}^n(-1)^i\omega_{2i}[m]_{11,2i}\big).\end{aligned}\ ] ] as @xmath301_{21,13}=[\tilde{m}]_{11,23}$ ] and @xmath301_{11,2i}=[m]_{11,2i}$ ] for @xmath350 then @xmath351 . since @xmath352 then this implies @xmath353 verifying the claim as @xmath354 can be removed by multiplying by its inverse in @xmath1 . an argument similar to that in the proof of proposition [ prop1 ] ( see equation ( [ eq3 ] ) ) implies @xmath277 and @xmath255 differ at most by @xmath14 . and if it is the case that @xmath355 a similar argument implies the same result completing the proof . we note that the preceding two proofs are enough to prove theorem [ theorem1 ] via the method described in section 4.3 . we now turn to the proof of theorem [ theorem3 ] for which we will need the following . [ lemma-1 ] if @xmath3 is a graph in @xmath207 where @xmath79 then @xmath356 . moreover , if @xmath38 where @xmath357 then @xmath358 for all @xmath359 . let @xmath360 , @xmath361 , and @xmath362 be elements of @xmath1 such that @xmath363 , @xmath364 . as @xmath365 @xmath366 @xmath367 are each nonpositive then the first statement of lemma [ lemma-1 ] follows from definition [ branchprod ] and [ reductiondef ] . consider then a single vertex @xmath74 of @xmath3 where @xmath6 is assumed to be in @xmath207 . as @xmath74 can not induce any cycles in @xmath58 and @xmath368 given @xmath369 then @xmath358 so long as @xmath4 has more than one vertex . the result then follows . let @xmath3 be a graph in @xmath207 with @xmath38 . taking the perspective that reductions amount to removing vertices from a graph then , by way of notation , if @xmath370 then inductively define @xmath371 note that the graph @xmath372 is well defined by lemma [ lemma-1 ] for each @xmath373 . [ lemma4 ] let @xmath3 where @xmath202 . if @xmath79 , @xmath38 , and @xmath374 where @xmath375 then for any permutation @xmath137 on @xmath376 , @xmath377 . that is , any reduction over a structural set @xmath10 can be accomplished by sequentially removing vertices in @xmath50 in any order without effecting the resulting graph . in the proof of this lemma we use the following . if @xmath378 where @xmath379 and @xmath380 then let @xmath381 be the _ concatenation _ of these branches given by @xmath382 . note that the branch @xmath383 is the _ empty branch _ with branch product 0 . we now give a proof of lemma [ lemma4 ] . let @xmath6 and @xmath10 be as in the statement of lemma [ lemma4 ] . for the sake of notation let @xmath384 and note that as @xmath50 induces no cycles in @xmath58 then the same is true of @xmath385 implying @xmath386 . from this and lemma [ lemma-1 ] it then follows that @xmath387 . therefore , let @xmath388 , @xmath389 , and @xmath11 . also for any graph @xmath390 with structural set @xmath140 let @xmath391 be the branches in @xmath392 containing the vertex @xmath393 . then for @xmath394 note that @xmath395 consists of at most two branches @xmath396 where @xmath397 where either branch is possibly empty . hence , @xmath398 . in what follows let @xmath399 be the direct product @xmath400 . as @xmath401 and @xmath402 if each are nonempty then @xmath403 on the other hand , for @xmath394 @xmath404 note that if @xmath405 then there exists a unique pair @xmath406 such that @xmath407 . hence , @xmath408 . moreover , as @xmath409 then @xmath410 for all @xmath394 implying @xmath411 to see how this is useful let @xmath412 for @xmath413 . as @xmath414 for the same reason @xmath386 then by adjusting the indices in equation ( [ eq5 ] ) it follows that @xmath415 by inserting this back into ( [ eq5 ] ) , @xmath416 @xmath417 continuing in this way we arrive at @xmath418 where the final equality follows from the fact that @xmath419 . as the labeling of @xmath50 was arbitrary then @xmath377 for any permutation @xmath137 on @xmath376 . [ lemma5 ] let @xmath202 with vertex set @xmath38 and @xmath420 . for any permutation @xmath137 on @xmath421 where @xmath422 , @xmath423 . given @xmath3 let @xmath424 , @xmath425 , and @xmath426 . for @xmath427 let @xmath428 and note that if @xmath429 where @xmath185 then @xmath430 this implies in particular that @xmath431 @xmath432 as this equation is symmetric in the indices 1 and 2 then @xmath433 for @xmath429 , @xmath185 . if @xmath434 then it follows that @xmath435 . hence , @xmath436 note that this can be used to show @xmath437 or that any transposition in the order of the removal of vertices @xmath438 does not effect the reduction . as any permutation @xmath439 on @xmath421 can be achieved via some number of transpositions it follows that @xmath440 . we now give a proof of theorem [ theorem3 ] . let @xmath191 and @xmath208 both induce a sequence of reductions on the graph @xmath2 where @xmath209 . if @xmath441 and @xmath442 then label @xmath443 , @xmath444 where @xmath445 for @xmath446 , @xmath447 . by repeated use of lemma [ lemma4 ] it follows that @xmath448 since the order in which the vertices of @xmath449 and @xmath450 are removed does not matter so long as the vertices in @xmath449 are removed before those in @xmath451 and the vertices in @xmath450 before @xmath452 . since @xmath453 then under the relabeling of @xmath454 to @xmath455 let @xmath439 be the permutation on @xmath456 such that @xmath457 . lemma [ lemma5 ] then implies @xmath458 . it then follows that @xmath210 . for a proof of theorem [ theorem-1 ] we give the following . let @xmath3 be a graph in @xmath207 where @xmath459 is any nonempty set . if @xmath460 then lemma [ lemma-1 ] implies that @xmath461 is well defined completing the proof of the first statement of the theorem . the proof of the second statement follows by a direct application of theorem [ theorem3 ] . recall from section 5 that the strongly connected components of a graph @xmath2 carry all the information necessary to compute the spectrum of @xmath12 . a natural question then is to what degree graph reductions respect the structure of strongly connected components . our first result in this direction is the following . [ lemma6 ] for the weighted digraph @xmath3 suppose @xmath79 and @xmath270 are the strongly connected components of @xmath6 . if @xmath213 is the nonempty set of vertices of @xmath263 contained in @xmath10 then @xmath462 are strongly connected components of @xmath71 for @xmath446 where @xmath463 . suppose @xmath267 are vertices in @xmath96 which are possibly nondistinct . assuming @xmath464 has more than one vertex then there is a path or cycle @xmath51 , with positive length , contained in @xmath464 from @xmath74 to @xmath75 . as every edge in this strongly connected component must be on a cycle then @xmath465 where each @xmath466 for @xmath467 since every cycle in @xmath6 contains at least one vertex in @xmath10 . if @xmath468 is the vertex joining @xmath469 and @xmath470 there is a path or cycle @xmath471 in @xmath71 from @xmath74 to @xmath75 . switching the roles of @xmath74 and @xmath75 it follows that both the vertices @xmath96 belong to the same strongly connected component of @xmath71 . to see that these are the only vertices in this strongly connected component suppose that there is no path from @xmath74 to @xmath75 in @xmath6 . that is , if @xmath472 then @xmath473 , @xmath474 where @xmath475 . by similar reasoning then both vertices must be in different strongly connected components of @xmath71 as there can be no concatenation of branches @xmath476 from @xmath74 to @xmath75 . hence , the strongly connected component in @xmath477 containing @xmath96 contains only the vertices in @xmath96 implying @xmath478 is a strongly connected component of @xmath477 . if @xmath464 consists of a single vertex @xmath479 then @xmath96 is empty implying @xmath478 is as well . on the otherhand if @xmath480 then , from the above , @xmath481 . the result then follows . lemma [ lemma6 ] implies that the adjacency matrix of @xmath71 can be written in the block diagonal form @xmath482\ ] ] where @xmath483 is the adjacency matrix of @xmath462 if @xmath463 . that is , reductions respect the strongly connected component structure of the graph in the sense that the reduction of a strongly connected component is a again a strongly connected component in the reduced graph . another characteristic of strongly connected components mentioned in section 5 is that all edges off of these components @xmath2 can be removed without effecting @xmath12 . that is , if @xmath3 and @xmath484 where @xmath485 is the set of edges in @xmath5 belonging to some strongly connected component of @xmath6 then @xmath486 . this suggests an alternate approach to finding a reduction of the graph @xmath6 by first removing the edges of @xmath6 which do not belong to any strongly connected component then reducing as usual . let @xmath2 and @xmath79 . then @xmath487 and both @xmath488 and @xmath489 differ from @xmath12 by at most @xmath14 . let @xmath3 and @xmath79 . as @xmath6 and @xmath490 have the same strongly connected components then the fact that @xmath491 implies via lemma [ lemma6 ] that @xmath71 and @xmath492 have the same strongly connected components as well . hence , the same is true of @xmath492 and @xmath493 . note that @xmath492 has the property that each of its edges is contained in some strongly connected component of this graph . this follows from the fact that every branch in @xmath494 is contained in some strongly connected component of @xmath490 so lemma [ lemma6 ] implies there are no edges between the associated components in @xmath492 . as there are no edges between strongly connected components in @xmath492 and @xmath493 then these graphs are identical . to compare spectra note that both @xmath486 and @xmath495 + @xmath496 . furthermore , an application of theorem [ theorem1 ] implies @xmath497 and @xmath488 differ at most by @xmath14 since every loop of @xmath5 is contained in @xmath485 . similarly , as @xmath12 and @xmath13 differ at most by @xmath14 the proof is complete . a typical graph will usually have many different structural sets which on the one hand adds flexibility to the process of graph reductions but on the other means some choice of structural set must be made before a graph can be reduced . in this section we consider a particular type of structural set that is defined independent of any specific graph . the advantage in this is foremost that every graph in @xmath0 will have a reduction with respect to this particular type of vertex set . this will allow us not only to relate each graph s spectrum to this particular reduction but also to partition the graphs in @xmath0 according to their reductions . for a vertex @xmath498 let the _ out - degree _ of this vertex be denoted by @xmath499 . for the graph @xmath19 define @xmath500 we say a cycle of @xmath6 is a _ simple cycle _ if it contains no vertices in @xmath501 . we call the set of vertices @xmath501 together with the vertices on the simple cycles of @xmath6 the _ basic structural set _ of @xmath6 and denote this by @xmath502 to justify our use of the term _ basic structural set _ we give the following proposition . [ prop4 ] for @xmath2 the set @xmath503 where @xmath504 . also the relation of having a reduction in common with respect to a graphs basic structural sets is an equivalence relation on @xmath0 . for @xmath2 it follows that @xmath503 as every cycle of @xmath6 contains some vertex in @xmath505 . this can be seen from the fact that every cycle of @xmath6 is or is not simple which in either case implies that the cycle must contain a vertex in @xmath505 . also , because no branch in @xmath506 contains a nonzero loop on an interior vertex then @xmath504 . next , if @xmath507 such that the pair @xmath6 and @xmath139 and the pair @xmath139 and @xmath508 have a reduction in common with respect to their basic structural sets then @xmath509 implying in particular that @xmath510 . hence , having a common reductions with respect to a graphs regular vertices is a transitive relation on @xmath0 . from this one can also see that the relation is also reflexive and symmetric . we denote by @xmath511 $ ] the equivalence class of graphs in @xmath0 having a common reduction with respect to their basic structural sets which contains the graph @xmath6 . this will be of use in section 7.1 . it is also important to note that other criteria induce an equivalence relation on @xmath0 or @xmath207 all that is required is some rule that results in a unique reduction or sequence of reductions of any graph in @xmath0 or @xmath207 respectively . this can be summarized as follows . * ( uniqueness and equivalence relations ) * suppose for any graph @xmath3 in @xmath207 that @xmath512 is a rule that selects a unique nonempty subset @xmath513 . then @xmath512 induces an equivalence relation @xmath514 on the set @xmath207 where @xmath515 if @xmath516\simeq\mathcal{r}_{\tau(w)}[h]$ ] where @xmath517 @xmath518 are the vertex sets of @xmath6 and @xmath139 respectively . as an example consider the following rule . for a graph @xmath3 where @xmath202 let @xmath519 be the set of vertices of minimal out degree . if @xmath520 then by theorem [ theorem-1 ] @xmath521 $ ] is uniquely defined and this process may be repeated until all vertices of the resulting graph have the same out degree . as the vertex set @xmath522 of the graph resulting from this sequence of reductions is unique then the relation of having an isomorphic reduction via this rule induces an equivalence relation on @xmath207 . unlike the weight set @xmath1 used in this paper , it is more typical to consider weighted digraphs @xmath6 having weights in some subset of @xmath523 . the tradeoff then for considering @xmath71 is that although the graph structure is simpler the weights become rational functions . in some sense the weights begin to take on the shape of the characteristic polynomial of @xmath45 . on the other hand , if we wish to reduce the size of the graph , i.e. number of vertices , while maintaining its spectrum along with a particular set of edge weights the following is possible . [ theorem5 ] let @xmath524 be a unital subring and suppose @xmath2 has weights in @xmath525 . if @xmath526 and @xmath527 is the length of the longest branch in @xmath528 for all @xmath529 then there exists a graph @xmath335 with the following properties : + ( 1 ) @xmath530 $ ] implying @xmath12 and @xmath277 differ at most by @xmath531 . + ( 2 ) @xmath335 has weights in @xmath525 . + ( 3 ) @xmath335 has @xmath532 vertices . we prove the theorem by explicitly constructing the graph @xmath335 with the required properties . to do so , let @xmath3 where @xmath6 has weights in @xmath525 some unital subring of @xmath1 . for @xmath526 let @xmath533 and let @xmath534 be a branch of maximal length for all @xmath446 . furthermore , let @xmath535 and note each @xmath536 is given by the sequence of vertices @xmath537 where @xmath527 is its length . we construct @xmath335 via the following . first , let @xmath538 where @xmath539 and each @xmath540 for @xmath541 . moreover , each @xmath536 is independent of each @xmath542 , @xmath185 . if @xmath543 is the edge from @xmath544 to @xmath545 then set @xmath546 and note that each @xmath547 . lastly , for each @xmath548 , if @xmath83 has length @xmath94 then add an edge @xmath260 from @xmath74 to @xmath549 to the edge set @xmath550 . make @xmath551 where @xmath552 is the weight sequence of @xmath83 . if multiple edges of @xmath335 are reduced to single edges by summing the corresponding weights then @xmath335 has the following properties . the vertex set @xmath550 consists of all distinct vertices in @xmath553 of which there are @xmath532 . moreover , each weight of @xmath335 is the product or sum of products of elements of @xmath525 implying @xmath335 has weights in this set . finally , there is a one - to - one correspondence between branches in @xmath554 and @xmath555 where corresponding branches have the same branch product . hence , @xmath556 $ ] . fig1.eps ( 10,-3)@xmath139 ( 0,0)@xmath149 ( 0,20)@xmath150 ( 20,20)@xmath151 ( 20,0)@xmath152 ( 71,-3)@xmath6 ( 40,10)@xmath62 ( 57,14)@xmath63 ( 71,20)@xmath64 ( 71,5)@xmath65 ( 85,14)@xmath66 ( 102,10)@xmath67 an example of this construction is the graph @xmath139 in figure 8 which is constructed from the graph @xmath6 ( in the same figure ) over the weight set @xmath557 . @xmath139 is a reduction over the weight set @xmath557 of @xmath6 in the sense that it has fewer vertices than the graph @xmath6 from which it is constructed . furthermore , one can compute @xmath558 and @xmath559 . a more complicated problem involves finding the graph @xmath560 $ ] with the least number of vertices where both @xmath139 and @xmath6 have weights in some set @xmath524 . the main results of this paper are concerned with the way in which the structure of a graph influences the spectrum of the graph s adjacency matrix and to what extent this spectrum can be maintained if this structure is simplified . for the most part these results give algorithmic methods whereby a graph can be reduced in size but do not mention if such methods might be useful in determining the spectrum of a given graph . as is shown in @xcite , the graph reductions considered here do indeed help in the estimation of a graph s spectrum . for example , it is possible to extend the eigenvalue estimates given by the classical result of gershgorin @xcite to matrices with entries in @xmath1 . the main results of @xcite is that eigenvalue estimates via this extension improve as the graph is reduced . analogous results and extensions also hold for the work done by brauer , brualdi and varga @xcitewhose original results are each improvements of gershgorin s . gershgorin s original result is in fact equivalent to a nonsingularity result for diagonally dominant matrices ( see theorem 1.4 @xcite ) which can be traced back to earlier work done by lvy , desplanques , minkowski , and hadamard @xcite . importantly , we note that via our method of graph reductions we obtain better estimates than those given by all of the previous existing methods . moreover , graph reductions can be used to obtain estimates of the spectrum of a matrix with increasing precision depending on how much one is willing to reduce the associated graph . if the graph is completely reduced the corresponding eigenvalue estimates give the exact spectum of the matrix along with some finite set of points . these techniques can furthermore be used for the estimation of spectra for combinatorial and normalized laplacian matrices as well as giving bounds on the spectral radius of a given matrix . in fact it is in such applications that the flexibility of isospectral graph reductions is particularly useful . the results of the present paper demonstrate various approaches to simplifying a graph s structure while maintaining its spectrum . therefore , these techniques can be used for optimal design , in the sense of structure simplicity of dynamical networks with prescribed dynamical properties ranging from synchronizability to chaoticity @xcite . we would like to thank c. kirst and m. timme for valuable comments and discussions . this work was partially supported by the nsf grant dms-0900945 and the humboldt foundation .

in this review some recent multi - loop results obtained in the framework of perturbative quantum chromodynamics ( qcd ) and quantum electrodynamics ( qed ) are discussed . after reviewing the most advanced techniques used for the computation of renormalization group functions , we consider the decoupling of heavy quarks . in particular , an effective method for the evaluation of the decoupling constants is presented and explicit results are given . furthermore the connection to observables involving a scalar higgs boson is worked out in detail . an all - order low energy theorem is derived which establishes a relation between the coefficient functions in the hadronic higgs decay and the decoupling constants . we review the radiative corrections of a higgs boson into gluons and quarks and present explicit results up to order and , respectively . in this review special emphasis is put on the applications of asymptotic expansions . a method is described which combines expansion terms of different kinematical regions with the help of conformal mapping and pad approximation . this method allows us to proceed beyond the present scope of exact multi - loop calculations . as far as physical processes are concerned , we review the computation of three - loop current correlators in qcd taking into account the full mass - dependence . in particular , we concentrate on the evaluation of the total cross section for the production of hadrons in annihilation . the knowledge of the complete mass dependence at order has triggered a bunch of theory - driven analyses of the hadronic contribution to the electromagnetic coupling evaluated at high energy scales . the status is summarized in this review . in a further application four - loop diagrams are considered which contribute to the order qed corrections to the decay . its relevance for the determination of the fermi constant is discussed . finally the calculation of the three - loop relation between the and on - shell quark mass definitions is presented and physical applications are given . to complete the presentation , some technical details are presented in the appendix , where also explicit analytical results are listed . ( to appear in physics reports )

hep-ph0201075

nowadays the standard model ( sm ) of elementary particle physics is well established . some parts of it ( e.g. the properties of the @xmath8 boson ) have been tested to an accuracy far below the percent level mostly at the cern large - electron - positron collider ( lep , geneva ) , at the slac linear collider ( slc , stanford ) and at the fermilab tevatron ( chicago ) . up to now no significant deviation between theory and experiment has been found . for other parts of the sm , related to cp violation and quark mixing , the @xmath9 factories like babar at slac , belle at kek ( tsukuba ) or hera - b at desy ( hamburg ) will provide deeper insight , and significant improvements in the determination of the corresponding parameters will be obtained . currently mainly the scalar sector of the sm eludes from direct experimental observation . this affects both the generation of the particle masses and the existence of the higgs boson itself . probably run ii of the tevatron and certainly the large hadron collider ( lhc ) at cern will provide more sureness not only in connection to the higgs sector but also to possible extensions of the sm . once the higgs boson would be discovered it immediately would become subject to precision measurements . in particular at a future @xmath2 linear collider such as desy tesla , a precise study of its properties would be possible . in the recent years there has been an enormous development in the evaluation of radiative corrections . it is fair to say that the major part of it was initiated by the fundamental works of t hooft and veltman in 1972 @xcite where dimensional regularization ( see also @xcite ) was established as a powerful tool in the evaluation of multi - loop diagrams . since that time a whole industry has been formed to develop techniques for the computation of complicated feynman integrals . at one - loop order the procedure of the computation has been systematically studied quite some time ago @xcite . nevertheless also nowadays it is not completely straightforward to evaluate an arbitrary one - loop diagram in particular if many legs and complicated momentum configurations are involved . one can easily imagine that at two and more loops one arrives quite soon at the limit where the occuring mathematical expressions can not be solved . at two - loop order certain classes of diagrams can still be treated by either using a combination of analytical simplifications and fast numerical routines , like in the case of two - point function with several non - zero masses @xcite , or applying purely analytical methods , like in the case of massless digrams with four external legs @xcite . however , at three - loop order it is essentially only possible to solve one - scale integrals . a systematic study at four or more loops is still missing . qcd , the field theoretical realization of the strong interaction , constitutes an important part of the sm and also of most of its extensions . at low energies the coupling constant of qcd , @xmath10 , is large and perturbative calculations are not possible . however , due to the phenomenon of asymptotic freedom the value of @xmath10 gets smaller with raising energy and perturbation theory is an appropriate tool to evaluate radiative corrections . up to now the vast majority of the multi - loop calculations have been performed in the framework of qed and qcd . one reason is certainly that the calculations are simpler as compared to the full sm since there are less parameters . furthermore , there is a strong hierarchy both in the quark and lepton masses which also simplifies the calculations . on the other hand , the higher - order corrections are indeed necessary . in qed there exist precise experiments which require high theoretical precision and although the coupling constant is quite small sometimes high loop orders are necessary . for example , in the case of the anomalous magnetic moment of the electron , four - loop corrections are needed to match the experimental precision . in qcd the coupling is roughly a factor of ten bigger . nevertheless it is often still small enough to perform a perturbative expansion . however , the higher order terms are significant and can not be neglected in the cases where high precision is required . in this work some recent developments in the calculation of multi - loop diagrams are reviewed . thereby we will mention the most important methods which have been used in the computation of higher order quantum corrections and explain a few selected ones in greater detail . at the same time , we discuss the present theoretical status of important physical quantities . in particular , the renormalization group functions in the modified minimal subtraction scheme @xcite ( @xmath7 ) are provided up to the four - loop order . as is well - known , four loop running must be accompanied by three - loop matching at quark thresholds . the corresponding decoupling relations are presented in section [ sec : dec ] . the hadronic higgs decay is closely connected to the decoupling relations as we will show is section [ sec : dim4 ] . parts of the quantum corrections can be computed in the framework of an effective lagrangian where the coefficients can be determined from the decoupling relations . the origin of this miraculous connection lies in the use of the dimension - four operators , which constitute an important ingredient of the effective lagrangian . another application of the dimension - four operators are the quartic mass corrections to the cross section @xmath11 , which is also discussed in section [ sec : dim4 ] . the last issue is again picked up in section [ sec : pade ] , where also qcd corrections to the production of hadrons in @xmath2 annihilation are computed . putting together all terms one arrives at a complete picture up to the quartic mass corrections of order @xmath1 . the main purposes of section [ sec : pade ] are practical applications of asymptotic expansions for details . ] . besides the diagonal correlators also the non - diagonal ones are considered . the fermi constant , @xmath6 , the mass of the @xmath8 boson , @xmath12 , and the electromagnetic coupling constant , @xmath13 , are the best known parameters of the sm . @xmath12 has been measured at lep with an accuracy of a few per mille to be @xmath14 gev . in this review we want to discuss quantum corrections to the other two parameters . an essential ingredient to the running of the electromagnetic coupling from @xmath15 to @xmath16 is provided by the cross section @xmath11 . the correction terms discussed earlier have been used to obtain so - called theory - driven results . the different approaches are discussed . as further applications we present the status of the qed corrections to the muon decay and the relation between the @xmath7 and the on - shell quark mass . let us in the following discuss the individual issues in more detail . as far as radiative corrections are concerned , a crucial role is played by the renormalization group functions . in particular the functions @xmath17 and @xmath18 governing the running of the coupling and the quark masses comprise a significant part of the higher quantum corrections . they are in particular very important to re - sum large logarithms to all orders in perturbation theory . only a few years ago the four - loop terms of order @xmath0 have been evaluated for @xmath17 @xcite and @xmath18 @xcite . the latter has been computed by two groups using completely independent methods . both methods are based on the fact that the pole part of a logarithmically divergent diagram is independent of the masses or momenta . in @xcite this is exploited together with the technique of infra - red re - arrangement ( irr ) in order to obtain a factorization of the four - loop integrals into massless three - loop and massive one - loop ones . in refs . @xcite all lines were assigned to the same mass , and all external momenta were set to zero , which leads to a special class of bubble diagrams . from them only the pole parts have to be computed . in section [ sec : rge ] we want to review both methods and explicitly demonstrate the way they work . in this review special emphasis is put on the construction of effective theories in the framework of qcd . in section [ sec : dec ] an effective qcd lagrangian is constructed for the case where one of the quarks is much heavier than the others . the construction is made explicit by specifying the relations between the parameters in the full and effective theories . these relations provide at the same time the well - known decoupling constants which have to be applied in qcd every time a particle threshold is crossed . the most prominent example for their necessity is probably the computation of @xmath19 from @xmath20 or vice versa . in the latter case five quarks are active whereas in the former one only three quarks are present in the effective qcd lagrangian . in section [ sec : dim4 ] a slightly different point of view is adopted . here the scalar operators of dimension four are considered in qcd . in a first step they are used to construct an effective lagrangian describing the coupling of an intermediate - mass higgs boson to quarks and gluons . the top quark is considered as heavy and manifests itself in the coefficient functions of the effective lagrangian . once the latter has been found , the imaginary part of the higgs boson correlator in the effective theory leads to the total decay rate . as a central result of section [ sec : dim4 ] we derive a low - energy theorem which considerably simplifies the computation of the coefficient functions as they are related to the decoupling constants of qcd evaluated in section [ sec : dec ] . in section [ sub : hggbfm ] the background field method is introduced as a convenient tool for the computation of higher - order corrections . as an example , the coefficient functions describing the decay of the higgs boson into gluons is also computed in this framework . in the second part of section [ sec : dim4 ] another important application of the scalar dimension four operators is discussed , namely the quartic mass corrections to the cross section @xmath11 . mass corrections of order @xmath21 and @xmath22 are obtained relatively easy as in qcd there are no non - trivial operators of dimension less than four . however , the quartic corrections require the inclusion of the dimension - four operators with all their renormalization and mixing properties . we will explain the techniques and present results obtained recently at order @xmath1 . the last part of this review , section [ sec : pade ] , is devoted to the discussion of some results obtained with the help of asymptotic expansion accompanied with conformal mapping and pad approximation . this method has been developed in the recent years and has been applied successfully to a number of important processes . the underlying idea is the following : only in rare cases it is possible to compute three - loop diagrams involving more than one scale . however , if a certain hierarchy exists between the scales it is promising to apply an asymptotic expansion . this effectively reduces the number of scales present in the integrals which are subsequently significantly simplified . in particular we will discuss the corrections of order @xmath3 to the photon polarization function . its imaginary part is directly connected to the physical quantity @xmath23 . the application of our method leads to the full mass dependence . combining the results with the quartic corrections given in section [ sub : as3m4 ] , one obtains a prediction for @xmath24 up to and including @xmath25 . only recently also the non - diagonal current correlator formed by a massive and a massless quark has been computed . in this application special emphasis lies on the extraction of information about the threshold behaviour which has some relevance in the framework of heavy - quark effective qcd . as an application of the knowledge of @xmath24 to high perturbative order , we discuss the evaluation of @xmath26 . the electromagnetic coupling is defined at vanishing momentum transfer . however , its evolution to high energies constitutes the dominant part of the radiative corrections to electroweak observables . the accurate determination of @xmath26 is thus essential for any precise test of the theory . at the same time the indirect determination of the masses of heavy , hitherto unobserved particles , e.g. the higgs boson or supersymmertic particles , depends critically on this parameter . of particular importance in this context is the hadronic vacuum polarization . it is nearly as large as the leptonic contribution , but can not yet be computed perturbatively . however , it may be related through dispersion relations to the cross section for hadron production in electron - positron annihilation . the integrand can thus be obtained from data , phenomenological models and/or perturbative qcd , whenever applicable . in section [ subsub : delal ] we will discuss the developments in the evaluation of @xmath26 which took place in the recent two to three years due to the knowledge of the complete mass dependence of @xmath24 at order @xmath3 ( cf . section [ subsub : r ] ) . @xmath6 is defined through the muon lifetime , and the decay of the muon , as a purely leptonic process , is rather clean both experimentally and theoretically . the one - loop corrections of order @xmath13 were computed more than 40 years ago @xcite , whereas only recently the two - loop corrections of order @xmath4 have been evaluated @xcite . the large gap in time shows that this calculation is highly non - trivial . the inclusion of the two - loop terms greatly reduced the relative theoretical error of @xmath27 which was an estimate of the size of the missing corrections . the remaining error on @xmath6 now reads @xmath28 and is of pure experimental nature . upcoming experiments will further improve the accuracy of the muon lifetime measurement and therefore the @xmath29 corrections to the muon decay are very important and constitute a crucial ingredient from the theoretical side . in section [ sub : mudec ] we discuss the results obtained with the help of asymptotic expansion . in the sm the quark masses have still relatively big uncertainties . this is mainly due to the confinement property of qcd which prevents the production of free quarks . it is also important to have a convenient definition of the quark mass in order to perform a comparison between theory and experiment . recently there has been quite some activity connected to the precise determination of the bottom- and top - quark masses . the bottom - quark mass is determined with the help of qcd sum rules where a proper mass definition helps to reduce the error . in the case of the top quark , studies have been performed for an @xmath2 collider with a center - of - mass energy in the threshold region of top - quark - pair production . an energy scan which provides the measurement of the total production cross section would provide an error of about 100 mev in the top - quark mass . also here a special mass definition has to be employed . in both cases the three - loop on - shell@xmath7 conversion formula is needed in order to obtain the corresponding @xmath7 quark mass . the latter is important for processes not connected to the threshold . in section [ sub : msos ] these issues are discussed in detail . in perturbative qcd the renormalization group functions play a very important role . in particular the @xmath30 and @xmath31 functions governing the running of the strong coupling and the quark masses , respectively , are indispensable when evaluating physical observables . for this reason we decided to discuss the techniques used for the computation of the four - loop contributions to @xmath30 @xcite and @xmath31 @xcite in more detail . in the @xmath7 scheme the knowledge of the renormalization group functions is equivalent to the knowledge of the corresponding renormalization constant . thus , in order to compute a renormalization constant at @xmath32-loop order it is sufficient to evaluate the ultra - violet ( uv ) poles of @xmath32-loop diagrams . nevertheless it is often also quite useful to have also a handle on the infra - red ( ir ) poles . more details on the uv and ir structure of feynman diagrams are given in section [ sub : uvir ] . afterwards , in sections [ sub : irr ] and [ sub : massint ] , two practical methods are described which have been applied at the four - loop level . before we want to consider explicit examples in the next two sections the theoretical background needed for the higher - loop calculation of renormalization group functions is introduced in this subsection . in particular we want to demonstrate the interplay between uv and ir divergences in dimensional regularization and show their connection to asymptotic expansions . we refrain from presenting the material in a mathematical rigorous framework , for which we refer to the original literature , but exemplify the important points at explicit diagrams . as the properties discussed in this section are independent of the particle type we consider only scalar propagators of the form @xmath33 where also @xmath34 is allowed . we furthermore assume that the integration momenta are denoted by @xmath35 and introduce the abbreviation @xmath36 . in order to remove the uv divergences the so - called @xmath37 operation has been introduced @xcite . it is a recursive subtraction scheme where the uv divergences are removed from the feynman integrals in a way compatible to adding local counterterms to the lagrangian . formally , the @xmath37 operation can be written as a sum where each term is a product of operators acting on subsets of disjoint one - particle - irreducible subgraphs . in this way counterterms are generated . they have to be inserted into the vertices of the diagrams which remain after shrinking the corresponding subgraphs to a point . non - trivial diagrammatic examples can , e.g. , be found in the text books @xcite . the @xmath37 operation applied to a one - loop propagator - type integral with external momentum @xmath38 leads to the equation @xmath39[m^2-(k_1+q)^2 ] } \right ] & = & \int_1\frac{1}{[m^2-k_1 ^ 2][m^2-(k_1+q)^2 ] } + z^{(1 ) } \nonumber\\ & = & \mbox{finite } \ , , \label{eq : z1}\end{aligned}\ ] ] from which the renormalization constant @xmath40 is determined . in the @xmath7 scheme it reads @xmath41 once @xmath40 is known one can turn to the two - loop order . the application of the @xmath37 operation to the diagram shown in fig . [ fig : rex ] leads to @xmath42[m^2-(k_1+q)^2][m^2-k_2 ^ 2]^2 } \right ] } \nonumber\\&=&\mbox { } \int_1\int_2 \frac{1}{[m^2-(k_1-k_2)^2][m^2-(k_1+q)^2][m^2-k_2 ^ 2]^2 } + z^{(1 ) } \int_2\frac{1}{[m^2-k_2 ^ 2]^2 } + z^{(2 ) } \nonumber\\ & = & \mbox{finite } \ , , \label{eq : z2}\end{aligned}\ ] ] which fixes @xmath43 to @xmath44 \ , . \label{eq : z2_2}\end{aligned}\ ] ] a formal definition of the @xmath37 operation can , e.g. , be found in @xcite . [ cols="^,^,^ " , ] an example demonstrating the practical application of the large - momentum expansion was already presented in section [ sec : dim4 ] ( cf . [ fig : lmp1l ] ; see also @xcite ) . as an application of the hard - mass procedure let us consider the double - bubble diagram of fig . [ fig : db ] with the hierarchy @xmath45 . the imaginary part leads to contributions for the total cross section @xmath11 . one may think of charm quark production ( @xmath46 ) in the presence of a virtual bottom quark ( @xmath47 ) . it turns out that already the first term provides a very good approximation almost up to the threshold @xmath48 @xcite . for simplicity we set @xmath49 and @xmath50 in the following . = 2.cm = 3.cm + = 2.cm = 3.cm = 3.cm the corresponding diagrammatic representation is shown in fig . [ fig : dbhmp ] . there are three subdiagrams , where one again corresponds to the naive taylor expansion of the integrand in the external momentum @xmath38 . after taylor expansion , the subdiagrams are reduced to tadpole integrals with mass scale @xmath51 . the scale of the co - subgraphs is given by @xmath38 , thus leading to massless propagator - type integrals . the result for the first three terms reads @xcite @xmath52 + \cdots\ , , \label{eq : dbhmpres}\end{aligned}\ ] ] with @xmath53 and @xmath54 . we want to stress that the main simplification , which is common to all kinds of asymtotic expansions , comes from the fact that the expansions in the small parameters are done before any momentum integration is performed . the proof that this leads to correct results is based on the so - called strategy of regions @xcite . there different regions of each loop momentum are selected and in each of them taylor expansions with respect to the small parameters are performed . in the limits of the hard - mass and large - momentum procedures an interpretation of the different regions in terms subgraphs and cosub - graphs is possible ( see above ) . this is different in the case of the threshold expansion @xcite where a graphical representation becomes much less transparent . however , the application of the strategy of regions @xcite leads to correct results . the problem of evaluating one - loop feynman diagrams is at least in principle solved ( see , e.g. , refs . however , in case many legs and lots of different masses appear also one - loop computations can become very tedious , in particular if degenerate momentum configurations are involved . at two - loop order the class of feynman diagrams which have been studied in detail is much more restricted . there is a good understanding of two - point functions ( see , e.g. , @xcite ) which also has found important physical applications @xcite . concerning three- and four - point functions one is essentially restricted to the massless case . in this context we want to draw the attention to the recent activity in the computation of the two - loop box diagrams ( for a brief overview see @xcite ) . within the last two years the basis has been established to compute two - loop virtual corrections to the four - point feynman amplitudes where all internal lines are massless and at most one external leg is off - shell . this opens the door to investigate next - to - next - to - leading order processes like the two - jet production at hadron colliders or three - jet production in @xmath2 annihilation . it is obvious that the complexity of the computation of a feynman diagram strongly depends on the number of different scales involved . there is one class of diagrams which is studied in great detail up to three loops , namely diagrams which only depend on one dimensionful scale . next to massless propagator - type diagrams with one external momentum , @xmath38 , we have in mind vacuum integrals with one non - zero mass , @xmath55 and so - called on - shell integrals for which the condition @xmath56 is fulfilled . the basic idea for the computation of the integrals is common to all three types : after the numerator is decomposed in terms of the denominator recurrence relations are applied which express the diagram as a linear combination of so - called master integrals . only for the latter a hard computation is necessary . however , since in the case of single - scale diagrams the master integrals are essentially pure numbers it is also possible to use high - precision numerical methods in case an analytical calculation is not possible . we want to mention that for the massless propagator - type and the massive vacuum integrals two and nine master integrals are needed , respectively . in the case of the three - loop on - shell integrals a list of all master integrals can be found in the appendix of ref . counting also those integrals which are composed of products of lower - order diagrams they amount to 18 . for convenience we want to provide the analytical results for the massless one - loop two - point functions , @xmath57 , the on - shell two - point functions , @xmath58 , and the one- ( @xmath59 ) and two - loop ( @xmath60 ) vacuum integrals in euclidian space . @xmath61 the results of a general three - loop diagram can not be expressed in terms of @xmath62 functions . moreover , due to the large number of contributing diagrams and the complexity of intermediate expressions it is absolutely necessary to use computer algebra programs for the computation of multi - loop diagrams . it is thus also hardly possible to provide intermediate results which eventually could be used in other calculations . therefore , on one side one is left with the description of the method used for the evaluation of the diagrams . on the other hand it is possible to provide the program code which was used for the computation . thus everybody can repeat the calculation or apply it to own problems . both for the massless propagator - type integrals and the massive vacuum integrals form @xcite packages have been published : massless integrals up to three loops can be computed using mincer @xcite ; the package for the massive integrals is called matad @xcite . in the following we will present an example which demonstrates the use of matad . the use of mincer is very similar . actually , in the package geficom @xcite matad and mincer are used in parallel using the same notation for the input . let us consider the triangle diagram as pictured in fig . [ fig : hgg ] . it is one of the 657 diagrams which contribute to the coefficient function @xmath63 appearing in the effective lagrangian of eq . ( [ eq : eff ] ) . according to the lorentz structure the result can be written as @xmath64 where @xmath65 and @xmath66 are the momenta of the gluons with polarization vectors @xmath67 and @xmath68 . thus the vertex diagrams have to be expanded up to linear order both in @xmath65 and @xmath66 , and an appropriate projector has to be applied in order to get @xmath69 . + matad requires one file containing the diagrams and the projectors which have to be applied . in our case it could look as follows .... * -- # [ treat0 : multiply , ( a*deno(2,-2)*(q1.q2*d_(mu , nu)-q2(nu)*q1(mu)-q2(mu)*q1(nu ) ) + b*deno(2,-2)*(-q1.q2*d_(mu , nu)+(3 - 2*ep)*q2(nu)*q1(mu)+q2(mu)*q1(nu ) ) ) ; .sort * -- # ] treat0 : * -- # [ treat1 : * -- # ] treat1 : * -- # [ treat2 : * -- # ] treat2 : * -- # [ treatmain : * -- # ] treatmain : * -- # [ d3l335 : ( ( -1 ) * m * dg(nu1,nu2,p1 ) * dg(nu7,nu8,-p4 ) * dg(nu3,nu4,q1,-p1 ) * dg(nu5,nu6,q2,p1 ) * s(-q1,-p3m , nu7,-q1,p5m , nu4,-p2m , nu6,q2,p5m , nu8,q2,-p3 m ) * v3g(mu , q1,nu1,-p1,nu3,p1-q1 ) * v3g(nu , q2,nu2,p1,nu5,-p1-q2 ) * 1 ) ; # define topology " o4 " * -- # ] d3l335 : .... where the diagram d3l335 corresponds to the one shown in fig . [ fig : hgg ] ( for details concerning the nomenclature of the momenta see @xcite ) . the fold treat0 contains ( up to an overall factor @xmath70 ) the projector on the coefficients in front of the structures @xmath71 and @xmath72 of eq . ( [ eq : k1 ] ) . they are marked by the symbols ` a ` and ` b ` , respectively . thus the transversality of eq . ( [ eq : k1 ] ) can be explicitly checked in the sum of all contributing diagrams ( the result of a single diagram does in general not have a transverse structure ) . a second file which is required , the so - called main - file , reads .... # define prb " hgg " # define problem0 " 1 " # define dala12 " 1 " # define gauge " 0 " # define power " 2 " # define cut " 0 " # define folder " hgg " # define diagram " d3l335 " # - # include main.gen .... the fifth line ensures that an expansion of the integrand up to the second order in the external momenta is performed and the third line sets @xmath73 and @xmath74 to zero and factors out the scalar product @xmath75 . ` # define cut " 0 " ` sets @xmath76 to zero in the final result . in this example we choose feynman gauge which is achieved with ` # define gauge " 0 " ` . after calling matad it takes of the order of a minute to obtain the result : .... d3l335 = + ep^-2 * ( 40*q1.q2*m^2*a + 344/9*q1.q2 ^ 2*a - 232/9*q1.q2 ^ 2*b ) + ep^-1 * ( - 308/3*q1.q2*m^2*a - 3530/27*q1.q2 ^ 2*a + 1786/27*q1.q2 ^ 2 * b ) + 60*q1.q2*m^2*z2*a + 734/3*q1.q2*m^2*a - 1936/9*q1.q2 ^ 2*z3*a + 1136/9 * q1.q2 ^ 2*z3*b + 172/3*q1.q2 ^ 2*z2*a - 116/3*q1.q2 ^ 2*z2*b + 46817/81 * q1.q2 ^ 2*a - 26239/81*q1.q2 ^ 2*b ; .... note that the terms proportional to ` q1.q2*m^2 ` cancel after adding all contributing diagrams . in this section we briefly want to mention the program packages which have been used to obtain most of the results discussed in this review . for a general overview concerning the automation of feynman diagram computation we refer to @xcite . the large number of diagrams which occurs in particular if one considers higher loop orders makes it necessary to generate the diagrams automatically . the fortran program qgraf @xcite provides the possiblility to implement own models in a simple way . furthermore it is quite fast and generates several thousand diagrams in a few seconds . one of the disadvantages of qgraf is that the user has to put the feynman rules himself . on the other hand , this provides quite some flexibility in the choices of the vertices . e.g. , it is straightforward to implement the vertex involving the coupling of the operator @xmath77 to four gluons ( cf . ( [ eq : op1 ] ) ) . in general , the application of asymptotic expansions , in particular if serveral of them are applied successively , generates many subdiagram which have to be expanded in several small quantities . even for a single multi - loop diagram this becomes very tedious if it has to be performed by hand . for this reason the programs lmp @xcite and exp @xcite have been developed . lmp was especially developed in order to apply the large - momentum procedure to the diagonal current correlators @xcite . in some sense exp can be considered as the successor of lmp . next to the hard - mass procedure also the succesive application of large - mometum and/or hard - mass procedure is possible . this broadens the area of applications . here we just want to mention as examples the correction of @xmath78 to the @xmath8 boson decay @xcite , @xmath79 corrections to the top quark decay @xcite or two - loop qed corrections to the muon decay @xcite ( cf . section [ sub : mudec ] ) . the very computation of the integrals is performed with the program packages mincer @xcite and matad @xcite ( see appendix [ sub : single ] ) . the former deals with massless propagator - type integrals up to three loops and matad was written to deal with vacuum diagrams at one- , two- and three - loop order where several of the internal lines may have a common mass . the area of application for each of the individual packages seems to be quite restrictive . however , in particular the combined application offers a quite flexible use . in order to handle problems where a large number of diagrams are involved and where eventually an asymptotic expansion has to be applied in a convenient way the program package geficom has been written . a very limited number of small input files allows the user to rule the flow of the computation . qgraf is called to generate the diagrams . a mathematica @xcite script determines the toplogy of each individual diagram and provides input which either can be directly read from mincer and/or matad or can be passed to exp or lmp . at the end the results of the individual diagrams are summed and the bare result is stored . moreover a convenient environment is provided which , e.g. , makes sure that all result files are up - to - date . thus , processes involving a large number of ( sub-)diagrams can be treated without taking care of each individual result . transforming the decoupling constants of eqs . ( [ eq : zetamos ] ) and ( [ eq : zetagos ] ) to the @xmath7 scheme one obtains @xmath80 \nonumber\\ & \approx&1 + 0.2060\left(\frac{\alpha_s^{(n_f)}(\mu_h)}{\pi}\right)^2 + \left(1.8476 + 0.0247\,n_l\right ) \left(\frac{\alpha_s^{(n_f)}(\mu_h)}{\pi}\right)^3 , \label{eq : zetamms}\end{aligned}\ ] ] @xmath81 \nonumber\\ & \approx&1 + 0.1528\left(\frac{\alpha_s^{(n_f)}(\mu_h)}{\pi}\right)^2 + \left(0.9721 - 0.0847\,n_l\right ) \left(\frac{\alpha_s^{(n_f)}(\mu_h)}{\pi}\right)^3 \ , . \label{eq : zetagms}\end{aligned}\ ] ] in the following , we list the decoupling constants @xmath82 and @xmath83 appropriate for the general gauge group su(@xmath84 ) . the results read @xmath85\ln\frac{\mu^2}{m_h^2 } + \left(-\frac{1}{32n_c^2 } -\frac{5}{576n_c } -\frac{5}{384 } + \frac{5}{576}n_c + \frac{17}{384}n_c^2\right ) \ln^2\frac{\mu^2}{m_h^2 } \nonumber\\ & & \mbox { } + \left(\frac{1}{144n_c } -\frac{11}{576 } -\frac{1}{144}n_c + \frac{11}{576}n_c^2\right ) \ln^3\frac{\mu^2}{m_h^2 } \nonumber\\ & & \mbox { } + n_l\left(\frac{1}{n_c}-n_c\right)\left ( -\frac{1327}{31104 } + \frac{1}{36}\zeta_3 + \frac{53}{1152}\ln\frac{\mu^2}{m_h^2 } + \frac{1}{288}\ln^3\frac{\mu^2}{m_h^2}\right ) \bigg\ } \ , , \label{eq : zetammsnc}\end{aligned}\ ] ] @xmath86 \nonumber\\ & & \mbox { } + \left(\frac{\alpha_s^{(n_f)}(\mu)}{\pi}\right)^3 \bigg\ { \frac{1}{n_c^2}\left ( -\frac{97}{2304 } + \frac{95}{1536}\zeta_3\right ) + \frac{1}{n_c}\left ( -\frac{103}{10368 } + \frac{7}{512}\zeta_3\right ) \nonumber\\ & & \mbox { } -\frac{1063}{5184 } + \frac{893}{3072}\zeta_3 + n_c\left ( \frac{451}{20736 } -\frac{7}{256}\zeta_3\right ) + n_c^2\left ( \frac{7199}{13824 } -\frac{17}{48}\zeta_3 \right ) \nonumber\\ & & \mbox { } + \left(-\frac{9}{256n_c^2 } -\frac{5}{192n_c } -\frac{119}{1152 } -\frac{23}{3456}n_c -\frac{1169}{6912}n_c^2\right ) \ln\frac{\mu^2}{m_h^2 } \nonumber\\ & & \mbox { } + \left(\frac{5}{192n_c } -\frac{11}{384 } + \frac{35}{576}n_c -\frac{1}{128}n_c^2\right ) \ln^2\frac{\mu^2}{m_h^2 } -\frac{1}{216}\ln^3\frac{\mu^2}{m_h^2 } \nonumber\\ & & \mbox { } + n_l\bigg [ \frac{41}{1296n_c } -\frac{329}{10368}n_c + \left(-\frac{5}{384n_c } + \frac{139}{3456}n_c\right ) \ln\frac{\mu^2}{m_h^2 } \nonumber\\ & & \mbox { } + \left(\frac{1}{96n_c } -\frac{1}{96}n_c\right ) \ln^2\frac{\mu^2}{m_h^2}\bigg]\bigg\ } \ , . \label{eq : zetagmsnc}\end{aligned}\ ] ] for @xmath87 , we recover eqs . ( [ eq : zetamms ] ) and ( [ eq : zetagms ] ) . the renormalized decoupling constants @xmath88 and @xmath89 for the quark and gluon fields , respectively , arise from eqs . ( [ eq : zeta20 ] ) and ( [ eq : zeta30 ] ) . of course , @xmath88 and @xmath89 are both gauge dependent . restricting ourselves to the case @xmath87 , we find in the covariant gauge ( [ eq : gluprop ] ) @xmath90 \ , , \label{eq : zeta2res } \\ \zeta_3^{\rm ms}&=&1 + \frac{\alpha_s^{(n_f)}(\mu)}{\pi}\bigg ( { 1\over 6}\ln\frac{\mu^2}{m_h^2}\bigg ) + \left(\frac{\alpha_s^{(n_f)}(\mu)}{\pi}\right)^2\bigg ( { 91\over 1152 } + { 29\over 96}\ln\frac{\mu^2}{m_h^2 } + { 3\over 32}\ln^{2}\frac{\mu^2}{m_h^2}\bigg ) \nonumber\\ & & \mbox { } + \left(\frac{\alpha_s^{(n_f)}(\mu)}{\pi}\right)^3\bigg [ -{284023\over 62208 } + { 86183\over 27648}\zeta_3 + { 99\over 128}\zeta_4 -{1\over 32}b_4 \nonumber\\ & & \mbox { } + \left({52433\over 27648 } -{33\over 64}\zeta_3\right)\ln\frac{\mu^2}{m_h^2 } + { 383\over 2304}\ln^2\frac{\mu^2}{m_h^2 } + { 119\over 768}\ln^3\frac{\mu^2}{m_h^2 } \nonumber\\ & & \mbox { } + n_l\bigg ( { 3307\over 15552 } -{1\over 12}\zeta_3 -{293\over 1152}\ln\frac{\mu^2}{m_h^2 } + { 1\over 36}\ln^2\frac{\mu^2}{m_h^2 } -{1\over 96}\ln^3\frac{\mu^2}{m_h^2}\bigg ) \nonumber\\ & & \mbox { } + \xi\bigg ( -{677\over 1536 } + { 3\over 32}\zeta_3 + { 233\over 1024}\ln\frac{\mu^2}{m_h^2 } -{3\over 32}\ln^2\frac{\mu^2}{m_h^2 } + { 3\over 256}\ln^3\frac{\mu^2}{m_h^2}\bigg)\bigg ] \,.\end{aligned}\ ] ] in ref . @xcite the yukawa corrections to @xmath82 and @xmath83 enhanced by the top quark mass have been evaluated . they are conveniently expressed in terms of the variable @xmath91 where @xmath92 is the top quark mass defined in the @xmath7 scheme . corrections proportional to @xmath93 arise if in addition to the pure qcd lagrangian also the couplings of the higgs boson ( @xmath94 ) and the neutral ( @xmath95 ) and charged ( @xmath96 ) goldstone boson to the top quark are considered . sample diagrams contributing to @xmath82 and @xmath83 are shown in figs . [ fig : sig ] and [ fig : z3 ] , respectively . c c the decoupling constant for the @xmath97 and @xmath98 quark mass reads are repeated from eq . ( [ eq : zetamms ] ) . ] @xmath99 \bigg\ } \ , , \label{eq : zetaml}\end{aligned}\ ] ] where @xmath100 is the third component of the weak isospin , i.e. @xmath101 for up - type quarks and @xmath102 for down - type quark flavours . for the bottom quark one receives @xmath103 \nonumber\\&&\mbox { } + \left(\frac{\alpha_s^{(6)}(\mu)}{\pi}\right)^2 \left [ \frac{472933}{12096 } - \frac{6133}{168}\zeta_2 + \frac{905}{72}\zeta_3 + \frac{383}{18}\zeta_4 + \frac{1251}{112}s_2 + \frac{19}{72}d_3 \right.\nonumber\\&&\mbox{}\left . - \frac{7}{9}b_4 + \left ( \frac{763}{18 } - \frac{55}{3}\zeta_2 - \frac{43}{4}\zeta_3 \right)\ln\frac{\mu^2}{m_t^2 } + \frac{529}{48}\ln^2\frac{\mu^2}{m_t^2 } + \frac{29}{12}\ln^3\frac{\mu^2}{m_t^2 } \right.\nonumber\\&&\mbox{}\left . + n_l\left ( - \frac{23}{24 } + \frac{31}{36}\zeta_2 - 2\zeta_3 + \left ( - \frac{241}{144 } + \frac{2}{3}\zeta_2 \right)\ln\frac{\mu^2}{m_t^2 } - \frac{1}{2}\ln^2\frac{\mu^2}{m_t^2 } \right.\right.\nonumber\\&&\mbox{}\left.\left . - \frac{1}{12}\ln^3\frac{\mu^2}{m_t^2 } \right ) \right ] \bigg\ } \ , , \label{eq : zetamb}\end{aligned}\ ] ] where we have used @xmath104 , @xmath105 and @xmath106 . the constants @xmath107 where @xmath108 , @xmath109 is clausen s function and @xmath110 is the quadrilogarithm , occur in the evaluation of the three - loop master diagrams @xcite . finally , for @xmath83 we obtain the following result : @xmath111 + x_t\bigg\ { - \frac{2}{3 } + \ln\frac{\mu^2}{m_t^2 } \nonumber\\&&\mbox { } + \frac{\alpha_s^{(6)}(\mu)}{\pi } \left [ c_f\,\left ( - \frac{17}{16 } + \frac{5}{4}\zeta_2 + \frac{25}{8}\zeta_3 - 3 \ln\frac{\mu^2}{m_t^2 } + \frac{3}{4 } \ln^2\frac{\mu^2}{m_t^2 } \right ) \right.\nonumber\\&&\left.\mbox { } + c_a\,\left ( - \frac{5}{4 } + \frac{3}{8}\zeta_2 - \frac{95}{64}\zeta_3 + \frac{7}{4 } \ln\frac{\mu^2}{m_t^2 } \right ) + t\,\left ( \frac{5}{4 } + \frac{7}{8}\zeta_3 + \frac{4}{9 } \ln\frac{\mu^2}{m_t^2 } \right.\right.\nonumber\\&&\left.\left.\mbox { } - \frac{2}{3 } \ln^2\frac{\mu^2}{m_t^2 } \right ) - \frac{7}{2}\zeta_3 t \right ] \bigg\ } \bigg\ } \label{eq : zetag } \,,\end{aligned}\ ] ] where the contribution of the diagrams in fig . [ fig : z3](c ) corresponds to the last entry in the last line of eq . ( [ eq : zetag ] ) . for convenience also the pure qcd result of @xmath79 is listed . the corresponding three - loop terms can be found in @xcite . in the remainder of this appendix we want to provide the analytical results for the @xmath93-enhanced corrections of order @xmath112 to the coefficient functions @xmath113 and @xmath114 @xcite . for @xmath113 we have @xmath115 \bigg\ } \ , , \label{eqc1}\end{aligned}\ ] ] where @xmath92 is the @xmath7 top quark . the @xmath116 terms can be found in @xcite and the @xmath79 results were computed in @xcite the corrections of @xmath117 are taken from @xcite . for the light quarks we get for @xmath114 @xmath118 \ , , \nonumber\\\end{aligned}\ ] ] and in the case of the bottom quark the coefficient function reads : @xmath119 + \left(\frac{\alpha_s^{(6)}(\mu)}{\pi}\right)^2 \left [ - \frac{12169}{144 } + \frac{110}{3}\zeta_2 \right.\nonumber\\&&\left.\mbox { } + \frac{43}{2}\zeta_3 - \frac{89}{2}\ln\frac{\mu^2}{m_t^2 } - \frac{55}{4}\ln^2\frac{\mu^2}{m_t^2 } + n_l\left ( \frac{241}{72 } - \frac{4}{3}\zeta_2 + 2\ln\frac{\mu^2}{m_t^2 } \right.\right.\nonumber\\&&\left.\left.\mbox { } + \frac{1}{2}\ln^2\frac{\mu^2}{m_t^2 } \right ) \right ] \bigg\ } \,.\end{aligned}\ ] ] in @xcite @xmath120 and @xmath121 are listed for general gauge group @xmath122 . as in the literature the quadratic and quartic correction terms to @xmath24 are not yet available in analytic form we want to provide the corresponding results using the notation introduced in eqs . ( [ eq : rq ] ) and ( [ eq : rq2 ] ) . for completeness also the massless approximation is given . of course , it is the same in the @xmath7 and on - shell schemes @xmath123 \nonumber\\&&\mbox { } + \left({\alpha_s\over \pi}\right)^3\,\bigg [ { 87029\over 288 } - { 121\over 8}\,\zeta_2 - { 1103\over 4}\,\zeta_3 + { 275\over 6}\,\zeta_5 + n_f\,\bigg ( -{7847\over 216 } + { 11\over 6}\,\zeta_2 \nonumber\\&&\mbox { } + { 262\over 9}\,\zeta_3 - { 25\over 9}\,\zeta_5 \bigg ) + n_f^2\,\bigg ( { 151\over 162 } - { 1\over 18}\,\zeta_2 - { 19\over 27}\,\zeta_3 \bigg ) \bigg ] \,,\end{aligned}\ ] ] @xmath124 \ , , \nonumber\\ r_{qq,2}^{\rm os } & = & { m_q^2\over s}\,\left({\alpha_s\over \pi}\right)^3\ , \bigg [ -80 + 60\zeta_3 + n_f\left(\frac{32}{9}-\frac{8}{3}\zeta_3\right ) \bigg ] \,,\end{aligned}\ ] ] where @xmath125 . @xmath126 \ , , \nonumber\\ r_{qq,4}^{\rm os } & = & \left({m_q^2\over s}\right)^2\ , \left({\alpha_s\over \pi}\right)^2\,\bigg [ \frac{13}{3 } - 4\zeta_3 - { l_{ms}}+ { \alpha_s\over \pi } \ , \left ( - \frac{4217}{48 } + 15\zeta_2 + \frac{139}{3}\zeta_3 + \frac{50}{3}\zeta_5 \right.\nonumber\\&&\left.\mbox { } + { l_{ms}}\left(\frac{97}{4 } - 38\zeta_3\right ) - 2{l_{ms}}^2 + n_f\left ( \frac{457}{108 } - \frac{2}{3}\zeta_2 - \frac{22}{9}\zeta_3 \right.\right.\nonumber\\&&\left.\left.\mbox { } + { l_{ms}}\left ( - \frac{13}{18 } + \frac{4}{3}\zeta_3 \right ) \right ) \right ) \bigg ] \,,\end{aligned}\ ] ] with @xmath127 . # 1#2#3 ann . b * # 1 * ( # 2 ) # 3 # 1#2#3 act . phys . pol . b * # 1 * ( # 2 ) # 3 # 1#2#3 comp . commun . * # 1 * ( # 2 ) # 3 # 1#2#3 comm . math . # 1 * ( # 2 ) # 3 # 1#2#3 eur . j. c * # 1 * ( # 2 ) # 3 # 1#2#3fortschr . phys . * # 1 * ( # 2 ) # 3 # 1#2#3 j. comp . * # 1 * ( # 2 ) # 3 # 1#2#3 nucl . a * # 1 * ( # 2 ) # 3 # 1#2#3 nucl . b * # 1 * ( # 2 ) # 3 # 1#2#3 nuovo cim . a * # 1 * ( # 2 ) # 3 # 1#2#3 phys . b * # 1 * ( # 2 ) # 3 # 1#2#3 phys . reports * # 1 * ( # 2 ) # 3 # 1#2#3 phys . d * # 1 * ( # 2 ) # 3 # 1#2#3 phys . * # 1 * ( # 2 ) # 3 # 1#2#3 phys . * # 1 * ( # 2 ) # 3 # 1#2#3 phys . reports * # 1 * ( # 2 ) # 3 # 1#2#3 prog . part . # 1 * ( # 2 ) # 3 # 1#2#3sov . j. nucl . # 1 * ( # 2 ) # 3 # 1#2#3 teor . mat . * # 1 * ( # 2 ) # 3 # 1#2#3yad . * # 1 * ( # 2 ) # 3 # 1#2#3 z. phys . c * # 1 * ( # 2 ) # 3 # 1#2#3 ibid . * # 1 * ( # 2 ) # 3 g. t hooft and m. veltman , nucl . b * 44 * ( 1972 ) 189 . g. t hooft and m. veltman , nucl . b * 50 * ( 1972 ) 318 . g. t hooft and m. veltman , nucl . b * 153 * ( 1979 ) 365 . bollini and j.j . giambiagi , phys . b * 40 * ( 1972 ) 566 ; nuovo cim . b * 12 * ( 1972 ) 20 . r. gastmans and r. meuldermans , nucl . phys . b * 63 * ( 1973 ) 277 ; + w. j. marciano and a. sirlin , nucl . b * 88 * , 86 ( 1975 ) ; + w. j. marciano , phys . d * 12 * , 3861 ( 1975 ) . a. sirlin , phys . d * 22 * ( 1980 ) 971 . g. passarino and m. veltman , nucl . b * 160 * ( 1979 ) 151 . w. f. hollik , fortsch . * 38 * ( 1990 ) 165 . a. denner , fortsch . phys . * 41 * ( 1993 ) 307 . g. weiglein , r. scharf , and m. bhm , nucl . b * 416 * ( 1994 ) 606 ; + s. bauberger , f.a . berends , m. bhm and m. buza , nucl . b * 434 * ( 1995 ) 383 . t. gehrmann and e. remiddi , report no . : hep - ph/0101147 ; + and references therein . g. t hooft , . larin , report nos . : preprint nikhef - h/92 - 18 ( 1992 ) , hep - ph/9302240 ; in proc . of the int . baksan school `` particles and cosmology '' ( april 22 - 27 , 1993 , kabardino - balkaria , russia ) , eds . alexeev , v.a . matveev , kh.s . nirov , v.a . rubakov ( world scientific , singapore , 1994 ) . k.g . chetyrkin and m. steinhauser , unpublished ; + m. steinhauser , published in the proceedings of _ 4@xmath128 international symposium on radiative corrections ( radcor 98 ) : applications of quantum field theory to phenomenology _ , barcelona , catalonia , spain , 8 - 12 sep 1998 . report no . : hep - ph/9811342 . s. wolfram , _ mathematica a system for doing mathematics by computer _ ( addison - wesley , new york , 1988 ) . k.g . chetyrkin , m. misiak , and m. mnz , nucl . b * 518 * ( 1998 ) 473 . gross and f. wilczek , ; ; + h.d . politzer , . kataev and v.v . starshenko , mod . lett . a * 10 * ( 1995 ) 235 ; phys . d * 52 * ( 1995 ) 402 ; + m.a . samuel , j. ellis , and m. karliner , phys . lett . 74 ( 1995 ) 4380 raczka and a. szymacha , z. phys . c * 70 * ( 1996 ) 125 ; phys . d * 54 * ( 1996 ) 3073 ; + j. ellis , e. gardi , m. karliner , and m.a . samuel , phys . b * 366 * ( 1996 ) 268 ; phys . d * 54 * ( 1996 ) 6986 ; + s. groote , j.g . krner , a.a . pivovarov , and k. schilcher , phys . * 79 * ( 1997 ) 2763 ; + s. groote , j.g . krner , and a.a . pivovarov , phys . b * 407 * ( 1997 ) 66 . harlander , phys . b * 492 * ( 2000 ) 74 . harlander and w.b . kilgore , phys . d * 64 * ( 2001 ) 013015 . s. catani , d. de florian and m. grazzini , jhep * 0105 * ( 2001 ) 025 . dewitt , phys . rev . * 162 * ( 1967 ) 1195 ; + g. t hooft , in _ karpacz 1975 , proceedings , acta universitatis wratislaviensis no.368 , vol.1 _ , wroclaw 1976 , 345 - 369 ; + h. kluberg - stern and j.b . zuber , ; ; + s. ichinose and m. omote , ; + d.m . capper and a. maclean , ; + d.g . boulware , . kniehl and m. spira , z. phys . c * 69 * ( 1995 ) 77 . kniehl , phys . * 240 * ( 1994 ) 211 . m. spira , fortsch . * 46 * ( 1998 ) 203 . e. braaten and j.p . leveille , phys . d * 22 * ( 1980 ) 715 . m. drees and k. hikasa , phys . b * 240 * ( 1990 ) 455 . ( e ) ibid . b * 262 * ( 1991 ) 497 . r. harlander and m. steinhauser , phys . d * 56 * ( 1997 ) 3980 . kataev and v.t . kim , report no . : hep - ph/9304282 . kataev , jetp lett . * 66 * ( 1997 ) 327 . larin , t. van ritbergen , and j.a.m . vermaseren , . t. inami , t. kubota , and y. okada , z. phys . c * 18 * ( 1983 ) 69 . a. djouadi , m. spira , and p.m. zerwas , phys . b * 264 * ( 1991 ) 440 ; + m. spira , a. djouadi , d. graudenz , and p.m. zerwas , nucl . b * 453 * ( 1995 ) 17 . see , e.g. , + j. ellis , i. jack , d.r . jones , m. karliner , and m.a . samuel , phys . d * 57 * ( 1998 ) 2665 ; + v. elias , f.a . chishtie , and t.g . steele , j. phys . g * 26 * ( 2000 ) 1239 . smirnov , _ renormalization and asymptotic expansion _ ( birkhuser , basel , 1991 ) ; + v.a . smirnov , mod . lett . a * 10 * ( 1995 ) 1485 ; + and references therein . baikov and d.j . broadhurst , _ 4th international workshop on software engineering and artificial intelligence for high energy and nuclear physics ( aihenp95 ) _ , pisa , italy , 3 - 8 april 1995 . published in pisa aihenp ( 1995 ) 167 . chetyrkin , a.h . hoang , j.h . khn , m. steinhauser , and t. teubner , phys . b * 384 * ( 1996 ) 233 . khn , t. hahn , and r. harlander , report nos . : ttp99 - 50,ka - tp-26 - 1999 , bnl - het-99/42 and hep - ph/9912262 . f. jegerlehner , proceedings of the _ ivth international symposium on radiative corrections ( radcor 98 ) : applications of quantum field theory to phenomenology _ , barcelona , catalonia , spain , sept . 8 - 12 , 1998 , report no . : hep - ph/9901386 . bai _ et al . _ [ bes collaboration ] , phys . * 84 * ( 2000 ) 594 ; hep - ex/0102003 . chetyrkin and m. steinhauser , phys . b * 502 * ( 2001 ) 104 . chetyrkin and m. steinhauser , eur . j. c * 21 * ( 2001 ) 319 . smirnov , phys . b * 465 * ( 1999 ) 226 . van oldenborgh and j.a . vermaseren , z. phys . c * 46 * ( 1990 ) 425 . g. weiglein , report no . : hep - ph/0109237 ; + and references therein . r. harlander , t. seidensticker , and m. steinhauser , phys . b * 426 * ( 1998 ) 125 . l. avdeev , j. fleischer , s. mikhailov , and o. tarasov , ; ( e ) .

the stability of the chromomagnetic savvidy vacuum in qcd under the influence of positive riemannian curvature is studied . the heat traces of the operators relevant to so(2 ) gauge - invariant yang - mills fields and faddeev - popov ghosts are calculated on product spaces of and . it is shown that the chromomagnetic vacuum with covariantly constant chromomagnetic field is stable in a certain set of radii and field strengths . this thesis would not have been possible without the help of my thesis committee , whose collective tolerance of my last - minute defense allowed me to complete this work . i would especially like to thank my adviser , ivan avramidi , who spent countless hours explaining and often re - explaining every detail of every answer to every question that i ever posed to him . i owe this thesis to his efforts .

0908.4456

quantum chromodynamics is the highly successful theory of the strong interaction of elementary particles . it is based on @xmath2 yang - mills theory , which is a non - abelian gauge theory invariant under the gauge group @xmath2 , and a set of spin-@xmath3 quarks , which form the fundamental representation of the gauge group . the @xmath2 degrees of freedom are referred to as `` color . '' the high - energy behavior of yang - mills theory is well - understood and leads to a renormalizable quantum field theory . gross , wilczek@xcite , and politzer @xcite discovered that at high energies ( or short distance scales ) , the interaction strength of yang - mills fields decreases to zero , a property known as asymptotic freedom . however , qcd is not well understood at low energies or large distance scales , i.e. , those comparable to @xmath4 ( @xmath5 cm ) , at which point perturbation theory breaks down . the interaction strength increases , and the energy required to separate quarks becomes infinite . this leads to the property of confinement , which is exhibited by the experimental absence of free quarks , but has not been demonstrated theoretically . at energy scales less than @xmath4 , qcd has to be replaced by an effective theory of composites of quarks in the form of color - neutral hadrons . the low - energy behavior of qcd can be examined through the effective potential . the effective potential is a function of the background yang - mills field which is minimized by the absolute lowest energy state of the physical system . this minimum defines the physical vacuum . the physical vacuum is considered trivial when the effective potential takes a minimum with a zero background yang - mills field . if the minimum of the effective potential occurs when the yang - mills field is non - zero , the vacuum state will consist of a non - zero background field . the first attempt to study the yang - mills vacuum in this manner was by savvidy@xcite in 1977 . savvidy introduced a background chromomagnetic field of constant field strength lying in the cartan algebra of the lie group @xmath6 . he found that the minimum of the one - loop effective action occurs at a non - zero value of the background field , which causes the vacuum to be infrared unstable . it was later pointed out by h.b . nielsen and p. olesen @xcite that the chromomagnetic vacuum discovered by savvidy has an energy density with an imaginary part , which implies that it has a tachyonic mode , leading to instability of the vacuum . further corrections@xcite were made to the chromomagnetic vacuum to show that the energy is lower when the chromomagnetic vacuum consists of tube - like domain structures , with the chromomagnetic field pointed along the axis of each tube . the finite width of the tubes serves as an infrared cutoff , which destroys the low - energy instability . the minimum energy density of this type of state has been found by @xcite to be a superposition of domains separated by a fixed distance . this model is known as the `` spaghetti vacuum . '' it is important to note that these calculations pertain to a chromomagnetic field in flat space . in this paper , we will consider the related problem of the stability of the yang - mills vacuum on a curved space . the addition of the curvature will make the operator corresponding to the second variation of the action positive definite for large enough values of curvature , which will change the tachyonic mode into a physical state and cause the vacuum to stabilize . in particular , we will consider non - zero covariantly constant @xmath7 yang - mills fields on product spaces of spheres , valued in the sub - algebra @xmath8 . by analyzing the spectrum of the second variation of the action and computing the effective action , it will be shown that the vacuum will stabilize on spaces that have sufficiently strong curvature . this paper is organized as follows . chapter two of this paper discusses the effective action approach to quantum field theory and how it applies to gauge theories , as well as the technique of zeta - function regularization . in order to compute the effective action , we need the spectrum of the yang - mills and faddeev - popov ghost operators on spheres , which are calculated in chapter three . in chapter four , we apply the results of chapter three to find the heat trace of yang - mills and ghost operators on @xmath0 and @xmath1 . chapter five is devoted to finding the total heat kernel on the products of spheres and determining in what cases the vacuum is stable . in this chapter , we will use the proper time method of schwinger and dewitt , and so the notation will follow that of dewitt@xcite . [ [ spacetime - geometry ] ] spacetime geometry + + + + + + + + + + + + + + + + + + the spacetime under consideration is an @xmath9-dimensional pseudo - riemannian manifold @xmath10 endowed with a globally hyperbolic metric @xmath11 with signature @xmath12 . we will assume that the spacetime manifold has a global time - like killing vector so that @xmath13 , where @xmath14 is an @xmath15-dimensional compact oriented spin manifold without boundary . local coordinates @xmath16 on @xmath10 are labeled by greek indices that run over @xmath17 . the coordinate basis @xmath18 for the tangent space @xmath19 at the point @xmath20 has dual basis @xmath21 in @xmath22 . the christoffel symbols can be found from the metric ^ _ = g^(_g _ + _ g _ - _ g _ ) . the curvature of the metric @xmath23 is described by the riemann curvature tensor r^ _ = _ ^ _ - _ ^ _ + ^_^ _ - ^_^ _ , and its contractions : the ricci tensor , r _ = r^ _ , and the ricci scalar , r = g^r_. [ [ orthonormal - frame ] ] orthonormal frame + + + + + + + + + + + + + + + + + an orthonormal frame @xmath24 can be constructed at every point on the manifold and is labeled by lower case greek indices in parentheses . the orthonormal frame of @xmath19 can be constructed as a set of vector fields over @xmath10 , where @xmath25 so that = _ ( ) ( ) , where @xmath26 . the orthonormal basis @xmath24 may be expanded in the coordinate basis @xmath18 , e _ ( ) = e_()^_. the inverse matrix @xmath27 of @xmath28 defines the dual basis e^ ( ) = e^()_dx^ in the cotangent space @xmath29 . then g^e^ ( ) _ e^ ( ) _ = ^ ( ) ( ) , g_e_()^e_()^= _ ( ) ( ) . yang - mills theory describes the dynamics of a vector bundle over @xmath10 . to say that the theory is gauge invariant is to impose the restriction that the action does not change under transformations by a gauge group . in particular , consider a compact simple lie group @xmath30 attached to every point of @xmath10 so that any neighborhood in a fiber bundle has the local structure @xmath31 . in other words let @xmath32 be coordinates on @xmath30 , so that for any coordinate patch on @xmath10 , a point in the fiber bundle is described by the set of coordinates @xmath33 . group indices are labeled by lower case latin letters , which run over @xmath34 . an element @xmath35 of a compact simple gauge group may be written in the form u = , where @xmath32 are parameters and @xmath36 are the generators of the lie group @xmath30 , which lie in the lie algebra . it is clear that the expression t_a = u|_k^a = 0 is an equivalent definition of the generators of @xmath30 . the generators of a simple compact lie algebra satisfy the relation = c^c_abt_c , where @xmath37 are the structure constants of the lie algebra @xmath30 . the adjoint representation of the lie algebra is defined by taking the generators to be ( t_a)^b_c = c^b_ac . to form inner products between algebra - valued tensors , we must introduce an inner product on the lie algebra . we define the cartan - killing metric e_ab = -c^c_adc^d_bc = -(t_a t_b ) to raise and lower group indices . in the case of compact simple lie groups , this metric can be normalized by e_ab = _ ad . to make the action invariant under gauge transformations , the covariant derivative @xmath38 of a field must satisfy the condition _^= u ( _ ^ ) . the gauge matrix @xmath39 is local , i.e. , depends on the coordinates , and primed quantities denote the transformed quantities . to do this , we let _ ^= ( ^lc _ + _ ) ^ , where @xmath40 is an algebra - valued vector field that transforms as _= u _ u^-1 -(_u ) u^-1 for any gauge transformation @xmath35 . in general , @xmath41 depends on the representation of the group @xmath30 . the strength @xmath42 of the yang - mills field @xmath41 is defined by _ = _ a_- _ _ + [ _ , _ ] . the strength of the field can be used to define the action functional s_ym = _ m dx(^ _ ) , where @xmath43 is a coupling constant and @xmath44 denotes the trace over the lie algebra . a scalar field @xmath45 is invariant under diffeomorphisms and has a covariant derivative of _ = ( _ + _ ) , with @xmath46 in an appropriate representation . the action for a scalar field must be constructed out of a scalar potential term @xmath47 and the quantity @xmath48 , where t denotes transpose and @xmath49 is the dalembert operator ^_. the effective action approach to quantum field theory is a highly useful approach that was developed by dewitt and others @xcite . this section follows the method as developed for boson fields . consider two causally connected in- and out- regions of spacetime that lie in the past and future of a region @xmath50 in which physical dynamics will take place . the goal of quantum field theory is to compute the amplitude @xmath51 for some initial state @xmath52 in the in- region to evolve into some final state @xmath53 in the out region . to calculate this , consider a change in the action @xmath54 . the schwinger variational principle states that the amplitude @xmath51 will change according to = . let @xmath55 be the boson fields relevant to the problem , where @xmath56 is taken to run over both continuous ( i.e. spacetime ) and discrete ( i.e. spinor , tensor , field ) indices . change the action by adding a linear interaction of @xmath55 with some classical sources @xmath57 that vanish in the in- and out- regions @xmath58 , where the contraction over @xmath56 is taken as both a summation over discrete indices and integration over spacetime ^i j_i = _ m dx _ ( a)j^(a ) , where @xmath59 . with this variation , the solution to the schwinger variational principle is expressed in terms of the feynman path integral [ pathint ] | in = \ { [ s()+j^k_k ] } , where @xmath60 represents the feynman measure . the generating functional for connected diagrams @xmath61 is defined in terms of the in - out transition amplitude by [ defw ] ( w(j ) ) . the first functional derivative of @xmath62 gives the background field @xmath63 ^i(j ) = w(j ) , the second functional derivative produces the propagator ^i_1 i_2(j ) = w(j ) , and the higher derivatives produce the many - point green functions ^i_1 i_k(j ) = w(j ) . in order to calculate vertex functions , we define the effective action @xmath64 by the functional legendre transform [ effac ] ( ) = w(j ( ) ) - j_i()^i , where the sources @xmath65 are expressed in terms of the background fields . the first functional derivative of @xmath66 is equal to the sources ( ) = -j_i ( ) , the second derivative defines the inverse propagator ( ) = d_ij ( ) , d_ij^jk = -_i^k(x , x ) , and the higher derivatives determine the vertex functions _ i_1 i_k ( ) ( ) . we see from ( [ pathint ] ) , ( [ defw ] ) , and ( [ effac ] ) that the effective action satisfies \ { ( ) } = \{}. in gauge field theories , the above formalism can not be applied immediately . instead , there will be problems arising from the fact that the measure will include an integral over non - physical fields . to be more explicit , consider an action functional @xmath67 that is invariant under some vector fields @xmath68 on the configuration space . transformations of the fields _ ^i = r^i_a^a , with @xmath69 being some parameters , that do not affect any real physics are called gauge transformations . in the case of gauge theories , the feynman path integral ( [ pathint ] ) will be carried over both physical and non - physical degrees of freedom . this adds divergences to the path integral that can only be removed by the dewitt - fadeev - popov method . we can separate the field variables @xmath45 into physical variables @xmath70 and gauge variables @xmath71 , so that the action @xmath72 does not depend on the group variables @xmath73 , that is , it is invariant under the variations with respect to @xmath71 , but not under the variations with respect to @xmath70 . to get rid of the excess degrees of freedom , we change the variables @xmath74 in the path integral and omit the integration over the group variables @xmath73 ( in other words , we divide out the volume of the gauge group ) , so that the integral is over the physical variables @xmath75 only . if the inverse change of variables is given by @xmath76 , then the jacobian of the change of variables is @xmath77 , where q^a_b ( ) = r^i_b ( ) is the faddeev - popov operator . the gauge condition is defined by a surface in the configuration space ^a()=^a where @xmath78 are some constants . this surface intersects all orbits of the gauge group transversally , and thus has a one - to - one correspondence with the set of all physical states . with these changes , we get the feynman measure i = q ( ) ( ^b ( ) - ^b ) , where @xmath79 is the functional delta function . the path integral becomes ( ( ) ) & = & ( ^b ( ) - ^b)q ( ) + & & \{}. in order to write the effective action in terms of gauge - invariant quantities , we integrate over @xmath78 with gaussian weight to get the expression & & ( ( ) ) = q()(h())^ + & & [ effacc_gauge ] \ { } , + where @xmath80 is some non - degenerate operator that does not depend on @xmath45 . the determinant @xmath81 is typically calculated in terms of faddeev - popov ghost fields , and @xmath82 is calculated in terms of the nielsen - kallosh ghost . in the case that @xmath83 does not depend on the background field @xmath84 , @xmath85 is simply an infinite constant that can be factored into the measure . in order to calculate the effective action , we introduce perturbation theory . in this section , we follow @xcite . perturbation theory is based on the idea that the largest contributions to the effective action come from fields @xmath55 close to the background field @xmath63 in the sense that they can be split into a background part and a quantum part [ phisplit ] ^i = ^i + h^i , with @xmath86 being the quantum fluctuations of the background field . with this change of variables , the effective action will have an expansion of the form [ gammaseries ] ( ) = s_ym + _ ( 1 ) + o(^2 ) , which is known as the loop expansion . splitting the fields as in ( [ phisplit ] ) , the path integral ( [ effacc_gauge ] ) may be expanded using ( [ gammaseries ] ) to give the first order correction [ gamma1gauge ] _ ( 1 ) = l_ym -h -il_fp , where [ lym_secvar ] ( l_ym)_ik = -s_,ik ( ) + _ a_,i()h^ab_b_,k ( ) , and l_fp = q ( ) . the quantity @xmath87 is the one - loop effective action . [ [ wick - rotation ] ] wick rotation + + + + + + + + + + + + + the determinants of differential operators @xmath88 are problematic for two reasons . one reason is that @xmath88 is divergent and must be regularized in order to make sense . this will be dealt with in the next section . before that can be accomplished we must first deal with the fact that @xmath89 is , in general , not a positive definite operator . to get rid of this latter problem , we wick rotate time in the complex plane t = -i . under this transformation , the metric is changed to have the riemannian signature @xmath90 , so the operator @xmath91 becomes an elliptic operator rather than a hyperbolic one . in addition , the measure picks up an additional factor of @xmath92 : dtd^n-1x -idd^n-1x , which in turn causes the action to pick up the same factor . for the remainder of the paper , all quantities will be assumed to be wick rotated . to describe finite temperature effects , the euclidean time is compactified to @xmath93 , with radius given by the inverse temperature @xmath94 . the quantity @xmath87 in ( [ gamma1gauge ] ) is the functional determinant of an elliptic differential operator . this quantity will always be infinite and therefore must be regularized in order for any physical calculation to make sense . following @xcite , this section will show that using the heat kernel representation , the effective action can be expressed in terms of a zeta function . then by analytic continuation , the zeta function can be regularized and made to yield finite physical results . [ [ green - functions ] ] green functions + + + + + + + + + + + + + + + for a bosonic field , the second functional derivative of the action may be brought by choice of gauge to the form [ minimal ] l + m^2= -+ q + m^2 , where @xmath95 is a matrix - valued function acting on the fields @xmath55 , @xmath96 is the mass ( which may be zero ) , and @xmath97 is the identity matrix . green functions are solutions @xmath98 of the equation ( l + m^2)(x , x ) = ( x , x ) , with ( x , x ) = g^-1/2(x)(x - x ) . they can be constructed in terms of a contour integral of the heat kernel @xmath99 [ greenheat ] ( x , x ) = _ 0^t(-tm^2)u(t|x , x ) . the heat kernel satisfies the heat equation [ heateq ] ( + l)u(t ) = 0 with the boundary condition [ heatbound ] .u(t|x , x)|_t=0 = ( x , x ) . [ [ from - green - functions - to - effective - action ] ] from green functions to effective action + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + the heat equation ( [ heateq ] ) has the formal solution u(t ) = ( -tl ) . in terms of the eigenfunctions @xmath100 corresponding to eigenvalues @xmath101 of the operator @xmath89 , i.e. l_n = _ n_n , the heat kernel can be written u(t|x , x ) = ^_n=1_n(x)^_n(x)e^-t_n . the heat kernel diagonal is defined by taking the coincidence limit of this expression u(t|x , x ) = ^_n=1_n(x)_n^(x)e^-t_n , and the functional @xmath102 heat trace is the trace of the diagonal over all indices ( -tl ) & = & _ m dx u(t|x , x ) + [ heattrace ] & = & ^_n=1e^-t_n , in the case of yang - mills theory , the trace @xmath44 is over both group indices and tangent space indices . if the eigenvalues are degenerate , we may express the heat kernel in terms of the eigenvalues @xmath103and degeneracies @xmath104 [ heat_trace ] ( -tl ) = ^_n=1d_n e^-t_n if the mass is sufficiently large so that @xmath105 , the quantity @xmath106 can similarly be expressed = ^_n=0(_n+m^2 ) . we can use the identity = -^_0e^-t + c with @xmath107 an infinite constant , and the expression for the heat trace ( [ heattrace ] ) to find [ lndet ] ( l + m^2 ) = -_0^(-tm^2)_m dx u(t|x , x ) + . the infinite constant has no effect on the dynamics , and can be dropped . the one - loop effective action is expressed completely in terms of the logarithms of determinants of operators , so calculation of the heat kernel @xmath108 for various operators gives all of the information needed to calculate @xmath87 . this reduces the task of calculating the one - loop effective action to that of finding the eigenvalues of the second variation of the action . [ [ zeta - function - regularization ] ] zeta - function regularization + + + + + + + + + + + + + + + + + + + + + + + + + + + + the quantity @xmath109 ( [ lndet ] ) is infinite . in order to make it finite , it must be regularized in terms of the @xmath110function [ lndetzeta ] ( l + m^2 ) = -(0 ) , (0 ) = ( p)|_p=0 . the @xmath110function of a differential operator @xmath10 is defined in terms of the heat kernel by _ m(p ) = ^2pm^-p = _ 0^dt t^p-1(-tm ) , where @xmath111 is a renormalization parameter with dimension of inverse length . the @xmath110function is analytic at @xmath112 , so the expression ( [ lndetzeta ] ) is finite and well - defined . with this method of regularization in mind , we may return to yang - mills theory . the one - loop effective action for yang - mills theory ( [ gamma1gauge ] ) in a general covariant gauge is the sum of contributions from yang - mills fields and ghosts . in terms of functional determinants in euclidean space , [ gamma1ym ] _ ( 1 ) = l_ym ( ) - 2l_fp ( ) , where @xmath113 is a gauge - fixing parameter , and @xmath114 is the operator acting on gauge fields as found in ( [ lym_secvar ] ) l_ym ( ) = l_ym + h , ( l_ym)^ = -^ - 2^_^+ r^_^[lym ] ( h)^ = ^_^ , @xmath115 is the faddeev - popov ghost operator acting on anti - commuting scalar fields l_fp ( ) = l_fp , [ lfp ] l_fp= - . we can regularize the gauge - fixed @xmath87 by expressing it in terms of the @xmath110function [ gamma1zeta ] _ ( 1 ) = -_tot(0 ) , where [ zeta_total ] _ tot(p ) = _ l_ym(p ) -2_l_fp(p ) is the total @xmath110function . the zeta - function can be analytically continued to give a renormalized expression for the effective action . the factor @xmath116 guarantees gauge independence of the regularized effective action on the mass shell@xcite . it can also be proven @xcite that @xmath117 is independent of gauge , and so we may choose @xmath118 so that we are left with minimal differential operators as in ( [ minimal ] ) . it was shown by savvidy that the one - loop effective action for @xmath6 yang - mills in flat space takes a minimum for a non - zero field . consider a covariantly constant yang - mills background in flat space _ _ = 0 . one flat space solution to this equation is ^a_= -f_x^n^a , so that @xmath119 takes the form @xmath120 , where @xmath121 is a unit vector in the cartan subalgebra of the lie algebra of @xmath30 , @xmath122 . to make this a `` magnetic '' background , conditions on group invariants are imposed f_f^ = ( h^2 - e^2 ) > 0 , ^f_f _ = he = 0 . expanding the one - loop effective action in terms of momenta for a magnetic - type field and taking the first term@xcite , the one - loop correction is @xmath123 _(1)h = _ 0^ ( + 2hs(hs ) ) . renormalization of this expression gives _(1)h = - to which the corresponding energy density is = + . it is easily seen @xcite that the energy density has a minimum at h_min = ^2(- ) . it was later pointed out by nielsen and olesen @xcite that the energy density for this model has an imaginary part , which leads to instability of this model . in the spirit of savvidy @xcite , we consider a covariantly constant field strength tensor @xmath124 that takes values in the center of the gauge lie algebra . the condition that the field is constant gives rise to the equation _ - [ _ , _ ] _ = 0 . it can be shown@xcite that this yields the integrability condition [ integrable ] [ _ , _ ] + r _ - r _ = 0 . by taking @xmath124 to be in the center of the cartan algebra , @xmath125 = 0 $ ] , writing the riemann tensor of the @xmath126sphere as r^ _ = ( ^_^ _ -^_^ _ ) for @xmath127 , and contracting over @xmath111 and @xmath128 in ( [ integrable ] ) , we find ( n-2)f _ = 0 . therefore , non - zero covariantly constant magnetic fields in the center of the algebra of the lie group can only exist on @xmath0 or @xmath129 . each of the manifolds that we consider has the product manifold structure @xmath130 , where @xmath131 are submanifolds of @xmath10 . in this case , we have the decomposition of the operator @xmath89 l = l_1 _ 2 _ n-1 l_n , where @xmath132 is the projection of @xmath89 onto the submanifold @xmath131 . in this case , the heat kernel has the form ( -tl ) = ( -tl_1 ) ( -tl_n ) . to calculate the heat kernel on a general product manifold , we only need to calculate the heat kernel on each submanifold and multiply the results e^-tl = _ m_1(-tl_1 ) _ m_n(-tl_n ) , where @xmath133 denotes the functional trace , which is also taken to be the trace over group and coordinate indices . on @xmath93 , a circle of radius @xmath134 , the laplacian acting on any function is simply the operator l = -= -^2 _ , where @xmath135 is the coordinate along the circle , @xmath136 . the eigenvalues for this operator are _ n = , n = 0 , 1 , the multiplicities are & & d_0 = 1 + & & d_n = 2 , n = 1,2 , + the heat kernel trace for a function on @xmath93 can then be calculated using the formula for the heat kernel ( [ heat_trace ] ) ( -tl ) = 1 + 2_n=1^e^-tn^2/r^2 . we define this to be the function s ( ) = 1 + 2_n=1^e^-tn^2/r^2 . there is no difference between a scalar and a @xmath137-form on @xmath93 , so this heat trace applies to all geometric objects on @xmath93 . the heat trace for @xmath0 is non - trivial and its calculation is a significantly more complicated problem . on @xmath0 , scalars and one - forms will be distinct objects and will form different representations of both the gauge group and the isotropy group , which will determine the covariant derivative and thus the form of the laplacian . in addition , the existence of non - zero chromomagnetic fields will cause the the eigenvalues to split , leading to a much more complicated spectrum . consider the 2-sphere @xmath0 of radius @xmath138 endowed with the standard riemannian metric ds^2 = e^()_e^()_dx^dx^ = r^2(d^2 + ^2d^2 ) , where @xmath139 , and @xmath136 . greek letters without parentheses denote the coordinate indices , which range over the two values @xmath140 and @xmath135 . the orthonormal basis one - forms @xmath141 are given by e^(1 ) = r d , e^(2 ) = r d. greek indices with parentheses range over @xmath142 and denote indices of the orthonormal basis . the volume form is given by dvol = r^2dd . the components of the spin connection 1-form can be found by the cartan method to be _ ( ) ( ) = - _ ( ) ( ) d . the curvature tensor components are r^ ( ) ( ) _ ( ) ( ) = ( ^ ( ) _ ( ) ^ ( ) _ ( ) - ^ ( ) _ ( ) ^ ( ) _ ( ) ) . the sphere @xmath0 is diffeomorphic to the quotient space @xmath143 . here , @xmath7 is the isometry group of @xmath0 and @xmath8 is the isotropy group of @xmath0 . the rotation group @xmath7 maps @xmath0 to itself and @xmath8 rotations centered around a point will leave the point unmoved . let @xmath144 be a field which transforms under a representation of the group @xmath8 . let @xmath145 be the generators of the group @xmath8 in the representation acting on @xmath45 . because @xmath8 is a one - dimensional , abelian group , there is only one generator @xmath146 . then the covariant derivative of @xmath45 is _ = ( _ + _ ) , where _ 1 = 0_2 = - . now assume that the field @xmath45 also transforms under another representation @xmath147 of the group @xmath8 , which we call a gauge representation . then @xmath45 transforms under the product of two representations of the group @xmath8 . let @xmath46 be the corresponding gauge connection and @xmath148 be the curvature of this connection . the gauge curvature @xmath149 is a 2-form on a 2-dimensional space , and so @xmath149 must be proportional to the volume form @xmath150 = dd , where , in general , @xmath83 is some function of the coordinates . by expressing @xmath149 in components [ f_coords ] ^_= e^ _ , where @xmath151 is the invariant volume form , we see _ _ = t e__h . requiring the curvature to be covariantly constant yields the condition that @xmath83 is a constant . physically , @xmath83 can be interpreted as the charge of a monopole at the center of the sphere . because @xmath149 is precisely the chern form of a line bundle over @xmath0 @xcite , we have _ s^2= 2nt , n . therefore , the monopole charge @xmath83 can only take integer values h = n , n . the corresponding gauge connection is found by solving the equation = d , which results in = dwe see that @xmath46 is proportional to the spin connection @xmath152 . the covariant derivative invariant under both space rotations and gauge transformations is then _ = ( _ + _ ) , where [ generator ] = + , and @xmath147 is the generator of the gauge group @xmath8 . in order to calculate the heat traces ( [ heat_trace ] ) on @xmath0 , we need to analyze the spectrum of the operator @xmath153 . we know from ( [ lym ] ) that this operator is equal to the negative laplacian plus yang - mills strength and ricci curvature terms . the field strength and curvature tensors are covariantly constant . therefore , the eigenfunctions of @xmath153 are proportional to the eigenfunctions of the laplacian . this causes the eigenvalues of @xmath153 to be the eigenvalues of the laplacian , shifted by the eigenvalues of the sum of the other two operators . the laplacian acting on general spin - tensor is given by the expression [ laplacian ] = g^ _ _ = |g|^-1/2(_+ _ ) |g|^1/2g^(_+ _ ) . in the case of @xmath0 , the laplacian becomes = it can be noted that this will yield the standard laplacian for a particle in a magnetic field @xcite in the limit @xmath154 . using polar coordinates near @xmath155 with @xmath156 , along with using the choice of gauge @xmath157 and denoting the generator of @xmath8 by @xmath56 , the connection becomes _ ( 1 ) = 0,_(2 ) = ( - 1 ) , where @xmath10 is the magnetic field @xmath158 . taking the limit @xmath154 then gives the standard laplacian on @xmath159 [ l_low ] = _ ^2 + _ + _ ^2 + _ - ^2 . for 1-forms , the generator @xmath160 is the matrix with components ^ ( ) a _ ( ) b = ^()_()^a_b + ^a_b^ ( ) _ ( ) . the eigenvalues of the matrix @xmath160 are @xmath161 k_1 = 1 + , k_2 = 1 - , k_3 = -1 + , k_4 = -1 - . in the same basis , the field strength tensor can be diagonalized with corresponding eigenvalues f_1 = -,f_2 = , f_3 = , f_4 = - , and the ricci tensor will be proportional to the identity , with all eigenvalues @xmath162 given by r_i = diagonalizing the matrix @xmath160 will cause the yang - mills operator @xmath153 to break into four separate operators of the form l^(i ) = --2 + , where @xmath163 is a half - integer that takes one of the four values @xmath164 . the values @xmath165 are the corresponding eigenvalues of the matrix @xmath149 and are given by the values f_1 = -,f_2 = , f_3 = , f_4 = - . it is clear that the spectrum of the laplacian will be invariant under change of sign of the field @xmath83 . for the remainder of the paper , we will then assume without loss of generality that @xmath83 is positive . the spectrum of the operator @xmath166 is defined by regular normalized solutions of the equation l^(i)u = u , where @xmath167 is a complex spectral parameter . this equation has regular solutions only for certain real discrete values of @xmath113 , which determine the spectrum of @xmath166 . the operators @xmath149 and @xmath138 have no dependence on the coordinates , and so the eigenfunctions of the operator @xmath166 are the same as for the laplacian . the eigenvalues of @xmath166 are obtained from the eigenvalues of the laplacian @xmath168 by shifting [ shift ] ( l^(i ) ) = ( -^(i ) ) + - , so we find the spectrum of the laplacian first . separating variables with the substitution u = e^i m h_m(),m , we obtain an ordinary differential equation for @xmath169 \{_^2 + _ - ( m - k_i)^2 + r^2}h_m ( ) = 0 . let us introduce the notation a^_ml = + || + || ( 1 + 4r^2)^1/2 . the index @xmath170 labels the eigenvalues , as will be described below . as explained in appendix a , this equation has regular normalized solutions given by h^l_mk ( ) & = & ( 1-)^|| ( 1+)^|| + & & f(a^+_ml , a^-_ml ; 1+|m - k| ; ) , + & & where @xmath171 is the hypergeometric function . in the case of integer @xmath163 , these solutions exist for the following values of @xmath172 and @xmath96 [ lambda_int ] _ l = ( |k| + l)(|k|+l+1 ) , -l m l , where @xmath170 is an integer greater than or equal to @xmath173 : l 0 . by counting all possible values of @xmath96 we obtain the multiplicities of the eigenvalues @xmath172 for integer @xmath163 [ deg_int ] d_l = 2(l + |k|)+1 . in the case of half - integer @xmath163 , there are two series of solutions . the first series is given by the following values of @xmath172 and @xmath96 : & & [ lambda_half1 ] _ l = ( |k| + l)(|k|+l+1 ) , + & & -|k| + m |k|- , where @xmath170 is an integer greater than or equal to @xmath173 l 0 , giving degeneracies [ deg_half1 ] d_l = 2|k| . the second series is given by & & [ lambda_half2 ] _ n = ( |k| + + n)(|k|++n ) , + & & -(|k| + + n ) m -(|k| + ) + & & ( |k| + ) m ( |k| + + n ) , which gives the degeneracies [ deg_half2 ] d_n = 2n+2,n 0 . therefore , the eigenvalues of the operator @xmath174 are @xmath172 and the corresponding eigenfunctions are u^l_m ( , ) = e^im h^l_m ( ) . it should be noted here that the eigenvalues of the laplacian here are very different from those of the laplacian for flat space ( [ l_low ] ) . in the flat space , the eigenvalues are that of a harmonic oscillator@xcite = hr^-2(n + ) , whereas in our case , the eigenvalues are quadratic in @xmath175 , and so will increase quadratically in @xmath83 as @xmath83 becomes large . the spectrum of the operator @xmath153 is then obtained by the shift of the laplacian s eigenvalues ( [ shift ] ) . the heat kernel of the operator @xmath153 acting on one - forms is found using the eigenvalues of the laplacian , which are the values of @xmath113 in ( [ lambda_int])- ( [ lambda_half2 ] ) plus the value @xmath176 . the eigenvalues for even magnetic field are given by _ il = ( |k_i|+l)(|k_i|+l+1 ) + - , where @xmath56 runs over the tangent space and group indices @xmath177 , @xmath178 , and the degeneracies are given by d_il = 2(l+|k_i| ) + 1 . the eigenvalues for odd magnetic field are in two series for each value of @xmath56 . the first series is given by _ il = ( |k_i| + l)(|k_i| + l + 1 ) + - with @xmath179 , @xmath178 , and degeneracies given by d_il = 2|k_i| . the second series is given by _ in = ( |k_i| + + n)(|k_i| + + n ) , with @xmath179 , @xmath180 , and degeneracies d_in = 2n+2 . in addition to the yang - mills field , we must consider the scalar faddeev - popov ghost field to compute the effective action . in this case , the relevant operator is @xmath181 . scalar fields are invariant under coordinate transformations , so the generator @xmath160 as given in ( [ generator ] ) will only transform under the gauge group so(2 ) , which means that the total generator for the ghosts will be x = ^a_b , which has the two eigenvalues @xmath182 with [ ghost_k ] _ 1 = , _ 2 = - . the values that the parameter @xmath113 takes on will be exactly the same as in ( [ lambda_int ] ) , ( [ lambda_half1 ] ) , and ( [ lambda_half2 ] ) , except with these values of @xmath183 to replace the values of @xmath163 . the eigenvalues are again characteristically different for even and odd magnetic field . for even values of @xmath83 , the eigenvalues of @xmath115 are given by _ l = ( l + ) ( l + + 1 ) the eigenvalues do not depend on the sign of @xmath183 , so the degeneracies are just doubled to account for the two values : d_l = 4(l + ) + 2 . for odd values of @xmath83 , the eigenvalues are in two series . the first series is given by _ l = ( l + ) ( l + + 1 ) , with the degeneracies doubled to account for positive and negative @xmath184 : d_l = 2h . the second series is given by _ n = ( n + + ) ( n + + ) , with degeneracies d_n = 4n + 4 . by using these values , we will be able to calculate the heat kernel in the next chapter . the heat trace of the total yang - mills and ghost operators can be computed on products of spheres using the factorization property of the heat kernel . in this chapter , we calculate the heat traces of each operator on the spaces @xmath185 and @xmath0 , from which we will be able to compute the total heat trace for the product manifolds in the next chapter . the heat kernel can readily be calculated from the eigenvalues and degeneracies that have been found , as well as using the defining formula for the heat trace ( [ heattrace ] ) , written in terms of eigenvalues @xmath186 and their degeneracies , [ sum ] ( e^-tl ) = _ l d_l e^-t_l(l ) . the simplest space to consider is the two - torus , @xmath187 , with each copy of @xmath93 having a different radius @xmath188 . the two - torus has no curvature and can support no covariantly constant chromomagnetic field due to topological constraints . thus , the operator @xmath153 is just the laplacian l_ym = - . it is then straightforward to find that the heat trace is [ t2_ym ] ( -tl_ym ) = 4s()s ( ) . the operator for ghosts in this case is simply the scalar laplacian @xmath189 , [ t2_fp ] ( -tl_fp ) = 2s()s ( ) , with the factor of @xmath190 coming from the trace over group indices . the heat trace for yang - mills theory on @xmath159 has been found @xcite to be [ r2_ym ] e^-tl_ym = _ ^2 dx ( 4t)^-1 and the heat trace for the corresponding ghost operator is [ r2_fp ] e^-tl_fp = _ ^2 dx ( 4t)^-1 . the heat kernel for the yang - mills field can be found by performing the spectral sum ( [ sum ] ) , using the eigenvalues and degeneracies for the operator @xmath153 . for the case of even magnetic charge @xmath83 , the heat kernel for the operator @xmath153 acting on one - forms is tr(e^-tl_ym ) & = & _ j=1 ^ 4 _ l=0^(2l + 2|k_j| + 1 ) + & & \{-}. for odd @xmath83 , & & tr(e^-tl_ym ) = _ j=1 ^ 4 + & & these sums can be expressed in terms of the functions _ j(t ) = ^_l=1 l^j e^-tl(l+1 ) , _ j(t ) = ^_l=1 l^j e^-tl^2 . these functions are regular in the limit @xmath191 . + for @xmath192 , the heat trace is given by [ s2_h0 ] ( e^-tl_ym ) = 4e^-t / r^2 . for @xmath193 , ( e^-tl_ym ) & = & ( 4e^-2t / r^2 + 4)_1 ( ) -4e^-2t / r^2_0 ( ) + & & + ( 6e^-9t/4r^2 + 2e^-t/4r^2)_0 ( ) -6e^-13t/4r^2 for @xmath194 , [ s2_h2 ] ( e^-tl_ym ) = 2(e^t / r^2 + e^-3t / r^2 ) + 2e^t / r^2 - 6e^-5t / r^2 . for @xmath195 , ( e^-tl_ym ) & = & ( 2e^7t/4r^2 + 10e^-19t/4r^2)_0 ( ) + ( 4e^2t / r^2 + 4e^-4t / r^2)_1 ( ) + & & -10e^-23t/4r^2 - 4e^-6t / r^2 - 10e^-35t/4r^2 - 4e^-10t / r^2 . [ s2_h3 ] for @xmath196 , ( e^-tl_ym ) = 2(e^3t / r^2 + e^-5t / r^2 ) -6e^-7t / r^2 - 10e^-11t / r^2 . for @xmath197 with @xmath198 odd , & & ( e^-tl_ym ) = 8(th / r^2)e^-t / r^2_1 ( ) + & & + [ ( 6 - 2h)e^t(h-1)/r^2 - 2(h+1)e^-t(h+1)/r^2]_0 ( ) + & & + [ ( 2h+4)e^t(h-3/4)/r^2 + ( 2h+4)e^-t(h+3/4 ) - 8e^t(h-5/4)/r^2]_0 ( ) + & & -2e^-t(h+3/4)/r^2(h+2)_l=1^+e^-tl^2/r^2 + & & -2e^t(h-3/4)/r^2(h-2)_l=1 ^ -e^-tl^2/r^2 + & & -2e^-t(h+1)/r^2_l=1^+(2l - h-1)e^-tl(l+1)/r^2 + & & -2e^t(h-1)/r^2_l=1 ^ -(2l+3-h)e^-tl(l+1)/r^2 [ s2_h5 ] for @xmath199 with @xmath198 even , ( e^-tl_ym ) & = & 4(th / r^2)e^-t / r^2 + & & -2e^-t(h+1)/r^2_l=1^(2l+1)e^-tl(l+1)/r^2 + & & -2e^t(h-1)/r^2_l=1 ^ -2(2l+1)e^-tl(l+1)/r^2 . [ s2_h6 ] in addition to the yang - mills field itself , there are ghost fields to eliminate the extra degrees of freedom caused by gauge invariance . the ghost fields on @xmath0 are scalar fields , which means that the eigenvalues and degeneracies are given by ( [ lambda_int])-([deg_half2 ] ) with the values @xmath200 . with these values , it is straightforward to calculate the heat trace . + for h = 0 , [ s2_h0 g ] ( e^-tl_fp ) = 4_1 ( ) + 2_0 ( ) + 2 . for @xmath193 , [ s2_h1 g ] ( e^-tl_fp ) = 2e^-t/4r^2_0 ( ) + 4_1 ( ) . for @xmath194 , [ s2_h2 g ] ( e^-tl_fp ) = 4_1 ( ) + 2_0 ( ) . for @xmath83 odd , with @xmath201 , ( e^-tl_fp ) & = & 2he^t/4r^2_0 ( ) + 4_1 ( ) + 2(1-h)_0 ( ) + & & -2he^t/4r^2_l=1 ^ -e^-tl^2/r^2 -_l=1 ^ -(4l+2 - 2h)e^-tl(l+1)/r^2 + & & [ s2_h3 g ] for even @xmath83 , @xmath202 , we have [ s2_h4 g ] e^t_0 = 4_1 ( ) + 2_0 ( ) -2_l=1 ^ -1(2l+1)e^-l(l+1 ) . with the heat trace of both the yang - mills and ghost operators calculated on all relevant submanifolds , it is possible to assemble the total heat trace the gauge - fixed yang - mills field on four - dimensional manifolds by using the factorization property of the heat kernel and calculating the total heat trace @xmath203 . we can characterize the stability of the yang - mills vacuum by examining the large @xmath204 behavior if this function . a negative eigenvalue corresponds to an unstable mode , which would indicate that the vacuum is unstable . however , we find that with a sufficiently strong positive curvature on @xmath0 , we can make all eigenvalues positive , and the vacuum becomes stable . this case is the analog of the problem studied by savvidy . spacetime has zero curvature and a constant chromomagnetic field exists . the total heat kernel is given by u_tot(t ) & = & ( -tl_ym)_^2(-tl_ym)_s^1 s^1 + & & -2(-tl_fp)_^2(-tl_fp)_s^1 s^1 . + & & this is evaluated using the heat traces ( [ r2_ym ] ) and ( [ r2_fp ] ) : u_tot(t ) = _ ^2 dx ( 4t)^-1 s()s ( ) , where @xmath206 and @xmath207 are the radii of the two copies of @xmath93 . superficially , it would seem that there are two different configurations of chromomagnetic field that can exist on the manifold @xmath208 either with the chromomagnetic field polarized along the torus @xmath1 or along the sphere @xmath0 . the first case can not be realized because a covariantly constant chromomagnetic field can not exist on @xmath1 . however , we can consider the related problem of having the non - zero field polarized along @xmath159 on the manifold @xmath209 . this case is not physical because it leaves the time direction to be incorporated into @xmath0 , which means that spacetime can no longer be deforemed to have the structure @xmath210 . this case has been investigated by elizalde , et . al . @xcite . in the limit that the curvature is small , it has been determined that the vacuum stabilizes for some radius of @xmath0 . we can also analyze this problem from our standpoint . in the case of a chromomagnetic field directed along @xmath159 , the operator @xmath153 will have the block diagonal form l_ym = ( -_1(1 ) - _ 1(2 ) ) ( cc & 0 + 0 & ) + ( cc 0 & 0 + 0 & _ 2 ) - 2 ( cc _ 1 & 0 + 0 & 0 ) , where @xmath211 is the laplacian acting on one - forms on @xmath159 , @xmath212 is the laplacian acting on one - forms on @xmath0 , @xmath213 is the ricci tensor on @xmath0 , and @xmath214 is the chromomagnetic field restricted to @xmath159 . the operator @xmath115 is simply the laplacian acting on scalars l_fp = -_0(1 ) - _ 0(2 ) , where @xmath215 is the laplacian acting on scalars on @xmath159 and @xmath212 is the laplacian acting on scalars on @xmath0 . the total heat kernel @xmath216 is then u_tot(t ) & = & ( -tl_ym)_s^2(-tl_ym)_^2 + & & -2(-tl_fp)_s^2(-tl_fp)_^2 using the heat trace expressions ( [ r2_ym ] ) , ( [ r2_fp ] ) , ( [ s2_h0 ] ) , ( [ s2_h0 g ] ) , this gives the result u_tot(t ) & = & _ ^2dx ( 4t)^-1 \ { + & & ( \{4e^-t / r^2 -4 } + 8e^-t / r^2 - 4 ) + & & -4 } another allowable configuration is to let the chromomagnetic field lie along @xmath0 . in this case , the total heat kernel is given by & & u_tot(t ) = ( -tl_ym)_s^2(-tl_ym)_s^1 s^1 + & & -2(-tl_fp)_s^2(-tl_fp)_s^1 s^1 using the @xmath185 heat trace expressions ( [ t2_ym]),([t2_fp ] ) with h = 0 , and the @xmath0 heat trace expressions ( [ s2_h0])-([s2_h6 ] ) calculated in chapter 4 , this gives us the following results : + for @xmath217 , u_tot(t ) & = & 8s()s ( ) + & & \{-1}. for @xmath218 , u_tot(t ) & = & 8s()s ( ) + & & \{3e^-(9/4)t / r^2_0 ( ) + ( 4(t / r^2)e^-t / r^2 -2)_1 ( ) + & & -2e^-2t / r^2_0 ( ) -6e^-9t/4r^2 } . + & & for @xmath219 , u_tot(t ) & = & 4s()s ( ) + & & \ { ( 4e^-t / r^2(2t / r^2 ) - 4 ) + & & + 2e^t / r^2 - 6e^-3t / r^2 } . for @xmath220 , & & u_tot(t ) = 4s()s ( ) + & & for @xmath221 , u_tot(t ) & = & 4s()s ( ) + & & \ { + & & -6e^-7t / r^2 + 10e^-11t / r^2 + 6e^-2t / r^2 } . for @xmath83 odd , @xmath222 , u_tot(t ) & = & 4s()s ( ) + & & \ { ( 8(th / r^2)e^-t / r^2 - 4)_1 ( ) + & & + _ 0 ( ) + & & e^t/4r^2_0 ( ) + & & -2e^-t(h+3/4)/r^2(h+2)_l=1^+e^-tl^2/r^2 + & & -2e^t(h-3/4)/r^2(h-2 ) _ l=1 ^ -e^-tl^2/r^2 + & & -2e^-t(h+1)/r^2_l=1^+(2l+h-1)e^-tl(l+1)/r^2 + & & -2e^t(h-1)/r^2 _ l=1 ^ -(2l+3-h)e^-tl(l+1)/r^2 + & & + 2he^t/4r^2_l=1 ^ -e^-tl^2/r^2 + _ l=1 ^ -(4l+2 - 2h)e^-tl(l+1)/r^2 } for @xmath83 even , @xmath223 , u_tot(t ) & = & 4s()s ( ) + & & \ { ( 4e^-t / r^2(th / r^2 ) - 2 ) + & & -2e^t(h-1)/r^2_l=1 ^ -2(2l+1)e^-tl(l+1 ) -2e^-t(h+1)/r^2_l=1^(2l+1)e^-tl(l+1 ) + & & + 2_l=1 ^ -1(2l+1)e^-tl(l+1 ) } when the heat traces above contain an exponential that grows or stays constant with @xmath204 , then the yang - mills vacuum will be unstable , causing the configuration with constant chromomagnetic field to decay into another state . however , if all exponentials are decreasing , then the vacuum will be stable . if the chromomagnetic field is polarized along @xmath0 on the manifold , the stability of the the constant chromomagnetic state will depend on the strength @xmath83 of the chromomagnetic field and the radius @xmath138 of the sphere @xmath0 . the case @xmath218 will always be stable , but @xmath219 , and @xmath220 will not be stable for any radii . in the case of @xmath218 , the lowest eigenvalue is given by _ min = ( ) , which implies that the vacuum stabilizes for @xmath218 . similarly , for @xmath219 , the minimum eigenvalue is _ min = - , and the vacuum is unstable . for @xmath220 , the minimum eigenvalue is _ min = - and the vacuum is unstable . for @xmath224 , the lowest mode will correspond to the eigenvalue _ the vacuum will be stable when this eigenvalue is positive , which occurs when the condition - + 10 is satisfied . this occurs for h 6 . thus , the vacuum is unstable for @xmath225 and stable for @xmath226 and @xmath227 . because @xmath83 is the dimensionless parameter relating to the magnetic field @xmath10 , @xmath158 , this implies that we can make a configuration stable by increasing either the magnetic field or the radius . a large radius would intuitively return us to the saviddy flat - space case , but we instead see that it actually increases the lowest eigenvalue . the local behavior of these cases is the same , so this must be a topological phenomenon . if we consider the radii of the spheres to be variable , then the lowest energy state is state that minimizes the effective action , which will be a function of both the chromomagnetic field and the radius of the spheres . the one - loop effective action is written in terms of the heat kernel as _ ( 1 ) = -_p=0 the effective action , then , to first order is = s + _ ( 1 ) where @xmath228 is the classical action s = -_m dx , which in our case can be integrated over the manifold to give s = ( m ) . calculating the effective action and finding a global minimum for all @xmath138 and @xmath83 will reveal the vacuum with minimum energy . we have calculated the heat traces for pure yang - mills on products of spheres with a covariantly constant chromomagnetic field , and have shown that for a space with certain values of curvature and magnetic field , the covariantly constant chromomagnetic vacuum forms a local minimum of the effective action . this lends creedence to the possibility that the savvidy - type vacuum , with a covariantly constant magnetic field will form an absolute minimum on the relevant spaces . contrary to expectations , the limit in which the radius of the sphere becomes infinite does not yield the standard flat - space results for eigenvalues of the laplacian . instead of having eigenvalues that are linear in the magnetic field , our results show that on the two - sphere , the lowest eigenvalue of the laplacian will increase quadratically with the magnetic field and quadratically with the radius of the sphere , leading to the counter - intuitive result that the space most closely approximating the flat - space saviddy vacuum will have a minimum eigenvalue that will be farthest from being unstable . this effect is a topological phenomenon that requires further study . the next step in examining this model should be to calculate the full effective potential as a function of both the strength of the chromomagnetic field and the radius of the sphere . the absolute minimum of the effective potential would yield the absolute vacuum state of yang - mills . in our case , we would find a state that would be at least a local minimum of the vacuum , and possibly the absolute minimum . in this section , we find the eigenfunctions and the corresponding eigenvalues @xmath172 of the operator @xmath189 . they are given by regular solutions of the equation \{-- _ l } u ( ) = 0 . the label @xmath170 on @xmath172 is only a label here . it s allowed values will be found later . introducing the change of variables @xmath229 , the equation becomes u(x ) = 0 . we may make the substitution u(x ) = ( 1-x)^(1+x)^f(x ) , where = || , = || . to get the equation & & \{(1-x^2 ) - [ ( 2- 2 ) + ( -2 - 2 - 2)x ] + & & + } f(x ) = 0 . by switching variables to @xmath230 , we obtain the hypergeometric equation & & \{z(1-z ) + [ ( 1 + 2 ) - 2(1 + + ) z ] + & & + [ ---(+ ) ^2 + r^2 ] } f(z ) = 0 . finally , introducing the notation a^= + + ( 1 + 4r^2)^1/2 , we obtain the solution f(z ) = f(a^+,a^- ; 1+|m - k| ; z ) , where the function @xmath231 is the hypergeometric function @xcite [ hypergeofn ] f(a , b;c;z ) = _ n=0^z^n . the notation @xmath232 denotes the pochhammer symbol ( a)_n = . in our case , it is easy to show that @xmath233 and @xmath234 , so the expression ( [ hypergeofn ] ) will diverge at @xmath235 unless the series terminates @xcite . regular solutions will exist only in the degenerate case when the series terminates and the hypergeometric function becomes a polynomial . this happens when at least one of the first two arguments of @xmath231 is a negative integer . thus , the only regular solutions occur when a^- = + + - ( 1 + 4r^2)^1/2 = -q , q = 0,1,2, where @xmath236 is a non - negative integer . this yields the eigenvalues [ lambda ] = ( + + q ) ( + + q + 1 ) . the quantity @xmath237 takes on the values + = \{|m+k| + |m - k| } = \{|m|,|k|}. then it can be directly seen from ( [ lambda ] ) that the lowest eigenvalue @xmath238 is _ 0 = for integer values of @xmath163 , the quantity @xmath237 is always integer , and so the eigenvalues are [ eval_int ] _ l = ( |k|+l)(|k|+l+1),l = 0,1,2, degeneracies of these eigenvalues can be counted using ( [ lambda ] ) , ( [ eval_int ] ) , and the fact that @xmath96 is an integer . for any given value of @xmath170 , the @xmath239 cases @xmath240 correspond to @xmath241 . the @xmath242 cases @xmath243 correspond to @xmath244 , respectively . counting these cases , the total degeneracy @xmath245 of @xmath172 from ( [ eval_int ] ) is d_l = 2(|k|+l ) + 1 . first consider the case when @xmath237 is half - integer . the eigenvalues are then _ l = ( |k|+l)(|k|+l+1),l = 0,1,2, in this case , we must have @xmath246 , which corresponds to the cases @xmath247 . the number of integer values of @xmath96 that satisfy this inequality is d_l = 2|k| . in the second case , @xmath248 is integer , in which case @xmath249 . because @xmath237 and @xmath236 are integers , the eigenvalues written in terms of @xmath163 are _ l = ( |k|+l + ) ( |k|+l+),l = 0,1,2, the @xmath250 cases m = ( |k|+ ) , ( |k|+ + 1 ) , , ( |k|+ + 2 ) correspond to @xmath251 , respectively . counting these gives the degeneracies d_l = 2l+2 .

the hard state of x - ray binaries ( xrbs ) is characterized by a power law spectrum in the x - ray band , and a flat / inverted radio / ir spectrum associated with occasionally imaged compact jets . it has generally been thought that the hard x - rays result from compton upscattering of thermal accretion disk photons by a hot , coronal plasma whose properties are inferred via spectral fitting . interestingly , these properties especially those from certain magnetized corona models are very similar to the derived plasma conditions at the jet footpoints . here we explore the question of whether the ` corona ' and ` jet base ' are in fact related , starting by testing the strongest premise that they are synonymous . in such models , the radio through the soft x - rays are dominated by synchrotron emission , while the hard x - rays are dominated by inverse compton at the jet base with both disk and synchrotron photons acting as seed photons . the conditions at the jet base fix the conditions along the rest of the jet , thus creating a direct link between the x - ray and radio emission . we also add to this model a simple iron line and convolve the spectrum with neutral reflection . after forward - folding the predicted spectra through the detector response functions , we compare the results to simultaneous radio / x - ray data obtained from the hard states of the galactic xrbs gx 339 and cygnus x-1 . results from simple compton corona model fits are also presented for comparison . we demonstrate that the jet model fits are statistically as good as the single - component corona model x - ray fits , yet are also able to address the simultaneous radio data .

astro-ph0509028

bipolar plasma outflows , loosely termed jets , are a feature common to a variety of different astrophysical objects . jets are observed in accreting compact objects of all scales , as well as at stellar birth and death . they seem equally able to form out of massive accretion disks as well as from quasi - spherically collapsing plasma , which suggests that their creation is a basic byproduct of some routinely occurring ingredients . these necessary inputs seem to be rotation , collapsing or infalling / accreting plasma , and magnetic fields . jets likely are formed at least in part from a necessity to shed the system of excess angular momentum ( e.g. , * ? ? ? the exact details of their formation , as well as the nature of the coupling between the infalling / collapsing plasma and the outflow is still an active area of controversy and research on many fronts . however , the fact that so many types of systems share this mechanism extends hope that clues gained from one class of objects can be applied to the others in the search for a unified picture . the advantage of looking for answers in x - ray binary ( xrb ) jets is twofold . first , their formation seems to be a recurrent phenomenon on time scales we can observe repeatedly over the course of our own human lifetimes . for instance in the galactic xrb gx 339@xmath04 and cyg x-1 , accretion without observable jets proceeds into accretion with discernible jets and back again every few years @xcite . thus the same object can provide tests of theories developed based on its earlier activity . secondly , the existence of quality simultaneous broadband data for many sources linking the `` traditional '' jet outflow bands ( radio and likely also infrared , or ir ) and inflow bands ( x - ray and optical ) gives us a way of studying the relationship and interplay between the two sides of accretion . we use the word `` traditional '' to refer to the picture which existed up until recently that the jets and accretion flow radiate in different energy bands and can generally be studied as distinct phenomena . this picture has rapidly evolved in the last few years , beginning with the discovery of an unexpectedly intimate connection between the low and high frequency wavebands in the hard state ( hs ) of xrbs ( see , e.g. , @xcite , for a detailed definition ) . using radio observations in conjunction with 1.3 - 12.2kev data from the all sky monitor ( asm ) on - board the rossi x - ray timing explorer ( rxte ; * ? ? ? * ) , @xcite first discovered a near - linear correlation between the radio and x - ray emission of gx 339@xmath04 . these studies were followed by joint radio observations and pointed x - ray observations with rxte @xcite . this intensive simultaneous monitoring showed that , in fact , the correlation is non - linear , with @xmath1 @xcite . additionally , this non - linear correlation holds over orders of magnitude changes in the source luminosity with time . in fact , even after fading into quiescence , gx 339@xmath04 seemingly returns to the same correlation in subsequent outbursts @xcite , although it is not clear whether the correlation always maintains the same normalization in each instance @xcite . despite some possible variation in its normalization , the correlation between radio and x - ray fluxes appears to be universal to all hs xrbs with comparable broadband data @xcite . the @xmath2 slope of the @xmath3-@xmath4 correlation follows directly from analytic predictions of scaling synchrotron jet models @xcite , and can be generalized for other x - ray emission processes in terms of their dependence on the accretion rate @xcite . the correlation s normalization in the @xmath5 plane also depends on the central engine mass . when this scaling is accounted for , unbeamed , extra - galactic , supermassive black holes sources agree remarkably well with the same radio / x - ray correlation found in galactic , stellar - mass sources @xcite . the underlying physics governing the connection between the radio / ir and x - ray bands must therefore be fundamental to accreting sources regardless of mass . in @xcite we showed that jet synchrotron emission could account for the broad continuum features of the simultaneous radio through x - ray hard state observation of the galactic xrb , xte j1118 + 480 . in a later work , we showed that this same model could also explain the broad spectral features of thirteen simultaneous or quasi - simultaneous radio / x - ray observations of gx 339@xmath04 , and that the radio / x - ray correlation was naturally produced by only changing the power input into the jet @xcite . these models , however , did not attempt to address the fine features present in the x - ray spectrum which are the hallmarks of an optically thick accretion disk : in particular fluorescent iron lines and a characteristic flattening above 10kev attributed to compton reflection @xcite . the reflection is generally assumed to result from a hard x - ray continuum originating above the cooler accretion disk . we began to explore the interaction of the jet emission with the putative accretion disk in @xcite . in the hard state ( hs ) , the spectrum is hard enough and the observed reflection signature weak enough to be problematic for coronal models where there is significant coverage of the disk by hard x - ray - emitting material ( e.g. , * ? ? ? * and references therein ) . various mechanisms have been proposed to decrease both the cooling of the hard x - ray - emitting material by soft disk photons and the fraction of reflected x - rays . these mechanisms include patchy coronae @xcite , high disk ionization @xcite and beaming of the coronae away from the disk with mildly relativistic velocities @xcite . this latter approach seems extremely close in principle to the characteristics of the base of a jet . compact accretion disk coronae are theoretical concepts based on observations that suggest the presence of hot electrons radiating near the inner parts of a thin accretion disk . the existence of hot electrons in the same region is also empirically required by the jets we image in the radio wavebands . the high brightness temperatures and occasionally measured linear polarization ( e.g. , * ? ? ? * ) argue strongly that nonthermal synchrotron is the dominant radio - emitting process . because the base of the jet is significantly more compact than the outer region which dominates the cm - bands , conservation arguments imply an even hotter , denser medium near the accretion disk . we therefore wish to determine whether the base of the jets can `` subsume '' the role of the corona . this idea was proposed already back in @xcite , and by subsume we mean provide the spectral characteristics ( both direct and reflected ) which are traditionally attributed to a compact comptonizing corona . thus we will compare coronal fits using a standard comptonizing model ( see below ) to fits made by our jet model . regardless of the ultimate relationship determined to exist between the base of the jets and the corona , this is a critical step in the road to understanding . the results of this study will provide valuable clues as to the nature of the corona , both in terms of geometrical as well as internal characteristics which differ from the base of the jets . as an additional note , the concept of cyclo - synchrotron photons feeding inverse compton processes in a corona has existed for over 20 years @xcite , inspired by multiwavelength observations showing similar optical / x - ray behavior @xcite . it has recently been reconsidered for several sources @xcite in the context of magnetic flares in the corona . while a valuable step towards understanding the role of magnetic fields , these models do not address the radio emission from the jets with the same model . the remainder of this paper is structured as follows . in section [ sec : model ] we very briefly summarize the model , with a full description included in appendix [ app ] . in section [ sec : data ] we present our data analysis and fitting techniques , the results of which will be presented in section [ sec : results ] . we discuss our conclusions in section [ sec : discuss ] . our model for the jets in hs xrbs is based upon five main assumptions : 1 ) the total power in the jet scales with the total accretion power at the inner edge of the disk , @xmath6 , 2 ) the jet is expanding freely and is only weakly accelerated via the resulting pressure gradient , 3 ) the jet contains cold protons which carry most of the kinetic energy while leptons do most of the radiating , 4 ) particles are eventually accelerated into a power - law distribution , and 5 ) this power - law is maintained along the length of the jet thereafter . these assumptions are motivated by observations which are discussed further in appendix [ app ] , where we have also included a detailed description of the model and its development . in this section we only summarize the key parameters and assumptions of the jet model . beyond a small nozzle region of constant radius , which establishes the base of the jet , the jet expands sideways with the initial proper sound speed for a relativistic electron / proton plasma , @xmath7 . the plasma is weakly accelerated along the resulting longitudinal pressure gradient , which allows an exact solution for the velocity profile via the euler equation ( see , e.g. , * ? ? ? this results in a roughly logarithmic dependence of velocity upon distance @xmath8 . after a period of more rapid acceleration immediately beyond the nozzle , the velocity gradient lessens , and the velocity saturates at lorentz factors of @xmath92 - 3 . the size of the base of the jet , @xmath10 , is a free parameter and once fixed determines the radius as a function of distance along the jet , @xmath11 . the the model is most sensitive to the fitted parameter @xmath12 , which acts as a normalization . it dictates the power initially given to the particles and magnetic field at the base of the jet , and is expressed in terms of a fraction of the eddington luminosity @xmath13 erg s@xmath14 . the total power input at the base of the jets is in fact approximately an order of magnitude larger than @xmath15 , due to the requirement of distributed acceleration along the jets ( see appendix ) . once @xmath12 is specified and conservation is assumed , the macroscopic physical parameters along the jet are determined . we assume that the jet power is evenly shared between the internal and external pressures , and that the radiating particles are close to equipartition with the magnetic field . the radiating particles enter the base of the jet where the bulk velocities are lowest , with a quasi - thermal distribution . around @xmath16 @xmath17 , a significant fraction of the particles are accelerated into a power - law tail whose maximum energy is determined self - consistently with the local cooling rates . the particles in the jet radiatively cool via adiabatic expansion , the synchrotron process , and inverse compton upscattering ; however , adiabatic expansion is assumed to dominate the observed effects of cooling because of distributed acceleration . while thermal photons from the accretion disk are included as seed photons in our compton calculations , the beaming reduces their energy density compared to the rest frame synchrotron photons ( synchrotron self - compton ; ssc ) , except at the very base of the jet where they can be of the same order . reprocessed disk radiation will contribute even less and thus its feedback on the x - ray spectrum will be negligible . we therefore do not include this latter component in our calculations . in the case of near - equipartition of energy between the magnetic field and particles , inverse compton processes only will dominate the direct synchrotron emission close to the compact object where relativistic beaming is at its minimum . as the jet plasma is only weakly beamed , some fraction of the radiation will impact upon the cooler material in the accretion disk and lead to fluorescent line emission and a reflection hump ( e.g. , * ? ? ? * ) . in @xcite we calculated the resulting reflection spectrum for typical jet models , and found that when synchrotron emission from the acceleration region dominates the hard x - ray spectrum , the fraction of reflected emission is very small ( @xmath18few percent ) . in contrast , when compton processes ( predominantly ssc ) originating near the base dominate the hard x - rays , ` reflection fractions ' of @xmath19 are possible . as discussed by @xcite , calculating the reflection spectrum for our jet models is complex , as one needs to calculate a reflected spectrum associated with each emission region along the axis of the jet . furthermore , for our detailed jet models the disk sees a qualitatively different jet spectrum than the direct spectrum viewed by a distant observer . this is in contrast to the simple ` moving corona ' models of , e.g. , @xcite , where both the direct spectrum and the spectrum impinging upon the disk have the same simple power law shape , and the total reflected spectrum can be calculated analytically . for the initial spectral fit studies presented here , a detailed reflection calculation for our jet model was deemed too computationally prohibitive . for the purposes of the study presented here , we adopt a simplified approach . in our fits , we add a single gaussian line to our continuum model , and we allow the line energy to vary between 67 kev and the line width , @xmath20 , to vary between 01.5kev . we then convolve the entire _ directly viewed _ spectrum with a non - relativistic reflection model derived from the green s functions of @xcite . the amplitude of the reflected spectrum is left as a fit parameter , expressed in the usual manner as a fractional solid angle , @xmath21 , subtended by the reflector . ( @xmath22 indicates that the radiation impinging upon the reflector is equal to that directly viewed by the distant observer . ) this general approach of phenomenologically fitting a reflection spectrum is similar to that employed by many pure comptonization models ( e.g. * ? ? ? * ; * ? ? ? * ) , albeit here it is less likely that the directly viewed spectrum is a completely adequate proxy for the spectrum viewed by the cold disk . we will consider more sophisticated approaches to calculating the reflection features in future works . in our previous papers ( e.g. , * ? ? ? * ; * ? ? ? * ) , we only compared the jet model to spectra that had been ` unfolded ' with ` xspec ` using simple exponentially cutoff , broken power - law models . this is in fact the usual practice among many researchers when comparing complicated theoretical models to multiwavelength data ( e.g. , * ? ? ? * ; * ? ? ? unfortunately , such a procedure does not allow the determination of a statistical goodness - of - fit , and more importantly it does not allow one to compare the model to the fine features of the x - ray spectral density distribution ( sed ) . here we address this issue by forward - folding the jet model through the detector response matrices of the x - ray instrument , and then comparing to the data in ` detector space ' . specifically , we have imported our jet + multi - color blackbody thermal disk ( similar to diskbb ; * ? ? ? * ; * ? ? ? * ) model as a subroutine for use in standard x - ray data analysis packages . the current routine works in both xspec @xcite and isis @xcite . in this work , we perform our fits using isis v.1.2.6 as this latter package can read in and fit lower frequency simultaneous data sets ( e.g. , radio through optical data ) as ascii files , without the need for creating dummy response matrices . furthermore , the isis ` unfolded spectra ' ( shown in the figures throughout this work ) are independent of the assumed spectral model ( i.e. , the unfolding is done solely with the response matrix and effective area files ; see @xcite for details ) . all shown residuals , however , are for comparisons between the properly forward - folded model and the data in detector space . see @xcite for a discussion of further differences between ` isis ` and xspec . we are making detailed comparisons between our jet model and detector space data for the first time . indeed , to our knowledge , this is the first time that such a complex , multi - wavelength model has been compared properly to the x - ray data in detector space . for this paper , we therefore have chosen to initially leave most jet model parameters free so that we can fully explore their effect on the fits . we can then determine which parameters are most fundamental for describing the data and which can effectively be frozen in future applications . this is essentially in the same spirit as prior studies with comptonization models , where models such as ` eqpair ` @xcite have significantly more parameters than are used in a typical fit . for such comptonization models , even fairly basic parameters , such as thermal vs. nonthermal compactness ( i.e. , the ratio of coronal energy to coronal radius ) , can be degenerate with one another in real data fits , and one often chooses to freeze the nonthermal parameters to negligible values . for a detailed discussion of these points , see the review article by @xcite . currently , our multi - component model runs significantly slower than single - component models which calculate comptonization in the corona . determining meaningful error bars for all parameters was not always possible because for some fit parameters the resulting spectrum can vary noncontinuously . given that this is the first time that broadband data are being properly compared to our forward - folded model , such initial difficulties are not surprising . nevertheless , the model fits presented here represent an exploration of a significant amount of the parameter space , especially given the fitting time scales involved . we model the data from rossi x - ray timing explorer ( rxte ) observations for the hard state galactic black hole candidates ( bhcs ) gx 339@xmath04 and cygnus x-1 . we use data from both the rxte proportional counter array ( pca ; @xcite ) and from the high energy x - ray timing experiment ( hexte ; @xcite ) . the data have been extracted with the recent ` heasoft 5.3.1 ` software release . compared to earlier releases of this software , the relative calibration of the pca and the hexte is now in excellent agreement ( see analysis in , e.g. , * ? ? ? * ) . power - law fits to spectra of the crab pulsar and nebula show the pca data to have a systematic uncertainty of 0.5% , which we added in quadrature to the data . we used the energy bands 322kev for the pca ( and grouped the data to a minimum of 30 counts / bin ) and 18200kev for the hexte ( where we added data from the two clusters and grouped the counts to a minimum signal - to - noise , after background subtraction , of 10 in each bin ) . we use the top layer of the pca only . during part of the observations , some proportional counter units of the pca were switched off . in these cases we added the spectra extracted for the different pcu combinations , appropriately weighting the response matrices . the cygnus x-1 data consist of three of the many observations taken during our several years long rxte monitoring campaign ( * ? ? ? * ; * ? ? ? * ; * ? ? ? * and references therein ) and were chosen to be representative spectra bracketing the typical hard state spectral variations of cyg x-1 . the data from gx 339@xmath04 also have been discussed by us elsewhere @xcite , and bracket the typical variations in the overall hardness and luminosity of the gx 339@xmath04 hard state . ( they do not , however , encompass the most extreme bright and relatively soft hs data discussed by @xcite,@xcite , and @xcite . ) table [ tab : obslog ] contains a log of the observations . cclcc cyg x-1 & 40099 - 01 - 19 & 1999 september 25 & 8080 & 13 + cyg x-1 & 60090 - 01 - 26 & 2003 february 23 & 9696 & 1/4 + cyg x-1 & 60090 - 01 - 41 & 2003 september 22 & 11600 & 13/14 + gx 339@xmath04 & 20181 - 01 - 02 & 1997 february 10 & 10528 & + gx 339@xmath04 & 40108 - 02 - 01 & 1999 april 2 & 9152 & + gx 339@xmath04 & 40108 - 02 - 03 & 1999 may 14 & 9776 & + in order to compare our jet models to the more standard comptonization models , we also fit our observations with the hybrid thermal / non - thermal comptonization model ` eqpair ` @xcite . we slightly changed the publicly available version of this code to make the model run under xspec v.11.3.1 , and hence also under isis v.1.2.6 . to allow for a partial covering of the disk by the compton corona , we add emission from a disk using the diskpn model @xcite . the peak temperature of this disk model is also set equal to the peak temperature of the seed photons input to the compton corona . these seed photons are similarly assumed to have a spectral energy distribution that follows that of the diskpn model . the corona is presumed to be comprised solely of thermal electrons , and its properties are described by two parameters : a seed electron scattering optical depth , @xmath23 ( pair production can increase the total optical depth from this value , although pair production is negligible for the fits described in this paper ) , and the coronal compactness ( i.e. , coronal energy content divided by coronal radius ) , expressed as a ratio relative to the soft , seed photon compactness , @xmath24 . we also allow for an additive gaussian fe k@xmath25 emission line whose centroid energy is constrained to be between 6 and 7kev and whose width is constrained to be @xmath26kev . in our use of the eqpair model , the reflection component is relativistically smeared using the velocity profile of the ` diskpn ` disk model , which in turn follows the rotational velocity of a standard thin disk circularly rotating under the influence of a paczyski - wita pseudo - newtonian potential @xcite . the reflecting medium is assumed to have solar abundances ; however , the reflecting medium can be ionized , as parameterized by the ionization parameter , @xmath27 @xcite . table [ tab : eqpair ] gives the best fit parameters of these ` eqpair ` fits , and the fits are shown in the third column of figs . [ gx339_all ] & [ cyg_all ] . the reduced @xmath28 values are close to unity , indicating very good agreement between the model and the data . the parameters found for the comptonization model are in general agreement with those reported for earlier observations . the fits yield moderate optical electron depths of @xmath29 with coronal compactnesses of @xmath30@xmath31 , which correspond to coronal temperatures of @xmath32 - 100kev . ( note that the @xmath33 for obs i d 40108 - 02 - 03 has rather large error bars , and is partly being driven by the poor statistics in the hexte band . ) all of the coronal model fits yield reflection covering fractions @xmath19 . none of the fits strongly require ionized reflection ; however , the reflection model used by the eqpair code is very simplified ( being based upon the pexriv code ; @xcite ) . the fitted equivalent widths of the fe k@xmath25 lines are on the order of 150ev , and consistent with the highly smeared reflection models employed , the lines are generally found to be rather broad . as will be discussed elsewhere in more detail @xcite , we believe that part ( although not all ) of the fitted line broadness is caused by the comptonization model employed here having too simplified a description of the transition between the accretion disk emission and the comptonization continuum . finally , the changes in accretion disk temperature and flux , coupled with the changes in coronal compactness and optical depth , are in response to the fact that we have deliberately chosen observations that span a wide range of flux and spectral hardnesses . detailed discussions of coronal model fits to cyg x-1 data ( e.g. , comparison to phenomenological broken power - law fits , comparison to timing data ) are discussed in @xcite . their are three main points that we wish to emphasize with the fits presented here . first , the eqpair fits describe the data well as well as any coronal models that we have explored ( although not better than simple exponentially cutoff , broken power - law models ; @xcite ) . second , the fits do not require comparatively as large reflection fractions . third , and related to the previous point , the coronal models themselves describe a continuum spectrum more complicated than a simple , exponentially cutoff power - law . some fraction of the spectral hardening above 10kev , normally attributed to the ` reflection hump ' , is in fact described by curvature of the comptonization continuum . in compton coronal models , this curvature is in part due to transiting from a spectrum strongly affected by the soft seed photons to one strongly dominated by photons that have been multiply compton scattered . as we discuss below , this continuum curvature an additional interpretation in jet models . as mentioned above , our strategy in testing our jet models against the data was to allow the maximum number of model parameters to vary freely during the fit . this allowed us to explore which parameters had the greatest influence on the fits . several parameters settled on fairly similar values for both sources and all observations , which suggests that those parameters could be fixed in future applications of this model . we will discuss this possibility explicitly below . fits with a complicated model can easily lead to false , local minima ; therefore , we began the fitting procedure outside of isis , using unfolded data sets in order to find a set of starting parameters that would yield reduced @xmath34 . for this paper we have chosen to explore only models with rough equipartition between the magnetic and radiating particle pressures ( @xmath35 ) . this assumption is obviously not applicable for poynting flux dominated jets ; however , we consider the very weak acceleration required by observations to be an indication that magnetic domination ( which would generally imply stronger acceleration mechanisms ) is not likely to be very extreme . we use the results of @xcite as a rough guide for consistency , and only consider models with reflection fractions @xmath19 for jet nozzle radii @xmath36 to be ` successful ' fits . for smaller values of the radii , slightly lower values for the reflection fraction would be expected . in all cases the jet models seemed to naturally prefer reflection fractions @xmath37 , so in practice this restriction on reflection fraction was not an issue . [ gx339_comps ] shows a representative model for gx 339@xmath04 in order to illustrate the contribution from the various components which go into the later figures showing actual fits . the radio through ir originates exclusively from self - absorbed synchrotron radiation beginning at @xmath38 and continuing outwards along the jet . optically thin synchrotron emission from the accelerated power - law tail of leptons also contributes to the soft x - ray band . the base of the jet radiates direct synchrotron photons , giving a slight hump in the optical / uv , which are then included as seed photons ( along with disk photons ) for upscattering by the emitting electrons into an ssc / ec `` hump '' in the hard x - rays . the shape of this hump is rounded , due to the quasi - thermal particle distribution assumed in the base , and thus reduces the need for a large fraction of disk - reflected photons to contribute to the spectral break / hardening at @xmath1810 kev . the possibility that there are two correlated continuum components in the x - ray band is supported by our ensemble of simultaneous radio / x - ray observations of cyg x-1 , whose spectra are very well - described by exponentially cut - off , broken power - laws ( @xcite ) . in these models , the soft x - ray spectral slope is very well - correlated with the break of the hard x - ray spectral slope , and the hard x - rays are very well - correlated with the radio @xcite . the multicolor blackbody included in the jet model is also shown in the soft x - ray band . the photons from this component are included in the inverse compton upscattering within the jet . the statistical fits are shown in figs . [ gx339_all ] & [ cyg_all ] , for gx 339@xmath04 and cyg x-1 , respectively . each figure encompasses nine panels , with each row representing a single simultaneous radio and x - ray observation of the source . the first column shows the entire radio through x - ray jet model + soft disk + reflection fit , and the second column focuses just on the x - ray band . the last column shows the thermal compton corona model from eqpair for comparison . values for all fitted parameters are given in tables 24 . the results presented in figs . [ gx339_all ] & [ cyg_all ] and tables [ tab : mffjeti ] , [ tab : mffjetii ] , & [ tab : eqpair ] immediately demonstrate three important results of this work . * * jet models describe the data equally well as pure compton corona models , even when employing the broad - band and high statistics of the rxte data . * there has been question in the literature ( e.g. , @xcite ) as to whether or not jet models can adequately describe the spectral cutoff of the hard tail . while pure synchrotron jet models may not be able to describe the steep cutoff present in some ( but not all ) observations , jets can also successfully account for this hard tail cutoff via comptonization . the primary differences between ` traditional ' coronal models and jet coronal models are that the coronal temperatures are higher in the latter ( @xmath39k vs. @xmath40k ) , and the disk seed photons in jet models are significantly augmented if not dominated by synchrotron seed photons . * * the jet models , similar to the compton corona models , describe a complex continuum with curvature specifically , a hardening above 10kev . * again , the hardening often predominantly ascribed to a ` reflection hump ' is being somewhat subsumed by continuum emission . in the jet model , the steeper slope of the soft x - rays is primarily due to synchrotron emission from the jet , rather than being influenced by the disk photons , while the hardening is due to the ssc / ec component . reflection is present , but represents a smaller fraction of the total hard flux compared to the corona models . note that the fits discussed here are in contrast to earlier incarnations of the jet model which attempted to describe observations solely via the synchrotron component ( e.g. , the studies of xte j1118 + 480 ; @xcite ) . for xte j1118 + 480 , the lack of any discernible break or hardening near 10kev in the continuum spectrum was used to argue for the lack of a reflection component @xcite , which would be consistent with a jet spectrum dominated by the synchrotron component @xcite . in contrast , the fact that all of the xrb spectra shown here harden above 10kev argue for the importance of both the synchrotron and ssc / ec components of the jet spectrum for these fits , as well as the presence of reflection . * * although the jet and compton coronal models describe the x - ray data equally well , only the jet model naturally describes both the amplitude and slope of the radio data without the need for additional free parameters . * it is important to note that the radio spectrum , as well as the bulk of the 3 - 200kev x - ray continuum ( i.e. , the overall amplitude and continuum curvature ) , are almost solely driven by the parameters of the jet : energy input to the jet ( @xmath41 ) , radius of the jet base ( @xmath10 ) , electron temperature at the jet base ( @xmath42 ) , slope of the power - law component of the electron energy distribution ( @xmath43 ) , the equipartition parameter ( @xmath44 ) , and the location of the particle acceleration zone in the jet ( @xmath45 ) . if one begins instead with a corona model and attempts to fit the amplitude and slope of the radio data by assuming a relationship to an extant jet model , this will in essence require the addition of many more parameters to describe just two physical quantities . we argue that nothing can be gained by this approach , which is why we have chosen to study the test case of full coupling . from our set of jet model fits that successfully describe the gx 339@xmath04 and cyg x-1 data , we can now explore the implications of the fitted parameters . cyg x-1 and gx 339@xmath04 are somewhat different sources while in their hard states . the luminosity of gx 339@xmath04 varies much more than that of cyg x-1 . this is apparent in the fitted values for the power normalization @xmath12 and output power @xmath46 shown in the first two columns of table [ tab : mffjeti ] . both sources show best fit @xmath12 in the range of @xmath47 @xmath48 , but gx 339@xmath04 shows significantly more variation in this parameter . differences can also be seen in the other main parameters , for instance on average gx 339@xmath04 favors slightly larger values of the nozzle radius , @xmath10 . this reflects the slightly higher ratio of x - ray to radio flux observed in cyg x-1 . a smaller scale jet base , for a given jet power , increases the jet compactness which in turn gives a slightly higher flux ( with the x - ray flux being more sensitive to this effect ) and pushes all jet emission to slightly higher frequencies . other fit parameters also indicate differences between these two sources ; however , the parameters surprisingly fall roughly in the same range given the _ a priori _ possibility for much greater parameter variations . note that the lowest luminosity gx 339@xmath04 observation , obs i d 40108 - 02 - 03 , has poor enough statistics that the following general statements about the gx 339@xmath04 fits will be based mostly on the brighter two observations . for example , both the electron temperature , @xmath42 , and the location of the start of the acceleration zone , @xmath38 , tend to be larger in gx 339@xmath04 than in cyg x-1 . ( although we were unable to find adequate error bars for the latter parameter , the trend of finding larger values in gx 339@xmath04 was persistent . ) additionally , we note that the electron power law index , @xmath43 , shows more variation in gx 339@xmath04 than in cyg x-1 . although the range in electron temperatures at the base of the jets falls within a range of a factor of @xmath49 for both sources , one of the criticisms of this class of models has been that @xmath50-@xmath51 k is not as `` natural '' a value as the @xmath52kev typically used in thermal comptonization models . the jet model electron temperature , however , is comparable to or slightly greater than the value typically derived for radiatively inefficient accretion flows ( e.g. , * ? ? ? if some fraction of the accreting plasma is fed directly into the jets , and also perhaps heated slightly in the process , we would expect such a temperature . on the other hand , several jet model free parameters seemed to settle quite quickly into similar values for both cyg x-1 and gx 339@xmath04 : the equipartition factor @xmath44 , the fraction of accelerated particles @xmath53 , and the ratio of nozzle length to radius @xmath54 . physically , this suggests that these jets either are close to equipartition , or we have found a local minima of solutions for @xmath55 , and we may find other reasonable minima for @xmath56 , e.g. , magnetically - dominated jets . we plan to explore this particular question in a future work . it seems reasonable , however , to freeze the fraction of accelerated particles at @xmath57 and make the statement that reasonably efficient acceleration is expected to occur in hard state bhc jets . similarly , a compact base / corona region with scale height similar to the radius seems a reasonable assumption . it is likely that this parameter can also be frozen in future applications of the jet model . we listed the two `` acceleration parameters '' , the shock speed relative to the bulk plasma flow , @xmath58 , and the ratio of the scattering mean free path to the gyroradius , @xmath59 , as separate fit parameters . however , as discussed in the appendix , these parameters are perhaps not physically meaningful as currently defined , since we are no longer as convinced that the acceleration process is diffusive fermi acceleration . these parameters enter into the acceleration rate as the factor @xmath60 , which is compared to the sum of the cooling rates from synchrotron and inverse compton radiation , and adiabatic expansion in order to calculate the local maximum accelerated lepton energy . therefore , this parameter can be loosely interpreted as a factor related to the efficiency of the acceleration process and merged for future fits into a single parameter . for the current fits , the individual components were left free to vary , with mixed results . for cyg x-1 , @xmath61 , suggesting a meaningful range of fits . however , for gx 339@xmath04 , @xmath62 ranges over two orders of magnitude , and we suspect that we did not fully explore parameter space meaningfully for this source . in future applications of the model , it is likely that we will combine these two parameters into a single parameter that will absorb our uncertainty about the acceleration process . while we list best fit values for the multicolor blackbody disk model parameters @xmath63 and @xmath64 , as well as the derived inner radius @xmath65 , it is important to keep in mind that because the data only extend down to a few kev , we can not meaningfully constrain this component from spectral fitting . as described in the appendix , the disk photons are not as important to the overall photon field as the locally produced synchrotron photons and thus can mainly be constrained by their direct spectral contribution . this contribution , although weak , is in fact required for a good fit , but it is not unique . @xcite have also fit some of the x - ray data presented here , and for similar reasons fixed the thermal disk parameters . on the other hand , for the disk emission plus comptonization models presented here ( as well as for those presented in @xcite ) , the temperature of the disk was tied to the temperature of the seed photons i.e . , the thermal photons were the _ only _ source of seed photons for comptonization . thus by virtue of these imposed restrictions , much more stringent formal limits for the disk components were attainable . the main result to take away about the accretion flow modeling is that our fits are generally consistent with a sub - eddington accretion disk with temperatures somewhat less than 1 kev . similarly , the total power entering into the jets is roughly consistent with being of the same order as the observed luminosity required in the disk to be consistent with the data . the presence of a weak disk component is necessary for a good fit , and thus disk photons will contribute to the inverse compton component from the jet . we do not , however , feel we can confidently make any statements about the disk geometry and thus the accretion rate at the inner radius assumedly feeding the jets . for both gx 339@xmath04 and cyg x-1 , the amount of reflection required by the jet models is roughly comparable to , although in general slightly lower than , that required by comptonization models . this trend can be understood by the fact that the jet base ssc / ec component dominates above 10kev , and to some extent subsumes the role played by the `` compton hump '' normally attributed to disk reflection . this effect of course brings up an interesting point : there is a clear degeneracy in how the spectral hardening above 10kev can be understood in terms of continuum models . the fact that the jet ssc / ec component has a similar appearance to the compton reflection hump does not preclude the presence of both . clearly , the fluorescent fe line implies that there must be a degree of reflection . what these results ( as well as the comptonization fits ; @xcite ) do suggest is that one can not uniquely determine a reflection fraction independently of the presumed continuum model . the key result of this work is that we have clearly demonstrated that even the high statistics of rxte x - ray data can not distinguish between the jet model and thermal comptonization . ( indeed , the rxte data also can not distinguish between thermal and nonthermal comptonization models , as has been previously discussed by , e.g. , @xcite ) . in this sense , the base of the jets can be said to effectively `` subsume '' , at least spectrally , the role created for the corona . the main difference between these two pictures then comes down to geometry , inflow versus outflow , and the relationship of the corona to the lower - frequency - radiating regions of the outer jets . the jet model presented here is consistent with the physical picture seen in relativistic mhd simulations @xcite , where the corona is not static but instead is a windy hot material blowing away from the inner regions of the accretion flow . this wind / corona is likely what enters into , or is collimated and becomes , the jets and therefore it is intimately related to the larger scale outflows , similar to what we propose here . in addition , the jet base provides a natural illuminator for the disk . our ability to constrain the role of the accretion disk is significantly limited by the lack of low - energy data . ideally we would like to be able to determine the inner temperature , and perhaps also assess the need for a recessed inner edge , explicitly from the observations . having this information would allow us to better constrain the ratio of external to synchrotron - produced photons in the jet base . additionally , we expect differences between jet models and compton corona models to become more pronounced as one considers data above the nominal comptonization cutoff at 100200kev . from both of these perspectives , the observational situation will likely soon improve with the launch of astro - e2 , which will have the ability to observe from 0.5 - 600 kev and thus more stringently constrain both jet and compton corona models . additionally , although the broad - band effects of reflection , e.g. , the reflection hump , can not be fit independently from the assumed continuum , there is some hope that finer spectral resolution ( perhaps coupled to correlated studies of the continuum variability ) will be able to more uniquely to determine the reflection fraction based upon detailed modeling of the fe line @xcite . in the scenario considered here , the nozzle / base of the jet is the dominant region for the creation of the hard x - ray emission via comptonization processes . constructing a completely realistic description of this region , however , is difficult given the that the physics of jet formation , acceleration , and collimation are not yet understood . the nozzle essentially represents the initial conditions which fix parameters along the rest of the jet . we have shown that the form we have developed for this region over the last several years can successfully mimic the corona in purpose , but it is not as well - determined feature as we would like . an alternative , but also currently rather ad - hoc , possibility would be to invoke special conditions linking an outflowing corona with more typical thermal plasma conditions to an outer jet . this would require some kind of acceleration and/or heating of the particles at the interface ( and thus more free parameters ) , but may satisfy many of the spectral requirements . one possible way to discern between these two situations , or at the very least limit the possible flux from a nozzle , could be to calculate the amount of local ionization expected from nozzle uv photons and compare that to what is observed ( j. krolik , priv . if it can be shown for a given source that all observed line emission can be accounted for by the known companion , this limit may be rather stringent . we note , however , that in high mass x - ray binaries such as cyg x-1 , the companion luminosity , especially in the uv , is extremely high . it is therefore more likely that meaningful limits can be obtained for low mass x - ray binaries , such as gx 339@xmath04 . while we have focused on modeling simultaneous radio / x - ray data sets in this paper , for gx 339@xmath04 and several other sources observations do exist which in addition have simultaneous ir / optical data . near - ir data in particular can provide a valuable constraint on the jet synchrotron component , especially when the transition turnover from optically thick to thin is resolved ( e.g. * ? ? ? * ) . using the work presented here as groundwork , we will explore elsewhere whether more parameters can be constrained by such data sets . similarly , with _ hess _ already online and _ glast _ imminent , predictions of @xmath66-ray fluxes will be another important way of testing not only the current model , but models which incorporate hadronic processes . however , the next major frontier for application of our jet model clearly is to consider questions of time - dependence . by studying the lag between the radio and x - ray correlated variability , we can constrain the plasma speed and/or distance between the two regions . the existing radio / x - ray flux correlations constrain the long time scale , while simultaneous radio / rxte pointed observations constrain thousand of second time scales @xcite . we are currently using long duration ( several hundred ksec ) , quasi - continuous radio / x - ray observations of cyg x-1 to explore intermediate time scales ( markoff et al . , in prep . ) . the models considered here , however , are steady - state and focus on the hard accretion state only . some of the most interesting and revealing behavior is seen during transitions between these states . for instance , during transition into the hard state we observe a hard x - ray tail before the jet forms as a detectable radio structure @xcite . the presence of the tail suggests that the jet base / corona may form as a viable region before the outer stable radio - emitting structures are built up . there are also significant clues coming from noise and faster scaled time variability such as qpos @xcite which need to be integrated into the picture . in general the state transitions reveal many interesting properties which are clearly not due to a steady state model , but which do share several similar properties to the more persistent hard states . studying the time - dependent behavior of these systems will likely be an effective way to further constrain the conditions at the jet base / corona . we acknowledge partial funding and travel grants from the deutscher akademischer austauschdienst ( daad ) and the national science foundation , under nsf grant int-0233441 . m.a.n . is also supported by nasa grant sv3 - 73016 . is supported by an nsf astronomy & astrophysics postdoctoral fellowship , under nsf award ast-0201597 . we thank the aspen center for physics for its hospitality during the final stages of the preparation of this paper , as well as the anonymous referee for helpful comments . like all models for optically thick outflows , our jet model builds on the initial work of @xcite . their paper demonstrated how a superposition of idealized , self - absorbed , conical jet components results in a flat synchrotron spectrum in the radio bands . in reality , compact jets usually deviate from this idealized case , and show a slight spectral inversion in the radio wavebands , with spectral index @xmath67 ( @xmath68 ) . a more realistic treatment of the internal physics can account for this inversion . using the framework for hydrodynamical jets developed in @xcite , @xcite developed a `` free jet '' model which is more self - consistent . for instance , the assumption that the jet is roughly conical via lateral expansion with a constant sound speed requires weak longitudinal acceleration . the euler equation can then be solved for the longitudinal velocity gradient , assuming only adiabatic losses , and is analogous to a pressure driven wind solution with the proper jet speed replacing the wind speed . this weak acceleration combined with relativistic beaming effects will result in the spectral inversion typical of compact jet sources . these earlier models were very useful for studying the general physical scalings predicted by compact jet models . however , without full particle distributions and radiative transfer , they can not be tested against observed spectral data . our models have been developed on the basis of these earlier papers , but specifically focus on the problem of spectral predictions . these models are only appropriate for systems where the jets are not expected to be highly collimated nor highly accelerated , e.g. , the jets in hard state xrbs and llagn . we start by assuming that the base of the jets are at the speed of sound for a relativistic gas with adiabatic index 4/3 , giving a proper sound speed @xmath69 . furthermore , we assume the jets expands laterally with the constant sound speed , which is appropriate under the assumption of maximally efficient jets ( see * ? ? ? the resulting longitudinal acceleration along the axes is largest near the bases , and asymptotically approaches a maximum of lorenz factor @xmath70 in the outer radio emitting regions , and @xmath71 for the synchrotron emission . such weak beaming factors are suggested by observations of hard state jets @xcite , in contrast to the near light - speed velocities implied by superluminal motion in higher luminosity radio - emitting states . the main free parameters of this model are the power normalization @xmath12 , the radius of the jet nozzle @xmath10 , the equipartition of energy between the radiating leptons and the magnetic field @xmath72 , and the parameters determining the initial lepton distribution as described below . we also left the length of the nozzle region ( with fixed radius ) , @xmath54 free , but it turns out the fits are not very sensitive to its value . other parameters which come in , such as the central mass @xmath73 , the inclination of the jets to the line of sight @xmath74 and the distance to the source @xmath75 , are determined from observations and remain fixed for any given source . finally there are two parameters related to the presence of weak disk emission , the total luminosity radiated by the disk and its inner radius temperature , @xmath76 and @xmath77 respectively . these can not be well constrained by our data because although they are spectrally fit assuming a multi - color disk blackbody @xcite , the low - energy x - ray spectrum is not well covered in our observations . similarly , although they do contribute to the photon field upscattered by energetic jet leptons , they are not the dominate component . the most important free parameter is @xmath12 , parameterized in terms of the eddington luminosity , which determines the power initially input into the particles and magnetic field at the base of the jets . the total power input at the base of the jets is in fact approximately an order of magnitude larger than @xmath78 , and for our model can be estimated only _ a posteriori _ from the resulting emission in the emitting rest frame . the difference between @xmath41 and the total output radiation @xmath46 expresses the lack of knowledge of what initially collimates the nozzle , as well as what causes the distributed reenergization and consequent energy requirement to maintain the observed ( e.g. * ? ? ? * ) particle distributions along the jet ( see below ) . @xmath12 plays a similar role to the compactness parameter in thermal comptonization models , which parameterizes the total energetics in the absence of an understanding of the mechanism for energizing the corona @xcite , as a consequence of assuming maximal jets , there should be approximate equipartition between the kinetic and internal energy . once @xmath79 is fixed , therefore , the energetics along the jets are fixed as well . for a given @xmath10 and @xmath44 at the base , conservation of energy and particle flux together with the proper velocity profile along the jet , @xmath80 , determine the radius , density , and magnetic field profiles along the entire jets . what is left to be determined is how the energy in the radiating particles is distributed . as mentioned earlier , we assume that the jets contain both leptons and hadrons , but that the hadrons remain non - relativistic and serve only to carry the bulk kinetic energy . the leptons , which are assumed to be mildly relativistic and quasi - thermally distributed ( see , e.g. , * ? ? ? * ) are thus more likely to be accelerated . in any case , we assume that leptons are predominantly responsible for the observed radiation . this is begging the question of hadronic acceleration , because the end effect of significant hadronic acceleration and subsequent inelastic collisions will invariably be relativistic leptons . the expected energy distributions , whether from direct acceleration or creation via hadronic collisions , are not trivial to discern from x - ray emission alone . therefore we do not specify whether the radiating particles are electrons or positrons , since this will not effect the observable outcome . eventually the consistency with these results can be checked against hadronic models , but this is a very difficult problem ( see , e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) and is not considered here . the one hint for what will be required beyond neutral matter is that we do not require @xmath81 , and once a fit is made , we can see if additional pair creation is necessary . the current model was developed in simplified form by @xcite , where we first incorporated simple particle distributions for the radiating leptons and calculated the radiative transfer along the jets . our goal was to determine if the same kind of model which could explain the inverted radio core of the galactic center supermassive black hole , sgr a * , also predicted significant x - ray emission . this idea was motivated by the first identification of sgr a * in the x - ray band by the _ chandra _ observatory @xcite . we explored two `` canonical '' types of particle distributions , power - laws and maxwellians , for sgr a * in its quiescent state ( see , e.g. , * ? ? ? * ) . these two ostensibly different distributions result in similar fits as long as the characteristic particle energy , ( @xmath82 for the power - law or @xmath83 for the maxwellian ) is similar . unlike `` typical '' low - luminosity agn ( llagn ; * ? ? ? * ) , the quiescent sgr a * does not show any indication of a power - law of optically thin synchrotron emission after the break from its flat / inverted radio spectrum . therefore if the radiating particles have a power - law distribution , it must be so steep as to be indistinguishable from a maxwellian in the optically thin regime . either scenario strongly suggests , for example , that acceleration in the jets of sgr a * is absent or very inefficient at @xmath84 . based on sgr a * , we therefore assume that the baseline particle distribution , below a certain accretion rate which must be very low , is quasi - thermal . this scenario is consistent with particles being advected into the jets directly from the accretion flow , or created and roughly thermalized near the base . sources with jets at higher luminosity , such as agn and xrbs , do , however , show the standard indicator of particle acceleration : optically thin synchrotron emission from a power - law distribution of particles ( e.g. , * ? ? ? * ; * ? ? ? * and refs . therein ) . interestingly , sgr a * shows daily flaring in which the x - ray spectrum rapidly increases in flux and hardens dramatically @xcite . the flare emission stems from energization of the particles via acceleration , heating or both at once @xcite , resulting in enhanced synchrotron and/or ssc emission in the x - rays . because the increased x - ray flux during the flares brings sgr a * more in line with typical levels of llagn activity @xcite , one interpretation is that acceleration of particles is the main factor contributing to the comparative weakness of sgr a*. we also therefore assume that above some critical accretion rate above that of sgr a * , particle acceleration will be commonplace in the jets . this is also necessary to explain the high synchrotron efficiency in xrbs ( see , e.g. , * ? ? ? * ) . in the only xrb where the optically thick - to - thin turnover from compact jet synchrotron has been directly observed ( gx 339@xmath04 ; * ? ? ? * ; * ? ? ? * ) , it occurs at a significantly lower frequency than would be expected if the acceleration began in the jet base . we therefore assume that when particle acceleration occurs , it begins at some location @xmath38 along the jet , which is a free parameter but which appears to fall in the range of a few @xmath85 @xmath17 for most sources we have so far considered . the fraction of particles accelerated into the tail is a free parameter in our model , but is generally driven to fairly high values so we will likely fix it at some value above 80% in future applications . the power - law index of the tail @xmath43 ( @xmath86 ) is also a free parameter , and can be constrained by the data . particle acceleration must compete with radiative and adiabatic cooling to energize the quasi - thermal particles into a power - law tail . where the sum of the cooling rates equals the acceleration rate defines the maximum achievable particle energy , @xmath87 . unfortunately since the acceleration process is still an open question , the acceleration rate is open to significant interpretation . in @xcite and @xcite , we used acceleration rates appropriate for diffusive shock acceleration in the most conservative case of the magnetic field direction parallel to the shock normal ( see * ? ? ? the rate of diffusive shock acceleration is always proportional to the magnetic field . if synchrotron cooling dominates the cooling term , @xmath87 will thus be independent of the magnetic field and can give information about the plasma parameters if the location of the cutoff can be determined . in @xcite we showed that this results in a synchrotron emission cutoff around 100 kev for @xmath88 and if the ratio between the particle s mean free path for diffusive scattering to the gyroradius , @xmath89 . in general @xmath27 is thought to be limited by the ratio of the particle to shock velocities , and so likely no higher than @xmath90 , but this value is certainly not fixed . at the time of writing , other processes such as stochastic resonant acceleration ( e.g. , * ? ? ? * ) are coming into favor . given this uncertainty in the mechanism , we are loosening the constraints on the acceleration rate we used previously and are letting @xmath27 and @xmath91 vary . if the acceleration rate we are using is not really physical , then these parameters have lost their meaning and become essentially fudge factors absorbing the free parameters controlling the acceleration rate for other mechanisms , to be determined later . if shock acceleration does still hold , then these tell us about the plasma conditions in the jets . regardless of the process , the inferred cooling times for the accelerated leptons will be too high to maintain the power - law along the jets unless there is continuous distributed acceleration . distributed , continuous _ in situ _ injection is a standard requirement for explaining the persistence of particle distributions along agn jets @xcite and seems to be a common feature in outflows . so to summarize the particle distribution discussion , we assume that the particles enter the jet with a relativistic quasi - maxwellian distribution , the temperature of which , @xmath92 , is a fitted parameter . some fraction of the particles are then assumed to be accelerated continuously beyond a location in the jets , @xmath38 . where exactly can be constrained by the frequency of the optically thick - to - thin break in the synchrotron spectrum , if detected , as well as from the synchrotron component contributing to the soft x - rays . as the leptons travel outwards along the jets , they interact with the local magnetic field and photon fields and cool via synchrotron and inverse compton radiation . inverse compton processes are strongest near the base of the jets , where the density is highest , and thus one x - ray component is due to the upscattering of both jet synchrotron photons as well as thermal disk photons by the quasi - thermal leptons in the base . for the assumptions in this paper ( the thin disk is recessed with inner radius @xmath65 calculated from fitted parameters @xmath76 and @xmath77 ) , the disk thermal photon field is only an important factor at the very base of the jets . this is because even the mild beaming in the jets quickly serves to weaken the disk photon field s energy density compared to the locally produced synchrotron photons . once the acceleration zone is reached , it is difficult to suppress x - ray synchrotron radiation from the tail of accelerated leptons in the diffusive acceleration case ( see * ? ? ? as described above , we are relaxing the constraint on the synchrotron cutoff by allowing for an alternate acceleration process . in this scenario , a power - law of synchrotron radiation will contribute significantly at least to the soft x - ray band , but not necessarily to the hard x - rays which are in this model dominated by inverse comptonized disk and synchrotron photons . this distinction is important because , as shown by @xcite , a very typical hard state model where the x - rays are exclusively due to synchrotron radiation can not easily reproduce disk reflection fractions greater than a few percent ( assuming perpendicular geometry and no disk flaring ) . the larger distance of the first synchrotron emission region @xmath38 from the disk , combined with moderate beaming , reduces its importance for disk feedback . on the other hand , inverse compton processes near the base of the jets can produce reflection fractions of up to @xmath9320% . this value is actually a lower limit because it does not include the effects of light bending , which will serve to increase the reflection fraction possibly quite significantly @xcite beyond relaxing the assumption about the acceleration rate , there are a few other differences between this model and the synchrotron - dominated models considered in @xcite and @xcite . the comparatively weak inverse compton component in prior models followed mainly from our choice of a small jet nozzle radius , @xmath94 , which was based on the assumption that the jet radius was on the order of the event horizon . if , on the other hand , the jet base is contiguous with , or generated in , an extended corona , a larger scale seems more sensible . with this in mind , we here consider models with with typical values of @xmath95 . the dependence of the calculated spectrum upon the model parameters , and their interdependence , are complex . increasing the scale of the jet base decreases the electron density as well as the magnetic field , for a fixed equipartition relationship . this allows one to consider electron temperatures several times higher than those used in our previous models ( to make up for lost synchrotron flux ) . the higher electron temperatures lead to greater inverse compton emission relative to synchrotron processes in the x - ray band . alternately , one can compensate for the weakened radiating power of the decreased particle density by increasing the total power input into the jet , @xmath96 . again , for a fixed equipartition relationship , the particle density is more sensitive to this change than the magnetic field ( because the pressure of the field is @xmath97 ) and thus the inverse compton component will experience a greater boost relatively . , k. , swank , j. h. , giles , a. b. , stark , m. j. , strohmayer , t. , zhang , w. , & morgan , e. h. 1996 , in proc . 2808 , p. 59 - 70 , euv , x - ray , and gamma - ray instrumentation for astronomy vii , oswald h. siegmund ; mark a. gummin ; eds . , 5970 rccccccccc 20181 - 01 - 02 & & _ 7.9 _ & & & & & & _ 25 _ & + 40108 - 02 - 01 & & _ 4.7 _ & & & & & & _ 302 _ & + 40108 - 02 - 03 & & _ 2.1 _ & & _ 4.0 _ & & _ 2.2 _ & _ 0.7 _ & _ 87 _ & + 40099 - 01 - 19 & & _ 7.4 _ & & & & & & _ 16 _ & + 60090 - 01 - 26 & & _ 7.1 _ & & & & & & _ 9 _ & + 60090 - 01 - 41 & & _ 6.4 _ & & & & & & _ 14 _ & + rcccccccccl 20181 - 01 - 02 & & & & & _ 486 _ & & & & & 179/109 + 40108 - 02 - 01 & & & & & _ 0.04 _ & & & & & 118/87 + 40108 - 02 - 03 & _ 0.57 _ & _ 230 _ & _ 0.1 _ & _ 0.36 _ & _ 0.4 _ & & & & & 22/36 + 40099 - 01 - 19 & & & & & _ 0.5 _ & & & & & 225/154 + 60090 - 01 - 26 & & & & & _ 0.1 _ & & & & & 207/177 + 60090 - 01 - 41 & & & & & _ 0.2 _ & & & & & 186/179 + rccccccccccl 20181 - 01 - 02 & & & & & & & & & & & 170/111 + 40108 - 02 - 01 & & & & & & & & & & & 117/90 + 40108 - 02 - 03 & & & & & & & & & & & 17/37 + 40099 - 01 - 19 & & & & & & & & & & & 226/158 + 60090 - 01 - 26 & & & & & & & & & & & 191/181 + 60090 - 01 - 41 & & & & & & & & & & & 186/183 +

we present results of near - ir spectroscopic measurements of 7 star - forming galaxies at . drawn from a large spectroscopic survey of galaxies photometrically pre - selected by their colors to lie at , these galaxies were chosen for their bright rest - frame optical luminosities ( ) . most strikingly , the majority of the sample of 7 galaxies exhibit [ nii]/ nebular emission line ratios indicative of at least solar h ii region metallicities , at a lookback time of 10.5 gyr . the broadband colors of the-bright sample indicate that most have been forming stars for more than a gyr at , and have already formed stellar masses in excess of . the descendants of these galaxies in the local universe are most likely metal - rich and massive spiral and elliptical galaxies , while plausible progenitors for them can be found among the population of lyman break galaxies . while the-bright galaxies appear to be highly evolved systems , their large luminosities and uncorrected star - formation rates of indicate that active star formation is still ongoing . the luminous uv - selected objects presented here comprise more than half of the high - redshift ( ) tails of current-band - selected samples such as the k20 and gemini deep deep surveys .

astro-ph0405187

rest - frame optical emission lines from extragalactic h ii regions provide an important diagnostic of galactic chemical evolution and therefore serve as an independent gauge of the history of star formation in the universe . furthermore , the chemical evolution of galaxies and of the igm are interconnected processes since feedback from star formation in galaxies may be responsible for polluting the igm with heavy elements over a wide range of redshifts @xcite . now , with the advent of sensitive near - infrared spectrographs on @xmath11 m class telescopes , it is possible to measure rest - frame optical nebular emission lines out to @xmath12 , and therefore to trace galactic chemical evolution over the bulk of cosmic history . in the local universe , star - forming spiral and irregular galaxies display a strong correlation between their h ii region oxygen abundances and optical luminosities , according to which more luminous and massive galaxies exhibit a higher degree of metal enrichment @xcite . an analogous correlation is observed in elliptical galaxies , where chemical abundances are measured from the strengths of stellar magnesium and iron absorption line indices @xcite . as we move to higher redshifts , the metallicity - luminosity relationship may evolve in both zero point and slope @xcite . using a sample of 64 star - forming galaxies at @xmath13 drawn from the deep extragalactic evolutionary probe groth strip survey ( dgss ) , @xcite demonstrate that galaxies at @xmath14 are @xmath15 magnitudes more luminous than local galaxies of similar metallicity . in contrast , @xcite find that a majority of a sample of 66 star - forming galaxies at @xmath16 drawn from the canada - france redshift survey ( cfrs ) have similar metallicities to galaxies in the local universe with comparable luminosities . at much higher redshift , @xmath17 , most of the information about the abundances of galaxies has been derived from absorption lines in damped lyman alpha systems ( dlas ) ( e.g. , * ? ? ? * ; * ? ? ? . however , relating line - of - sight dla absorption abundances to abundances measured from global galactic h ii emission - line spectra is not trivial , at least in part because the relationship of dlas themselves to high - redshift uv - selected galaxies has not been established . there are only limited data on the h ii region oxygen abundances of star - forming galaxies at @xmath18 . a sample of six galaxies has been assembled with near - infrared spectroscopic observations of the strong nebular emission lines that enable abundance estimates @xcite . while there are only rough constraints on the abundances in these galaxies , with @xmath19 , they are seen to be @xmath20 magnitudes more luminous than local galaxies of similar metallicity . until now , there have been virtually no observations of the h ii region metallicities in star - forming galaxies at @xmath21 , even though this critical epoch may host the production of a large fraction of the heavy elements present in the local universe . galaxies in the `` redshift desert '' range have only recently become accessible for detailed study due , in large part , to improvements in both near - uv and @xmath22 spectroscopic capabilities on @xmath11 m class telescopes . employing new technologies , several current surveys utilizing complementary selection techniques contain spectroscopically confirmed galaxies at these redshifts . the selection criteria include optical colors @xcite , @xmath10-band magnitude @xcite , @xmath23 color @xcite , and submillimeter flux @xcite . a complete picture of the galaxy population at @xmath24 will require an understanding of how these several selection techniques overlap and complement each other . as shown by @xcite and @xcite , h ii region metallicities can be measured in redshift desert galaxies using the ratio of the [ nii ] @xmath25 and @xmath4 nebular emission lines . here we present near - infrared ( @xmath10-band ) spectroscopic observations of [ nii ] and @xmath4 emission for a small sample of uv - selected star - forming galaxies at @xmath26 . these galaxies were selected for observation because they are luminous in the rest - frame optical ( @xmath3 ) , and therefore application of the same observational strategy yields much higher signal - to - noise ( s / n ) spectra than previous @xmath2 near - infrared spectroscopic observations of fainter galaxies @xcite . along with the spectroscopic observations , which probe the metal content of the galaxies , we also use broad - band optical / near - infrared colors to estimate their stellar populations and masses . with such information , it is possible to consider the build - up of stellar mass in the early universe , and the epoch at which the progenitors of massive galaxies in the local universe were mostly assembled . the sample of @xmath2 galaxies and near - infrared spectroscopic observations are described in [ sec : sample ] . in [ sec : results ] , the results of the @xmath4 and [ nii ] spectroscopic measurements are presented . oxygen abundances are derived from these results in [ sec : abundance ] . in [ sec : discussion ] , the discussion is extended to the galaxy metallicity - luminosity relationship , the stellar populations and masses of the luminous @xmath2 sample , a comparison of this sample with other recent surveys , and a comment on inferring stellar masses from population synthesis modeling versus dynamical masses from @xmath4 line - widths . finally , in [ sec : summary ] , we summarize our main conclusions . for all luminosity , star - formation rate , and stellar mass calculations , we adopt a cosmology with @xmath27 , @xmath28 , and @xmath29 . llllllccccc q1623-bx274 & 16 25 38.202 & 26 45 57.141 & 2.415 & 2.408 & 2.4100 & 23.23 & 0.25 & 0.89 & 3.66 & 3@xmath30900 + q1623-md66 & 16 25 40.392 & 26 50 08.878 & & 2.111 & 2.1075 & 23.95 & 0.37 & 1.40 & 4.04 & 3@xmath30900 + q1623-bx341 & 16 25 43.554 & 26 46 36.942 & & 2.377 & & 24.83 & 0.47 & 0.90 & 4.81 & 2@xmath30900 + q1623-bx344 & 16 25 43.931 & 26 43 41.977 & & 2.422 & 2.4224 & 24.42 & 0.39 & 1.25 & 4.41 & 2@xmath30900 + q1623-bx453 & 16 25 50.836 & 26 49 31.399 & 2.183 & 2.171 & 2.1816 & 23.38 & 0.48 & 0.99 & 3.65 & 3@xmath30900 + q1623-bx513 & 16 25 55.856 & 26 46 50.304 & 2.249 & 2.244 & 2.2473 & 23.25 & 0.26 & 0.68 & 3.33 & 2@xmath30900 + q1623-bx528 & 16 25 56.439 & 26 50 15.444 & & 2.266 & 2.2682 & 23.56 & 0.25 & 0.71 & 3.87 & 4@xmath30900 + q1623-bx599 & 16 26 02.545 & 26 45 31.900 & & 2.329 & 2.3304 & 23.44 & 0.22 & 0.80 & 3.65 & 4@xmath30900 + q1623-bx663 & 16 26 04.576 & 26 48 00.202 & 2.435 & & 2.4333 & 24.14 & 0.24 & 1.02 & 4.26 & 3@xmath30900 + the objects discussed here were drawn from the survey presented by @xcite . this survey photometrically identifies star - forming galaxies at @xmath2 by their locations in @xmath31 vs. @xmath32 color space . objects satisfying the optical color criteria presented in @xcite have been followed up spectroscopically using the low resolution imaging spectrometer on the keck i telescope @xcite . to date , 692 galaxies have been spectroscopically confirmed with @xmath33 . the survey has been conducted in seven separate fields , five of which contain one or more bright background qsos at @xmath34 in order to study the evolving relationship between star - forming galaxies , gas and metals in the intergalactic medium ( igm ) . the remaining two pointings are the goods - n and westphal / groth fields , both of which have extensive multi - wavelength datasets that are or will soon be publicly available . in addition to optical imaging and spectroscopic data , in june and october 2003 we obtained near - infrared @xmath5-band photometry for galaxies in three of the survey qso fields using the wide field infrared camera ( wirc ) on the palomar 5.1 m hale telescope . @xmath5 magnitudes have been measured for 283 spectroscopically confirmed galaxies at @xmath35 , with @xmath36 . one of these three fields , q1623 ( 1623 + 27 ) , contains 167 spectroscopically confirmed galaxies at @xmath35 in the region covered by wirc , 121 of which were detected at @xmath5 . at the redshifts probed in the sample , @xmath5-band corresponds most closely to rest - frame optical @xmath37-band , and for @xmath38 @xmath5-band spectroscopic observations cover the @xmath4 and [ n ii ] nebular emission lines . using nirspec @xcite on the keck ii telescope , we have assembled a sample of more than 50 @xmath4 measurements in the @xmath5-band @xcite . since the majority of these objects were chosen primarily for their close proximity to a qso sightline ( see * ? ? ? * for a discussion ) , the nirspec sample contains a fairly unbiased view of the range of rest - frame uv and optical properties of galaxies in the @xcite survey . with this unbiased sample ( with respect to photometric properties ) in place as a control , we used the @xmath5 photometry to select galaxies for nirspec rest - frame optical spectroscopic follow - up . in figure [ fig : krk ] , we present the distribution of @xmath5 magnitudes and @xmath39 colors for spectroscopically confirmed galaxies in the q1623 field . as shown in the figure , 10% of the galaxies in the q1623 field with wirc photometry have @xmath3 , ( and would be included in the @xmath40 survey @xcite , the high - redshift tail of which overlaps with the uv - selected sample here under consideration ) , while 8% have @xmath41 . in order to measure the rest - frame optical spectra of objects with bright rest - frame optical luminosities , we targeted galaxies with @xmath3 . the keck ii / nirspec observations were conducted during the course of a single observing run spanning 9 13 september 2003 . we targeted a total of nine spectroscopically confirmed @xmath2 objects in the q1623 field satisfying @xmath3 , four of which also had @xmath41 . all but one of these galaxies satisfy the `` bx '' color criteria presented in @xcite , while the remaining galaxy falls in the adjacent region of @xmath1 color space occupied by `` md '' objects @xcite . the single md object in the sample falls on the border between the two color criteria , lies at the same redshift as the other eight galaxies , and is therefore treated identically for the remainder of the analysis and interpretation . we followed the same observational procedures described in detail in @xcite . in brief , we used the nirspec6 filter , which transmits from @xmath42 @xmath43 m . since the medium - dispersion mode in which we were observing only allows for @xmath44 @xmath43 m of coverage to fall on the nirspec detector at once , we set the nirspec cross - disperser angle to cover @xmath45 @xmath43 m , which encompassed @xmath4 and [ nii ] for the redshifts of all of our target galaxies . for all objects except bx344 , we used a 076 @xmath30 42 long - slit , affording a spectral resolution of @xmath46 , whereas for bx344 , we used a 056 @xmath30 42 long - slit , affording @xmath47 spectral resolution , as measured from the widths of skylines . the same method of blind - offset object acquisition was used as described in @xcite . in most cases , the slit position angle was determined by the attempt to fit two galaxies on the slit ( usually our primary near - ir - selected target and a nearby companion with or without a keck i / lris - confirmed redshift ) , though in the case of bx513 , which appeared elongated in the wirc @xmath5 image , the slit position angle was chosen to lie along the major axis of the galaxy light . targets were observed for 2 , 3 , or 4 @xmath30 900 seconds . we detected eight out of nine of our target galaxies , and the one undetected galaxy , bx341 , was only observed for @xmath48 seconds . conditions were photometric throughout the run with approximately 05 fwhm seeing in the nirspec6 band . a summary of the observations including target coordinates , redshifts , optical and near - ir photometry , and total exposure times is given in table [ tab : obs ] . the two - dimensional galaxy spectral images were then reduced , extracted to one dimension along with 1-@xmath49 error spectra , and flux - calibrated with observations of a stars , according to the procedures outlined in detail in @xcite . the one - dimensional , flux - calibrated nirspec spectra are shown in figure [ fig : halphaspec ] . @xmath4 and [ nii ] @xmath50 6584 emission are very significantly detected in all spectra except that of bx513 . the redshift of this object places the @xmath4 emission line directly on top of a sky oh emission feature at 2.1318 @xmath43 m , thus making flux measurements extremely uncertain . we therefore exclude bx513 from all subsequent analysis , leaving us with a robust sample of seven galaxies . is not detected in the bx513 spectrum , the limits on [ nii]/@xmath4 are not very restrictive , given the low s / n and sky contamination of the @xmath4 line . ] [ sii ] emission is also significantly detected in the spectra of md66 and bx453 , and marginally detected for bx274 , bx344 , bx599 , and bx528 . @xmath4 and [ nii ] emission line fluxes were determined by first fitting a one dimensional gaussian profile to the higher s / n @xmath4 feature to obtain the redshift and fwhm ( in wavelength ) . the @xmath4 redshift and fwhm were then used to constrain the fit to the [ nii ] emission line . this procedure is based on the assumption that the @xmath4 and [ nii ] lines have exactly the same redshift and fwhm , but that the @xmath4 line offers a higher s / n estimate of these parameters . such an assumption appears to describe all of the spectra well except that of bx453 , in which the [ nii ] line has broader wings than the @xmath4 line , and the fit based on the @xmath4 parameters underestimates the [ nii ] flux by @xmath51% . @xcite find an average velocity dispersion of @xmath52 km s@xmath53 for the sample of 61 uv - selected @xmath2 galaxies with @xmath4 measurements from which the current sample is drawn . five of the seven @xmath5-selected objects have @xmath54 km s@xmath53 , though two of them , bx453 and bx344 , have velocity dispersions significantly below the @xcite average . we list the measured @xmath4 velocity dispersions , @xmath4 and [ nii ] fluxes , [ nii]/@xmath4 flux ratios , and random statistical uncertainties on these quantities in table [ tab : abun ] . even with relatively short exposure times , we obtained highly significant detections of @xmath4 emission for our target objects . the mean @xmath4 flux is @xmath55 with a standard deviation among the seven galaxies of @xmath56 ; the corresponding average @xmath4 star - formation rate is @xmath57 using the calibration of @xcite . these values are almost three times larger than the average @xmath4 flux and star - formation rate of the sample considered in @xcite . the average rest - frame optical continuum luminosity for these galaxies is also roughly three times brighter than that of the @xcite sample , so the average @xmath4 equivalents widths of the two samples are very similar . even after correcting the broadband @xmath5 magnitudes for the contribution from @xmath4 emission lines , five of seven galaxies in our sample would still have been selected as @xmath58 , whereas the remaining two galaxies , bx344 and md66 , would have almost made the cut , with @xmath59 and @xmath60 , respectively . in other words , for five of the seven galaxies in the sample presented here , the @xmath4 emission line contributes less than 0.15 of the flux in the @xmath5-band . for the remaining two galaxies , the @xmath4 line contributes between 0.15 and 0.20 of the @xmath5-band flux . most of the galaxies have spatially unresolved @xmath4 profiles , with the exception of bx528 and bx663 . as shown in figure [ fig : ha2d ] , bx528 has a spatially resolved and tilted @xmath4 line , even though the slit position angle was not selected to lie along any morphologically - preferred axis . the extent of the velocity shear in bx528 is @xmath61 km s@xmath53 . bx663 has a double - peaked @xmath4 emission line , with a velocity separation of @xmath62 km s@xmath53between the two peaks . the higher - redshift @xmath4 component is @xmath63 times more luminous than the lower - redshift component and fairly extended , with detected flux extending @xmath64 along the slit . while there is a tentative detection of [ nii ] for the fainter , lower - redshift @xmath4 component , we only list the [ nii]/@xmath4 ratio for the more - significantly detected , higher - redshift component . the @xmath4 flux for bx663 listed in table [ tab : abun ] represents the sum of the two components . the most striking result , shown in figure [ fig : halphaspec ] and table [ tab : abun ] , is the high [ nii]/@xmath4 ratios measured for the @xmath5-selected galaxies . the average ratio is @xmath65}/{\,{\rm h\alpha}}= 0.27 $ ] , and the measured values range from @xmath66 . it is relevant to consider the lowest [ nii]/@xmath4 ratio that could have been measured , given an @xmath4 s / n . in the sample of @xcite , while [ nii ] is detected for only one object ( q1700-bx691 , which has an above - average @xmath4 flux ) , the [ nii ] non - detections for the remainder of the sample are not very restrictive on the value of [ nii]/@xmath4 . since the @xmath4 emission lines in the @xcite sample are typically detected with a significance of @xmath67 times the random noise , [ nii ] will only be significantly detected if it attains @xmath68 the strength of @xmath4 . to estimate the typical [ nii]/@xmath4 for the galaxies in @xcite , we constructed an average composite spectrum from the individual nirspec spectra , and found @xmath69}/{\,{\rm h\alpha}}\rangle = 0.10 $ ] . in the @xmath5-selected sample , the higher significance of the @xmath4 detections allows for a more sensitive probe of individual galaxies [ nii]/@xmath4 ratios . we could have detected @xmath65}/{\,{\rm h\alpha } } < 0.10 $ ] for bx453 , md66 , and bx599 , and @xmath65}/{\,{\rm h\alpha}}= 0.12 - 0.13 $ ] for bx274 , bx344 , and bx663 . bx528 is the only galaxy whose [ nii]/@xmath4 ratio of @xmath70 is at the limit of what we could have detected . lccrcccc q1623-bx274 & 2.4100 & @xmath71 & @xmath72 & @xmath73 & @xmath74 & @xmath75 & @xmath76 + q1623-md66 & 2.1075 & @xmath77 & @xmath78 & @xmath79 & @xmath80 & @xmath81 & @xmath82 + q1623-bx344 & 2.4224 & @xmath83 & @xmath84 & @xmath85 & @xmath86 & @xmath87 & @xmath88 + q1623-bx453 & 2.1816 & @xmath89 & @xmath90 & @xmath91 & @xmath92 & @xmath93 & @xmath94 + q1623-bx513 & 2.2473 & & @xmath95 & @xmath96 & @xmath97 & @xmath98 & @xmath99 + q1623-bx528 & 2.2682 & @xmath100 & @xmath101 & @xmath102 & @xmath103 & @xmath104 & @xmath105 + q1623-bx599 & 2.3304 & @xmath106 & @xmath107 & @xmath108 & @xmath109 & @xmath110 & @xmath111 + q1623-bx663 & 2.4333 & @xmath112 & @xmath113 & @xmath114 & @xmath115 & @xmath116 & @xmath117 + the most robust measurements of chemical abundances from nebular emission lines are derived from the temperature - sensitive ratio of an upper- and intermediate - level transition from the same ion . the ratio of transitions such as [ oiii ] @xmath118 to [ oiii ] @xmath119 , indicates the nebular electron temperature ( @xmath120 ) , and , along with the electron density , yields a direct estimate of the chemical abundance . unfortunately , the relatively weak [ oiii ] @xmath118 line is only observed in the hottest , most metal - poor h ii regions , and is not detected in more metal - rich environments . to address this issue , h ii region photoionization models that predict the relative strengths of @xmath4 , @xmath121 , and strong forbidden lines of [ oii ] , [ oiii ] , [ nii ] , [ sii ] , and [ siii ] , have been calibrated against h ii regions with direct abundance estimates @xcite . with these empirically calibrated models , it is possible to infer approximate chemical abundances over a wide range of metallicity , using only the ratios of strong , easily detectable h ii region emission lines . the most commonly - used strong - line abundance indicator is the @xmath122 method , which relates the nebular oxygen abundance to the ratio of ( [ oii ] + [ oiii])/@xmath121 @xcite . the @xmath123 method has been employed to determine the metallicities of individual h ii regions in the local universe , and the average metallicities of entire star - forming galaxies in the distant universe @xcite . the double - valued nature of the @xmath123 calibration limits its utility unless an independent line - ratio can be used to determine whether a galaxy is on the upper , metal - rich branch or the lower , metal - poor branch . furthermore , in objects for which [ oiii ] @xmath118 or other temperature - sensitive auroral lines are detected , metallicities inferred from @xmath123 are systematically higher by @xmath124 dex than those computed directly from the @xmath120 method @xcite . as presented in section [ sec : results ] , we have measured [ nii]/@xmath4 from the sample of uv - selected star - forming @xmath2 galaxies . this line ratio can also be used to infer approximate chemical abundances . unlike @xmath123 , the [ nii]/@xmath4 ratio increases monotonically with increasing metallicity ( at least up to solar metallicity ) . another benefit of the [ nii]/@xmath4 ratio is the small difference in wavelength between the two features , which reduces the importance of accurate flux - calibration and dust extinction correction . the utility of the @xmath125 " index @xmath126}\lambda 6584/{\,{\rm h\alpha } } ) ) $ ] has recently been revisited empirically by @xcite , building on work by @xcite . @xcite demonstrate theoretically how @xmath125 depends on metallicity using detailed photoionization modeling . the @xmath125 metallicity dependence stems from several effects . as the metallicity increases , the ionizing stellar spectrum softens , and the overall ionization decreases , as demonstrated by the increasing ratio of [ oii]/[oiii ] . as the ratio of [ oii]/[oiii ] increases , so does the ratio of [ nii]/[niii ] . furthermore , at ( o / h ) metallicities above @xmath127 , the ratio of n / o increases due to the secondary nature of nitrogen in this regime @xcite . limitations of the @xmath125 method include the fact that @xmath125 is sensitive to the value of the ionization parameter as well as to metallicity . also , the n / o ratio can vary as a function age or star - formation history ( not simply abundance ) . finally , @xcite show with detailed photoionization models that while @xmath125 is a monotonically increasing function of metallicity at less than or equal to solar ( o / h ) , it saturates at solar metallicity and even turns over towards lower values ( at @xmath128 ) . despite these limitations , the @xmath125 indicator offers us rough estimates of the h ii region abundances , especially using the latest empirical calibration of @xcite . this calibration only includes ( o / h ) determinations from the direct @xmath120 method or detailed photoionization modeling . the linear relationship between @xmath129 and @xmath125 thus derived is : @xmath130 for any individual @xmath125 measurement , equation [ eq : n2 ] should yield the correct ( o / h ) abundance to within @xmath131 dex ( 68% confidence ) . using this relationship , and assuming its validity for interpreting the integrated spectra of high - redshift star - forming galaxies , we determine ( o / h ) abundances for the sample of @xmath2 galaxies with nirspec [ nii]/@xmath4 measurements . the ( o / h ) abundances and associated uncertainties are listed in table [ tab : abun ] . for comparison with other recent work , we also list the ( o / h ) abundances implied by the relation presented in @xcite @xmath132 which are systematically higher than those computed from the @xcite relation . in the remaining discussion , however , we emphasize the results from the @xcite calibration , since it is based on a more homogeneous sample of ( o / h ) abundances . all of the galaxies for which we detected [ nii ] have ( o / h ) abundances consistent with the most recent estimates of the solar abundance , @xmath133 , and that of the orion nebula , @xmath134 @xcite . furthermore , [ nii]/@xmath4 saturates at @xmath135 and even turns over towards lower values as the increased cooling at the highest metallicities lowers the h ii region @xmath120 to the point that the collisionally - excited [ nii ] emission is reduced @xcite . since equation [ eq : n2 ] is a linear fit , which does not take this turn - over into account , the highest [ nii]/@xmath4 measurements in the current sample those of bx599 , bx453 , bx344 , and bx663 could very well correspond to super - solar metallicity values . to address this uncertainty , @xcite propose the use of the @xmath136 " indicator , which consists of the ratio of two line ratios : ( [ o iii ] @xmath137)/([n ii ] @xmath138 ) . while [ nii]/@xmath4 saturates at high metallicity , [ oiii]/@xmath121 continues to decrease such that @xmath136 also continues to decrease . the monotonically decreasing behavior of @xmath136 with increasing metallicity has been calibrated from @xmath139 , and exhibits significantly less scatter than the corresponding relation using @xmath125 alone . a clear next step is to obtain @xmath140-band [ oiii]/@xmath121 spectra for the galaxies in this sample . in the preceding discussion , we have interpreted the [ nii]/@xmath4 ratios in terms of the h ii region oxygen abundances that they imply . this interpretation rests on the assumption that the observed emission lines originate in h ii regions excited by the photoionizing radiation from massive o and b stars . the contribution from diffuse ionized gas ( dig ) to the integrated emission - line spectra of local star - forming galaxies gas can enhance the [ nii]/@xmath4 ratio by different amounts ranging from less than 0.1 dex to as much as a factor of two , relative to what is measured in individual h ii regions @xcite . the dig component could potentially bias our interpretation of [ nii]/@xmath4 ratios towards higher metallicities . currently , we have no constraints on the contribution of dig to the spectra presented here , which would require spatially - resolved @xmath4 narrow - band imaging of the cosmologically - distant galaxies in our sample . other possible excitation mechanisms include photoionization by an agn power - law continuum , or shock - heating , as seen in low - ionization nuclear emission regions ( liners ) and supernova remnants . in the future , combining [ nii]/@xmath4 with measurements of [ oiii]/@xmath121 will clearly discriminate among the different excitation mechanisms for these galaxies . for now , there is no evidence pointing to an agn as the source of excitation . in addition to the nirspec rest - frame optical spectra , we have also analyzed the keck i / lris rest - frame uv spectra of these galaxies . an agn manifests itself in the rest - frame uv with strong ly@xmath141 emission , and high - ionization emission lines including si iv , ci iv , and n v @xcite . only one of the seven objects in the @xmath5 selected sample , bx663 , displays strong ly@xmath141 emission in its rest - frame uv spectrum , but no other emission lines are detected . in the remainder of the sample , ly@xmath141 is detected in either strong absorption or with an equivalent width of @xmath142 , and none of these rest - frame uv spectra show evidence for high - ionization emission lines . a consistency check on the possible agn contribution is provided by the nirspec spectra . different types of extragalactic emission - line regions can be separated according to the relative values of their emission - line strengths . for example , @xcite and @xcite show that agn and liners tend to have higher [ nii]/@xmath4 ratios ( i.e. on the order [ nii]/@xmath143 ) than star - forming galaxies , and that agn have higher [ oiii]/@xmath121 ratios than star - forming galaxies for a given [ nii]/@xmath4 ratio . while we currently do not have @xmath140-band [ oiii]/@xmath121 spectra for the @xmath2 sample , another way in which agn and star - forming galaxies can be distinguished is according to the [ oi ] 6300/@xmath4 ratio . agn and shock - heated regions have [ oi ] 6300/@xmath144 , whereas star - forming galaxies have [ oi ] 6300/@xmath4 ratios of @xmath145 . since the detection of [ oi ] at less than 0.10 times the strength of @xmath4 pushes the limit of what can be achieved with the individual galaxy spectra , we construct higher - s / n composite nirspec spectra of the entire sample of seven galaxies , and the subset of four of these with the highest [ nii]/@xmath4 ratios . [ oi ] 6300 is not detected in either case , yielding a limit of [ oi ] 6300/@xmath146 . the lack of significant [ oi ] 6300 emission in the composite nirspec spectra lends further support to the interpretation of the emission line strengths in the context of stellar h ii regions . we have presented the rest - frame optical spectra of a sample of uv - selected star - forming galaxies that are also bright in the rest - frame optical . the [ nii]/@xmath4 line ratios of the majority of these galaxies indicate that they have already been enriched to at least solar oxygen abundance at a lookback time of @xmath67 gyr . it is worthwhile to consider their metallicities and luminosities along with those of local star - forming and elliptical galaxies . over a factor of 100 in ( o / h ) abundance and 11 magnitudes in blue luminosity , nearby star - forming spiral and irregular galaxies exhibit a strong trend of higher metallicity with increasing luminosity @xcite . elliptical galaxies show a similar correlation between metallicity and luminosity , although their metallicities are measured from absorption indices in late - type stars rather than h ii emission line strengths @xcite . this correlation can be explained if less luminous ( and less massive ) galaxies have larger gas fractions , either because they have lower star - formation rates per unit mass than more massive galaxies , or because they are younger . another way of explaining the correlation is through galactic - scale outflows , which could remove a larger fraction of gas and metals from less massive galaxies , which have shallower gravitational potentials , thereby reducing the effective heavy element yield @xcite . both @xcite and @xcite have shown that @xmath147 lyman break galaxies ( lbgs ) are @xmath148 magnitudes overluminous for their metallicities , when compared with the local metallicity - luminosity relation . despite large uncertainties in lbg metallicities , the upper bound of the abundances allowed by the observed lbg @xmath123 values still falls a factor of three in metallicity below the local metallicity luminosity relationship . in figure [ fig : zmb ] we plot the local metallicity@xmath149 relationship ( with data compiled by @xcite ) along with the shaded region allowed for lbgs . in this figure , we also include the luminosity and metallicity measurements for the rest - frame optically luminous @xmath26 star - forming galaxies presented here . for both the lbgs and @xmath2 galaxies , @xmath149 was calculated using best - fit model spectral energy distributions ( see section [ sec : masses ] ) to interpolate the observed broadband @xmath150 colors to rest - frame @xmath151 . while the specific ( o / h ) values and errors implied by the linear @xcite relationship place the @xmath2 galaxies below the metallicity - luminosity relationship for local galaxies , it must also be borne in mind that our estimates of ( o / h ) represent lower limits because of the relative insensitivity of the @xmath125 index when ( o / h ) is higher than solar . furthermore , since ( o / h ) abundances in this plot are derived with the @xmath123 method for the @xmath7 and @xmath152 galaxies and with the @xmath125 method for the @xmath153 @xmath2 galaxies , it is important to acknowledge possible systematic differences between the two methods . as mentioned above , these systematic differences are definitely relevant in the super - solar metallicity regime where [ nii]/@xmath4 is no longer sensitive to increasing ( o / h ) abundance . the oxygen abundance derived from [ nii]/@xmath4 may also be @xmath154 dex lower than the abundance derived from @xmath123 at lower metallicities , where [ nii]/@xmath4 _ is _ sensitive to metallicity @xcite . to estimate ( o / h ) from the @xmath2 [ nii]@xmath4 ratios in a manner more consistent with the @xmath123 abundance determinations for local h ii regions , we examined the emission line data for a set of h ii regions located in eight galaxies in the @xcite sample . uniformly measured @xmath125 and @xmath123 values for these h ii regions are contained in @xcite . the @xmath2 galaxies in our sample have [ nii]/@xmath4 ratios that span from @xmath155 to @xmath156 . h ii regions in the @xcite sample with [ nii]/@xmath4 in this range are found to have @xmath157 , using the upper - branch @xmath123 calibration of @xcite , which corresponds to @xmath158 . while the @xmath123 calibration is in itself untested at these very high values of metallicity and may well suffer from its own systematic uncertainties , it seems very likely that the @xmath159 star - forming galaxies we have observed in this work are enriched to levels comparable to those of the most metal - rich and massive spiral and elliptical galaxies in the local universe . whether solar or super - solar , the metallicities of these galaxies are likely to grow further from @xmath159 to the present time , since we observe them while they are still in the process of actively forming stars at rates of @xmath160 . in one possible scenario , proposed by @xcite to explain the range of ( o / h ) and @xmath149 measurements of star - forming galaxies at @xmath7 , @xmath161 , and @xmath162 , the rest - frame optically bright @xmath2 galaxies in our sample will evolve both in metallicity and luminosity by @xmath162 ; their oxygen abundances will increase by a factor of two , and their luminosities will fade by @xmath163 magnitudes unfortunately , at this point , we can not precisely determine the amount of future star - formation and enrichment in these galaxies ( and therefore tracks in metallicity - luminosity space as a function of redshift ) since we have no information about their molecular gas content or future episodes of gas accretion and infall . in any case , it is important to obtain deeper [ nii ] and @xmath4 measurements of @xmath164 star - forming @xmath2 galaxies in order to determine the @xmath2 metallicity - luminosity relationship over a larger range of rest - frame optical luminosities and relate star - forming @xmath2 galaxies to those at lower redshift . as a first step , we compare galaxies in the current sample with those presented in @xcite , in terms of their rest - frame optical luminosities ( corrected for @xmath4 emission ) and [ nii]/@xmath4 ratios . the current sample of @xmath3 galaxies is characterized by @xmath165 and @xmath69}/{\,{\rm h\alpha}}\rangle=0.27 $ ] , corresponding to @xmath166 . of the 17 galaxies with @xmath4 emission - line flux measurements in @xcite , 14 also have measured @xmath5 magnitudes , from which rest - frame optical luminosities can be calculated . the @xcite galaxies with @xmath5 measurements are fainter on average than the @xmath153 sample , with @xmath167 . it was not possible to estimate [ nii]/@xmath4 on an individual basis for most of the @xcite galaxies , since [ nii ] is not significantly detected in the majority of the spectra . to determine the average [ nii]/@xmath4 , we constructed a composite nirspec spectrum from the 14 galaxies with @xmath5 measurements . in this spectrum , we found [ nii]/@xmath4=0.10 , corresponding to @xmath168 . therefore , the galaxies from @xcite are @xmath169 mag fainter in the rest - frame optical and @xmath170 dex lower in @xmath171 than the current sample . this trend marks the preliminary establishment of a metallicity luminosity relationship at @xmath2 . lrrcccccc q1623-bx274 & 2.4100 & 4.3 & 34 & 28 & 75 & 0.12 & 1.3 & 1.9 + q1623-md66 & 2.1075 & 6.5 & 51 & 10 & 65 & 0.23 & 0.9 & 0.9 + q1623-bx344 & 2.4224 & 7.9 & 62 & 8 & 49 & 0.20 & 1.6 & 1.9 + q1623-bx453 & 2.1816 & 4.9 & 39 & 16 & 174 & 0.27 & 0.4 & 0.9 + q1623-bx528 & 2.2682 & 3.0 & 24 & 18 & 44 & 0.11 & 1.7 & 1.9 + q1623-bx599 & 2.3304 & 7.6 & 60 & 22 & 49 & 0.10 & 1.3 & 1.3 + q1623-bx663 & 2.4333 & 7.8 & 62 & 12 & 33 & 0.13 & 2.0 & 2.3 + the high h ii region metallicities of the @xmath5-bright star - forming @xmath2 galaxies suggest mature systems that have processed a significant fraction of their baryonic material into stars . modeling the broad - band optical and near - infrared @xmath172 spectral energy distributions of these galaxies with population synthesis codes provides an independent test of this hypothesis . following a similar procedure to that of @xcite , we use @xcite models with solar - metallicity and a salpeter initial mass function ( imf ) to deduce stellar masses and ages . dust extinction is taken into account with a @xcite starburst attenuation law . star - forming galaxies . ] in order to determine how well the stellar masses and ages can be constrained , we investigated a range of star - formation histories of the form @xmath173 , with e - folding times of @xmath174 gyr , as well as continuous star - formation models . before modeling the colors , we corrected the @xmath5 magnitudes for the effects of @xmath4 and [ nii ] nebular emission . five of the seven galaxies show evidence of having sustained active star formation over long timescales , are not well described by steeply declining star - formation histories , and appear to have formed at least @xmath175 of stars . only @xmath176 myr models provide acceptable fits to their seds , yielding ages that are significant fractions of the hubble time at @xmath2 . in fact , while constant - star - formation models provide statistically excellent fits to the photometry for these five galaxies , the associated best - fit ages are older than the ages of the universe at each galaxy s redshift . to avoid this contradiction , we only consider models which yield best - fit ages causally allowed by the finite age of the universe . the @xmath177 myr models yield stellar masses of @xmath178 and ages of @xmath179 gyr , whereas the maximally - allowed @xmath180 models for each object yield masses of @xmath181 and ages of @xmath182 gyr . both ground - based @xmath183 and @xmath140-band , and _ spitzer _ rest - frame near - infrared photometry will help to discriminate between @xmath177 and @xmath184 gyr models . the colors of the remaining two galaxies , bx453 and md66 , provide very little restriction on star - formation history . both of these galaxies can be fit by steeply - declining or constant star - formation models . the best - fit masses for bx453 range from @xmath185 with associated ages ranging from @xmath186 gyr , whereas the masses for md66 range from @xmath187 and ages from @xmath188 gyr . while less massive than the majority of the sample , bx453 and md66 have already formed a significant fraction of the stellar mass of an @xmath189 galaxy in the local universe @xcite . for all seven galaxies , the uncertainties in star - formation history lead to systematic uncertainties in the stellar mass of a factor of @xmath163 . furthermore , if there is an underlying maximally - old population , hidden at rest - frame uv - to - optical wavelengths by the current episode of star formation , the stellar masses derived here could underestimate the total stellar mass by up to a factor of five @xcite . both the inferred dust extinction , @xmath190 , and dust - corrected star - formation rate depend systematically on the assumed star - formation histories ; specifically , the best - fit models are characterized by more dust - extinction and higher bolometric star - formation rates when we assume larger @xmath180 values . assuming the maximally - allowed @xmath180 for each galaxy yields @xmath191 for the sample , and @xmath192 . in table [ tab : sfr ] , we list the best - fit stellar - population parameters , assuming @xmath193 gyr , which provides causally allowed ages for all galaxies in the sample . figure [ fig : models ] shows plots of the best - fit @xmath193 gyr models . while these objects were selected to be at @xmath2 because they display the colors of star - forming galaxies , it is clear that they are not only metal - rich , but also have large stellar masses . the @xmath2 star - forming galaxies in the sample of @xcite were selected to have intrinsic rest - frame uv colors very similar to those of star - forming lbgs at @xmath7 . in relating the two samples of galaxies , it is important to consider if plausible progenitors of the massive , metal - rich @xmath3 `` bx '' galaxies can be found among the sample of @xmath7 lbgs with near - ir photometry , whose stellar populations were modeled by @xcite . to address this question , we re - model the @xcite sample , using the most current release of the bruzual & charlot population synthesis codes @xcite . for a consistent comparison with the best - fit values presented in table [ tab : sfr ] we assume exponentially declining star - formation histories , with @xmath193 gyr . evolving all the @xmath7 best - fit models forward in time , we compute the star - formation rates and stellar masses formed by @xmath194 . assuming that @xmath190 does not change between @xmath7 and @xmath194 , we also compute the predicted @xmath37 and @xmath5 magnitudes , and @xmath195 , @xmath32 colors . the predicted uv colors can be used to determine if a galaxy would be recovered by the @xmath2 uv criteria of @xcite , and the predicted near - ir magnitudes indicate whether or not a galaxy would be included in the rest - frame optically bright sample presented here , with @xmath3 . at least 15% of the @xmath7 lbgs fit by @xcite have stellar population models time - evolved to @xmath194 with stellar masses , star - formation rates , @xmath190 and age within the range of properties of the @xmath3 bx galaxies presented here . the time - evolved @xmath1 colors of these lbgs satisfy the uv criteria of @xcite , and they also have @xmath153 . therefore , at least 15% of the @xmath196 lbgs modeled by @xcite constitute plausible progenitors for the @xmath5-bright bx objects . as shown below , the @xmath5-bright bx s have a comoving number density that s only 14% as high as that of lbgs in the @xcite sample , so we conclude that most of them could have been lbgs in the past . it is now possible to estimate the stellar mass density associated with @xmath197 uv - selected @xmath2 objects . first we estimate the average mass of the 7 objects with [ nii]/@xmath4 measurements . taking into account the systematic uncertainties due to the range of allowed star - formation histories , we find @xmath198 . this average mass should be representative of the entire sample of @xmath3 @xmath2 galaxies since the distributions of @xmath32 and @xmath39 colors of the seven galaxies with mass estimates are similar to those of the entire sample of rest - frame optically luminous objects . rcc @xmath199 & 0/48 & 39 + @xmath200 & 6/51 & 13 + @xmath201 & 145/301 & 7 + @xmath202 & 386/556 & 8 + next we estimate the comoving space density of @xmath3 @xmath2 uv - selected galaxies . in the q1623 wirc pointing , which covers @xmath203 , there are 87 `` bx '' objects with @xmath153 , including 13 galaxies with confirmed redshifts between @xmath204 , one broad - lined qso at @xmath205 , four stars , and two low - redshift galaxies at @xmath206 . the remaining 67 photometric @xmath2 candidates have either not been attempted spectroscopically or have been attempted unsuccessfully , with undetermined redshifts . to estimate the number of @xmath153 galaxies without spectroscopic redshifts that are likely to fall in the redshift range of interest , we use the redshift statistics as a function of @xmath37 magnitude from our total sample of 956 bx objects with spectroscopic information @xcite . based on the bx high - redshift fractions and the distribution of @xmath37 magnitudes for the 67 @xmath3 sources without redshifts presented in table [ tab : rzbx ] ( i.e. multiplying the second column of table [ tab : rzbx ] by the third column ) , we estimate that an additional 11 @xmath3 galaxies without measured redshifts are likely to be at @xmath204 . the total number of @xmath3 bxs at @xmath204 is therefore @xmath207 . weighted by the bx redshift selection function @xcite , the comoving volume between @xmath204 is @xmath208 . over the entire q1623 wirc pointing , the space - density of @xmath204 uv - selected objects with @xmath3 is therefore @xmath209 . the space density of massive galaxies is of course prone to large sample variance because such objects are expected to be strongly clustered . while not yet properly quantified , initial statistics from our two other fields with @xmath5-band data suggest variations of a factor of @xmath163 from field to field . multiplying this space density by the average mass of the rest - frame optically luminous objects we obtain @xmath210 , which is 4% of the @xmath152 stellar mass density determined by @xcite . the @xmath153 objects represent only @xmath67% of the sample of bx objects at @xmath2 , and since we expect to find massive systems down to significantly fainter @xmath5 magnitudes , this figure represents an extreme lower limit on the stellar mass density associated with uv - selected star - forming galaxies at @xmath2 . in the future , when we have the distribution of stellar masses over a much wider range in luminosity in hand , we will determine a more robust constraint on the stellar mass density at @xmath159 , extending the work of @xcite to a larger cosmic volume . there are other current surveys targeting massive galaxies in the same redshift range . it is instructive to compare the properties of the @xmath3 uv - selected objects presented here with those of the galaxies in the k20 survey @xcite and the gemini deep deep survey ( gdds ; * ? ? ? the k20 project is a spectroscopic survey of @xmath211 @xmath3 objects , 9 of which have been spectroscopically confirmed in the redshift range @xmath212 . @xcite demonstrate that this subsample of galaxies is characterized by dust - corrected star - formation rates of @xmath213 , constant star - formation ages of @xmath214 gyr , and stellar masses of @xmath215 . six of the galaxies in @xcite are at @xmath216 , and four of them have optical colors that satisfy the `` bx '' criteria presented in @xcite . these four galaxies have @xmath217 and constant star - formation ages of @xmath218 gyr , very similar to the properties of bx453 and md66 when constant star - formation histories are assumed . of the three galaxies in @xcite that are at @xmath219 , two would be recovered by the `` bm '' criteria of @xcite , which target star - forming @xmath220 galaxies . since the k20 survey uses a different optical filter set from that of the @xcite survey , we calculated the @xmath1 colors of the @xcite galaxies by passing the best - fit redshifted , dust - reddened , constant star - formation models ( from their table 1 ) through the @xmath1 filter set . there appears to be significant overlap between the @xmath3 uv - selected objects and the high - redshift tail of the k20 survey . furthermore , the space density of @xmath221 k20 objects is @xmath222 , which is very comparable to the space density of @xmath3 uv - selected objects at @xmath204 in the q1623 field , @xmath223 . finally , the distribution of stellar masses of the k20 and rest - frame optically luminous uv - selected objects are very similar ( with the exception of one k20 object that has @xmath224 and would not be selected as a bx ) . the additional uv - selection criteria appear to recover the majority of k20 objects , because most @xmath225 galaxies harbor active star formation at @xmath2 . while the majority of the 300 galaxies in the gdds survey are at @xmath226 , this infrared - selected @xmath227 survey contains a small high - redshift tail of @xmath228 galaxies @xcite . using gdds galaxies to trace the evolution of stellar mass - density in the universe , @xcite find a @xmath229 mass density for @xmath230 objects of @xmath231 . below @xmath232 , the @xmath233 gdds becomes incomplete for objects with the highest mass - to - light ratios . the gdds masses and mass densities were calculated assuming an imf that turns over below @xmath234 @xcite ; it is therefore necessary to multiply the gdds numbers by a factor of 1.82 for a meaningful comparison with the results presented here , derived using a salpeter imf . the range of masses of the spectroscopically confirmed gdds @xmath229 galaxies is very similar to that of the @xmath3 uv - selected objects , while the mass density , computed down to a limiting magnitude of @xmath227 is @xmath63 times higher . a more fair comparison in mass densities will result when we have determined the stellar masses of fainter uv - selected objects . nonetheless , it is still currently possible to determine how complementary the uv - selected and gdds surveys are in the @xmath235 redshift range , since one of the gdds survey areas , the ssa22 field , has also been imaged in the @xmath1 filter set as part of the @xmath7 lbg survey @xcite . in the ssa22a field , there are three spectroscopically confirmed gdds galaxies with @xmath229 , and four with uncertain redshifts of @xmath236 @xcite . all three of the @xmath229 galaxies are recovered with the bx / bm @xmath1-selection criteria , and three of the four putative @xmath216 objects would be selected as bxs . at @xmath237 , there are 12 gdds objects in the ssa22a field , half of which would be recovered by the bx / bm uv - selection criteria . the 50% overlap between gdds galaxies at @xmath237 and bx / bm objects is not especially meaningful since the bm color criteria are intended to select galaxies at @xmath220 , and the bx color criteria target galaxies at @xmath238i.e . @xmath237 lies at the edge of the bx / bm selection function , where the selection efficiency drops off . however , the large overlap of the gdds and bx / bm surveys at @xmath235 is again consistent with the idea that even @xmath5-band selected surveys that are supposed to be `` mass - selected '' include a significant fraction of star - forming galaxies at @xmath2 . lcccccc q1623-bx274 & 121 & 0.28 & 0.20 & 1.27 ( 0.70 ) & 5.00 & 2.76 ( 1.52 ) + q1623-md66 & 120 & 0.28 & 0.05 & 0.49 ( 0.27 ) & @xmath239 & 1.19 ( 0.65 ) + q1623-bx344 & 92 & 0.16 & 0.20 & 1.06 ( 0.58 ) & 2.00 & 2.18 ( 1.20 ) + q1623-bx453 & 61 & 0.07 & 0.01 & 0.49 ( 0.27 ) & @xmath239 & 0.92 ( 0.51 ) + q1623-bx528 & 142 & 0.58 & 0.20 & 1.09 ( 0.60 ) & 2.00 & 2.47 ( 1.36 ) + q1623-bx599 & 162 & 0.50 & 0.20 & 0.83 ( 0.46 ) & 5.00 & 1.69 ( 0.93 ) + q1623-bx663 & 132 & 0.45 & 0.20 & 1.06 ( 0.58 ) & 1.00 & 2.35 ( 1.29 ) + the preceding discussion relies on population synthesis models of the galaxy colors to infer stellar masses and therefore the comoving stellar mass density in large galaxies at @xmath2 . rest - frame optical spectroscopy allows for a completely independent probe of galaxy mass : the @xmath4 velocity dispersion . a physical scale - length and assumption of the geometric configuration are necessary to convert the velocity dispersion into a dynamical mass . in previous determinations of high - redshift star - forming galaxy dynamical masses @xcite , a spherically - symmetric distribution of matter was assumed , and the physical scale - length adopted was the typical @xmath7 or @xmath2 galaxy continuum half - light radius , as determined from _ hubble space telescope _ images , @xmath2400203 . from the excellent seeing conditions during our nirspec exposures , we also determine @xmath4 half - light radii of 02 - 03 . using these radii , and assuming a spherical geometry , we determine dynamical masses with the formula : @xmath241 . dynamical mass estimates are listed in table [ tab : mcomp ] , as well as stellar masses inferred from the smallest @xmath180 allowed by the galaxy photometry , and the largest @xmath180 allowed by the age of the universe constraint . as discussed in section [ sec : masses ] , the stellar mass of the best - fit model is an increasing function of the assumed time - constant , @xmath180 . the actual stellar mass for each object lies somewhere between the limits listed in table [ tab : mcomp ] ( neglecting the mass of an underlying maximally - old burst ) . for every galaxy , if a salpeter imf extending down to @xmath242 is assumed , the allowed range of stellar mass exceeds the calculated dynamical mass within @xmath243 . for bx663 , md66 , bx528 , and bx599 , this discrepancy is roughly a factor of @xmath148 , whereas for bx344 , bx274 , and bx453 , the stellar mass estimate exceeds the dynamical mass by an order of magnitude . perhaps this discrepancy is not surprising , given that the galaxy stellar mass was determined from luminosities and colors extending over a physical region in some cases several times as large in radius as the region enclosed by the half - light radius . a larger radius may be more suitable for inferring the dynamical mass . indeed , with the assumption of a constant @xmath244 ratio as a function of radius , the mass enclosed within @xmath243 should be multiplied by 2 in order to compare with the total stellar mass . furthermore , there are observational effects that may apply if the geometry is not the simple spherically - symmetric one assumed for the calculation . if the emission is extended along a specific axis , the possible misalignment of the slit with this axis must be taken into account . if the configuration of h ii regions is more disk - like , then inclination effects may also cause us to underestimate the the circular velocity , and therefore , velocity dispersion . @xcite estimate that the combination of these two effects may result in an underestimate of a factor of @xmath245 in circular velocity . of course , the numerical coefficient multiplying @xmath246 depends on the assumed geometry as well , and is unity for a disk geometry ( as opposed to 5 , assumed for the spherical configuration ) . independent of slit misalignment and disk inclination effects , the emission within the @xmath4 half - light radius may not sample the full range of rotational velocities of the underlying disk . using local starburst galaxies with well - determined rotation curves , @xcite have shown that the widths of optical emission - lines originating in the bright central regions of local starburst galaxies only sample the solid - body part of the rotation curve , and therefore do not reflect the full circular velocity attained on the flat part of the rotation curve at larger radii . both @xcite and @xcite have pointed out this limitation in interpreting the rest - frame optical linewidths of @xmath7 and @xmath24 galaxies in terms of masses . finally , stellar masses were determined assuming a salpeter imf extending down to @xmath242 , though we have no observational constraints on the low - mass regime of the imf at high redshift . an imf which turns over below @xmath234 , as assumed by @xcite , would lower the stellar mass estimates by a factor of @xmath163 . we assumed the standard salpeter form in section [ sec : masses ] for the purposes of comparing with earlier studies of the stellar mass content of galaxies at low and high redshift @xcite , though a more realistic form for the low - mass end of the imf would significantly reduce the systematic discrepancy between stellar and dynamical masses ( as shown by the values in parentheses in table [ tab : mcomp ] ) . most of the above systematic effects would have caused us either to underestimate the dynamical mass , or overestimate the stellar mass . there is enough uncertainty in our estimates of the low - mass end of the imf , the appropriate radius at which to evaluate @xmath247 , and the geometrical configuration of h ii regions , that the current discrepancy between stellar and dynamical mass estimates does not represent a crisis . however , the method of @xcite and @xcite , using simple assumptions to compute dynamical masses , clearly does not account for the amount of stellar mass derived from modeling the colors of these galaxies . we must understand the nature of these discrepancies before we can trace the evolution of mass in galaxies as a function of redshift . we have presented the rest - frame optical spectra of a sample of seven uv - selected @xmath2 galaxies that also have @xmath153 . we reach the following principal conclusions : \2 . the calibration of @xcite , which converts [ nii]/@xmath4 ratios into ( o / h ) abundances , indicates that the galaxies in the sample are over - luminous for their metallicities relative to the local metallicity - luminosity relationship . however , we may have underestimated the ( o / h ) abundances due to the insensitivity of the @xmath125 indicator at super - solar metallicity . in any case , given the active star - formation rates inferred from @xmath4 luminosities and population - synthesis modeling , these objects will likely become more metal - rich at later times , and are probably the progenitors of massive spiral and elliptical galaxies in the local universe . the sample of @xmath2 galaxies presented in @xcite has an average rest - frame optical luminosity @xmath169 mag fainter and average [ nii]/@xmath4 ratio three times smaller , relative to the corresponding values for the current sample , consistent with the existence of a metallicity luminosity relationship at @xmath2 . population - synthesis models of their broad - band rest - uv / optical spectral energy distributions suggest that these galaxies have been forming stars over timescales comparable to a hubble time , contain @xmath225 of stars , and are still actively forming stars . comparison with surveys such as the k20 and gdds projects indicates a significant overlap between near - infrared- and uv - selected objects . a large fraction of the most massive galaxies at @xmath2 are actively forming stars . \5 . dynamical mass estimates derived from @xmath4 velocity dispersions and galaxy half - light radii are systematically lower than stellar mass estimates deduced from the broad - band seds of the galaxies presented here . while several plausible effects can be invoked to explain the difference in mass estimates , this discrepancy must be understood if we are to trace the evolution of galaxy masses in the universe . we would like to thank naveen reddy and matthew hunt for their contributions to the optical imaging and spectroscopic observations that form the backbone of our survey of @xmath2 star - forming galaxies . we appreciate the assistance of an anonymous referee , whose comments improved the manuscript . we also acknowledge extremely helpful conversations with don garnett and lisa kewley . we wish to extend special thanks to those of hawaiian ancestry on whose sacred mountain we are privileged to be guests . without their generous hospitality , most of the observations presented herein would not have been possible . finally , we thank the staff at the w. m. keck observatory for their assistance with the nirspec observations , especially joel aycock , gary puniwai , and bob goodrich . ccs and dke have been supported by grants ast00 - 70773 and ast03 - 07263 from the u.s . national science foundation and by the david and lucile packard foundation . aes acknowledges support from the miller institute for basic research in science . , s. , norberg , p. , baugh , c. m. , frenk , c. s. , bland - hawthorn , j. , bridges , t. , cannon , r. , colless , m. , collins , c. , couch , w. , cross , n. , dalton , g. , de propris , r. , driver , s. p. , efstathiou , g. , ellis , r. s. , glazebrook , k. , jackson , c. , lahav , o. , lewis , i. , lumsden , s. , maddox , s. , madgwick , d. , peacock , j. a. , peterson , b. a. , sutherland , w. , & taylor , k. 2001 , , 326 , 255 , e. , cimatti , a. , renzini , a. , vernet , j. , conselice , c. , pozzetti , l. , mignoli , m. , tozzi , p. , broadhurst , t. , di serego alighieri , s. , fontana , a. , nonino , m. , rosati , p. , & zamorani , g. 2004 , , 600 , l127 , m. , labb ' e , i. , rudnick , g. , van dokkum , p. g. , daddi , e. , f " orster schreiber , n. m. , moorwood , a. , rix , h. , r " ottgering , h. , van de wel , a. , van der werf , p. , & van starkenburg , l. 2003 , , 587 , l79 , i. s. , becklin , e. e. , bendiksen , o. , brims , g. , canfield , j. , figer , d. f. , graham , j. r. , hare , j. , lacayanga , f. , larkin , j. e. , larson , s. b. , levenson , n. , magnone , n. , teplitz , h. , & wong , w. 1998 , in proc . 3354 , p. 566 - 578 , infrared astronomical instrumentation , albert m. fowler ; ed . 3354 , 566578

global three dimensional magnetohydrodynamic ( mhd ) simulations of turbulent accretion disks are presented which start from fully equilibrium initial conditions in which the magnetic forces are accounted for and the induction equation is satisfied . the local linear theory of the magnetorotational instability ( mri ) is used as a predictor of the growth of magnetic field perturbations in the global simulations . the linear growth estimates and global simulations diverge when non - linear motions - perhaps triggered by the onset of turbulence - upset the velocity perturbations used to excite the mri . the saturated state is found to be independent of the initially excited mri mode , showing that once the disk has expelled the initially net flux field and settled into quasi - periodic oscillations in the toroidal magnetic flux , the dynamo cycle regulates the global saturation stress level . furthermore , time - averaged measures of converged turbulence , such as the ratio of magnetic energies , are found to be in agreement with previous works . in particular , the globally averaged stress normalized to the gas pressure , , with notably higher values achieved for simulations with higher azimuthal resolution . supplementary tests are performed using different numerical algorithms and resolutions . convergence with resolution during the initial linear mri growth phase is found forcells per scaleheight ( in the vertical direction ) . [ firstpage ]

1212.1306

accretion disks are ubiquitous in astrophysics and play an essential part in the formation of stars and galaxies . for accretion through a disk to be effective , angular momentum must be transported radially outwards , allowing material to move radially inwards . one means of achieving this is through viscous torques @xcite , and considerable progress has been made using the phenomenological @xmath2-viscosity introduced by @xcite which assumes that viscosity is provided by turbulent stresses which scale with the gas pressure . however , despite its successes , the @xmath2-viscosity model provides little physical insight into the mechanism(s ) responsible for the turbulent stress . even prior to the work of @xcite , instabilities in magnetized rotating plasmas had been discovered by @xcite and @xcite . yet , it was not until the seminal work of @xcite that the so - called magnetorotational instability ( mri ) received widespread attention as the agent responsible for the onset of accretion disk turbulence . linear stability analysis has shown that the mri will amplify a seed magnetic field indefinitely until confronted by the strong - field limit or the diffusion scale @xcite . non - linear stability analysis finds that growth of the magnetic field by the linear phase of the mri is likely to be truncated by saturation resulting from secondary , or parasitic , instabilities ( e.g. * ? ? ? * ; * ? ? ? * ) . that saturation of the magnetic field occurs was clearly demonstrated by even the very first shearing box simulations @xcite . contemplating the next steps in magnetized disks studies is aided by summarising what we have already learned . for example , as mentioned above , it is clear that the magnetic field reaches saturation and that the resulting maxwell stress dominates the angular momentum transport . in numerical simulations this necessitates high resolution to ensure that the fastest growing mri modes are sufficiently well resolved ( see , e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? related to this point is the importance of stratification , which introduces a characteristic length scale , removing the problem of non - convergence with simulation resolution encountered in unstratified simulations @xcite . stratification could also play a role in the dynamo process which sets the saturation stress @xcite . however , the shearing box approximation used in a large number of numerical studies to - date has limitations ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) , including the use of shearing - periodic boundary conditions in the radial direction , and/or periodic boundary conditions in the vertical direction . there boundary conditions artificially trap magnetic flux , assisting the maintenance of the turbulent dynamo and obscuring the dependence of the saturated state on resolution . this is supported by a comparison of periodic and open boundary conditions in global models by @xcite where the former were found to assist the dynamo by preventing magnetic flux from being expelled from the domain . in this regard global models have the advantage of removing the unphysical influence of the shearing box boundary conditions , albeit at a much larger computational expense . other motivations for global models are the results from stability analyses of non - axisymmetric disturbances in magnetized accretion disks , where the most robust mri modes are localized and the most robust buoyant ( parker ) modes are global @xcite . therefore , large radial extents are required to accommodate the more global modes , and in this regard there is a limit to the radial periodicity adopted in most shearing box simulations . these factors point to the need for high resolution , global , stratified disk simulations to further unravel the complexities of magnetorotational turbulence . of the global simulation studies that have been performed a large number of the findings from local models have been maintained or have persisted ; the ratio of the maxwell and reynolds stress is @xmath3 , and variations in toroidal magnetic field with time are suggestive of a dynamo cycle @xcite . however , a large number of these simulations do not start from fully equilibrium initial conditions where the magnetic field is accounted for in the force balance and in the induction equation . both local and global models started with poloidal fields which do not satisfy the induction equation show rapid disruption and re - arrangement of the disk ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? this introduces a transient phase where channel flows are fueled by rapid shearing of the poloidal field lines . as such , extended run times are required to ensure that transients have subsided . to our knowledge , no previous global simulations of the mri in stratified disks have used a fully equilibrium initial disk ( i.e. satisfying force balance _ and _ the induction equation ) . we aim to explore the influence of magnetic fields on an accretion disk with global simulations . in this first paper we present equilibrium initial disk models with arbitrary radial density and temperature profiles . we then investigate the saturation ( both locally and globally ) of the growth of magnetic field perturbations . to this end we excite the mri in global simulations using linear mri calculations as a guide , and recover growth of magnetic field perturbations in agreement with estimates . in so doing we show that non - linear gas motions saturate the initial growth of the magnetic field and that at later times the turbulent state retains no knowledge of the initially excited mri mode(s ) . the plan of this paper is as follows : in [ sec : model ] we describe the equilibrium initial conditions and details of the numerical calculations and in [ sec : excite ] we perform a linear perturbation analysis for the non - axisymmetric mri . we present a suite of global magnetized disk simulations in [ sec : global_models ] , which explore the effect of different mri mode excitation and numerical algorithms . in [ sec : discussion ] we compare our results to previous work , and then close with conclusions in [ sec : conclusions ] . for our global disk simulations , we use a 3d spherical @xmath4 coordinate system with a domain which closely encapsulates the initial disk ( e.g. * ? ? ? * ) , and we solve the time - dependent equations of ideal mhd using the pluto code @xcite . we note that throughout this work we describe our results in terms of both spherical @xmath4 and/or cylindrical @xmath5 coordinates , with @xmath6 and @xmath7 . the relevant equations for mass , momentum , energy conservation , and magnetic field induction are : @xmath8 & = & 0 , \\ \frac{\partial\rho{\bf v}}{\partial t } + \nabla\cdot[\rho{\bf vv } - { \bf bb } + ( p+\frac{1}{2}|{\bf b}|^2){\bf i } ] & = & -\rho \nabla\phi,\\ \frac{\partial e}{\partial t } + \nabla\cdot[(e + p){\bf v } - ( { \bf v \cdot b}){\bf b } ] & = & -\rho { \bf v}\cdot \nabla\phi -\rho \lambda \nonumber\\\label{eqn : energy}\\ \frac{\partial\bf b}{\partial t } & = & \nabla \times ( { \bf v \times b } ) \label{eqn : induction}. \end{aligned}\ ] ] here @xmath9 , is the total gas energy density , @xmath10 is the internal energy per unit mass , @xmath11 is the gas velocity , @xmath12 is the mass density , and @xmath13 is the pressure . we use the ideal gas equation of state , @xmath14 , where the adiabatic index @xmath15 . the adopted scalings for density , velocity , temperature , and length are , respectively , @xmath16 where @xmath17 is the speed of light , and the value of @xmath18 corresponds to the gravitational radius of a @xmath19 black hole . the gravitational potential , @xmath20 of a central point mass ( ignoring self - gravity of the disk ) , @xmath20 is modelled using the pseudo - newtonian potential introduced by @xcite : @xmath21 note that we take the gravitational radius ( in scaled units ) , @xmath22 . the schwarzschild radius , @xmath23 for a spherical black hole and the innermost stable circular orbit ( isco ) lies at @xmath24 . the @xmath25 term on the rhs of eq ( [ eqn : energy ] ) is an ad - hoc cooling term used to keep the scaleheight of the disk approximately constant throughout the simulations ; without any explicit cooling in conjunction with an adiabatic equation of state , dissipation of magnetic and kinetic energy leads to an increase in gas pressure and , consequently , the disk scaleheight over time . the form of @xmath25 is particularly simple , @xmath26 where @xmath27 and @xmath28 are the position dependent initial and current temperature , respectively , @xmath29 is the rotational velocity , and @xmath30 is the cylindrical radius . this cooling function drives the temperature distribution in the disk back towards the initial one over a timescale of an orbital period and is similar in its purpose to the cooling functions used by @xcite , @xcite , and @xcite . note that we only apply cooling within @xmath31 , where @xmath32 is the scaleheight of the disk , allowing heating via dissipation to occur freely in the corona . our choice of an orbital period for the cooling timescale is somewhat arbitrary but is chosen as it represents a characteristic timescale for the disk . the pluto code was configured to use the five - wave hlld riemann solver of @xcite , piece - wise parabolic reconstruction ( ppm - * ? ? ? * ) , and second - order runge - kutta time - stepping . in order to maintain the @xmath33 constraint for the magnetic field we use the upwind constrained transport ( uct ) scheme of @xcite . such a configuration has been shown to be effective in recovering the linear growth rates of the axisymmetric mri by @xcite . in [ subsec : alg_test ] we test a number of different numerical setups : order of reconstruction , slope limiters , and simulation resolution . however , in all of the other global simulations presented in [ sec : global_models ] we use reconstruction on characteristic variables ( e.g. * ? ? ? a courant - friedrichs - lewy ( cfl ) value of 0.35 was used for all simulations . the grid used for the global simulations is uniform in the @xmath34 and @xmath35 directions and extends from @xmath36 and @xmath37 . in the @xmath38 direction we use a graded mesh which places slightly more than half of the cells within @xmath39 of the disk mid - plane with a uniform @xmath40 , where @xmath32 is the thermal disk scaleheight , and the remainder of the cells on a stretched mesh between @xmath41 . for example , for simulation gbl-@xmath42 the 256 cells in the @xmath38 direction are distributed so that 140 cells are uniformly spaced between @xmath43 . the respective grid resolutions and number of cells per scaleheight for the three global simulations are noted in table [ tab : alg_test ] . the grid cell aspect ratio at the mid - plane of the disk and at a radius of @xmath44 ( i.e. the disk midpoint ) are @xmath45 and @xmath46 for models gbl-@xmath42 and gbl-@xmath47 , respectively . the @xmath34 and @xmath38 boundary conditions depend on whether the cell adjacent to the boundary contains @xmath48 disk material - which we determine using a tracer variable . if this constraint is satisfied we use outflow boundary conditions on all hydrodynamic variables except @xmath29 which is determined from a zero - shear boundary condition ( i.e. @xmath49 ) and the normal velocity , for which we enforce zero inflow . if the condition on disk material at the boundary is not satisfied we use outflow boundary conditions on hydrodynamic variables with the limit that the values must lie between the floor values and the initial conditions for the background atmosphere - we find this choice to be useful in setting up a steady background inflow during the early stages of the simulation before material initially in the disk evolves to fill the domain . for the magnetic field we use zero gradient boundary conditions on the tangential field components and allow the uct algorithm to calculate the normal component so as to satisfy the divergence free constraint , with the exception that at the inner radial boundary we enforce a negative magnetic stress condition ( e.g. * ? ? ? a periodic boundary condition is used in the @xmath35 direction . finally , we use floor density and pressure values which scale linearly with radius and have values at the outer radial boundary of @xmath50 and @xmath51 , respectively . motivated by the fact that magnetorotationally turbulent disks are dominated by toroidal field , we start from an analytic equilibrium disk with a purely toroidal magnetic field . the disk equilibrium is derived in axisymmetric cylindrical coordinates @xmath52 ; further details can be found in appendix [ sec : appendix ] along with alternative disk solutions which may be of use in future work . in the following we briefly summarise the equations for the isothermal in height , @xmath53 , constant ratio of gas - to - magnetic pressure , @xmath54 , net magnetic flux disk adopted for the simulations presented in this paper . the choice of temperature and magnetic field lead to a density distribution , in scaled units , @xmath55 where the pressure , @xmath56 . for the radial profiles @xmath57 and @xmath58 we use simple functions inspired by the @xcite disk model , except with an additional truncation of the density profile at a specified outer radius : @xmath59 where @xmath60 sets the density scale , @xmath61 and @xmath62 are the radius of the inner and outer disk edge , respectively , @xmath63 is a tapering function and is described in appendix [ sec : appendix ] , and @xmath10 and @xmath64 set the slope of the density and temperature profiles , respectively . in all of the global simulations @xmath65 , @xmath66 , @xmath67 , @xmath68 , @xmath69 , and @xmath70 ( consistent with the radial scaling in the gas pressure and thomson - scattering opacity dominated region from * ? ? ? * ) , producing disks with an aspect ratio , @xmath71 . the rotational velocity of the disk is close to keplerian , with a minor modification due to the gas and magnetic pressure gradients , @xmath72 where , @xmath73 one advantage using such an equilibrium disk is that one begins with a disk that is close to the expected scale height and density . an isothermal disk , for example , has a scale height that is proportional to @xmath74 . finally , the region outside of the disk is set to be an initially stationary , spherically symmetric , hydrostatic atmosphere with a temperature and density given by , @xmath75 where @xmath76 and @xmath77 is a reference radius which we take to be @xmath78 , the radius of peak disk density ( see appendix [ sec : appendix ] ) . the transition between the disk and background atmosphere occurs where their total pressures balance . as an example , model gbl-@xmath42 corresponds to a disk with a peak density of @xmath79 and a peak temperature of @xmath80 . turbulence is by its very nature chaotic . therefore , averaged quantities are particularly useful diagnostics . in this section we describe how we calculate averages , and define the variables used to analyse the simulations . to compute shell - averaged values ( denoted by curly brackets ) of a variable @xmath81 at a radius @xmath34 we average in the @xmath38 and @xmath35 directions via , @xmath82 similarly , we calculate a horizontally averaged value ( denoted by square brackets ) as , @xmath83 = \frac{\int q r \sin \theta dr d\phi}{\int r \sin \theta dr d\phi}\ ] ] to attain a volume - averaged value ( denoted by angled brackets ) we integrate over the radial profile of shell - averaged values and normalize by the radial extent , @xmath84 time averages receive an overbar , such that a volume and time averaged quantity would read @xmath85 . ( note that density - weighted averages are computed , but only for _ hydrodynamical _ variables . ) for the analysis presented in [ sec : global_models ] we restrict the integration over @xmath34 and @xmath38 to the range @xmath86 and in @xmath87 , where @xmath88 . we define this region as the `` disk body '' and limit the integration over this region to allow comparison against recent global ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) and large local simulations ( e.g. * ? ? ? * ; * ? ? ? in order to keep a track of the fluctuations in the scaleheight of the disk during the simulation - which results from the interplay between adiabatic heating and our cooling function - a density - weighted average disk scaleheight is computed , where we take @xmath89 ( where @xmath90 is the sound speed ) , then perform a density - weighted shell - average followed by a radial averaging to acquire a volume averaged value , @xmath91 . for accretion to occur , angular momentum must be transported radially outwards by turbulent stresses , and a major focus of numerical simulations is quantifying the stress . to this end , we define the perturbed flow velocity as @xmath92 with @xmath93 , and compute the @xmath94 component of the combined reynolds and maxwell stress , @xmath95 which is normalized by the gas pressure to acquire , @xmath96 furthermore , we calculate the @xmath94 component of the maxwell stress normalized by the magnetic pressure , @xmath97 to examine the operation of dynamo activity in the disk we compute the toroidal magnetic flux , which is defined as , @xmath98 the ability of the simulations to resolve the fastest growing mri modes is quantified in the same fashion as @xcite and @xcite . the wavelength of the fastest growing mri modes with respect to the grid resolution in the @xmath99 and @xmath35 directions are , respectively , @xmath100 and , @xmath101 where @xmath102 and @xmath103 are the vertical and azimuthal alfv ' en speeds , respectively , @xmath40 and @xmath104 are the cell sizes in the @xmath38 and @xmath35 directions , respectively , and @xmath105 is the corresponding cell size in the @xmath99 direction . we define a single valued measure of resolvability as the fraction of cells in the disk body ( @xmath106 ) that have @xmath107 ( e.g. * ? ? ? * ) , @xmath108 where @xmath109 and @xmath110 represents a cell . the principal aim of calculating @xmath111 and @xmath112 is to quantify how well resolved the turbulent state is in a simulation , and consequently whether global simulations are approaching the region of convergence found from shearing box simulations @xcite . to allow a direct comparison between the growth of mri modes estimated from a linear perturbation analysis ( [ sec : excite ] ) and the results of global simulations ( [ sec : global_models ] ) we analyse the growth of magnetic field perturbations in fourier space . the procedure we follow is to remap the disk body ( which we define in [ subsec : diagnostics ] ) to a cylindrical mesh with uniform cell spacing in all directions , and a sufficiently fine resolution to ensure that the smallest cells from the spherical simulation grid are sampled . we then perform a 3d fourier transform of the data on the cylindrical grid . a detailed description of the cylindrical fourier transform can be found in appendix [ sec : cft ] . given that our global simulations commence with an equilibrium disk the mri requires a seed perturbation to excite the growth of the magnetic field and development of turbulence . for this purpose we have chosen to excite a specific fourier mode of the mri using poloidal velocity perturbations . in the following we present perturbation calculations for the local , linear , non - axisymmetric mri , the results of which are used in [ sec : global_models ] to elucidate the evolution of magnetic field perturbations in global numerical simulations . .parameters used for the linear mri growth calculations [ cols="<,<,<,<,<,<,<",options="header " , ] c + c + c + in this section we describe the results of global simulations using the initial conditions and simulation setup described in [ sec : model ] . the global simulations are listed along with grid dimensions , number of cells per scaleheight , and approximate mri growth rates in [ tab : alg_test ] . time and volume averaged variables quantifying the steady - state turbulence are given in table [ tab : global_models ] . in models gbl-@xmath42 and gbl-@xmath47 we excite a specific fourier mode using a plane wave , which takes the form of eq ( [ eqn : vz_pert ] ) , as described in [ sec : excite ] . these models use the same wavenumbers as model lin-@xmath42 so as to allow a direct comparison of magnetic field growth . the third model , gbl - rand , uses random pressure and poloidal velocity perturbations to initially seed the disk disturbance . all of the global models start with a purely toroidal magnetic field with @xmath113 . models gbl-@xmath47 and gbl - rand are computed on grids with lower poloidal resolution ( roughly 2/3 that of model gbl-@xmath42 ) , but with a factor of three better azimuthal resolution . in the following section we present some properties of our model disks and demonstrate that higher @xmath35 resolution to be a crucial ingredient in producing a sustained , high valued turbulent stress , @xmath114 . we begin with a description of the evolution of models gbl-@xmath42 and gbl-@xmath47 ( tables [ tab : alg_test ] and [ tab : global_models ] ) . in this model we adopt an azimuthal wavenumber which varies with cylindrical radius , @xmath115 . we give the azimuthal wavenumber a radial dependence of @xmath116 , where the critical for which disturbances grow most rapidly , which follows from equation ( 2.30 ) of @xcite . the radial dependence of eq ( [ eqn : mcrit ] ) stems from the @xcite potential ( eq ( [ eqn : pwpot ] ) ) . ] azimuthal wavenumber for the linear , non - axisymmetric mri , @xmath117 at @xmath118 in model gbl-@xmath42 , @xmath119 ; adopting @xmath120 ensures that the corresponding @xmath121 and @xmath122 are well resolved by the numerical grid . @xcite noted that the fastest growing non - axisymmetric modes occur for @xmath123 , and we also use this relation to calculate @xmath124 . given that our grid resolution is coarser in @xmath34 than it is in @xmath38 , we set @xmath125 . the initial poloidal velocity perturbations seed the growth of magnetic field perturbations via the mri and after roughly @xmath126 turbulent motions become apparent in the disk body . as the poloidal magnetic field becomes established throughout the disk the resulting maxwell stresses disrupt the disk equilibrium . the evolution of models gbl-@xmath42 and gbl-@xmath47 is largely similar during the first few orbits of the simulations . examining the normalized stress , @xmath114 , shows that there is an initial transient phase which peaks after a simulation time of roughly @xmath127 ( fig . [ fig : global_stress ] ) . following this , @xmath114 gradually decreases until a steady - state is reached after roughly @xmath128 and the time - averaged stress for the remainder of the simulation , @xmath129 for gbl-@xmath42 . the time - averaged ratio of the maxwell stress to the magnetic energy , @xmath130 , which is below the values of roughly 0.4 quoted by , for example , @xcite for well resolved turbulence . to investigate the dependence of these values on the azimuthal resolution of the simulation we have also run model gbl-@xmath47 , which has 12.5 cells/@xmath32 in the @xmath35 direction ( and a lower resolution in the poloidal direction - see table [ tab : alg_test ] ) . the higher @xmath35 resolution clearly influences the turbulent stresses in the simulation and for model gbl-@xmath42 we find @xmath131 and @xmath132 , in agreement with high resolution shearing - box simulations . the resolvability of the fastest growing mri modes ( see eq ( [ eqn : resolvability ] ) and fig . [ fig : resolvability ] ) also clearly show that a higher azimuthal resolution helps to maintain ( or even strengthen ) the poloidal magnetic field - models gbl-@xmath42 and gbl-@xmath47 initially show similar values of @xmath111 but largely different values of @xmath112 , and combined with the evidence mentioned above is evident that azimuthal resolution is very important for maintaining a healthy turbulent state ( see also the discussion in * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) . in [ sec : discussion ] we compare further quantitative measures of the steady - state turbulence to previous works . the poloidal magnetic field develops in flux tubes with small spatial scale , which dissipate magnetic energy via reconnection , heating the disk . in fig . [ fig : h_r ] we show the density - weighted and volume - averaged scaleheight of the disk , @xmath91 as a function of time . in model gbl-@xmath42 the scaleheight of the disk increases initially until @xmath133 , after which it steadily declines . this shows that during the initial disk evolution , dissipation heats the disk more rapidly than the cooling function , @xmath25 ( see eq ( [ eqn : cooling_function ] ) ) can drive the temperature back to its initial value . in other words , the dissipative timescale is shorter than an orbital period . in contrast , for model gbl-@xmath47 , @xmath91 remains roughly constant after the initial rise , which shows that the higher @xmath114 in this model ( fig . [ fig : global_stress ] ) is causing more heating , and a marginally thicker disk . c + c + fig . [ fig : gbl_beta20 ] shows a poloidal slice through the disk in model gbl-@xmath47 at @xmath134 . during the turbulent steady state the disk is characterised by a dense , cold , sub - thermally magnetized core close to the mid - plane and a tenuous , hot , trans - to - super thermal magnetic field at @xmath135 ( the corona ) . turbulent motions are clearly evident in the plot of @xmath136 in fig . [ fig : gbl_beta20 ] with the dominant eddies appearing to have a larger size in the corona compared to the disk body . as noted by @xcite , such behaviour arises due to conservation of angular momentum in eddie motions - or wave action - as small scale eddies rise out of the dense disk mid - plane into the less dense coronal region . our intended purpose for the explicit cooling function , @xmath25 ( see [ subsec : hydromodel ] for details ) becomes more apparent from the temperature plot - we aim to take a step beyond the purely isothermal approximation and towards the observationally supported picture of a hot corona and cooler disk body . in fig . [ fig : beta ] we show the volume - averaged plasma-@xmath137 . in the disk body , we find @xmath138 and for the corona @xmath139 . the coronal value is higher than values of close to one found in previous isothermal @xcite and quasi - isothermal simulations @xcite , which may be attributable to the lack of any explicit cooling in the corona in our simulations . however , although the gas in the corona is heated by dissipation , it does not continually heat up through the simulation , and in fact remains quasi - steady through the latter half of the simulation . this contrasts with adiabatic shearing - box simulations with imposed periodic boundary conditions , in which the gas does heat up ( e.g. * ? ? ? * ; * ? ? ? * ) and demonstrates that when coronal gas is allowed to freely expand , adiabatic cooling can , to some extent , balance heating via turbulent dissipation . c + c + in fig . [ fig : bcomp ] we compare the evolution of @xmath140 for the @xmath141 , @xmath142 , @xmath143 mode ( measured in fourier space - see [ subsec : fourier ] ) for model lin-@xmath42 ( which describes linear mri growth - see tables [ tab : alg_test ] and [ tab : local_models ] ) and models gbl-@xmath42 and gbl-@xmath47 ( global simulations which allow fully non - linear evolution - see table [ tab : global_models ] ) . quantifying the initial growth by deriving approximate growth rates . ] , @xmath144 for the curves shown in fig . [ fig : bcomp ] , we find that the linear mri estimate is matched best by model gbl-@xmath47 , with gbl-@xmath42 ( and gbl - rand ) producing higher growth rates . the higher amplitude perturbation for model gbl-@xmath42 compared to gbl-@xmath47 originates from the larger amplitude initial velocity perturbation of @xmath145 ( compared to @xmath146 for gbl-@xmath47 and gbl - rand - see table [ tab : global_models ] ) . the agreement between the global simulations and linear growth estimate ( model lin-@xmath42 ) begins to falter after roughly @xmath147 and growth in @xmath148 for the global simulations levels off . from the linear mri calculations shown in fig . [ fig : local_limits ] one may anticipate that the growth in @xmath149 is halted by the magnetic pressure evolving to equipartition with the gas pressure - which is illustrated by the vertical dashed line in fig . [ fig : bcomp ] - and would mean that one can not rely on model lin-@xmath42 as a predictor for model gbl-@xmath42 . however , fig . [ fig : comp_fail ] shows that @xmath137 remains roughly constant in model gbl-@xmath42 over the first few orbits . what then causes the local mri estimates and the global simulations to diverge ? in the lower panel of fig . [ fig : comp_fail ] we plot the evolution of the following non - linear terms derived from the momentum equations , @xmath150 small spikes in these non - linear terms occur after @xmath151 orbit . however , after 1 - 2 orbits considerably larger fluctuations become apparent , particularly in @xmath152 which also appears to have the highest amplitude oscillations of the plotted terms thereafter . considering that the linear mri is seeded by velocity perturbations through the induction equation ( eqs ( [ eqn : dbrdt ] ) and ( [ eqn : dbzdt ] ) ) , the correlation in time between the non - linear velocity terms becoming active and the growth in @xmath153 departing from the linear mri growth predictions is highly suggestive of non - linear motions causing saturation in the growth of a specific mri mode . furthermore , the small amplitude kicks from these non - linear terms after 0.1 orbits may explain the early divergence between the @xmath137 values predicted from model lin-@xmath42 and those found from gbl-@xmath42 . in this sense the non - linear motions provide saturation to the initial phase of _ local _ @xmath154 growth . whether the non - linear motions are attributable to secondary instabilities feeding off the linear mri growth locally ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) , or are due to the onset of turbulence @xcite propagating radially outwards through the disk is unclear and would require an analysis of the non - linear growth of the non - axisymmetric mri , which we do not pursue here . + c + in summary , comparisons between linear growth calculations and global simulations highlights a number of potential saturation mechanisms . such as , growth of magnetic field perturbations beyond the weak field limit , and/or growth of the radial wavenumber beyond the finite limit of the simulation resolution . however , for the simulations performed in this work , we find that saturation of growth in magnetic field perturbations correlates well with the onset of non - linear motions . a major focus of magnetized accretion disk simulations is to study properties of the quasi - steady - state turbulence . a necessary test is whether the turbulent steady state depends on the mri mode initially excited , and also whether prohibitive transient behaviour arises due to the choice of exciting a specific mri mode . for this purpose we have computed model gbl - rand which uses the same initial disk as model gbl-@xmath47 with the difference that the disk is perturbed with random perturbations in the both the poloidal velocity ( amplitude @xmath155 ) and gas pressure ( @xmath156 amplitude ) which excite a range of mri modes . simulation resolution and time - averaged measures of the turbulent state are listed in tables [ tab : alg_test ] and [ tab : global_models ] , respectively . the evolution of model gbl - rand is very similar to that of model gbl-@xmath47 ; the initial perturbations excite the mri and lead to growth of @xmath157 . both models show similar growth in the @xmath141 , @xmath121 , @xmath143 mode ( fig . [ fig : bcomp ] ) which one would expect given that this mode is excited with the same amplitude perturbation . after @xmath158 ( @xmath159 ) values of @xmath157 become almost identical between the models irrespective of the differing initial perturbations . we illustrate this in figs . [ fig : kr2d_gbl - m10 + ] and [ fig : kr2d_gbl - rand ] in which we show the evolution of @xmath160 in fourier space for a range of @xmath122 values . the different panels in the figures correspond to low , moderate , and high wavenumber values for @xmath161 and @xmath121 ( relative to the size of the disk and the nyquist limit ) . as mentioned above , @xmath157 values are very similar between the two models at @xmath162 . furthermore , even though we excite a specific low wavenumber mode in model gbl-@xmath42 , a wide range of modes rapidly emerge . we attribute this behaviour to wave - wave coupling and the onset of a turbulent cascade . exciting larger wavenumbers should provide a larger _ initial _ mri growth rate ( see [ sec : excite ] ) , but how does this affect the evolution of magnetic field perturbations in the global simulations ? in particular , does the wavenumber of the excited mri mode affect the globally - averaged saturation stress ? in model gbl - rand a white noise spectrum of perturbations has been excited . therefore higher wavenumber modes can contribute to the initial growth phase in @xmath114 . there is an indication of this from fig . [ fig : kr2d_gbl - rand ] ) where the growth of @xmath153 at a range of wavenumbers means that the maxwell stress , and consequently @xmath114 will also grow across a range of wavenumbers . [ fig : global_stress ] shows that @xmath163 does grow faster for gbl - rand compared to gbl-@xmath47 ( which have identical grid resolution ) , supporting the notion that the growth in the globally averaged stress due to an ensemble of unstable modes is higher than for a single wavenumber mode . all three models start with a toroidal magnetic field with a net flux , and during the early evolution of the disk , the combination of magnetic buoyancy and accretion expels magnetic flux from the disk body such that by the time the turbulent steady state is reached the net toroidal magnetic flux of the disk , @xmath164 is close to zero . subsequently , @xmath164 oscillates about the zero - point with a period of roughly @xmath165 orbits ( upper panel of fig . [ fig : flux ] ) consistent with previous global simulations and suggestive of a dynamo cycle @xcite . all three models demonstrate this behaviour . however , minor differences in the toroidal magnetic flux , @xmath164 , are visible between models gbl-@xmath47 and gbl - rand ( fig . [ fig : flux ] ) . the different models are slightly out of phase , which is not surprising given the differences in the transient evolution at early simulation times ( fig . [ fig : global_stress ] ) . interestingly , model gbl - rand does not overshoot when expelling the initial net toroidal flux and thus settles into dynamo oscillations at a slightly earlier time , which may explain why the transient phase in @xmath114 takes a longer time to fade in this model . c + in conclusion , once the disk reaches a turbulent steady - state the disk retains no knowledge of the mri mode initially excited . this is supported by the almost identical time - averaged properties of the disk noted in table [ tab : global_models ] for models perturbed by a single low wavenumber mode or an ensemble of modes . in this section we examine the ability of different numerical algorithms to recover the growth of magnetic field perturbations resulting from the non - axisymmetric mri . comparisons between numerical simulations and analytical estimates for the _ axisymmetric _ mri have been presented by @xcite and @xcite . considering that mhd turbulence in accretion disks produces a predominantly toroidal field it is important to examine how well numerical algorithms can recover the growth of @xmath154 as a result of the _ non - axisymmetric _ mri . the setups used are listed in table [ tab : alg_test ] . the different combinations are intended to test different orders of reconstruction , parabolic limiters , and grid resolution . reconstruction refers to the order of accuracy used to interpolate cell interface values ( which are then used in the riemann solver to calculate fluxes of conserved variables ) . parabolic limiters are used to preserve monotonicity and prevent extrema from being introduced by the reconstruction step . we examine the original limiter for ppm proposed by @xcite , the extremum preserving limiters presented by @xcite , and limiters based on reconstruction via characteristic variables ( e.g. * ? ? ? the aforementioned slope limiters are respectively denoted `` cw84 '' , `` cs08 '' , and `` char '' in table [ tab : alg_test ] . the last parameter we vary is the grid resolution , as this places a constraint on the maximum resolvable wavenumber , and for this purpose we compute models gbl-@xmath42-hr , gbl-@xmath42 , gbl-@xmath42-lr , gbl-@xmath42-llr , and gbl-@xmath42-lllr ( which have decreasing resolution ) . note that with the exception of models gbl-@xmath47 and gbl - rand , all models have the same cell aspect ratio and the same ratio of cells in the disk body to cells in the corona as gbl-@xmath42 . as in [ subsec : comp ] , we calculate approximate growth rates of the magnetic field perturbation , @xmath166 for the @xmath141 , @xmath142 , @xmath143 mode via a fourier analysis of the initial simulation evolution . the results are shown in table [ tab : alg_test ] , which we summarise as follows : * the best agreement with the linear mri growth rate comes from model gbl-@xmath47 , showing that azimuthal resolution is important for properly capturing mri growth . * within errors the choice of limiter does not make a considerable difference to the resulting growth rate . * linear reconstruction produces a comparable growth rate to parabolic reconstruction . * there is a consistent trend that growth rates increase with increasing resolution ( see models gbl-@xmath42-lllr , gbl-@xmath42-llr , gbl-@xmath42-lr , gbl-@xmath42 , and gbl-@xmath42-hr ) . with 4 cells/@xmath32 in the @xmath35 direction the growth rates are converged ( within errors ) for roughly 23 - 35 cells/@xmath32 in the @xmath99 direction . this is a slightly lower threshold than the @xmath167 cells/@xmath32 ( in the vertical direction ) found to achieve convergence in the time - averaged turbulent stress in stratified shearing - box simulations @xcite . comparison with model gbl-@xmath47 ( which has 12.5 cells/@xmath32 in the @xmath35 direction ) suggests that convergence in global disks may be achieved at lower resolutions when the cell aspect ratio is closer to unity . with a growing number of studies using stratified shearing boxes with high resolution and/or a large spatial extent @xcite and higher resolution global models @xcite , quantifying the steady state turbulence and making direct comparisons between simulations provides a check of both consistency and convergence . one of the most popular measures of the steady state is @xmath168 . in this regard , models gbl-@xmath47 and gbl - rand produce values of @xmath169 which is higher than recently reported by @xcite and , as noted by these authors , higher than previous global models and a number of high resolution shearing - box simulations ( see * ? ? ? * and references there - in ) . we attribute the larger @xmath168 in our models to a higher azimuthal resolution than used by @xcite , but also note the possible indication that higher @xmath168 s may be more readily achievable in global simulations . our average @xmath170 ( for models gbl-@xmath47 and gbl - rand ) is in good agreement with the @xmath171 achieved by the highest resolution shearing - box simulations to - date @xcite . considering that our models have a lower number of cells/@xmath32 than the aforementioned shearing - box models , there may also be an indication that convergence may be achieved at lower grid resolutions than in localized models , potentially due to averaging over a larger volume , and capturing lower wavenumber eddies . comparing models gbl-@xmath42 and gbl-@xmath47 , strong evidence points to the grid cell aspect ratio and , in particular , the resolution in the @xmath35 direction as an important parameter in achieving a high @xmath168 and @xmath172 ( see the discussion in @xcite @xcite , @xcite @xcite , @xcite @xcite and @xcite @xcite ) . a possible explanation for this is that the dynamo cycle - which helps to sustain the turbulent state and involves the mri as a driving agent - can operate more effectively at higher frequencies when the cell aspect ratio is closer to unity . related to @xmath172 is the tilt angle , @xmath173 , where @xmath174 @xcite . it has been argued by @xcite that this parameter provides a better measure of convergence than @xmath168 , at least in the case of unstratified turbulence for which the question of convergence in the absence of explicit dissipation was raised by @xcite . we find @xmath175 for model gbl-@xmath42 which is consistent with previous findings for stratified global disks @xcite . however , models gbl-@xmath47 and gbl - rand have @xmath176 which is comparable to values of @xmath177 for shearing box simulations ( both unstratified and stratified , e.g. , * ? ? ? * ; * ? ? ? * ) and also for recent stratified global disks calculations performed with an orbital advection algorithm by @xcite . these results are encouraging as they show that global simulations are reaching sufficient grid resolution to reproduce shearing - box results . the ratio of directional magnetic energy also provides insight into convergence and correspondence between simulations . we find , @xmath178 , @xmath179 and @xmath180 for models gbl-@xmath47 and gbl - rand . these values are higher than obtained by @xcite for their global disk simulations , and in some cases only slightly lower than values found from high resolution shearing - box simulations @xcite . interestingly , model gbl-@xmath42 produces a sustained stress , albeit with a lower value than model gbl-@xmath47 , but with only 4 cells/@xmath32 in the @xmath35-direction . @xcite found that at least 8 cells/@xmath32 were required to produce a sustained turbulent stress ( see also * ? ? ? * ) . however , these authors used linear reconstruction , whereas we have used parabolic reconstruction which may permit a sustained stress at a slightly lower resolution . finally , we note that we do not see any prominent evidence of recurring transient phenomena due to linear growth revivals in the mean magnetic fields , as recently reported by @xcite . this may be due to differences in the numerical setup between our models and those of @xcite , or perhaps this phenomena occurs at later times that we have not reached with the simulation runtimes of our models . we have performed global 3d mhd simulations of turbulent accretion disks which start from fully equilibrium mhd initial conditions . the local linear theory of the mri is used as a predictor of the growth of magnetic field perturbations in the global simulations . additional tests have also been performed to compare results obtained from global simulations using a number of different numerical algorithms and resolutions to the linear growth estimates . our main findings are : a. the growth of magnetic field perturbations in the global simulations shows good agreement with the linear mri growth estimates during approximately the first orbit of the disk . subsequently , the overwhelming influence of non - linear motions , which may be due either to the onset of turbulence or to secondary instabilities , saturates the _ local _ growth . b. the saturated state is found to be independent of the initially excited mri mode , showing that once the disk has expelled the initial net flux field and settled into quasi - periodic oscillations in the toroidal magnetic flux , the dynamo cycle regulates the global saturation stress level . furthermore , time - averaged measures of quasi - steady turbulence are found to be in agreement with previous work . in particular , the time averaged stress , @xmath181 . c. we find @xmath178 for global stratified simulations with 12.5 cells/@xmath32 in the @xmath35 direction , which is in good agreement with value found from high resolution , stratified , shearing - box simulations . higher @xmath35 resolution in the simulation ( at least @xmath182 cells/@xmath32 ) is required to maintain stronger radial and vertical magnetic field , and consequently a larger @xmath168 . d. from the numerical algorithms that we tested , the choice of reconstruction order or limiter does not significantly alter the resulting linear mri growth rate . convergence with resolution ( for the linear mri growth tests ) is found for resolutions of roughly @xmath1cells per scaleheight ( in the vertical direction ) . however , above all , a higher azimuthal resolution contributes to a much better agreement with linear growth estimates , supporting the push for low cell aspect ratio ( close to one ) in global accretion disk simulations . we thank the referee for a particularly useful report which helped to significantly improve the paper . this research was supported under the australian research council s discovery projects funding scheme ( project number dp1096417 ) . e. r. p thanks the arc for funding through this project . this work was supported by the nci facility at the anu and by the ivec facility at the pawsey centre , perth , wa . 67 natexlab#1#1 , s. a. , & hawley , j. f. 1991 , , 376 , 214 . 1992 , , 400 , 610 . 1998 , reviews of modern physics , 70 , 1 , d. s. , & meyer , c. 2010 , arxiv:1003.0018v1 , k. , armitage , p. j. , & simon , j. b. 2011 , , 416 , 361 , g. , cattaneo , f. , ferrari , a. , mignone , a. , & rossi , p. 2011 , , 739 , 82 , g. , mignone , a. , cattaneo , f. , rossi , p. , & ferrari , a. 2008 , , 487 , 1 , a. 2005 , an , 326 , 787 , a. , nordlund , a. , stein , r. f. , & torkelsson , u. 1995 , , 446 , 741 , s. 1960 , , 46 , 253 , p. , & sekora , m. d. 2008 , , 227 , 7069 , p. , & woodward , p. r. 1984 , , 54 , 174 , j. l. , bicknell , g. v. , sutherland , r. s. , & bland - hawthorn , j. 2008 , , 674 , 157 , s. w. , stone , j. m. , & pessah , m. e. 2010 , , 713 , 52 , m. , dzyurkevich , n. , klahr , h. , & mignone , a. 2010 , , 516 , a26 , m. , dzyurkevich , n. , klahr , h. , turner , n. , & henning , t. 2012 , , 744 , 144 , m. , dzyurkevich , n. , klahr , h. , turner , n. j. , & henning , t. 2011 , , 735 , 122 , s. , & nelson , r. p. 2006 , , 457 , 343 . 2009 , , 496 , 597 , s. , & papaloizou , j. 2007 , , 476 , 1113 , t. a. , & stone , j. m. 2008 , , 227 , 4123 , p. , & lynden - bell , d. 1965 , , 130 , 125 , j. , & xu , g. 1994 , , 432 , 213 , o. 2010 , , 405 , 41 , x. , & gammie , c. f. 2011 , , 728 , 130 , x. , gammie , c. f. , simon , j. b. , & johnson , b. m. 2009 , , 694 , 1010 , j. f. 2000 , , 528 , 462 , j. f. , & balbus , s. a. 1991 , , 376 , 223 , j. f. , gammie , c. f. , & balbus , s. a. 1995 , , 440 , 742 . 1996 , , 464 , 690 , j. f. , guan , x. , & krolik , j. h. 2011 , , 738 , 84 , j. f. , & krolik , j. h. 2002 , , 566 , 164 , h. n. , lesaffre , p. , & balbus , s. a. 2009 , , 394 , 715 , g. , & longaretti , p .- y . 2007 , , 378 , 1471 , d. , & pringle , j. e. 1974 , , 168 , 603 , w. , johansen , a. , klahr , h. , & piskunov , n. 2008 , , 479 , 883 , j. c. , tchekhovskoy , a. , & blandford , r. d. 2012 , , 423 , 3083 , a. , bodo , g. , massaglia , s. , matsakos , t. , tesileanu , o. , zanni , c. , & ferrari , a. 2007 , , 170 , 228 , a. , flock , m. , stute , m. , kolb , s. m. , & muscianisi , g. 2012 , , 545 , 152 , k. a. , & stone , j. m. 2000 , , 534 , 398 , t. , & kusano , k. 2005 , , 208 , 315 , s. c. , krolik , j. h. , & hawley , j. f. 2010 , , 711 , 959 , s. m. , reynolds , c. s. , miller , m. c. , & sorathia , k. a. 2011 , , 736 , 107 , b. , & wiita , p. j. 1980 , , 88 , 23 , j. c. b. , & terquem , c. 1997 , , 287 , 771 , m. e. 2010 , , 716 , 1012 , m. e. , chan , c .- k . , & psaltis , d. 2007 , , 668 , l51 , m. e. , & goodman , j. 2009 , , 698 , l72 , w. h. , flannery , b. p. , & teukolsky , s. a. 1986 , numerical recipes . the art of scientific computing , ed . press , w. h. , flannery , b. p. , & teukolsky , s. a. , o. , & umurhan , o. m. 2008 , , 481 , 21 , w. j. , greenough , j. a. , & kamm , j. r. 2007 , , 225 , 1827 , m. m. , ustyugova , g. v. , koldoba , a. v. , & lovelace , r. v. e. 2012 , , 421 , 63 , t. , inutsuka , s .- i . , turner , n. j. , & stone , j. m. 2004 , , 605 , 321 , r. , mckinney , j. c. , narayan , r. , tchekhovskoy , a. , gammie , c. f. , & mcclintock , j. e. 2008 , , 687 , l25 , n. i. , & sunyaev , r. a. 1973 , , 24 , 337 , j. , krolik , j. h. , & hirose , s. 2010 , , 708 , 1716 , j. b. , beckwith , k. , & armitage , p. j. 2012 , , 422 , 2685 , j. b. , hawley , j. f. , & beckwith , k. 2009 , , 690 , 974 . 2011 , , 730 , 94 , k. a. , reynolds , c. s. , & armitage , p. j. 2010 , , 712 , 1241 , k. a. , reynolds , c. s. , stone , j. m. , & beckwith , k. 2012 , , 749 , 189 , j. m. , hawley , j. f. , gammie , c. f. , & balbus , s. a. 1996 , , 463 , 656 , j. m. , & pringle , j. e. 2001 , , 322 , 461 , c. , & papaloizou , j. c. b. 1996 , , 279 , 767 , c. a. , & pringle , j. e. 1992 , , 259 , 604 , e. p. 1959 , , 36 , 1398 , e. t. 2009 , , 696 , 1021 the initial conditions for our simulations are an equilibrium thin disk with a purely toriodal magnetic field . in the following we present some analytic solutions which are of use for numerical simulations of accretion disks and for disks in other environments , such as starburst galaxies ( e.g. * ? ? ? these solutions incorporate more or less arbitrary radial profiles of density and temperature and a toroidal magnetic field . the latter involves either a constant ratio of gas - to - magnetic pressure , @xmath137 , radially dependent @xmath137 , constant @xmath183 , and variants with net / zero toroidal magnetic flux . in axisymmetric cylindrical coordinates ( @xmath184 ) , in steady - state , and with @xmath185 , the induction equation is identically satisfied and we are left with the two momentum equations : @xmath186 where the pressure , @xmath56 , with @xmath187 in scaled units . we derive a compatibility condition for the above equations by subtracting @xmath188 of eq ( [ eqn : zmom_app ] ) from @xmath189 of eq ( [ eqn : rmom_app ] ) , to obtain , @xmath190 to solve for the disk equilibrium we take the approach of using eq ( [ eqn : zmom_app ] ) to derive an equation for @xmath191 , eq ( [ eqn : comp_cond_app ] ) to acquire @xmath192 , and eq ( [ eqn : rmom_app ] ) to obtain an expression for @xmath193 . the resulting equations require boundary conditions for the run of @xmath12 and @xmath187 at the disk midplane , which can be chosen arbitrarily . from here on we take the disk to be isothermal in height , @xmath53 , and firstly consider a disk with a constant , @xmath194 . eq ( [ eqn : zmom_app ] ) then becomes , @xmath195 which integrates to give an expression for the density in terms of its midplane value , @xmath196 turning to the rotational velocity , the compatibility relation ( eq ( [ eqn : comp_cond_app ] ) ) reduces to , @xmath197 which upon integrating and using eq ( [ eqn : rho2_app ] ) leads to an expression for the azimuthal velocity in terms of its midplane value , @xmath198 the model is completed with a midplane rotational velocity , which is determined by substituting eq ( [ eqn : rho_app ] ) into eq ( [ eqn : rmom_app ] ) . this gives @xmath199 the first term is the square of the keplerian velocity ; the remaining terms are proportional to the square of the sound speed so that eq ( [ eqn : vphi0_app ] ) represents a minor departure from a keplerian disk . a possible variation to the aforementioned disk would be to make @xmath137 radially dependent , i.e. , @xmath200 . for example , one may choose to make @xmath201 , where @xmath202 is a radial wavenumber . in such a case eq ( [ eqn : rho_app ] ) for @xmath191 is unchanged . however , the expression for the rotational velocity becomes , @xmath203 where , following substitution in eq ( [ eqn : rmom_app ] ) , @xmath204 alternatively , one may desire a disk with a constant @xmath205 , in which case the magnetic pressure does not influence the density profile , leading to , @xmath206 and a corresponding velocity profile of , @xmath207 with , @xmath208 and where the alfvn speed , @xmath209 . as mentioned above , the radial profiles @xmath57 and @xmath58 required to complete the disk model may be chosen arbitrarily , subject to boundary constraints at the outer disk edge . as an example , we use simple functions inspired by the @xcite disk model , modified by truncation of the density profile at a specified outer radius : @xmath210 where @xmath60 sets the density scale , @xmath61 and @xmath62 are the radius of the inner and outer disk edge , respectively , and @xmath10 and @xmath64 set the slope of the density and temperature profiles , respectively . the tapering function , @xmath211 is used to truncate the disk at an inner and outer radius . in practice this function is normalized to give @xmath212 , where the radius of peak density , @xmath78 is given by the positive root of the quadratic resulting from taking @xmath213 of eq ( [ eqn : density_profile_app ] ) , namely , @xmath214 where @xmath215 . once @xmath78 is known it is straightforward to renormalize the density profile . finally , studies of turbulent dynamos in magnetized disks are often concerned with the net flux of the magnetic field ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? for the initially purely toroidal field we have adopted in this paper the net flux of the disk is given by @xmath216 . noting that in the above derivations we have used @xmath137 to relate @xmath217 to @xmath13 , meaning @xmath218 , i.e. we are free to choose the sign of @xmath183 . therefore , if a net flux field is required then one may set the sign of @xmath183 the same everywhere , whereas if one desires a zero - net flux field then , for example , one may choose to make @xmath183 anti - symmetric about the disk midplane . we wish to evaluate the fourier transform @xmath219 of a function @xmath220 expressed in terms of cylindrical polar coordinates @xmath5 . the definition of the fourier transform is @xmath221 cylindrical coordinates in real and fourier space are expressed via the following equations : @xmath222 hence , @xmath223 and @xmath224 \ , f(r,\phi , z ) \ > r dr d\phi dz\ ] ] where @xmath225 is the computational region , usually of the form : @xmath226 we begin by constructing a fourier _ series _ in the periodic azimuthal coordinate @xmath35 : @xmath227 where the coefficients @xmath228 are given by : @xmath229 we now make the change of angular variable @xmath230 ; the integration over @xmath64 is still over the interval @xmath231 $ ] since all of the angular functions within the integrand have period @xmath232 . the fourier transform can now be expressed as : @xmath233 \ > \ > dz \right ] r dr\ ] ] the angular integral can be expressed in terms of bessel functions ( @xmath234 ) : @xmath235 hence , @xmath236 \ > r \ , dr \label{fkcyl}\ ] ] 1 . evaluate the angular coefficients : @xmath237 2 . then perform the integration in the @xmath99 direction : @xmath238 3 . finally , perform the ( truncated ) hankel transform in the radial direction : @xmath239 4 . the fourier transform of @xmath240 is @xmath241 since the input data for @xmath240 are on a grid , the azimuthal , vertical and radial wave numbers , @xmath242 and @xmath122 , are limited by the nyquist limit . let the number of intervals in each coordinate direction be @xmath243 and the grid increments be @xmath244 $ ] . the grid coordinates are @xmath245 where : @xmath246 the expressions for the azimuthal @xmath228 and vertical @xmath247 parts of the fourier transform can be approximated by discrete fourier transforms as follows : @xmath248 & \quad m & = 0,~1, ... ,~n_\phi/2 \\ f_m ( r_u , k_{\rm z } ) & \approx \frac { 2 z_0}{n_{\rm z } } \exp[-i k_{\rm z } z_0 ] \ > \sum_{w=0}^{n_{\rm z}-1 } f_m ( r_u , z_w ) \ > \exp [ 2 \pi l w / n_{\rm z } ] & \quad l & = 0,~1, ... ,~n_{\rm z}/2 & \quad k_{\rm z } & = \frac { \pi}{z_0 } \ , l \end{aligned } \label{ft_phiz}\ ] ] the radial transform can be evaluated from @xmath249

this article is aimed at improving the description of lepton - nucleus interactions in the sub - gev energy range . approximate spectral functions for oxygen , calcium , and argon are constructed and used to obtain the electron cross sections in a given scattering angle . comparison with a sample of available experimental data shows satisfactory agreement . discrepancy between the presented model and the systematic computations available for oxygen [ o. benhar _ et al . _ , phys . rev . d * 72 * , 053005 ( 2005 ) ] is also found to be very small . analysis of appropriate kinematical regions leads to the conclusion that the obtained argon spectral function should describe well neutrino scattering in the 800-mev energy region . several approximations used in the model are critically reviewed . all the details needed to implement the presented approach in monte carlo simulations are given .

0711.2031

in the impulse approximation ( ia ) , lepton - nucleus interaction is described as a two step process : in the first step , the lepton interacts with a single bound nucleon , and in the second one , the resulting particles propagate inside the nucleus . the ia formalism is the basic framework in which @xmath01-gev leptons scattering is described : on one hand , it is the approach applied to understand electron scattering , and , on the other , it is used to model neutrino interactions @xcite . the status of these two cases is quite different : for electrons a lot of experimental data exist , whereas for neutrinos precise measurements are still missing . due to this lack of knowledge , reliable theoretical models are needed in the next generation of precise neutrino oscillation experiments @xcite , possibly with liquid argon target . to construct a successful model of neutrino - nucleus scattering , the following procedure seems to be well justified : * relevant kinematical region in energy and momentum transfer has to be identified , * description of a nucleus for electron scattering should be formulated in this kinematical region , * performance of the electron scattering model must be confronted with the existing data , and , if the agreement is satisfactory , * the same treatment of nuclear effects should be applied to neutrino interactions . this is the basic logic of this paper , in which we propose a model to describe the @xmath01-gev neutrino scattering off medium - mass nuclei , such as calcium and argon . the article reports continuation of the research started in ref . , where a less sophisticated description was applied to argon , and ref . @xcite , where the model was introduced . in the ia regime , a nucleus is described by means of the spectral function ( sf ) . the sf contains information about the momentum distribution in conjunction with the distribution of binding energy of nucleons inside the the nucleus . evaluation of the sf for medium nuclei requires several approximations . in our presentation , we try to identify and justify all the theoretical assumptions , but the most important argument for the correctness of our model is the agreement of its predictions with the data for electron scattering . detailed verification of the description is performed using two targets , namely oxygen and calcium . oxygen was selected because of additional opportunity to compare results with a more systematic theoretical approach to modeling of the sf , whereas the calcium nucleus is most similar to argon s , for which precise measurements have been performed . our description of is confronted also with the theoretical results of butkevich and mikheyev . finally , a comparison with the few known data for electron scattering off argon @xcite is done as well . basic computations of quasielastic inclusive scattering are standard and can be found elsewhere . the outcome of numerical calculations depends on several assumptions which specify the implementation of the sf model . we apply the recent bbba05 parameterization of the proton and neutron form factors @xcite . the off - shell hadronic current matrix elements are evaluated with the use of the standard de forest prescription @xcite . furthermore , in the electromagnetic case , we adopt a procedure to impose electromagnetic current conservation . such prescription is not unique , and for this reason , in the case of weak interactions , we avoid analogous manipulations with the vector part of the current and rely on the de forest approach only . an important ingredient in calculations is the treatment of final - state interactions ( fsi ) . there are several approaches to deal with them , e.g. , the wentzel - kramers - brillouin method , relativistic mean - field approach , or correlated glauber approximation . in the previous article , we adopted the plane wave impulse approximation ( pwia ) and disregarded fsi beyond the pauli blocking , arguing that this approach is sufficient to describe neutrino scattering . in this paper , the presented model is validated by confronting it with a large sample of electron scattering data and this comparison requires inclusion of fsi . we consider two fsi effects : pauli blocking and reinteractions of the struck nucleon with the spectator system described by means of the time - independent optical potential . in the lepton energy range of @xmath01 gev , two dynamical mechanisms are most important : quasielastic ( qe ) scattering ( throughout this article we use the terminology of the neutrino community ) and single pion production through the @xmath1 excitation . they clearly manifest themselves as two peaks in the electron differential cross section in energy transfer for fixed scattering angle . our numerical computations include the qe dynamics only . the reason is that our concern is to define a systematic procedure to construct sf , and it is sufficient to test it in the case of the qe process . thank to this , we avoid dynamical issues in which the theoretical situation is not completely clear , namely the nonresonant background and two - particles two - holes excitations . however , we have to pay the price for the constraint on the dynamics we adopted ; in comparisons of our predictions with the experimental data , for higher values of the energy transfer some strength is systematically missing . strictly speaking , we verify our model only in the kinematical region of energy transfers below the qe peak . this paper is organized as follows . in sec . [ sec : description ] , the construction of our model is described : in sec . [ sec : general ] , we present basic formulas for the lepton - nucleus cross section and introduce notation used throughout the article . section [ sec : selection ] discusses a relation between kinematical regions in electron and neutrino scattering . in sec . [ sec : model ] , a method to approximate sfs is given . the treatment of fsi effects in our model is covered in sec . [ sec : fsi ] . section [ sec : implementation ] provides the parameterization of the sfs for oxygen , calcium , and argon . the information is detailed enough for everybody to be able to implement our results in their own numerical codes . in sec . [ sec : results ] , our results are compared to large data samples for electron scattering off oxygen , calcium , and argon , selected according to the conclusions of sec . [ sec : selection ] . our predictions are also confronted with other theoretical approaches . we observe that the performance of the presented description of nuclei is satisfactory and arrive at the conclusion that when applied to neutrino scattering , the model should produce reliable results . [ sec : precision ] is devoted to a discussion of the approximations used in this paper . we consider plausible modifications of the adopted parameters and try to understand how uncertain our results for the cross section are . finally , in sec . [ sec : summary ] , we summarize the conclusions of this article . since the most important features of the predictions seem to follow from the very basic assumption of the ia , the failure of the model in some kinematical situations may be interpreted as a failure of the ia itself . the presented results suggest that the ia starts to be unreliable when the typical value of momentum transfer is smaller than @xmath0350400 mev . we consider quasielastic ( qe ) electron scattering off a ( @xmath2 , @xmath3 ) nucleus of mass @xmath4 , which changes its four - momentum from @xmath5 to @xmath6 . associated with this interaction energy and momentum transfers are @xmath7 and @xmath8 , respectively . when the impulse approximation holds , i.e. , when only one nucleon is involved in the primary vertex , nuclear effects can be described by means of the spectral function . the proton spectral function ( sf ) of a given nucleus @xmath9 is the probability distribution of removing from this nucleus a proton with momentum @xmath10 and leaving the residual nucleus with energy @xmath11 which includes recoil energy of the residual nucleus @xmath12 , compare refs . . the neutron sf is defined in an analogous way . energy balance of qe production of a free nucleon carrying four - momentum @xmath13 , @xmath14 may be rewritten in a useful form using _ removal _ energy @xmath15 , which is an argument of the sf : @xmath16 in sec . [ sec : model ] , we will justify that the recoil energy can be neglected , and therefore from now on , the energy balance @xmath17 is used . according to the ia , the inclusive electron - nucleus cross section is the sum of contributions from protons and neutrons : @xmath18 each term is expressed by the standard formula @xmath19 where the index @xmath20 denotes the nucleon isospin . the leptonic tensor is given by @xmath21 due to negligible mass of electron , and the hadronic tensor is @xmath22 with the scalar coefficients @xmath23 and @xmath24 depending on @xmath25 and @xmath26 in the following way : @xmath27 the form factors @xmath28 are in turn expressed by the appropriate electric @xmath29 and magnetic @xmath30 form factors @xcite . to handle the problem with the off - shell kinematics , we use the de forest prescription @xcite : treat interacting nucleon as free and use free form factors but modify the energy conservation to take into account that a part of energy transferred by the probe is absorbed by the spectator system . comparing eq . to the energy balance @xmath31 where the part of energy transfer which goes to the on - shell interacting nucleon with @xmath32 is denoted by @xmath33 , one can find momentum - dependent binding energy : @xmath34 replacing @xmath35 by @xmath36 in the hadronic tensor , @xmath37 we obtain the standard description of the off - shell kinematics . however , this procedure violates the conservation of the electromagnetic current , because @xmath38 . to restore it , we have to add a correction to the contraction of the tensors : @xmath39 which is equal to @xmath40 the coefficients @xmath41 and @xmath42 can be expressed as @xmath43~,\\ c_2&=&c_1\mathcal{p}^2 + 4(\omega-\widetilde\omega)\mathcal{pq}\:{\left({\ensuremath{\mathbf{k}}}\cdot{\ensuremath{\mathbf{p}}}-\frac{{\ensuremath{\mathbf{p}}}\cdot{\ensuremath{\mathbf{q}}}}{{\ensuremath{\mathbf{q}}}^2}{\ensuremath{\mathbf{k}}}\cdot{\ensuremath{\mathbf{q}}}\right)}~,\\\end{aligned}\ ] ] with a shorthand notation introduced for @xmath44 when we consider qe muon neutrino scattering , its four - momentum is denoted as @xmath45 , four - momentum of the produced muon as @xmath46 , and @xmath47 . the cross section @xmath48 contains contraction of the leptonic and hadronic tensors @xmath49 where @xmath50 the tensors differ from the ones for electromagnetic interaction due to the axial contribution ( in our calculations axial mass @xmath51 gev ) and to the fact that , thanks to the conserved - vector - current hypothesis , @xmath52 and @xmath53 are expressed by differences of the proton and neutron form factors ; see ref . @xcite . considering neutrino interactions , we apply the de forest prescription but do not restore conservation of the vector current . all the other quantities are defined and denoted as in the case of electron interaction . according to the plan outlined in sec . [ sec : introduction ] , first we identify the region in the @xmath54 plane which is most important for qe _ neutrino _ scattering . the energy and momentum transfers are related to the muon production angle @xmath55 by the expression @xmath56 where @xmath57 . therefore fixing @xmath55 is equivalent to restricting a region in the @xmath54 plane . points in fig . [ fig : mapping ] show the neutrino differential cross section @xmath58 for neutrino energy @xmath59 gev . the peak at @xmath033@xmath60 is rather broad and @xmath050% of the cross section comes from @xmath61 $ ] . for @xmath62 gev , the maximum moves to @xmath022@xmath60 and the peak becomes narrower ( not shown in the figure ) . we want to map the allowed kinematical region for neutrino scattering , weighted by the cross section , to the corresponding region for _ electron _ scattering . for electron of energy @xmath63 , the relation analogous to eq . reads @xmath64 hence , for a given value of @xmath65 and selected @xmath63 , we can map the muon production angle to the electron scattering angle : @xmath66 to weight the electron scattering angles by the neutrino cross section , we calculated the quantity @xmath67 using the oxygen target , described by the benhar sf . ( we checked that the fermi gas model and the effective description lead to the same conclusions . ) analysis of the dependence of the @xmath68{o}(\nu_\mu,\mu^-)$ ] cross section on energy- and momentum - transfer . lines show what electron scattering angles @xmath69 in the process @xmath68{o}(e , e')$ ] correspond to the same kinematical region . the standard differential cross section in muon scattering angle for @xmath68{o}(\nu_\mu,\mu^-)$ ] is represented by points.,scaledwidth=46.0% ] from now on , we concentrate on @xmath59 mev . in fig . [ fig : mapping ] , we show the quantity for three selected values of electron energy : @xmath70 , 1.08 , and 1.2 gev . the conclusion is that to describe well the 0.8-gev neutrino scattering , our model should be verified with 1.2-gev electron data at @xmath71 , @xmath72-gev data at @xmath73 , or @xmath74-gev data at @xmath75 . let us go into more detail . we deduce that for @xmath76 gev , the range of muon scattering angle @xmath77 $ ] corresponds to @xmath78 $ ] with a maximum at 23@xmath60 ; for @xmath79 gev , to @xmath80 $ ] with a maximum at 25@xmath60 ; whereas for @xmath70 gev , to @xmath81 $ ] with a maximum at 30@xmath60 . the general rule is that for higher electron beam energies , the smaller scattering angles become significant . equation is well defined when @xmath82 . for lower @xmath63 , this equation may be applied only for the prize of a loss of normalization the form of the denominator excludes some of the points in the @xmath54 plane . for example , when @xmath83 gev is used , 5% of the strength is lost and @xmath61 $ ] corresponds to @xmath84 $ ] with a maximum at 35@xmath60 . in the case of electron scattering off oxygen , the measurements were performed for scattering angle @xmath85 using beam energies 700 , 880 , 1080 , 1200 , and 1500 mev @xcite , whereas 537- and 730-mev beams were used for angle @xmath86 @xcite . as follows from our analysis , to obtain the model which describes well qe neutrino - nucleus scattering at energy 800 mev , the most significant electron data are those for 880 and 730 mev . the relevance of the experimental points for 1080 and 700 mev is smaller . the energy and momentum transfers which characterize scattering with electron beams of energies 1200 and 537 mev are least similar to what is needed but these energies are still in the region of interest . the set of data for 1500 mev was collected at too high scattering angle for our applications . among a few papers reporting results of electron scattering experiments with a calcium target @xcite , the most suitable for testing of our model is ref . @xcite , containing data at the lowest scattering angle , namely 45.5@xmath60 , in conjunction with the highest values of beam energy up to 841 mev . we have checked that all the measurements at 45.5@xmath60 correspond to our region of interest in the @xmath54 plane . obviously , only the data for @xmath87 mev cover the whole region , and the lower electron energy is , the more normalization is lost . for example , when one uses @xmath88 mev , @xmath61 $ ] corresponds to @xmath89 $ ] with a maximum at 46@xmath60 and 27.4% of the strength is lost . therefore , we rely mainly on comparisons with the experimental data for higher electron energies . finally we want to explain why we decided to study neutrino energy @xmath59 gev . the reason is that there is a lot of relevant electron scattering data to compare with . for higher @xmath65 , say 1.2 gev , the situation would be quite different the electron - scattering data at smaller angles would be required , but they are missing for the targets we are interested in . the spectral function describes distribution of nucleons in the @xmath90 plane . by integrating out the dependence on @xmath15 , the momentum distribution @xmath91 is obtained : @xmath92 our normalization convention is @xmath93 where the number of nucleons @xmath94 is @xmath2 for protons and @xmath3 for neutrons . approximately 80%90% of nucleons in a nucleus can be described as occupying shell - model states and moving freely in the mean - field ( mf ) potential . the rest of them take part in interactions . it is natural to decompose the sf into the sum of the mf and correlated parts : @xmath95~.\ ] ] by analogy to eq . , the mf and correlated momentum distributions are introduced : @xmath96 so the momentum distribution can be described as composed of two subdistributions : @xmath97~.\ ] ] the basic assumption underlying the presented approach is the ia , therefore the mf part of the sf can be written in the form ( compare ): @xmath98 with separated contributions from each shell - model state @xmath99 , @xmath99 ranging from 1 to @xmath94 . denoting spectroscopic factor by @xmath100 , wave function by @xmath101 , level energy by @xmath102 , and a function describing level width by @xmath103 , we have omitted the isospin index @xmath20 for clarity of the notation . if interactions between nucleons disappeared , the mf part would describe the whole sf ( equivalently , all @xmath100 s would become equal to 1 ) and each @xmath103 would be the @xmath104 function . in this article we are interested in a description of medium - sized nuclei , like calcium and argon . recoil energy of the residual nucleus @xmath105 may then be neglected in the mf part of the sf since it is typically @xmath00.5 mev ( see the average mf momenta in table [ tab : momdistrib ] ) . we assume that @xmath106 , what can be physically interpreted as the momentum independence of level widths . than the mf momentum distribution can be expressed as @xmath107 let us make the _ crucial assumption _ : each level contributes equally to the mf momentum distribution . it means that in eq . for each @xmath99 we can make the substitution @xmath108 the final form of the mf part of the sf , @xmath109 have to be further specified by the form of the function which describes level width . for a given half - width , the breit - wigner distribution has longer tails then the gaussian one , so we found the latter more suitable : @xmath110 therefore we refer to the proposed model as the _ gaussian spectral function _ ( gsf ) . the factor 8 in the argument of exponential function is introduced for further convenience . note that the sum in eq . extends to _ all _ occupied states . this approach differs from the one presented in ref . and allows to avoid singularities in the argon sf . to describe a specific nucleus by its gaussian sf , one needs to know the appropriate mf momentum distribution , the values of energy levels , and their widths @xmath111 . interacting nucleons are described by the correlated part of the sf . it is a known fact ( see ref . and references therein ) that the two - nucleon interactions dominate . these short - range correlations give rise to pairs of nucleons with high relative momentum . we follow the approach of kulagin and petti and do not include in the considerations interactions of higher order . than , @xmath112 can be expressed analytically in the form : @xmath113~.\qquad\ ] ] the constant @xmath99 appearing in the above formula is a shorthand notation for @xmath114 with @xmath115 and the mean square of the mf momentum @xmath116 defined as @xmath117 whereas @xmath118}\,\big\}^2,\\{{\ensuremath{\mathbf{p}}}}_\text{max}^2&=\big\{\beta { \ensuremath{|\mathbf{p}| } } + \sqrt{2m\beta[e - e^{(2)}-t_{a-1}]}\,\big\}^2.\end{split}\ ] ] the two - nucleon separation energy @xmath119 is an average excitation of the @xmath120 nucleon system . since by definition averaging should be carried out only over the low - lying states , it can be approximated by the mass difference @xmath121 . because an overwhelming contribution to the correlated part comes from the peak at @xmath122 and the recoil energy @xmath105 is less than @xmath123 by the factor @xmath124 , therefore eq . may be simplified to @xmath125}\,\big\}^2,\\{{\ensuremath{\mathbf{p}}}}_\text{max}^2&=\big\{\beta { \ensuremath{|\mathbf{p}| } } + \sqrt{2m\beta[e - e^{(2)}]}\,\big\}^2.\end{split}\ ] ] for the lightest considered here nucleus , i.e. , oxygen this simplification yields a @xmath126 change of the cross section . the struck nucleon moves in nuclear matter and may interact with surrounding spectators . such interactions make the nucleon an open system in the sense that measured @xmath127 is not equal to its energy in the interaction vertex . one can describe this situation in terms of a complex optical potential , @xmath128 , as proposed originally in ref . . we assume that the potential is time - independent . then the result is equivalent to making in eq . the substitution @xmath129 ^ 2}~,\ ] ] see refs . . the imaginary part of the optical potential may be approximated by @xmath130 this article s main interest is a description of medium nuclei , such as calcium and argon , therefore the nuclear matter density @xmath131 is assumed to be constant and equal to the saturation density @xmath132 @xmath133 . in the kinematical region of our interest , the typical proton kinetic energy is 100300 mev and the nucleon - nucleon cross section @xmath134 at @xmath135 varies between 16.2 and 19.1 mb . we set it to the value for 200-mev protons , i.e. , to 17.4 mb . optical potential used in this paper . dashed line represents its imaginary part obtained from eq . and solid line the real one from eq . . for comparison , the imaginary part calculated also from eq . is shown by dotted line.,scaledwidth=46.0% ] the real part of the potential we use is calculated in the following way : reference gives a dirac optical potential of fitted to proton - scattering data in the energy range 1611040 mev as a function of kinetic energy of proton and position in the nucleus . since what we need is the potential depending on energy only , averaging over spatial coordinate should be performed . we do it by evaluating the potential at the root mean square ( rms ) radii from ref . . as a result , we obtain a potential @xmath136 related to the scalar and vector part of the potential in ref . by @xmath137 ^ 2+{\ensuremath{\mathbf{p'}}}^2}+v(t_{\ensuremath{\mathbf{p'}}},\bar r_v)~.\ ] ] in the above equation , @xmath138 denotes two parameters , because the real and imaginary part of @xmath139 have different values of the rms radius . the same holds true for @xmath140 and @xmath141 . for @xmath142 @xmath143 , the real part of @xmath136 is positive , what is inconsistent with the correlated glauber theory . therefore when @xmath142 @xmath143 , we set its value to zero , as shown in fig . [ fig : pot ] . from the few parameterizations of the potential in ref . we decided to use the one called case 2 . we checked that the imaginary part of @xmath136 is then very close to @xmath144 obtained from eq . ( compare the dotted and dashed lines in fig . [ fig : pot ] ) , so our approach is self - consistent . the assumption that the optical potential is time - independent leads to folding of the cross section with the lorentzian function [ eq . ] . to cure the resulting problem with nonzero cross section for @xmath145 ( compare fig . 4 in ref . ) , we impose an additional constraint on the upper limit of the integration over @xmath15 , @xmath146 when fsi effects are not included , this restriction comes automatically from the energy - conserving @xmath104 function . influence of fsi on electron - nucleus cross section . dotted line shows the cross section without fsi , dashed line with only imaginary part applied , and solid line with full fsi.,scaledwidth=46.0% ] as can be seen in fig . [ fig : fsi ] , the essential effect of the imaginary part of the potential is to broaden the qe peak , whereas the real part mainly moves the strength to lower @xmath147 s . thanks to these two effects , the agreement of the calculated cross sections with the experimental data is significantly better . however , the time - independent imaginary part of the optical potential overestimates the fsi , namely too much strength is redistributed from the qe peak to its tails . we postpone the discussion of this point to sec . [ sec : precision ] . in this subsection , we want to cover all the details of description of three nuclei oxygen , calcium , and argon by their gaussian sfs . a procedure to divide momentum distributions given in ref . into the mf and correlated parts is presented and justified . we concentrate on @xmath148 and obtain the mf part from . then , we show the parameterization of the energy levels and comment the way , their widths are obtained . in ref . , the total momentum distributions for many nuclei are calculated . however , the model described in this paper requires a separation of the mf and correlated contributions [ see eqs . and ] . references contain plots with momentum distributions divided in the way we need . the conclusion from these articles is that above @xmath149 @xmath143 , the correlated part dominates overwhelmingly . we assume that above 2 @xmath143 this contribution is equal to the total distribution , and that @xmath150(p ) may be expressed as the correlated distributions given there , i.e. , as a sum of two exponential functions . moreover , smooth transition at 2 @xmath143 is imposed . distributions from refs . , denoted here by @xmath151 , are calculated up to @xmath152 @xmath143 . we extrapolated them smoothly to 5 @xmath143 , but it turned out to have very little influence on the cross sections . the correlated part of the momentum distribution is assumed to be of the following functional form : @xmath153&\text{for $ 0\leq{\ensuremath{|\mathbf{p}|}}\leq2.025$~fm$^{-1}$,}\\[5pt]\frac{\mathcal{f}}{(2\pi)^3}\frac{a}{n_t}n(t,{\ensuremath{\mathbf{p}}})&\text{for $ 2.025$~fm$^{-1}<{\ensuremath{|\mathbf{p}|}}\leq3.585$~fm$^{-1}$,}\\[5pt ] \frac{\mathcal{f}}{(2\pi)^3}\frac{a}{n_t}c_3\exp(-e_3{\ensuremath{\mathbf{p}}}^2)&\text{for 3.585~fm$^{-1}<{\ensuremath{|\mathbf{p}|}}\leq5.0$~fm$^{-1}$. } \end{cases}\ ] ] in the above equation , @xmath154 stands for the number of nucleons . we normalize the momentum distributions introducing the factor @xmath155 : @xmath156 to find the values of the parameters in , we assume that @xmath157 , so that only the @xmath158-containing term is responsible for the behavior of @xmath148 at large momenta . by demanding the continuity and smoothness of @xmath150 , one can determine @xmath159 and @xmath158 at @xmath160 @xmath143 , while @xmath161 and @xmath162 at @xmath152 @xmath143 . the values of @xmath163 are taken from ; in sec . [ sec : precision ] we will show that @xmath163 s do not affect the cross sections . the values of @xmath164 are fixed by the overall normalization of @xmath150 , which follows from ref . : the data contained there in tables ii and iii allow to calculate what fraction of nucleons can not be assigned to any shell - model state and , as a consequence , must be described by the correlated part of sf . the normalization of @xmath150 with respect to @xmath165 is the same as the normalization of the the correlated sf with respect to the total sf . proton momentum distribution in from ref . ( dots ) divided into the mf ( dashed line ) and correlated part ( dotted and solid line ) . solid line shows extrapolation according to eq . with parameters given in table [ tab : momdistrib].,scaledwidth=46.0% ] cccccc & & & + & & & & & + @xmath166 & 1.0200 & 1.0370 & 1.0370 & 1.0440 & 1.0200 + @xmath164 & 2.1280 & 4.2150 & 4.2700 & 4.0040 & 4.6700 + @xmath163 & 1.4000 & 1.7700 & 1.7700 & 1.7700 & 1.7700 + @xmath159 & 0.1427 & 0.1940 & 0.1855 & 0.1536 & 0.1656 + @xmath158 & 0.2260 & 0.2260 & 0.2142 & 0.2018 & 0.2065 + @xmath161 & 0.1678 & 0.2282 & 0.2451 & 0.2500 & 0.2960 + @xmath162 & 0.2410 & 0.2580 & 0.2648 & 0.2972 & 0.2940 + normal.&12.00%&16.20% & 16.20% & 17.10% & 13.64% + + @xmath167 ( mev ) & 174.4 & 189.1 & 187.1 & 180.8 & 196.4 + sample outcome of the described procedure is presented in fig . [ fig : momca40p ] . one can see that at the sewing points , @xmath160 @xmath143 and 3.585 @xmath143 , the correlated contribution is smooth . the total momentum distributions and the normalization of the correlated parts alike are taken from ref . . therefore the set of parameters collected in table [ tab : momdistrib ] can be considered as self - consistent . to handle the lack of knowledge of the momentum distributions for protons and neutrons in the argon nucleus , we apply in the sfs the appropriate distributions calculated for . in our approach , each shell - model state @xmath99 is fully characterized by two parameters : energy level @xmath168 and width @xmath111 , defined by means of eq . . l|dd|dd & & + & & d _ & & d _ + @xmath169 & 57.38 & 25 & 66.12 & 25 + @xmath170 & 36.52 & 15 & 43.80 & 15 + @xmath171 & 31.62 & 15 & 39.12 & 15 + @xmath172 & 14.95 & 4 & 22.48 & 6 + @xmath173 & 10.67 & 2 & 17.53 & 4 + @xmath174 & 8.88 & 2 & 15.79 & 4 + @xmath175 & 4.71 & & 12.0 & + l|dd|dd & & + & & d _ & & d _ + @xmath169 & 52 & 25 & 62 & 25 + @xmath170 & 32 & 15 & 40 & 15 + @xmath171 & 28 & 15 & 35 & 15 + @xmath172 & 11 & 4 & 18 & 5 + @xmath173 & 8 & 2 & 13.15 & 4 + @xmath174 & 6 & 2 & 11.45 & 3 + @xmath176 & & & 5.56 & 3 + @xmath175 & & & 8.0 & + l|dd|dd & & + & & d _ & & d _ + @xmath169 & 45.00 & 70 & 47.00 & 70 + @xmath170 & 18.44 & 4 & 21.80 & 4 + @xmath171 & 12.11 & 4 & 15.65 & 4 + the energy levels of calcium shown in table [ tab : caenergylev ] result from theoretical calculations in refs . ( for neutrons ) and ( for protons ) . a few available neutron levels of argon form a pattern very similar to the one of the neutron levels of calcium : the distance between @xmath174 and @xmath173 is 1.7 mev for ar and 1.74 mev for ca , whereas the distances between @xmath174 and the fermi level @xmath175 are 3.5 mev for ar and 3.8 mev for ca . to reconstruct the missing data , we assume that all the neutron levels follow the same pattern , see table [ tab : arenergylev ] . due to the lack of knowledge about the proton levels of argon , we use the modified values from calcium . the data for oxygen are collected in table [ tab : oxenergylev ] . the widths for most of the calcium levels can be determined by fitting to the plots of energy distribution in papers . we estimate the remaining ones using the fact that @xmath111 should be , approximately , a function of a distance from the fermi level : @xmath177 to get @xmath111 s for argon , we assume that their values lie on roughly the same curve as for ca . we did not find the energy distribution for oxygen , calculated in the same way as for calcium in refs . . since oxygen nucleus plays only the role of a testing ground for our model , we decided to obtain the proton @xmath111 s directly from the energy distribution in the benhar sf , and use the same values for neutrons . thus we avoided additional discrepancies between the two descriptions . the goal of this subsection is to confront the model presented in sec . [ sec : description ] with the existing electron scattering data . since description of the dip region and the @xmath1 excitation is ambiguous , our considerations include qe interactions only and we test predictions of the obtained sfs in energy transfers below the qe peak . figures [ fig : oxy][fig : ar ] show comparison with a wide spectrum of experimental points . the missing cross section for energy transfer above the qe peak may be attributed to the two - nucleon interactions , @xmath1 production , and nonresonant background . in the captions , we give the momentum transfer at the qe peak calculated according to the formula @xmath178 this value depends rather weakly on @xmath147 and therefore it provides quite good characteristics of the whole peak . we start with the oxygen target . figure [ fig : oxy ] presents the predictions of three models and the data from refs . the dotted line corresponds to the fermi gas ( fg ) model ( fermi momentum @xmath179 mev , binding energy @xmath180 mev , no fsi ) , the solid line shows the cross sections of the oxygen gsf with fsi as in sec . [ sec : fsi ] , and the dashed line depicts results for the benhar sf with the same fsi . differences between our model and the more systematic sf are of the size of the error bars . the main source of these differences is another momentum distribution , see fig . [ fig : oxbenhar ] and the discussion in sec . [ sec : precision ] . both sfs reproduce the shape and height of the qe peak quite well , but underestimate the cross section at low @xmath147 s . this discrepancy may be attributed to the unsatisfactory treatment of fsi effects , because they tend to increase the cross section in this region . the best agreement with the data is obtained for the 880-mev electron beam , whereas the worst one corresponds to @xmath181 mev . fortunately , the latter set of data is least relevant in our analysis ( see sec . [ sec : selection ] ) . for calcium , we compare in fig . [ fig : ca ] the cross sections obtained using the fg model ( @xmath182 mev , @xmath183 mev @xcite ; represented by the dotted line ) , the gsf ( solid line ) , and the calculations of butkevich and mikheyev ( dashed line ) to the sample of electron scattering data collected at scattering angle 45.5@xmath60 and various beam energies @xcite . only our model includes fsi effects . the fg model describes very well the position and size of the qe peak for the highest values of beam energy . when the energy is lower than 700 mev , it obviously fails . despite the fact that the approach of butkevich and mikheyev is based on the sf , it yields results very similar to these for the fg model when @xmath147 is near the value corresponding to the qe peak or higher . the reason of this behavior lies in too simple treatment of the mf part in their sf : the energy distribution was limited to a single @xmath104 function . for energies 628841 mev , the accuracy of the gsf is very good . the occurring discrepancies can be explained as a contribution from the @xmath1 production . at small values of energy transfer , the cross section is slightly overestimated . it means also that the qe peak is slightly underestimated , because fsis based on a folding function do not change the total cross section . note that the agreement with the data in our region of interest ( see sec . [ sec : selection ] ) is better than in the case of oxygen . it may be attributed to the way fsi is introduced : density of nucleus is assumed to be constant and equal to the saturation density of nuclear matter ; this approximation should work better for heavier nuclei . the real part of the optical potential used in our computations was obtained for calcium and should work better for this target than for oxygen . for @xmath184 mev , our model fails to describe the position and shape of the qe peak . however , the inaccuracy of the fg and the approach of ref . is visibly more severe . similar problems for oxygen occur when @xmath185 mev . at the first glance , there is no connection between these two cases . but when we have a closer look at the values of the momentum transfer at the qe peak , we will discover that our model starts to lose accuracy when momentum transfer is lower than @xmath0350400 mev . it could be related to simplifying assumptions of our approach , treatment of fsi effects or the very basic assumption the ia . the models based on different from our approximations apart from the ia and with more systematical treatment of fsi suffer a similar drawback . it suggests that it is the loss of reliability of the ia , what is responsible for the problem . before applying our model to neutrino interactions , we perform the final test by confronting it with the data for electron scattering off argon . we have found only one such experiment @xcite , which measured the cross section of @xmath186-mev electrons scattered at @xmath85 . in the left panel of fig . [ fig : ar ] , predictions of the argon gsf and the fg model ( @xmath187 mev and @xmath188 mev ) are presented . the accuracy of the gsf is clearly better than that of the fg model . the result for our model , shown by the solid line , does not describe properly only the cross section at very low values of energy transfer . we have faced the same problems for oxygen and calcium and interpreted it as a breakdown of the ia at @xmath189400 mev . in the considered case of scattering off argon , the momentum transfer at the qe peak is equal to 371 mev . when we compare the result for argon with the one for oxygen in exactly the same kinematical conditions ( see right panel of fig . [ fig : ar ] ) , we can see that the level of accuracy is comparable . the same holds true also for comparison with scattering off calcium for electron - beam energy 471 or 545 mev . therefore , we expect that if the typical @xmath190 was higher , the agreement with the data for argon would be better . we have observed that even for argon , the neutron sf may be approximated by the corresponding proton sf , as far as electron scattering is concerned . it can be explained by the fact that the contribution of neutrons to the inclusive cross section is small , what suppresses the differences between the sfs . this contribution is equal to 13% for 700-mev electrons scattered at 32@xmath60 and rises to 23% when the beam energy is increased to 1200 mev . quasielastic differential cross section @xmath191 of as a function of produced muon energy @xmath192 for the gsf ( solid line ) , approach of ref . ( dashed line ) , and the fg model ( dotted line).,scaledwidth=46.0% ] in the case of neutrino scattering , quantities of interest are the total cross section and the differential cross section in @xmath193 or in energy transfer ( equivalently : in energy of produced muon ) . figure [ fig : nuar800 ] depicts differences between @xmath191 for the argon gsf ( solid line ) , the sf we described in ref . ( dashed line ) , and the fg model ( dotted line ) . one can see that the sfs introduce significant reduction of the cross section , mainly in the region of low energy transfers . the line representing the predictions of the gsf model is slightly wiggly , because when @xmath147 increases , lower - lying energy levels consecutively start contributing to the cross section . there are no singularities in the cross section , and in this sense , the gsf is more realistic then the sf from ref . . effects of fsi are not taken into account except those from pauli blocking , but their influence on the cross section @xmath191 is rather small ( see fig . 14 in ref . showing the impact of introducing fsi on @xmath194 ) . the purpose of fig . [ fig : nuar800 ] is to show discrepancy of our description of argon nucleus and the fg model , commonly used in monte carlo simulations . the results for neutrinos can not be directly confronted with experimental data . therefore , we first identified , in sec . [ sec : selection ] , the region in the @xmath54 plane which is most important for the 800-mev neutrino scattering . than we substantiated accuracy of our approach : we showed in sec . [ sec : electrons ] that it describes well kinematical aspects of nuclear effects . this whole analysis allows us to expect that using the presented approximation of the sf , we model neutrino interactions at a similar level of accuracy as achieved in the case of electron scattering . our approach is based on many approximations and in this section , we would like to understand how uncertain our final predictions are . dependence of the cross section on the form factors . the dipole parametrization ( dashed line ) produces @xmath03% higher result than the bbba05 one @xcite.,scaledwidth=46.0% ] influence of the procedure restoring current conservation [ eq . ] on the cross section . the result obtained without it is represented by the dashed line.,scaledwidth=46.0% ] intrinsic inaccuracy of our model arising from the treatment of the mf part of the sf . calculation for the benhar s exact sf of oxygen ( dashed line ) are compared with result for the gsf with the same momentum distribution.,scaledwidth=46.0% ] [ [ form - factors . ] ] form factors . + + + + + + + + + + + + + different choices of parameterization of the electromagnetic form factors may change the results by a few percentages . as shown in fig . [ fig : cadip ] , the dipole parameterization yields the cross sections higher than the bbba05 one @xcite used in this article . the discrepancy at the qe peak is @xmath03.2% for beam energy 350 mev and @xmath03.5% for 841 mev . [ [ current - conservation . ] ] current conservation . + + + + + + + + + + + + + + + + + + + + + describing both electron and neutrino interactions , we applied the de forest prescription to describe the off - shell kinematics . however , this leads to a loss of conservation of the electromagnetic current in electron scattering and of the vector current in neutrino interactions . the procedure , by which we restore it in the electron case , modifies the cross section mainly above the qe peak ; see fig . [ fig : ca628nocec ] . for beam energy 628 mev , the effect is as small as 1.5% at the peak and it decreases when the energy becomes larger . [ [ simplifications - in - the - mean - field - sf . ] ] simplifications in the mean - field sf . + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + in the derivation of eq . , we made two simplifying assumptions : level widths do not depend on momentum and contribution to the momentum distribution of each level is the same [ see eq . ] . figure [ fig : oxbenhar ] illustrates the loss of accuracy due to these simplifications . to depict their influence , we use the momentum distribution calculated from the benhar sf instead of the one from ref . . since the level widths of oxygen are obtained by fitting to the energy distribution of the benhar sf , a slightly different shape of the predicted qe peak is the result of the simplifying assumptions only . we checked that for other values of beam energy discrepancy does not increase . therefore , we conclude that the gsf can be considered as quite good approximation of the more systematic approach . [ [ parameterization - of - the - momentum - distributions . ] ] parameterization of the momentum distributions . + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + application of the momentum distributions from ref . in our model requires dividing each of them into the mf and correlated parts . it involved introduction of a few parameters . to find out how much choice of these parameters influences the cross sections , we calculated first @xmath150 for the oxygen normalized as in table [ tab : momdistrib ] , but with @xmath195 ( instead of 1.400 ) : @xmath196\end{aligned}\ ] ] for @xmath197 @xmath143 . when it is applied , the cross sections change less than 0.1% in the considered energy range . thus , we do not need to pay much attention to parameter @xmath163 , as far as the same normalization is kept . second , we found the distribution with @xmath198 , but with the normalization 16.2% ( instead of 12.0% ) : @xmath199\end{aligned}\ ] ] at the interval @xmath197 the above distribution leads to the cross sections changed by up to 2.2% . we have analyzed a few such modifications and in each case we have found that the influence of the normalization is greater than that of @xmath163 . it is because variation of the parameter @xmath163 only redistributes the strength within given part of the momentum distribution [ and as a consequence modifies the parameter @xmath99 in eq . ] , whereas variation of the normalization changes the way some part of the strength is treated . uncertainty of the the cross section with respect to used momentum distribution . solid line shows result for momentum distribution from ref . ( used throughout this paper ) and the dashed one from ref . .,scaledwidth=46.0% ] same as fig . [ fig : cacda ] but for @xmath200 scattering off argon.,scaledwidth=46.0% ] [ [ momentum - distributions . ] ] momentum distributions . + + + + + + + + + + + + + + + + + + + + + + + both for oxygen and calcium , the momentum distributions are given by analytical formulas in ref . . in fig . [ fig : cacda ] , we show that even though they predict slightly higher qe peak , the yielded cross section is lower . because the calcium momentum distributions are used for argon , its description `` inherits '' the same uncertainties , see fig . [ fig : arcda ] . throughout this paper , we rely on the distributions from ref . , because they are obtained in more systematic calculations . estimation of the uncertainty due to unknown momentum distribution of argon . the cross section calculated using the sf of with momentum distribution of ( solid line ) and ( dashed line).,scaledwidth=46.0% ] influence of the level width on the cross section . solid line : calculation with the values from table [ tab : arenergylev ] . dashed line : values multiplied by 3.,scaledwidth=46.0% ] comparison of the cross sections obtained with the energy levels from table [ tab : arenergylev ] ( solid line ) , levels shifted by @xmath201 mev ( dotted line ) , and by @xmath202 mev ( dashed line).,scaledwidth=46.0% ] in addition to the already described sources of uncertainty , the description of argon nucleus suffers from the lack of the available momentum distributions and knowledge of energy levels . we estimate them using the information for . for this reason , a few words of comment on the accuracy for this specific nucleus are needed . [ [ momentum - distributions.-1 ] ] momentum distributions . + + + + + + + + + + + + + + + + + + + + + + + the surplus neutrons modify both the proton and neutron momentum distributions . a similar situation appears for , where the distributions are available . we have used the momentum distributions to estimate how these modifications can affect the argon cross sections ; see fig . [ fig : ca48toar ] . the proton cross section was increased by 4% and the neutron one was decreased by 3.8% . the overall increase is equal to 2.9% . the number of surplus neutrons in is smaller than in , therefore we expect this effect to be smaller too . [ [ level - widths . ] ] level widths . + + + + + + + + + + + + + due to the lack of any knowledge about the level widths of argon , we use the values for calcium . figure [ fig : arspreading ] presents that @xmath111 s three times larger than those given in table [ tab : arenergylev ] change the cross sections only up to 2% ( decrease at the peak ) . narrower levels gives barely noticeable difference : 0.23% for the widths divided by 3 , and 0.53% for divided by 100 ( increase at the peak ) . for @xmath191 , the more the levels overlap , the less wiggly the cross section is . [ [ energy - levels . ] ] energy levels . + + + + + + + + + + + + + + the argon energy levels may differ from the used ones . we may expect that the discrepancies in table [ tab : arenergylev ] are distributed randomly , and so a part of their influence on the cross section is diminished . figure [ fig : arlev ] shows that even if every level is shifted by the same value , chosen to be 5 mev , the cross section does not change dramatically the qe peak only moves a little bit . we conclude that the way to increase the accuracy of the presented argon sf is to apply the actual values of the energy levels ; the degree in which they are smeared has minor influence on the cross section , especially in the case of electrons . [ [ real - potential . ] ] real potential . + + + + + + + + + + + + + + + to find out if one can approximate the real part of the potential by a constant , we have applied the value 10 mev . in the case of oxygen , it slightly improved agreement with the experimental data . however , the same value employed to calcium decreased the level of accuracy of the model . it might suggest that the real potential for oxygen is deeper than the potential shown in fig . [ fig : pot ] . [ [ imaginary - potential . ] ] imaginary potential . + + + + + + + + + + + + + + + + + + + + the use of the imaginary part of the potential @xmath136 defined in eq . instead of approximation has minor influence on the obtained cross sections . typical change is a @xmath01%-increase . we conclude that for practical purposes these two approaches are equivalent . [ [ cross - section . ] ] cross section . + + + + + + + + + + + + + + when evaluating imaginary potential , we have fixed the nucleon - nucleon cross section to 17.4 mb , which corresponds to nucleon kinetic energy 200 mev . in principle , one should take into account the cross section s dependence on energy . therefore , to check validity of our approximation , we have used the exact nucleon - nucleon cross section in the energy range 100300 mev , most important for the discussed kinematical region . for 545-mev electron scattering off calcium , the result decreases by 1.1% . when beam energy is higher , the effect is even smaller . [ [ density - of - nucleus . ] ] density of nucleus . + + + + + + + + + + + + + + + + + + + we have assumed that the density of nucleus is equal to the saturation density , despite the fact that in reality its average value is smaller . however , the quantity of interest is not the density itself but rather @xmath203 . this product decreases by 7% ( 15.4% ) when @xmath131 changes to 0.14 @xmath133 ( 0.12 @xmath133 ) , i.e. , by 12.5% ( 25% ) . since the corresponding increase of the electron cross section is only 1% ( 2.4% ) at the qe peak , our approach seems to be well justified . [ [ folding - function . ] ] folding function . + + + + + + + + + + + + + + + + + employing lorentzian folding function , i.e. , neglecting correlations between nucleons in nucleus is a crude approximation . comparison to the results presented in ref . suggests that an accurate approach could yield the cross sections higher at the peak by up to @xmath015% and with lower tails . precise comparison is difficult because in our computations contribution of the @xmath1 resonance is missing . the main goal of the article is to improve description of neutrino scattering off argon in the 1-gev energy region . we have presented the way to calculate approximate spectral functions of medium nuclei and applied it to electron scattering off oxygen , calcium , and argon targets . for neutrino interactions precise experimental data are missing . therefore , we have identified the region of the @xmath54 plane which is most important for neutrino quasielastic interaction . the presented model to describe nuclear effects has been then tested using the electron scattering data which lie in this region . the obtained agreement is good in the case of oxygen and very good for calcium . moreover , our approximation reproduces results of the benhar sf for oxygen with a satisfactory degree of accuracy . detailed discussion of uncertainties due to many simplifications of our model have lead us to the conclusion that all of them are of the order of a few percentages . in addition , we have observed that when the typical value of the momentum transfer is less than @xmath0350400 mev , systematic discrepancies between the presented model and the electron data occur : the shape of the calculated cross section @xmath204 is not suitable to fit the data and also increasing amount of strength is missing at low energy transfers . a similar problem is present in other models what suggests that the source of the problem may be the loss of reliability of the impulse approximation . we would like to thank giampaolo co for the momentum distributions used in this paper . we also express our gratitude to omar benhar for providing us with his spectral function for oxygen . the authors were supported by mnisw under grants no . 3735/h03/2006/31 ( jts , ama ) and no . 3951/b / h03/2007/33 ( ama ) . 99 j. g. morfin , m. sakuda , and y. suzuki ( eds . ) , nucl . b , proc . suppl . * 112 * ( 2002 ) ; f. cavanna , p. lipari , c. keppel , and m. sakuda ( eds . ) _ ibid _ * 139 * ( 2005 ) ; f. cavanna , j. g. morfin , and t. nakaya ( eds . ) _ ibid _ * 159 * ( 2006 ) . a. m. ankowski and j. t. sobczyk , phys . c * 74 * , 054316 ( 2006 ) . a. m. ankowski , acta phys . pol . b * 37 * , 2259 ( 2006 ) . o. benhar , n. farina , h. nakamura , m. sakuda , and r. seki , phys . d * 72 * , 053005 ( 2005 ) . a. v. butkevich and s. p. mikheyev , phys . c * 72 * , 025501 ( 2005 ) . m. anghinolfi _ et al . _ , j. phys . g 21 , l9 ( 1995 ) . s. frullani and j. mougey , adv . * 14 * , 1 ( 1984 ) . r. bradford , a. bodek , h. budd , and j. arrington , nucl . b , proc . suppl . * 159 * , 127 ( 2006 ) . t. de forest jr . , nucl . phys . * a392 * , 232 ( 1983 ) . c. maieron , m. c. martnez , j. a. caballero , and j. m. udas , phys . c * 68 * , 048501 ( 2003 ) . o. benhar , a. fabrocini , s. fantoni , g. a. miller , v. r. pandharipande , and i. sick , phys . c * 44 * , 2328 ( 1991 ) . h. nakamura , r. seki , and m. sakuda , nucl . b , proc . suppl . * 139 * , 201 ( 2005 ) . g. co , nucl . b , proc . suppl . * 159 * , 192 ( 2006 ) . e. oset , l. l. salcedo , d. strottman , phys . lett . * b165 * , 13 ( 1985 ) . e. oset , l. l. salcedo , nucl . phys . * a468 * , 631 ( 1987 ) . d. h. e. gross and r. lipperheide , nucl . a150 * , 449 ( 1970 ) . m. anghinolfi _ et al . a602 * , 405 ( 1996 ) . j. s. oconnell _ et al . c * 35 * , 1063 ( 1987 ) . r. r. whitney , i. sick , j. r. ficenec , r. d. kephart , and w. p. trower , phys . c * 9 * , 2230 ( 1974 ) . z. e. meziani _ et al . * 54 * , 1233 ( 1985 ) . t. c. yates _ et al . _ , phys . lett . * b312 * , 382 ( 1993 ) . c. f. williamson _ c * 56 * , 3152 ( 1997 ) . c. ciofi degli atti , s. simula , l. l. frankfurt , and m. i. strikman , phys . c * 44 * , r7 ( 1991 ) . s. a. kulagin and r. petti , nucl . phys . * a765 * , 126 ( 2006 ) . o. benhar , a. fabrocini , s. fantoni , and i. sick , nucl . phys . * a579 * , 493 ( 1994 ) . c. ciofi degli atti , s. liuti , and s. simula , phys . c * 41 * , r2474 ( 1990 ) . y. horikawa , f. lenz , and n. c. mukhopadhyay , phys . c * 22 * , 1680 ( 1980 ) . v. r. pandharipande and s. c. pieper , phys . c * 45 * , 791 ( 1992 ) . e. d. cooper , b. c. clark , r. kozack , s. shim , s. hama , j. i. johansson , h. s. sherif , r. l. mercer , and d. b. serot , phys . c * 36 * , 2170 ( 1987 ) . e. d. cooper , s. hama , b. c. clark , and r. l. mercer , phys . c * 47 * , 297 ( 1993 ) . c. bisconti , f. arias de saavedra , and g. co , phys . c * 75 * , 054302 ( 2007 ) . g. co , private communication . o. benhar , c. ciofi degli atti , s. liuti , and g. salm , phys . lett . * b177 * , 135 ( 1986 ) . c. ciofi degli atti and s. simula , phys . c * 53 * , 1689 ( 1996 ) . c. h. johnson and c. mahaux , phys . c * 38 * , 2589 ( 1988 ) . w. tornow , z. p. chen , and j. p. delaroche , phys . c * 42 * , 693 ( 1990 ) . c. h. johnson , r. f. carlton , and r. r. winters , phys . c * 44 * , 657 ( 1991 ) . v. gillet and n. vinh mau , nucl . phys . * 54 * , 321 ( 1964 ) . a. bohr and b. r. mottelson , _ nuclear structure _ ( benjamin , new york , 1969 ) , vol . 1 , 232 , 326 . k. a. de witt huberts , j. phys . g 16 , 507 ( 1990 ) . a. gil , j. nieves , and e. oset , nucl . * a627 * , 543 ( 1997 ) . d. drechsel , o. hanstein , s. s. kamalov , and l. tiator , nucl . phys . * a645 * , 145 ( 1999 ) . o. benhar , d. day , and i. sick , rev . * 80 * , 189 ( 2008 ) . o. benhar , nucl . . suppl . * 159 * , 168 ( 2006 ) .

we present a new model for protoplanetary disc evolution . this model combines viscous evolution with photoevaporation of the disc , in a manner similar to . however in a companion paper we have shown that at late times such models must consider the effect of stellar radiation directly incident on the inner disc edge , and here we model the observational implications of this process . we find that the entire disc is dispersed on a time - scale of orderyr after a disc lifetime of a few myr , consistent with observations of t tauri ( tt ) stars . we use a simple prescription to model the spectral energy distribution of the evolving disc , and demonstrate that the model is consistent with observational data across a wide range of wavelengths . we note also that the model predicts a short `` inner hole '' phase in the evolution of all tt discs , and make predictions for future observations at mid - infrared and millimetre wavelengths . [ firstpage ] accretion , accretion discs circumstellar matter planetary systems : protoplanetary discs stars : pre - main - sequence

astro-ph0603254

for over twenty years the evolution , and eventual dispersal , of discs around young stars has been an important area of study , for theories of both star and planet formation . it is now well - established that at an age of @xmath1yr most stars are surrounded by discs that are optically thick at optical and infrared wavelengths @xcite . observations at millimetre wavelengths show that these discs have masses that are typically a few percent of a solar mass @xcite , and so discs are widely believed to be potential sites for planet formation . however at an age of @xmath2yr most stars are not seen to have discs , suggesting that disc lifetimes are typically a few myr ( e.g. * ? ? ? how stars lose their discs remains an unsolved question . the distribution of t tauri stars ( tts ) at infrared wavelengths provides some insight into this problem . these objects tend to fall into two distinct groups : those whose emission is consistent with a stellar photosphere plus an optically thick disc , and those which are compatible with purely photospheric emission . [ note that these two classes usually coincide with the spectroscopic classifications of classical and weak - lined t tauri stars ( ctts and wtts ) respectively . ] a number of authors have noted that very few transition objects are observed between the ctt and wtt loci @xcite . observations at longer wavelengths show a similar behaviour , both in the mid - infrared @xcite and at millimetre wavelengths @xcite . these observations suggest that discs are dispersed very rapidly , with the dispersal time estimated to be @xmath3yr @xcite . moreover the simultaneous decline is disc emission across such a wide range in wavelength suggests that the dispersal is essentially simultaneous across the entire radial extent of the disc ( see also discussion in @xcite ) . this `` two - time - scale '' behaviour is inconsistent with conventional models of disc evolution , such as viscous evolution models @xcite or models of magnetospheric clearing @xcite . such models predict power - law declines in disc properties , and therefore predict dispersal times which are always of the same order as the disc lifetime . however ( * ? ? ? * hereafter cgs01 ) showed that models which combine photoevaporation of the disc with viscous evolution can reproduce this two - time - scale behaviour . in this model , known as the `` uv - switch '' model ( cgs01 ) , ionizing radiation from the central star produces a photoevaporative wind at large radii . detailed models of photoevaporative winds were constructed by ( * ? ? ? * see also @xcite ) for the cases of both weak and strong stellar winds . in the case of tts we consider only the weak stellar wind case . in this case ionizing radiation from the star creates an ionized layer on the surface of the disc , with conditions akin to an hii region . beyond some critical radius , known as the gravitational radius , the local thermal energy of the ionized is greater than its gravitational energy and the gas escapes as a wind . the gravitational radius is therefore given by @xmath4 where @xmath5 is the sound speed of the ionized gas , typically 10kms@xmath6 . @xcite show that the wind rate is determined by the the density at the ionization front , and find an integrated mass - loss rate of @xmath7 where @xmath8 is the ionizing flux produced by the star . more recent studies have extended the study of the details in a number of ways @xcite , and recent hydrodynamic modelling has resulted in slight modification of the quantitative results . when hydrodynamic effects are considered the `` effective @xmath9 '' is reduced by a factor of 5 @xcite , and the mass - loss rate is reduced by a factor of around 3 @xcite . however the qualitative behaviour is unchanged from that of @xcite . the so - called `` uv - switch '' model of cgs01 couples a photoevaporative wind to a simple disc evolution model . at early times in the evolution the accretion rate through the disc is much larger than the wind rate , and the wind has a negligible effect . however at late times photoevaporation becomes important , depriving the disc of resupply inside @xmath9 . at this point the inner disc drains on its own , short , viscous time - scale , giving a dispersal time much shorter than the disc lifetime . a number of similar studies have now been conducted @xcite , and this class of models show a number of attractive properties . however cgs01 highlighted two key problems with the model . firstly , the model requires that tts produce a rather large ionizing flux , of order @xmath10 ionizing photons per second . they also found that the outer disc , beyond @xmath9 was dispersed much too slowly to satisfy millimetre observations of wtts , a finding re - affirmed by recent sub - millimetre observations @xcite . we have previously shown that it is reasonable to treat tt chromospheres as having a constant ionizing flux in the range @xmath11@xmath12s@xmath6 @xcite , and we now seek to address the `` outer disc problem '' . in a companion paper ( * hereafter paper i ) , we highlighted an important flaw in the uv - switch model . the uv - switch model relies on the wind parametrization of @xcite , which assumes that the disc is extremely optically thick to lyman continuum photons at all radii . @xcite find that the diffuse ( recombination ) field dominates the photoevaporation at all radii of interest , as the direct field suffers extremely strong attenuation by the disc atmosphere . however we note that at late times in the uv - switch model the inner disc is drained , and is therefore optically thin to ionizing radiation . consequently the direct field is important after the inner disc has drained . in paper i we constructed detailed hydrodynamic models of the wind driven by the direct field , and derived a functional form for the mass - loss rate : @xmath13 \ , \mathrm m_{\odot}\mathrm{yr}^{-1 } \ , .\end{aligned}\ ] ] here @xmath14 and @xmath15 are order - of - unity scaling constants , @xmath16 is the mean molecular weight of the gas , @xmath17 is a power - law index , @xmath18 is the ratio of the disc scale - height to radius , and @xmath19 and @xmath20 are the inner and outer disc radii respectively . our numerical analysis fixed the values of the scaling constants to be @xmath21 and @xmath22 ( for @xmath23 ) . as noted in paper i , both the geometry of the radiative transfer problem and the form of the wind are qualitatively similar to the strong wind case of @xcite . however the effect of the direct radiation field is to increase the efficiency of the wind . consequently the mass - loss rate due to direct photoevaporation is around an order of magnitude larger than that from the diffuse field , and is significant at late stages of the evolution . the diffusion equation for the evolution of disc surface density @xmath24 is @xcite @xmath25 - \dot{\sigma}_{\mathrm { wind}}(r , t ) \ , , \ ] ] where @xmath26 is the kinematic viscosity and the term @xmath27 represents the mass - loss due to photoevaporation . cgs01 solved this equation using the `` weak - wind '' profile of @xcite . at some point in the evolution the mass - loss rate from the wind falls to a level comparable to the accretion rate through the disc , and at this point the disc is rapidly drained inside @xmath9 . however , as mentioned above , cgs01 neglect the influence of the direct radiation field after this inner draining occurs . consequently they find that the time - scale for dispersal of the outer disc is limited by the time material takes to diffuse inward to @xmath9 ( as the most of the mass - loss occurs close to @xmath9 ) . further , the @xmath28 dependence of the wind profile at large radii means that the mass - loss rate due to photoevaporation decreases significantly with time as the inner edge of the disc moves outward . consequently the dispersal of the outer disc occurs on the viscous time - scale of the _ outer _ disc , and thus the outer disc is dispersed in a time comparable to the disc lifetime , much too slowly to satisfy observational constraints . we suggest that photoevaporation by the direct radiation field results in a dispersal time significantly shorter than that predicted by cgs01 , and now seek to model the effects of this process on the evolution of the outer disc . in this paper we seek to incorporate the result of paper i into models of disc evolution . we do this by solving the equation for the evolution of the surface density of a geometrically thin disc , including the photoevaporative wind as a sink term . in section [ sec : timescales ] we present a simple time - scale analysis , which demonstrates the significance of the direct radiation field . in section [ sec : disc_model ] we construct a numerical model of disc evolution in the presence of a photoevaporative wind . we the use a simple prescription to model the observed spectral energy distribution ( sed ) of the evolving disc , and construct a set of models which cover a broad range in parameter space ( section [ sec : sed_model ] ) . in section [ sec : results ] we present our results , comparing the predicted seds to recent observational data . in section [ sec : dis ] we discuss the implications and limitations of our results , and in section [ sec : summary ] we summarize our conclusions . it is useful at this point to consider the evolutionary time - scales predicted by the different models . in the original model of cgs01 , the inner disc drains on approximately the viscous time - scale at the draining radius , @xmath29 , which is significantly shorter than the disc lifetime to that point . however the outer disc drains on a much longer time - scale , comparable to the disc lifetime . after the inner disc has drained there is no accretion on to the star , so the only mass - loss is due to the wind . consequently , if we neglect viscous evolution during clearing , the time - scale for the wind to clear the disc out to a radius @xmath30 , @xmath31 , is given by @xmath32 where @xmath33 is the total disc mass at radii @xmath34 and @xmath35 is the total mass - loss rate from the wind , integrated from radius @xmath36 outwards . neglecting numerical factors of order unity , the viscous time - scale for the evolution of an accretion disc is given by @xmath37 where @xmath38 is the disc accretion rate . we can therefore combine equations [ eq : tc ] & [ eq : tnu ] and express the clearing time as @xmath39 clearing occurs when the disc accretion rate @xmath38 falls to a level comparable to the diffuse wind rate ( which is constant , see equation [ eq : wind_rate ] ) , and therefore the clearing time - scale depends only on @xmath35 . if @xmath40 is a decreasing function of @xmath36 then accretion will dominate over the wind as the disc evolves , whereas if @xmath35 increases with @xmath36 then the wind will dominate the evolution . in the model of cgs01 , where the diffuse field mass - loss profile is adopted throughout , the wind profile takes the form @xmath41 . consequently the total wind mass - loss rate , integrated from an inner edge radius @xmath36 outwards , is @xmath42 , and we can normalise by noting that @xmath43 . therefore the clearing time - scale is given by @xmath44 thus we see that in the uv - switch model ( cgs01 ) the time - scale to clear the disc out to a radius @xmath36 is longer than the viscous time - scale for all @xmath30 . consequently in this model viscosity dominates : mass is lost only after it has had time to diffuse inwards towards the inner disc edge . thus while the inner disc satisfies the two - time - scale behaviour demanded by observations ( see section [ sec : intro ] ) , the outer disc is dispersed much too slowly to satisfy the data . however if we consider direct photoevaporation of the outer disc we see a very different behaviour . in this case , as seen in equation [ eq : mout_anal ] ( see also paper i ) , the integrated mass - loss from a disc with an inner edge at radius @xmath36 is @xmath45 . consequently in this case the clearing time - scale is given by @xmath46 thus we see that in the case of direct photoevaporation viscosity becomes progressively _ less _ significant as the inner edge evolves outwards , and that the evolution is instead dominated by the wind . consequently we predict a much faster dispersal of the outer disc than that originally predicted by cgs01 . for tt parameters the draining radius is of order au , and tt discs are observed to be up to several hundred au in size . thus we predict a clearing time for the entire disc that is @xmath47110% of the viscous scaling time . such behaviour satisfies the two - time - scale constraint across the entire radial extent of the disc . we solve the diffusion equation for the disc surface density ( equation [ eq:1ddiff ] ) using a standard first - order explicit scheme ( e.g. * ? ? ? * ) , with a grid of points equispaced in @xmath48 . following @xcite and cgs01 , we adopt a kinematic viscosity @xmath26 which scales linearly with radius , so @xmath49 for some scaling value @xmath50 at a scale radius @xmath51 . such a viscosity law is consistent with an @xmath52-prescription @xcite if the disc temperature at the midplane scales as @xmath53 , and arguments in favour of this viscosity law are discussed in @xcite . we adopt an initial surface density profile consistent with the similarity solution of the diffusion equation @xcite . again following cgs01 , we define this profile to have the form @xmath54 for an initial disc mass @xmath55 . in this form @xmath56 of the disc mass is initially at @xmath57 , with an exponential decline in surface density at radii beyond @xmath51 . however in practice the results are not sensitive to the form of the initial profile , and this form is chosen primarily as a simple means of parametrizing the initial disc mass . with this viscosity law the viscous scaling time , which governs the evolution of the disc , is given by @xmath58 and consequently the initial accretion rate at the origin is given by @xmath59 this fixes the scaling constant @xmath60 , and thus the disc model is entirely specified by the three parameters @xmath55 , @xmath61 and @xmath51 . initially we adopt @xmath62m@xmath63 , @xmath64m@xmath63yr@xmath6 and @xmath65au . we use 1000 grid points , equispaced in @xmath48 , which span the radial range @xmath66 $ ] . we also adopt adopt zero - torque boundary conditions throughout ( i.e. we set @xmath67 at the grid boundaries ) but note that the spatial domain is always large enough that the outer boundary condition has no effect on the results . at early times in the evolution we use the wind profile of ( * ? ? ? * kindly provided in numerical form by ian mccarthy ) , which incorporates more detailed hydrodynamics than the original work of @xcite . however once the inner disc has been drained we alter the mass - loss profile to reflect the influence of direct photoevaporation on the outer disc . thus it is necessary to define a numerical criterion for `` draining '' , in order to indicate when to change to the direct profile . additionally , as the form of the direct profile is normalised at the inner disc edge it is necessary to define the inner edge numerically also . by inspection of the mass - loss profiles we see that in both the diffuse and direct cases the mass - loss rate scales as @xmath68 numerical analysis of the scaling constants shows that for equal ionizing fluxes the direct mass - loss rate exceeds that due to the diffuse field by a factor of @xmath69 ( for @xmath70m@xmath63 ) . thus if the direct flux reaching the disc is greater than approximately 0.01 of its total value the direct wind exceeds the diffuse wind . consequently our criterion for the `` transition '' between the two wind profiles is that the optical depth to ionizing photons , @xmath71 , along the disc midplane to the draining radius ( @xmath72 ) satisfies @xmath73 we evaluate the optical depth as @xmath74 , where @xmath75 is the column density along the disc midplane . additionally , once the inner disc has drained we define the inner edge @xmath19 to be the radius at which the optical depth reaches this critical value . after draining we adopt the mass - loss profile derived in paper i ( for @xmath23 ) : @xmath76 where @xmath77 here @xmath78 is the case b recombination coefficient for atomic hydrogen at @xmath79k , which has a value of @xmath80@xmath81s@xmath6 @xcite . we adopt the best - fitting scaling constants from paper i of @xmath82 and @xmath83 , and set @xmath84 ( following @xcite , cgs01 ) . fig.[fig : uv_switch_plus ] shows the evolution of the surface density for the reference model . this model has @xmath70m@xmath63 , @xmath62m@xmath63 , @xmath64m@xmath63yr@xmath6 and @xmath65au . the results shown are for @xmath85s@xmath6 . the inner disc begins to drain at @xmath86myr , and the disc is cleared to the outer grid radius ( 2500au ) at @xmath87myr . at the point where the inner disc begins to drain the total disc mass remaining is approximately 0.002m@xmath63 , or approximately 2m@xmath88 . in order to investigate the dependence on the value of @xmath18 the same disc model was run with @xmath89 ( and therefore best - fitting parameters of @xmath90 and @xmath91 , see paper i ) . in this case the inner disc draining is identical to the case of @xmath23 , as the early evolution depends only on the diffuse wind and is independent of @xmath18 . the outer disc is cleared somewhat more slowly by the direct wind than for @xmath23 , with the outer grid radius not reached until @xmath92myr . however this represents a change of less than a factor of two in the time required to clear the outer disc , so we are satisfied that the choice of @xmath18 is not a significant factor in evolution of the outer disc . a second version of this model was run with @xmath93 , in order to investigate the effect of varying the viscosity law . the numerical scaling constants in the similarity solution vary with the viscosity law @xcite , but while the details differ the qualitative behaviour of the model is unchanged . the steeper surface density profile that results from this new viscosity law means that the inner disc draining is somewhat slower than in the reference model , but this is countered by a more rapid clearing of the outer disc ( as a smaller fraction of the disc mass now resides at large radii ) . the disc is still dispersed on a time - scale some 12 orders of magnitude shorter than the disc lifetime , when the disc is dispersed . ] , and the only significant consequence of changing the viscosity law is the expected modification of the surface density profile . thus , as predicted in section [ sec : timescales ] , when direct photoevaporation is taken into account the entire disc is dispersed on a time - scale approximately 2 orders of magnitude shorter than the disc lifetime . thus it seems that this model has solved the `` outer disc problem '' that affected the original uv - switch model of cgs01 . in order to compare to observed data , however , it is necessary to model the observable properties of the disc . in order to compare the results of the model to observed data it is necessary to use the model to generate spectral energy distributions ( henceforth seds ) . following @xcite and cgs01 , we assume that the disc is vertically isothermal and emits as a blackbody . therefore the flux emitted by the disc at frequency @xmath26 is to denote both frequency and kinematic viscosity . @xmath26 denotes frequency only in equations [ eq : nu1][eq : nu2 ] , and it should be obvious from context to which quantity the symbol refers . ] @xmath94 dr \ , , \ ] ] where @xmath95 is the earth - star distance ( taken to be 140pc , the distance of the taurus - auriga cloud ) , @xmath96 is the planck function and @xmath97 is the inclination angle of the disc ( @xmath98 is face - on ) . @xmath99 is the radial temperature profile of the disc , and the optical depth @xmath100 is evaluated as @xmath101 we adopt a standard power - law for the dust opacity @xmath102 @xcite : @xmath103 we add the stellar contribution to the sed as the total flux emitted by a blackbody of temperature @xmath104 and radius @xmath105 . the values of these two parameters are taken from the pre - main sequence models of ( * ? ? ? * kindly provided in electronic form by chris tout ) . tt stars are contracting due to gravitational collapse , so for a given stellar mass both @xmath104 and @xmath105 vary somewhat with age . we adopt the median age of the taurus - auriga cloud , 2myr @xcite , throughout . obviously the disc temperature adopted is crucial to the resultant sed . further , we wish to investigate the effect of varying the stellar mass in our models , so it is necessary to define @xmath99 in such a way that it only depends on the stellar mass @xmath106 . we adopt a `` flared reprocessing disc '' @xcite power - law profile : @xmath107 where the normalisation condition is set by the dust destruction radius @xmath108 . at radii smaller than this the temperature is set to zero , as no dust can survive here and we assume that the opacity due to the gas alone is negligible . we adopt a dust destruction temperature of @xmath109k . additionally , we adopt a minimum disc temperature of 10k to account for external heating of the disc ( e.g. cosmic rays or diffuse uv , see @xcite ) . by assuming that the stellar irradiation balances the emission from the vertical disc edge at @xmath108 we find that @xmath110 where @xmath111 is a constant of proportionality that reflects how efficiently the disc radiates . 100% efficiency ( the `` small grains approximation '' ) gives @xmath112 , but by comparison to the observed `` median sed '' for taurus - auriga ( * ? ? ? * see fig.[fig : median_sed ] below ) we find that @xmath113 provides a better fit to the data . in adopting a value of @xmath114 we essentially assume that the disc edge at @xmath108 does not radiate as a blackbody ( i.e. it radiates with less than 100% efficiency ) , an assumption verified by detailed radiative transfer models ( e.g. * ? ? ? we note at this point that the disc temperature at the `` draining radius '' of @xmath115 does not vary significantly with stellar mass . obviously @xmath9 scales linearly with @xmath106 , but the effect of this on the disc temperature is offset by the variation of @xmath104 with stellar mass . we find that @xmath116k for @xmath117m@xmath63 , but is only slightly smaller ( 410k ) for @xmath118m@xmath63 . consequently , while the draining radius scales linearly with stellar mass , the colour change resulting from inner disc draining is not especially sensitive to stellar mass . in addition to evaluating the seds , we use the predicted seds to generate magnitudes in the various photometric bands of the _ 2-micron all sky survey _ ( henceforth 2mass ) and the _ spitzer space telescope_. we use the filter transmission functions given on the 2mass and _ spitzer science center _ websites , and use the given zero - point fluxes to convert these fluxes to magnitudes . in order to test this model we compare the predicted sed with the `` median sed '' of ctts in taurus - auriga @xcite . we adopt stellar parameters of @xmath119k and @xmath120r@xmath63 , consistent with @xmath1210.4m@xmath63 ( depending on age ) , and use the surface density profile of the reference disc at @xmath122 . as seen in fig.[fig : median_sed ] , the model reproduces the observed data well out to a wavelength of @xmath123@xmath16 m , but rather under - predicts the flux at millimetre wavelengths . this apparent error is not of great concern , however , as in the initial disc configuration mass is mostly confined to small radii , with the result that the disc is optically thick at millimetre wavelengths . at later times the disc expands and becomes optically thin , boosting the millimetre flux . additionally , once the disc is optically thin the emitted flux is very sensitive to the total disc mass , and our reference model has rather a low initial disc mass ( 0.05m@xmath63 ) . more massive discs at later times ( once the disc has spread beyond the scale radius @xmath51 ) match observed millimetre fluxes much better . in the _ spitzer _ irac bands , our model gives colours of @xmath124-[4.5]=0.46 $ ] , @xmath125-[5.8]=0.57 $ ] , and @xmath126-[8.0]=0.91 $ ] . these compare favourably with the values for the median sed ( @xmath124-[4.5]=0.40 $ ] , @xmath125-[5.8]=0.52 $ ] , and @xmath126-[8.0]=0.83 $ ] , @xcite ) , suggesting that the model predicts irac colours to an accuracy of around @xmath127 mag . in order to study the behaviour of the sed a series of disc models were run with different stellar masses . as seen above , the outer disc evolution is not especially sensitive to the value of @xmath18 , so when considering direct photoevaporation we adopt the best - fitting wind profile for @xmath23 . there are therefore five free parameters in the disc model : stellar mass @xmath106 , initial disc mass @xmath55 , initial accretion rate @xmath61 , scaling radius @xmath51 and ionizing flux @xmath8 . therefore it is necessary to evaluate these parameters self - consistently in order to study the effect of a single parameter . in order to achieve this we assume that both the initial disc mass and the disc scaling radius scale linearly with @xmath106 , and we adopt the normalisation conditions @xmath128 and @xmath129 the initial accretion rate is dependent on the disc viscosity . as noted above , we adopt a viscosity law of the form @xmath130 . we now fix the normalisation by assuming that that the viscosity parameter @xmath52 @xcite and disc aspect ratio @xmath18 are independent of @xmath106 . the orbital and viscous time - scales at @xmath51 are related by @xmath131 the relationships between the accretion rate , viscosity and viscous time - scale were given in equations [ eq : visc_time ] & [ eq : mdot_visc ] . we combine these expressions to fix the initial accretion rate as @xmath132 for typical parameters equation [ eq : tnu_torb ] gives @xmath133 , where @xmath134 is simply the keplerian orbital time at @xmath51 . ( this gives @xmath135yr at @xmath65au . ) thus the disc parameters are specified in a manner which depends only on the stellar mass @xmath106 , due to the manner in which @xmath55 and @xmath51 are specified . this fiducial disc model was evaluated for stellar masses of @xmath136 , 0.5 , 1.0 & 2.0m@xmath63 , and for each model the sed was evaluated as a function of time at an inclination angle of @xmath137 ( i.e. @xmath138 , the mean inclination of a random sample ) . we set @xmath139s@xmath6 in all of these models . in order to explore parameter space further four additional disc models were run for @xmath140m@xmath63 , with @xmath141au , @xmath142s@xmath6 , @xmath143 and @xmath144 . lastly the sed for the fiducial 1m@xmath63 model was also evaluated for an inclination angles of @xmath98 ( i.e. a face - on disc ) and @xmath145 ( i.e. a nearly edge - on disc ) . thus one group of models explores the effect of stellar mass on the observed disc emission , while the second group explores the consequences of varying the disc parameters for a fixed stellar mass . magnitudes were evaluated in the @xmath146 , @xmath147 and @xmath148 2mass bands , and the four irac bands ( which have central wavelengths of 3.6 , 4.5 , 5.8 & 8.0@xmath16 m respectively ) . in addition magnitudes were calculated in the _ spitzer _ 24@xmath16 m mips bandm mips band , 7.3jy , is rather uncertain . consequently the absolute value of the magnitudes in this band are subject to systematic errors . however this merely shifts the zero - point of the magnitude system , and any trends are unaffected . ] . the flux at @xmath149@xmath16 m was also measured , for comparison to sub - millimetre ( scuba ) observations . each of these four wavebands ( observed by 2mass , irac , mips and scuba ) probes a different region of the disc , with each providing different constraints . the 2mass bands primarily observe the stellar flux , and therefore provide a valuable normalisation condition . the irac bands probe the inner disc , at the draining radius and smaller radii , while the 24@xmath16 m mips band probes the emission at somewhat larger radii , beyond the initial draining radius . the emission in both the irac and mips bands is ( mostly ) optically thick ( the longer wavelength bands show weak optical depth effects ) , and so is sensitive only to the disc temperature and to whether or not the disc has drained . therefore small changes in the disc model do not have a significant effect here . lastly the flux at 850@xmath16 m , as measured by scuba , measures optically thin emission from the entire disc , and is therefore rather sensitive to a number of the parameters in the disc model . we now seek to compare the results from our disc models to observed data . in order to do this we have created a composite dataset , using observations taken from the literature . we use data from recent surveys of the taurus - auriga cloud , from the irac observations of @xcite and the scuba observations of @xcite . ( @xcite also list 2mass magnitudes for all sources . ) we include only the sources which are unambiguously included in both samples , rejecting any binaries which are resolved by irac but not by scuba . this leaves a total of 42 objects : 29 ctts , 12 wtts , and the possible transition object coku tau/4 ( see * ? ? ? * ; * ? ? ? all but 3 of the ctts are detected by scuba , with upper limits only found for dp tau , cida 11 and cida 12 . ( these latter two objects are close to the brown dwarf limit and may not be `` true '' tts . ) by contrast only 1 of the wtts ( lkh@xmath52332 g1 ) is detected at 850@xmath16 m , confirming that , in general , disc dispersal occurs simultaneously over the entire radial extent of the disc ( see discussion in * ? ? ? .table showing the parameters used in the various disc evolution models , and also the disc lifetimes found for each model . the column labelled @xmath150 indicates the time at which the inner disc was drained ( i.e. when the model switches from the diffuse to direct wind profile ) , while the column labelled @xmath151 indicates the time at which the entire disc was dispersed . [ cols="^,^,^,^,^,^,^ " , ] the results of our models are shown in table [ tab : disc_models ] and figs.[fig : k3k8][fig : k4k24 ] . table [ tab : disc_models ] shows the disc lifetimes predicted by the model , which are entirely consistent with disc lifetimes of 110myr and dispersal times of order @xmath0yr ( as derived from observations , e.g. * ? ? ? * ; * ? ? ? fig.[fig : k3k8 ] shows evolutionary tracks on a @xmath152 $ ] / @xmath153 $ ] two - colour diagram ( analogous to previously published @xmath154 / @xmath155 plots , e.g. * ? ? ? * ; * ? ? ? the data points show a clear gap between the loci of ctts and wtts which the tracks reproduce well , showing a rapid transition across the gap . the disc emission is optically thick , and so we see class ii colours before the disc is cleared followed by class iii colours afterwards . we also see that stellar mass and disc inclination angle are the dominant effects at these wavelengths . for a given stellar mass and inclination angle different disc models show very similar tracks , as the infrared emission is generally optically thick . consequently the tracks depend primarily on the temperature profile , and are insensitive to the parameters of the disc model . minor optical depth effects are seen at 8@xmath16 m in the models with @xmath142s@xmath6 and @xmath144 , but these are not significant . the models struggle to reproduce the extreme points in the ctt distribution , at both the red and blue ends , and indeed the @xmath156 $ ] colours of our model during the ctt phase are rather redder than the observed data . however there are several factors which can account for this . the sed is rather sensitive to the adopted stellar temperatures , and the disc emission can also be increased by so - called `` accretion luminosity '' . this arises due to viscous heating of the disc , and boosts the disc emission . it is omitted from our sed model but can be significant in ctts with high accretion rates ( e.g. * ? ? ? . additionally , our model does not include emission from the inner disc edge ( or `` wall '' ) at the dust destruction radius , which may contribute significantly to the sed at @xmath157@xmath16 m ( e.g. * ? ? ? we note also that our models predict rather bluer class iii colours than those observed . this is due to the rather unrealistic use of a blackbody stellar spectrum , and we do not consider it to be a serious problem with the model . more interesting is the evolution of the millimetre flux , as shown in fig.[fig : scuba ] . the millimetre emission is mostly optically thin , and so is much more sensitive to the disc mass than the emission at shorter wavelengths . however as it is optically thin , it is relatively insensitive to the inclination angle . the behaviour of the tracks in the @xmath158(850@xmath16 m ) / @xmath153 $ ] plane , shown in fig.[fig : scuba ] , is explained as follows . the initial evolution shows an increasing 850@xmath16 m flux at fixed infrared colour , where the infrared colour is dependent on both stellar mass and disc inclination angle . the increase in 850@xmath16 m flux occurs because the disc viscously expands from its initial configuration , which is optically thick at 850@xmath16 m . however soon accretion takes over and the flux , now optically thin and simply proportional to the disc mass , declines at the total disc mass decreases . the tracks fall vertically ( as the infrared emission is optically thick ) until the inner disc is drained , at which point the @xmath153 $ ] colour rapidly `` jumps '' to a stellar value at a fixed millimetre flux . the level of this flux is determined by the disc mass and temperature at this point , and does not change significantly as the inner disc drains . this is due to the @xmath130 viscosity law , which forces most of the disc mass to reside at large radii . draining occurs at a fixed value of the disc accretion rate , approximately equal to that of the diffuse wind . however the disc accretion rate can be expressed as @xmath159 @xcite . therefore for a fixed stellar mass we expect that the level of the 850@xmath16 m flux when the inner disc drains should depend only on the ionizing flux and the viscosity law . this is verified in fig.[fig : scuba ] , where the models with increased @xmath8 and lower viscosity show significantly larger 850@xmath16 m fluxes as the inner disc drains . ( the total disc mass at this point in the fiducial model is 0.002m@xmath63 . in the model with @xmath142s@xmath6 the disc mass at draining is 0.006m@xmath63 . ) at this point the evolution `` stalls '' for @xmath3yr . this occurs because most of the disc mass resides at large radii , so the 850@xmath16 m flux remains approximately constant while the inner disc is cleared . once direct photoevaporation starts to clear the disc at large radii the 850@xmath16 m flux falls rapidly to a very low level . the original uv - switch model ( cgs01 ) , which omitted direct photoevaporation , resulted in a millimetre flux of a few mjy which remained at very late times . this is clearly not a problem with our new model , which predicts millimetre fluxes of order @xmath160mjy ( the stellar contribution ) once the disc has been cleared . however the model predicts that for around @xmath0yr objects should show stellar near- to mid - infrared colours but retain millimetre fluxes of @xmath16110mjy . this may explain the two objects detected in this region by @xcite : coku tau/4 and lkh@xmath52332 g1 . these detections are near to the sensitivity limit of current observations . our model predicts that a factor of @xmath162 increase in sensitivity should results in the detection at millimetre wavelengths of an increased number of sources with class iii infrared seds , representing a few percent of the total population . this represents a valuable future test of this model . another interesting result of the models is the predicted behaviour in the mid - infrared at 2050@xmath16 m . this region of the spectrum probes the region outside the draining radius , but in a wavelength range where the emission is ( mostly ) expected to be optically thick . to date few observations have been made here , but this should be remedied in the near future by mips observations . fig.[fig : k4k24 ] shows the predicted tracks in the @xmath163 $ ] / @xmath164 $ ] two - colour diagram . here we initially see a slow decline in the 24@xmath16 m emission due to weak optical depth effects . the `` inside - out '' manner of the disc clearing causes the emission in the near infrared to decline more rapidly than that at 24@xmath16 m , with the models predicting that a significant population of sources , again at the few percent level , should show class iii colours in the irac bands but significant excesses at 24@xmath16 m . forthcoming observations from _ spitzer _ will therefore provide another valuable test of the model . lastly , we note that it is possible to constrain some disc parameters from the model results . by demanding that disc lifetimes be in the 110myr range , as seen in observations ( e.g. * ? ? ? * ) , it is possible to constrain the disc viscosity , albeit rather weakly . for initial disc masses of 0.15@xmath106 the ratio of viscous to orbital times must be @xmath165 , and certainly less than 5000 . for realistic @xmath18 ratios this suggests a viscosity parameter in the range @xmath166 . there is no real consensus as to what value of @xmath52 can be produced by models of angular momentum transport in discs , but fiducial values tend to be of order @xmath167 @xcite . thus these derived values of @xmath52 are near to the upper limit of those predicted by current models of angular momentum transport in discs , and may pose problems for these models . similarly , by comparing the models to observed millimetre fluxes ( as in fig.[fig : scuba ] ) it is possible to place indirect constraints on the ionizing flux @xmath8 . the fact that few ctts are not detected at 850@xmath16 m at the 10mjy level , and that similarly few wtts are detected at the same level , suggests that the millimetre flux during the `` transition '' phase must be in the 110mjy range for most objects . ( if this were not the case we would expect to see either many more wtts detected or many more ctts not detected at the 10mjy level at 850@xmath16 m . ) as seen in fig.[fig : scuba ] this level depends on both the disc and stellar parameters , and not solely on @xmath8 . however the results suggest ionizing fluxes in the range @xmath11@xmath168s@xmath6 , a range consistent with our previous estimates @xcite . there are obvious limitations to the model . firstly , we note that the manner in which we treat the disc temperature and viscosity profiles are not self - consistent . the derived wind profiles ( from paper i ) assume a constant @xmath18 , but throughout this paper we adopt a disc temperature profile which implies disc flaring . further , although the temperature profile adopted when evaluating the sed does result in a @xmath130 viscosity law , the normalisation of the two parametrizations is inconsistent . however the viscosity depends primarily on the midplane temperature , while the emitted sed depends on the surface temperature . these are most likely not the same ( e.g. * ? ? ? * ) , and so such an inconsistency is not unreasonable . therefore while there are minor inconsistencies in the disc model , they do have some physical motivation and they do not have a strong effect on the model results . in addition the sed model is rather simple , employing a power - law for the disc temperature structure and neglecting several possibly important factors , such as viscous heating or emission from the inner wall . further , the temperature normalisation is rather sensitive to the stellar temperature adopted , as seen in equation [ eq : temp_norm ] . the flux emitted by the disc is very sensitive to disc temperature , so the output seds are in turn rather sensitive to both the stellar temperature and the normalisation condition . our models struggle to reproduce the reddest and bluest ctts in the data and , as seen in fig[fig : k3k8 ] , our simple model predicts rather redder @xmath156 $ ] colours than those observed during the ctt phase . however this is easily remedied by small alterations to the sed model , and we note that our simple treatment of the disc neglects several possibly important effects ( see section [ sec : results ] ) , such as accretion luminosity or emission from the inner disc wall . additionally , our model uses a single temperature blackbody to model the stellar flux . model atmospheres can differ markedly from blackbody spectra in the near - infrared ( e.g. * ? ? ? * ) , and this may also affect the colours obtained from the sed model . the model also struggles to reproduce the handful of objects with the largest observed 850@xmath16 m fluxes . however , as seen above , the emission at millimetre wavelengths is extremely sensitive to the disc parameters . additionally , these objects tend to have rather unusual sed slopes in the sub - millimetre @xcite , and so we do not consider this to be a serious problem with the model . a further consideration is the manner in which the transition from the diffuse to direct regime is treated ( see section [ sec : disc_model ] ) . we define a critical value for the optical depth to ionizing photons along the disc midplane and use it to switch instantaneously between the two wind parametrizations . a more realistic treatment would gradually increase the strength of the direct field as the inner disc is drained . additionally , the behaviour of the diffuse field as the inner disc drains should be considered in more detail , as the diffuse wind model assumes that the underlying disc is always optically thick to ionizing photons . the transition occurs very rapidly , and its treatment does not have a strong effect on the overall evolution of the disc . the transition is important , however , when we consider the detailed conseqences for the observed sed during clearing . this is because the inner disc becomes optically thin to infrared emission ( perpendicular to the disc midplane ) rather earlier than it becomes optically thin to lyman continuum photons ( along the midplane ) . objects appear as `` inner hole '' sources during this phase , and the wavelength at which the disc emission `` cuts in '' is rather sensitive to the sed model . recently @xcite have observed a strong correlation in passive ( non - accreting ) discs , between stellar spectral type and the wavelength at which outer disc emission becomes visible . they define `` passive discs '' as sources which show excess emission at 1012@xmath16 m but lack any near - infrared excess , and in a survey of binary secondary stars they find that that stars of later spectral type are more likely to possess passive discs than stars of earlier spectral type . the binary separations are large , so within the framework of a photoevaporation model they attribute this to the linear scaling of the draining radius with stellar mass . thus low mass stars have smaller holes , which they argue should contribute disc emission at shorter wavelengths than the corresponding holes in higher mass stars . this interpretation is predicated on the assumption that the smaller holes are hotter . however , in our models , smaller holes turn out to be only modestly hotter , since the stellar photospheric temperature ( which sets the temperature of the reprocessing disc ) is lower for less massive stars . further , this interpretation remains valid only if the inner disc clearing takes significantly longer than the outer disc clearing . if the outer disc clearing dominates the dispersal time then the distribution of observed hole sizes at any given time is independent of stellar mass . therefore a more accurate treatment of the transition from the diffuse to direct wind is necessary in order to make such detailed predictions . we also stress that a better treatment of the disc sed , in particular the contribution from the inner wall , is needed before we can assess whether our models can reproduce the passive discs observed by @xcite . we note also that our model neglects the effects of non - ionizing far - ultraviolet ( fuv ) radiation on the evolution of the disc . fuv radiation heats the disc to lower temperatures than ionizing radiation ( @xmath165k cf . @xmath169k , * ? ? ? * ; * ? ? ? * ) and therefore fuv - driven winds are launched from rather larger radii than those driven by ionizing radiation , typically launching at radii from 20100au . @xcite have shown that external fuv radiation can produce a significant disc wind , which is is dominated by flow from the outer disc edge . however this wind relies on the presence of a strong interstellar fuv radiation field , which is only present in the immediate vicinity of massive o- or b - type stars . self - consistent models of fuv heating from the central object have not yet been completed , but it seems likely that fuv radiation can drive a wind from the outer disc edge in this case also . the interplay between mass - loss from the outer disc edge ( from fuv heating ) , mass - loss from smaller radii ( from lyman continuum heating ) and viscous evolution will provide an interesting area for future study . while the details of such a calculation are beyond the scope of this paper , we note that if the stellar fuv radiation field can drive a significant wind from the outer edge of the disc then this may significantly shorten the disc lifetime in any given disc model . consequently we expect that the inclusion of such a wind will reduce the viscosity ( @xmath52 ) required to satisfy observational constraints on disc lifetimes . another important consideration is dust . dust is responsible for all of the emission discussed above , and our sed modelling assumes a constant dust - to - gas ratio throughout . however it is not at all obvious that the dust in the disc remains coupled to the gas . there are several forces which act on dust grains in discs , all of which may be important here ( see e.g. * ? ? ? in addition to gravity the dust grains feel a drag force from the gas , and grains can both grow and be ground down through collisional processes . grains can also feel a force due to radiation pressure . in most cases the optical depth through the disc is such that this force is negligible , but near to the inner disc edge it may become significant . radiation from the star is also responsible for the poynting - roberston ( p - r ) effect , which can cause grains to lose angular momentum and slowly spiral inwards . again , this effect is not significant if the grains are shielded from the stellar radiation field , but may become significant once the gas disc begins to drain . additionally , as noted in paper i , any dust remaining in the `` inner hole '' may absorb some ionizing photons , reducing the efficiency of the wind and lengthening the dispersal time - scale . the net effect of these forces on the dust in the disc is difficult to predict without making detailed calculations . the evolution of dust in the original uv - switch model has been studied in some detail by @xcite . they find that gas drag causes millimetre - size dust grains to accrete on to the star more rapidly than the gas unless the grains are rather `` fluffy '' ( i.e. have rather low densities ) . in this model the grains are removed from the outer disc before the inner disc is drained , and the decline in disc emission is primarily due to the rapid accretion of the grains rather than global evolution of the disc . however the sub - millimetre observations by @xcite do not detect a significant fraction of objects which have no dust in the outer disc but are still accreting gas from their inner disc , suggesting that the dust is removed too quickly in the model of @xcite . additionally , @xcite find that the force exerted on the grains by the wind is always less than gravity , so little dust is carried away by the photoevaporative wind . this is not of great significance in their model , as the grains are removed from the outer disc by gas drag , but may be a problem when considering direct photoevaporation of the outer disc . in this case gas drag can not migrate the grains beyond the inner disc edge , and the wind can not carry away any dust grains found in the outer disc . consequently we would expect the dust grains in the outer disc to remain after the gas has been dispersed . however the time - scales for grains to be removed by either collisional processes or by p - r drag are very short , of order @xmath170@xmath79yr at radii of a few au ( e.g. * ? ? ? consequently any dust `` left behind '' in this manner is expected to be removed sufficiently quickly as to have a negligible effect on the evolution of the sed . however the analysis of @xcite does not consider the effects of radiation - powered forces on the grains . this is not a significant problem in their model as the dust is shielded from stellar irradiation , but may well become significant in our model once the inner disc begins to drain . @xcite found that radiation pressure in discs with low opacity can result in a clumping instability in the dust grains . in our model this effect may well become important when the inner disc drains , as the optical depth is essentially zero inside the inner disc edge . such clumping increases grain collision rates , and therefore may have important consequences for planet formation theories as well as on the observed sed . we also note that any dust exposed to the stellar radiation field in this manner will attenuate the direct radiation field responsible for the photoevaporative wind ( see discussion in paper i ) . additional effects , such as grain growth and dust replenishment , may also be significant . in short , it is unlikely that the disc evolves with a constant gas - to - dust ratio , as we have assumed . however the manner in which the dust evolves is not at all obvious , and we make no attempt to model it here . we also note in passing that a handful of gas - poor discs have recently been observed by @xcite , and suggest that photoevaporation may have played a significant role in their evolution . lastly , we note that one consequence of the model is that all objects are predicted to pass through an `` inner hole '' phase , where the disc is drained close to the star but remains `` normal '' at larger radii . several such objects have now been observed ( e.g. gm aur , @xcite ; dm tau , @xcite ; tw hya , @xcite ; coku tau/4 , @xcite ) , and it has been suggested that these may represent a class of `` transition objects '' , intermediate between the ctt and wtt states . it is not clear that these objects represent a homogeneous group ( in fact this seems unlikely ) , but it is interesting to compare them to the predictions of our model . despite the presence of holes in their ( dust ) discs , both gm aur and dm tau have large accretion rates ( on to the star ) , of order @xmath171m@xmath63yr@xmath6 @xcite additionally , in the case of gm aur the observed inner hole is incompatible with a standard gas - to - dust ratio @xcite . as such gm aur is obviously inconsistent with the photoevaporation model , and some other process , such as grain growth or the presence of a planet , must be at work here . both gm aur and dm tau have also been observed to have relatively large outer disc masses , of order 0.050.1m@xmath63 @xcite . in the photoevaporation model the disc mass at draining is approximately equal to the product of the ( diffuse ) wind mass - loss rate ( @xmath172m@xmath63yr@xmath6 ) and the viscous evolution timescale ( @xmath2yr ) , and , as seen in section [ sec : ref_results ] , is typically 0.001m@xmath63 ( plus or minus around an order of magnitude ) . therefore it is rather difficult ( though not completely impossible ) for photoevaporation to produce inner holes with outer disc masses as large as those observed in these two sources , again suggesting that photoevaporation is not responsible for the holes observed in these two discs . however we note that in deriving these disc masses @xcite have derived rather small viscosity parameters ( @xmath173 ) , and that larger values of @xmath52 would imply correspondingly smaller disc masses ( see also * ? ? ? tw hya has a `` hole '' in the disc inside 34au , but still shows some dust and gas emission from inside this radius @xcite . it also has a measured accretion rate of @xmath174m@xmath63yr@xmath6 @xcite , and it has been suggested that a planet may have formed ( or be forming ) in the disc . the disc mass is rather larger than that predicted by the photoevaporation model , but the low accretion rate is similar to that produced as the inner disc drains in the photoevaporation model . therefore in principle it may be possible to model the sed with a photoevaporation model , although in order to reproduce the observed accretion rate the model requires that the object be observed during a rather short `` window '' in the evolution ( while the inner disc is in the process of draining ) . consequently it is not yet clear whether or not photoevaporation plays a role here . coku tau/4 , however , is almost entirely devoid of material inside 10au and shows no evidence of accretion on to the star @xcite . the outer disc mass of 0.001m@xmath63 @xcite is similar to that predicted by our model ( see section [ sec : ref_results ] and discussion above ) . we have seen above that the evolutionary tracks produced by the photoevaporation model can reproduce the sed of coku tau/4 well , and at present this object seems to be entirely consistent with the predictions of the photoevaporation model . we note that much current work has invoked the presence of a formed or forming planet in order to explain such holes ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? whilst this may well turn out to be the case , we emphasise that other mechanisms , such as photoevaporation , can also produce holes in discs , and that holes do not necessarily imply the presence of a planet . indeed it seems likely that photoevaporation and planet formation occur simultaneously during the evolution of protoplanetary discs , and it is not yet clear whether photoevaporation acts to accelerate or retard planet formation . _ spitzer _ is expected to discover many more objects similar to those discussed here , and only with more data will the nature of these objects become apparent . however the current data suggest that the inner hole sources discovered to date are far from a homogeneous class of objects . we have presented a new model for the evolution of protoplanetary discs . our model combines viscous evolution with photoevaporation of the disc by stellar radiation , and includes the effects of the direct radiation field at late stages of the evolution ( using the wind prescription derived in paper i ) . we have constructed a numerical model for the evolving disc , and used a simple prescription to model the behaviour of the spectral energy distribution as the disc evolves . our fiducial model predicts that the disc is completely dispersed on a time - scale of order @xmath0yr after a lifetime of a few myr , consistent with observationally derived time - scales . our results are consistent with observational data across a broad range in wavelengths , and allow us to place weak constraints on some model parameters : we derive ionizing fluxes in the range @xmath11@xmath168s@xmath6 and viscosity parameters ( @xmath52 ) in the range @xmath1750.4 . we also make predictions as to what will be seen in future observations in the mid - infrared and at millimetre wavelengths . to date this is the only model of disc evolution which can reproduce the rapid disc dispersal seen in observations of t tauri discs . we note also that the model suggests that all evolving discs pass through a short `` inner hole '' phase . during this phase the outer disc mass is of order 0.001m@xmath63 , and we predict that inner hole sources should represent around 110% of the observed population of tts . we compare the predictions of the model with the handful of inner hole sources observed to date and show that some , but not all , of these objects are consistent with the predictions of our model . we thank doug johnstone , andreea font and ian mccarthy for providing numerical results from their hydrodynamic models . we also thank doug johnstone for a useful referee s report , which improved the clarity of the paper . we thank chris tout for providing data from his pre - main - sequence evolutionary models . rda acknowledges the support of a pparc phd studentship . cjc gratefully acknowledges support from the leverhulme trust in the form of a philip leverhulme prize . parts of this work were supported by nasa under grant nng05gi92 g from the beyond einstein foundation science program . 99 adams , f.c . , hollenbach , d. , laughlin , g. , gorti , u. , 2004 , apj , 611 , 360 alexander , r.d . , clarke c.j . , pringle , j.e . , 2005 , mnras , 358 , 283 alexander , r.d . , clarke c.j . , pringle , j.e . , 2006a , mnras , in press ( paper i ) andrews , s.m . , williams , j.p . , 2005 , apj , 631 , 1134 armitage , p.j . , clarke , c.j . , tout , c.a . , 1999 , mnras , 304 , 425 armitage p.j . , clarke c.j . , palla f. , 2003 , mnras , 342 , 1139 baraffe , i. , chabrier , g. , allard , f. , hauschildt , p.h . , 1998 , a&a , 337 , 403 beckwith , s.v.w . , sargent , a.i . , chini , r.s . , gsten , r. , 1990 , aj , 99 , 924 bontemps , s. , andr , p. , kaas , a.a . , et al . , 2001 , a&a , 372 , 173 calvet , n. , dalessio , p. , hartmann , l. , wilner , d. , walsh , a. , sitko , m. , 2002 , apj , 568 , 1008 calvet , n. , et al . , 2005 , apj , 630 , l185 clarke , c.j . , gendrin , a. , sotomayor , m. , 2001 , mnras , 328 , 485 ( cgs01 ) chiang , e.i . , goldreich , p. , 1997 , apj , 490 , 368 cox a.n . ( editor ) , 2000 , _ allen s astrophysical quantities _ , aip press , new york dalessio , p. , calvet , n. , hartmann , l. , lizano , s. , cant , j. , 1999 , apj , 527 , 893 dalessio , p. , calvet , n. , hartmann , l. , et al . 2005a , apj , 621 , 461 dalessio , p. , mern , b. , calvet , n. , hartmann , l. , montesinos , b. , 2005b , rev . mex . ast . , 41 , 61 duvert , g. , guilloteau , s. , mnard , f. , simon , m. , dutrey , a. , 2000 , a&a , 355 , 165 font , a.s . , mccarthy , i.g . , johnstone , d. , ballantyne , d.r . , 2004 , apj , 607 , 890 forrest , w.j . , sargent , b. , furlan , e. , et al . , 2004 , apjs , 154 , 443 gorti , u. , hollenbach , d. , 2004 , apj , 613 , 424 gustafson , b..s . , 1994 , ara&a , 22 , 553 haisch , k.e . , lada , e.a . , lada , c.j . , 2001 , apj , 553 , l153 hartmann , l. , 2001 , aj , 121 , 1030 hartmann , l. , calvet , n. , gullbring , e. , dalessio , p. , 1998 , apj , 495 , 385 hartmann , l. , megeath , s.t . , allen , l. , luhman , k. , calvet , n. , dalessio , p. , franco - hernandez , r. , fazio , g. , 2005 , apj , 629 , 881 hollenbach , d. , johnstone , d. , lizano , s. , shu , f. , 1994 , apj , 428 , 654 hollenbach , d.j . , yorke , h.w . , johnstone , d. , 2000 , in mannings , v. , boss . russell , s.s . , eds , _ protostars & planets iv _ , univ . arizona press , tuscon , p401 hueso , r. , guillot , t. , 2005 , a&a , 442 , 703 johnstone , d. , hollenbach , d. , bally , j. , 1998 , apj , 499 , 758 kenyon , s.j . , hartmann , l. , 1987 , apj , 323 , 714 kenyon , s.j . , hartmann , l. , 1995 , apjs , 101 , 117 klahr , h. , lin , d.n.c . , 2001 , apj , 554 , 1095 klahr , h. , lin , d.n.c . , 2005 , apj , 632 , 1113 liffman , k. , 2003 , pasa , 20 , 337 lynden - bell , d. , pringle , j.e . , 1974 , mnras , 168 , 603 matsuyama , i. , johnstone , d. , hartmann , l. , 2003 , apj , 582 , 893 mccabe , c. , ghez , a.m. , prato , l. , duchne , g. , fisher , r.s . , telesco , c. , 2006 , apj , 636 , 932 muzerolle , j. , calvet , n. , briceo , c. , hartmann , l. , hillenbrand , l. , 2000 , apj , 535 , l47 najita , j. , williams , j.p . , 2005 , apj , 635 , 625 natta , a. , prusti , t. , neri , r. , wooden , d. , grinin , v.p . , mannings , v. , 2001 , a&a , 371 , 186 palla , f. , stahler , s.w . , 2000 , apj , 540 , 255 persi , p. , marenzi , a.r . , olofsson , g. , et al . , 2000 , a&a , 357 , 219 pringle , j.e . , 1981 , ara&a , 19 , 137 pringle , j.e . , verbunt , f. , wade , r.a . , 1986 , mnras , 221 , 169 quillen , a.c . , blackman , e.g. , frank , a. , varnire , p. , 2004 , apj , 612 , l137 rice , w.k.m . , wood , k. , armitage , p.j . , whitney , b.a . , bjorkman , j.e . , 2003 , mnras , 342 , 79 richling s. , yorke , h.w . , 1997 , a&a , 327 , 317 ruden , s.p . , 2004 , apj , 605 , 880 simon , m. , prato , l. , 1995 , apj , 450 , 824 shakura , n.i . , sunyaev , r.a . , 1973 , a&a , 24 , 337 skrutskie , m. f. , dutkevitch , d. , strom , s.e . , edwards , s. , strom , k.m . , shure , m.a . , 1990 , aj , 99 , 1187 stone , j.m . , gammie , c.f . , balbus , s.a . , hawley , j.f . , 2000 , in mannings , v. , boss . russell , s.s . , eds , _ protostars & planets iv _ , univ . arizona press , tuscon , p589 strom , k.m . , strom , s.e . , edwards , s. , cabrit , s. , skrutskie , m.f . , 1989 , aj , 97 , 1451 takeuchi , t. , clarke , c.j . , lin , d.n.c . , 2005 , apj , 627 , 286 tout , c.a . , livio , m. , bonnell , i.a . , 1999 , mnras , 310 , 360 wolk , s.j . , walter , f.m . , 1996 , aj , 111 , 2066

in alloys exhibiting substitutional disorder , the variety of atomic environments manifests itself as a ` disorder broadening ' in their core level binding energy spectra . disorder broadening can be measured experimentally , and in principle can be used to deduce information about specific atomic environments within a sample . however , progress in this endeavor is hampered by the lack of a model for this phenomenon which can treat complex systems . in this work we describe such a model . the model is used to elucidate the relationship between charge transfer , atomic environment , and disorder broadening in complex systems , with a focus on the problem of characterizing the interface quality of cuni multilayers . we also validate the model against the results of _ ab initio _ electronic structure calculations . several counterintuitive aspects of the disorder broadening phenomenon are uncovered , an understanding of which is essential for the correct interpretation of experimental results . for instance , it is shown that systems with inhomogeneous concentration profiles can exhibit disorder broadenings significantly larger than random alloys . furthermore in some systems a ` disorder narrowing ' is even possible .

1402.2201

an atom s core level binding energies depend on its chemical species . this fact has been exploited for decades in order to determine the proportion of different species within a given sample via core level spectroscopy - which provides the distribution of core electron binding energies within a sample . an atom s binding energies also depend on its environment , i.e. , the species of its surrounding atoms . for instance , in a cu metal all atoms have the same environment , and hence the binding energies for a given cu core level type , e.g. 2p@xmath0 , will be the same for all atoms . by contrast , in a cupd alloy exhibiting substitutional disorder the cu atoms exhibit a variety of environments , and hence also a variety of 2p@xmath0 binding energies . such a dispersion in the binding energies has been observed in many alloy systems @xcite , and in theory can be used to deduce information about specific environments within a sample , including the concentration profile on the atomic scale . the viability of achieving this via high kinetic energy photoelectron spectroscopy ( hike ) has recently been demonstrated @xcite . this is promising because hike , unlike other widely - used techniques , is both bulk - sensitive and non - destructive . such environment - resolved spectroscopy would prove useful to the many research areas involving alloys in which segregation plays a crucial role , e.g. metal embrittlement @xcite , and nanocatalyst design @xcite . however , success in this endeavor hinges upon a solid understanding of the relationship between an atom s environment , its electronic structure , and its core level binding energies in alloys . the central quantity with regards to environment - resolved spectroscopy in alloys is the ( alloy - metal ) core level shift ( cls ) , which for a core level bound to an @xmath1 site @xmath2 is defined as @xmath3 where @xmath4 is the binding energy of the core level , and @xmath5 is the binding energy of the core level belonging to the type under consideration in a pure @xmath1 metal . note that this is a site - dependent quantity ; one must determine @xmath6 for all @xmath1 sites in the system under consideration in order to determine the @xmath1 cls distribution . this is problematic for systems exhibiting substitutional disorder on account of their lack of periodicity - which is a prerequisite for treatment within the conventional theoretical framework exploited by most _ ab initio _ methods . one way around this problem is to approximate the system under consideration as periodic , but with a large unit cell , i.e. , a supercell . in this _ supercell approximation _ one determines @xmath6 for all @xmath1 sites , and hopes that the range of environments exhibited by these sites is representative of the ` true ' ( non - periodic ) system , and hence will result in an accurate representation of the true @xmath1 cls distribution . random alloys - the archetype of disordered alloys in which there are no correlations between the species of sites - are the most tractable system exhibiting substitutional disorder to treat theoretically , and have been the focus of both experimental @xcite and theoretical @xcite attempts to understand the distribution of clss in alloys . in this context the dispersion of @xmath1 clss is known as ` disorder broadening ' on account of the increased width of , for example , the cu cls distribution for a cupd random alloy relative to that for a cu metal . sophisticated _ ab initio _ models utilizing the supercell approximation have provided insight into disorder broadening in these systems @xcite . however , most systems of practical interest can not be idealized as random alloys , and unfortunately the complexity of these systems is such that a description of their disorder broadenings using _ ab initio _ models is intractable within the supercell approximation . accurate methods do exist which do not resort to the supercell approximation . _ ab initio _ methods rooted in the coherent potential approximation ( cpa)@xcite have been shown to provide excellent agreement with experiment @xcite . furthermore , they can treat complex systems @xcite . however , cpa - based approaches can not provide detailed information regarding the disorder broadening in complex systems - nor even random alloys . while phenomenological models have been developed which can provide such information @xcite , their accuracy has been questioned @xcite . one criticism is that these models do not take into account ` final state effects ' associated with changes in the valence electron density after photoemission . it has even been claimed that the complexity of the relationship between clss and environment in alloys precludes an accurate alternative to _ ab initio _ methods @xcite . here we present an accurate phenomenological model for clss in alloys which relates @xmath6 - including the final state contribution - to the environment of site @xmath2 . the model , like previous approaches , is charge - transfer based , and provides a simple framework for rationalizing the disorder broadening phenomenon . the layout of this work is as follows . in sec . [ sec : model ] we review the theory which underpins the model , and derive expressions for clss which apply to a wide range of alloy systems . we then apply the model to the problem of characterizing the interface quality of metallic multilayers . this problem has received significant attention on account of its importance to nanotechnology @xcite , and can be restated as follows : what is the degree of ` interface roughening ' @xmath7 in a given sample ? [ fig : figure_cu5ni5](a ) provides an illustration of the multilayer system [ ni@xmath8/cu@xmath8 ] with various @xmath7 - where the square brackets signify that the system at @xmath10 consists of a 10 monolayer stack ni@xmath8/cu@xmath8 repeated throughout all space . in this regard , sec . [ sec : comp_details ] contains details of our calculations , and our results are presented in sec . [ sec : results ] . finally , in sec . [ sec : summary ] we summarize our main findings , and discuss the limitations of the model , and intentions for future work . note that throughout this work we use hartree atomic units unless otherwise stated . to transform energies within hartree atomic units to ev the former quantity should be multiplied by a factor 27.2114 . /cu@xmath8 ] at various @xmath7 ( a ) , and the corresponding model spectra ( b ) . in ( b ) , each red bar corresponds to a different value of @xmath11 : the ordinate of the bar is the cls corresponding to @xmath11 according to eqn . , and its height reflects the frequency of cu sites with that value of @xmath11 . the bars corresponding to @xmath12 , 6 , and 12 are indicated . the dotted curves in ( b ) are spectra for @xmath10.,scaledwidth=50.0% ] our model is based upon the charge - excess functional model @xcite within the non - random approximation @xcite ( nra - cefm ) . after briefly reviewing the nra - cefm in sec . [ sec : nracefm_review ] , in sec . [ sec : nracefm_deltaebi ] we use it to derive a relation between @xmath6 and the environment of site @xmath2 . in the subsequent subsections this relation is itself used to derive expressions for the mean and full width at half maximum ( fwhm ) of various cls distributions . while our expression for @xmath6 presented in sec . [ sec : nracefm_deltaebi ] is valid for _ any _ alloy - subject to the assumptions which underpin the nra - cefm - our expressions for means and fwhms apply only to the class of alloys which can be described as having a concentration profile which varies along one direction from monolayer to monolayer . to elaborate , for an alloy belonging to this class , the quantities @xmath13 for all @xmath1 and @xmath14 describe the concentration profile , where @xmath13 denotes the concentration of @xmath1 sites within monolayer @xmath14 , and we are assuming that the species of sites belonging to a given monolayer @xmath14 are assigned randomly in the appropriate concentrations . note that this class includes random alloys , for which @xmath15 for all @xmath14 , where @xmath16 denotes the global concentration of species @xmath1 . furthermore , note that we use the following convention for labeling the monolayers : monolayer @xmath14 is the @xmath14th monolayer in the direction in which the concentrations of each species are varying . the charge - excess functional model ( cefm ) @xcite has been shown to provide an accurate description of the charge distribution in disordered alloys @xcite . it has also been shown to provide an accurate description of energy differences between alloy configurations with the same _ composition _ @xcite - where by composition we mean a specification of the underlying lattice and the quantities @xmath16 for all @xmath1 . in the cefm the alloy energy is postulated to take the form @xmath17 where @xmath18 is the madelung energy , @xmath19 is the madelung potential of site @xmath2 , @xmath20 is the net charge on site @xmath2 , @xmath21 is the strength of the ` local interactions ' within site @xmath2 which act to keep the charge of site @xmath2 at its ` bare ' value @xmath22 , and @xmath23 is a constant . for all @xmath1 sites , @xmath21 and @xmath22 take the same values @xmath24 and @xmath25 respectively . minimizing @xmath26 subject to the constraint of charge neutrality leads to the following expression @xcite : @xmath27 where @xmath28 and @xmath29 is a lagrange multiplier added to enforce charge neutrality . eqn . describes the _ q - v _ relations , which are borne out in _ ab initio _ calculations to a high degree of accuracy @xcite . this partly explains the success of the cefm . eqn . can be derived within the class of generalized coherent potential approximations ( gcpas ) described by bruno _ @xcite , with the additional - essentially exact @xcite - assumption that charge transfer due to madelung interactions is small . here we consider the cefm within this framework , in which case @xmath25 is equivalent to the charge of an @xmath1 site embedded in the gcpa effective medium for the system under consideration , given the constraint that the site s madelung potential is 0 . in other words , @xmath25 is the charge of an @xmath1 site embedded in the effective medium if the madelung interactions are ` switched off ' . furthermore , @xmath24 is the linear response coefficient relating the madelung potential of the @xmath1 site to its perturbation from @xmath25 , and @xmath23 is the energy of the alloy if the madelung interactions between sites are switched off - which can be expressed as @xmath30 where @xmath31 is the energy of an @xmath1 site embedded in the effective medium if madelung interactions are switched off , and @xmath32 is the total number of sites in the system . conventional cpa calculations do not take into account madelung interactions , and hence for gcpa theories based upon cpa effective media @xmath25 is simply the charge of an @xmath1 site obtained from a conventional cpa calculation , and @xmath31 is the corresponding @xmath1 energy . calculations reveal that gcpa effective media are numerically indistinguishable @xcite for all alloys with the same composition . hence the same applies for the quantities @xmath25 , @xmath24 and @xmath23 : these quantities are transferable between such systems . the complexity of the cefm is significantly reduced if one makes the assumption that @xmath24 takes the same value @xmath33 for all species @xcite . this assumption is known as the non - random approximation ( nra ) , and we use it throughout this work . the non - random approximation is borne out calculations utilizing the single - site locally self - consistent green s function method @xcite - a gcpa method . in the nra - cefm @xmath20 for an @xmath1 site obeys @xcite @xmath34 where : @xmath35 is the number of @xmath36 sites in the @xmath37th nearest neighbor shell of site @xmath2 ; @xmath38 and the quantities @xmath39 and @xmath40 for all @xmath37 depend only on @xmath41 and the underlying lattice _ _ type__. ] - where @xmath42 is the wigner - seitz radius for the system under consideration - and are tabulated in ref . for the fcc , bcc and sc lattices . note that the free parameters @xmath33 and @xmath43 ( for all @xmath44 ) can be obtained from _ ab initio _ calculations or by other means @xcite . furthermore , the values of @xmath35 for all @xmath37 and @xmath36 characterize the environment of @xmath2 . therefore eqn . explicitly relates @xmath20 to the environment of site @xmath2 . eqn . also allows us to interpret an alloy s charge distribution in terms of charge transfer between pairs of _ unlike _ sites as follows @xcite : an @xmath1 site gains a charge @xmath45 from each @xmath36 site in its @xmath37th nearest neighbor shell , with the @xmath36 site losing the opposite amount . this picture allows us to attribute the following physical significance to the quantity @xmath46 : it is a measure of the _ electropositivity _ of species @xmath36 for the given composition . we will now use the nra - cefm to derive an expression for the cls of site @xmath2 . to do this , we first derive an expression for the total energy @xmath26 in terms of the bare charges @xmath22 of all sites . eqn . can be rewritten more explicitly as @xmath47 for the nra - cefm , where we have used eqn . . minimizing this with respect to the site charges , and subject to the constraint of global charge neutrality gives @xcite @xmath48 where @xmath49 denotes the mean value of @xmath22 over all @xmath2 . substituting the above into eqn . and simplifying gives @xmath50 where we have used the fact that @xmath51 now , the following expression holds for the charges at the minimum in @xmath26 @xcite : @xmath52 where @xmath53 @xmath54 denotes the madelung matrix , and @xmath55 denotes the identity matrix . this becomes @xmath56 since @xcite @xmath57 substituting eqn . into eqn . gives @xmath58 which becomes @xmath59 after separating out the terms containing @xmath60 and noting that @xmath61 is a symmetric matrix @xcite . we will use the above equation in a moment . the binding energy of a core level associated with site @xmath62 is @xmath63 where @xmath64 denotes the energy of the alloy s initial state - before photoemission from site @xmath62 , and @xmath65 denotes the energy of the alloy s final state - after photoemission from site @xmath62 . in the complete screening picture @xcite the valence electrons in the final state are assumed to be fully relaxed so to reach their minimum energy configuration . with this in mind , the energy of the final state is simply the energy of the initial state , but with the atomic core within @xmath62 replaced by its photo - ionized analogue . we choose site @xmath62 to belong to species @xmath1 , and denote the ` species ' corresponding to a photo - ionized @xmath1 site as @xmath66 . hence @xmath67 is the energy change if site @xmath62 , originally belonging to species @xmath1 , is transformed into species @xmath66 . eqn . allows us to evaluate the change in @xmath26 due to such a transformation . noting that the terms on the lower line of eqn . are unaffected by the transformation ( since @xmath33 and @xmath25 for all @xmath1 are composition - dependent , and the composition - which we defined in terms of macroscopic quantities - is unaffected by the transformation ) , that the transformation is such that @xmath68 , and also that @xmath69 ( see eqn . ) , we obtain @xmath70 this can be rearranged to give @xmath71 which in turn becomes @xmath72 after using eqn . and noting that @xmath73 @xcite . we emphasize that @xmath74 in the above equation refers to the charge of site @xmath62 _ before _ photoemission . finally , substituting the above equation into eqn . , and relabeling site @xmath62 as site @xmath2 , we obtain the following expression for the cls associated with site @xmath2 : @xmath75 where @xmath76 is composition - dependent . eqn . describes a species- and composition - dependent linear mapping between @xmath20 and @xmath6 . therefore changes in the @xmath1 cls distribution associated with configuration changes - such as an increase in the interface roughening @xmath7 in a multilayer system - directly reflect changes in the @xmath1 charge distribution . furthermore the _ shape _ of the cls distribution is the same as that of the charge distribution on account of the linear nature of the mapping between @xmath20 and @xmath6 . we will elaborate on these points later . henceforth we consider the mean and fwhm of the cls distributions for various groups of @xmath1 sites . note that the fwhm of any random variable @xmath77 is related to its variance by the equation @xmath78 we will use this fact several times below . we will also use the fact that the mean and fwhm of @xmath6 for _ any group _ of @xmath1 sites @xmath79 within a given alloy , as follows from eqn . , are given by @xmath80 and @xmath81 respectively , where @xmath82 and @xmath83 denote the mean and variance of the charge distribution for @xmath79 . consider the group of all sites within a random alloy . for random alloys @xcite @xmath84 and @xmath85 where @xmath86 denotes the variance of @xmath22 over _ all sites in the system _ , and @xmath87 depends only on @xmath41 and the underlying lattice type , and is tabulated in ref . for the fcc , bcc and sc lattices . therefore for random alloys eqns . and yield @xmath88 and @xmath89 where we have defined @xmath90 and used the fact that @xcite @xmath91 for binary alloys consisting of species @xmath92 and @xmath93 the above equations simplify to @xmath94 and @xmath95 for @xmath96 . henceforth we consider the class of systems described at the beginning of this section in which the species concentrations can vary from monolayer to monolayer . let @xmath79 be the set of @xmath1 sites within monolayer @xmath14 . furthermore , let @xmath97 and @xmath98 denote the mean and fwhm respectively of the cls distribution for @xmath79 . a similar notation will be used later for other quantities , e.g. @xmath99 and @xmath100 . we will now derive an expression for @xmath97 . consider a site @xmath2 within @xmath79 . from eqn . , @xmath20 can be expressed as @xmath101 where @xmath102 denotes the number of @xmath36 sites in the @xmath37th nearest neighbor shell of @xmath2 which are in monolayer @xmath103 . taking the mean over all @xmath104 gives @xmath105 now , @xmath102 over @xmath104 describes a random variable distributed according to the multinomial distribution . specifically , @xmath102 is the number of times outcome @xmath36 occurs in @xmath106 trials , given the probability of outcome @xmath36 in a single trial is @xmath107 , where @xmath108 is the total number of sites in the @xmath37th nearest neighbor shell of any site @xmath109 which also belong to _ one _ monolayer which is ` @xmath110 monolayers away ' from @xmath109 - with @xmath111 referring to the monolayer which contains site @xmath109 itself . the properties of the multinomial distribution are such that @xmath112 , and hence the above equation becomes @xmath113 this can be rewritten as @xmath114 where we have defined @xmath115 finally , substituting eqn . into eqn . gives @xmath116 for species @xmath92 in a binary alloy the above simplifies to @xmath117 the quantities @xmath118 depend on the lattice type , @xmath41 , and the set of lattice planes which constitute the monolayers . they determine the coupling between monolayers with regards to @xmath99 and @xmath97 . it can be shown that for the concentration profile @xmath15 for all @xmath14 , i.e. , a random alloy , the above equations become equivalent to those given earlier for random alloys . this is the case since @xmath119 which can be shown by noting that @xmath120 and @xcite @xmath121 we will now derive an expression for @xmath98 . taking the variance of eqn . over @xmath104 gives @xmath122 after noting that the random variables @xmath123 and @xmath124 , defined by considering @xmath102 and @xmath125 for @xmath104 , are independent if @xmath126 or @xmath127 . expanding the variance on the right - hand side gives @xmath128 , \end{split}\ ] ] where @xmath129 denotes the covariance of random variables @xmath77 and @xmath130 . because @xmath102 over @xmath104 form a multinomial distribution , @xmath131 and @xmath132 substituting these equations into eqn . gives @xmath133 , \end{split}\ ] ] where we have defined @xmath134 substituting eqn . into eqn . gives @xmath135}\ ] ] for species @xmath92 in a binary alloy the above equation simplifies significantly : @xmath136 the quantities @xmath137 determine the coupling between monolayers with regards to @xmath138 and @xmath139 . as was the case for @xmath140 , if @xmath15 for all @xmath14 then the above equations become equivalent to those given earlier for random alloys . this is the case since @xmath141 which can be shown by appealing to eqn . as well as the definition of @xmath87 : @xmath142 from the quantities @xmath140 for all @xmath14 , the mean cls @xmath143 over _ all _ @xmath1 sites in the system under consideration can be determined by using the following expression : @xmath144 where @xmath145 is the _ weight _ to be given to monolayer @xmath14 . for the ` true ' mean , @xmath145 is given by @xmath146 however , one is often interested in the mean of the cls distribution observed experimentally , which may differ from the ` true ' value on account of the fact that the probability @xmath147 of a photoelectron emitted from monolayer @xmath14 escaping the alloy is monolayer - dependent . in this case @xmath145 is given by @xmath148 eqn . follows trivially from the following theorem : if @xmath149 is a set of @xmath150 values whose mean is @xmath151 , then the mean of the superset @xmath152 formed by combining the sets @xmath149 for all @xmath2 is @xmath153 this result can be derived from the definitions of @xmath154 and @xmath151 : @xmath155 from the quantities @xmath143 , and @xmath140 and @xmath139 for all @xmath14 , the fwhm in the cls distribution over all @xmath1 sites @xmath156 can determined using the following expression : @xmath157}.\ ] ] eqn . can be derived by applying eqn . to the analogous equation to eqn . for the variance : @xmath158,\ ] ] where @xmath159 denotes the variance of @xmath149 , and @xmath160 denotes the variance of @xmath152 . the above equation can itself be derived somewhat similarly to eqn . : @xmath161 ^ 2 \\ = & \sum_i\frac{n_i}{n}\biggl[\frac{1}{n_i}\sum_{\varepsilon\in\sigma_i}(\varepsilon-\mu_i)^2+(\mu_i-\mu)^2 \\ & + 2\frac{1}{n_i}(\mu_i-\mu)\sum_{\varepsilon\in\sigma_i}(\varepsilon-\mu_i)\biggr ] , \end{split}\ ] ] which becomes eqn . after noting that the final term in the last equality vanishes . in the next section we apply the expressions of sec . [ sec : mean_fwhm ] to various multilayer systems . all systems we consider have an fcc underlying lattice , and the monolayers are the 001 planes . however , to apply these expressions we first had to determine the monolayer coupling parameters @xmath118 and @xmath137 , which are defined in eqns . and . in ref . , @xmath39 and @xmath40 are tabulated as a function of @xmath41 for the fcc , bcc and sc lattices . using this information , and after determining the quantities @xmath108 - which depend on the underlying geometry - one can tabulate @xmath118 and @xmath137 as a function of @xmath41 for any fcc , bcc or sc system . for other lattice types , @xmath39 and @xmath40 must be determined as a function of @xmath41 beforehand . a procedure to do this is described in ref . . we determined @xmath118 and @xmath137 for the 001 planes of the fcc lattice at selected @xmath41 . the results are shown in fig . [ fig : structural_params ] . we considered @xmath1621.4 , 1.6 , 2.0 , 2.6 and 3.0 , which reflects the range of @xmath41 found in the literature @xcite . note that the coupling between monolayers rapidly tends to zero with @xmath110 . and @xmath137 for the 001 planes in the fcc lattice for various values of @xmath41 . the top panel shows @xmath118 vs. @xmath110 ; while the bottom panel shows @xmath137 vs. @xmath110 . the @xmath41 to which each curve corresponds can be deduced from the key in the top panel.,scaledwidth=50.0% ] in addition to the aforementioned ` analytical ' results , we also performed supercell calculations for each of the systems we considered . in our supercells , each monolayer contained 200 sites , with the species of sites in each monolayer @xmath14 assigned randomly such that the desired species concentrations @xmath13 were obtained as closely as possible . @xmath6 was determined for each site using eqns . and . these @xmath6 where then used to determine @xmath163 , @xmath139 , @xmath143 and @xmath156 in the conventional manner , which allowed us to cross - check our analytical results . for all systems the analytical and supercell results were in excellent agreement , though we choose not to present the supercell means and fwhms for the sake of brevity and clarity of presentation . in addition to cross - checking , the supercell values of @xmath6 were used to simulate core level spectra for some systems . in this regard we used the following equation : @xmath164 where @xmath165 denotes the intensity of the @xmath1 spectrum at cls @xmath166 , @xmath167 is a lorentzian function with fwhm @xmath168 , and @xmath169 is the lifetime broadening of the core levels under consideration . the above equation does not take into account many features which are present in ` real ' spectra such as surface core - level shifts , experimental broadening , doniac - sunjic asymmetry , and inelastic scattering . however , since our simulated spectra are primarily for illustrative purposes , ignoring these complications is justified . our supercell calculations served one further purpose . later we present histograms of , for binary systems , the frequency of @xmath92 sites exhibiting each possible number of @xmath93 nearest neighbors , and for ternary systems , the frequency of @xmath92 sites exhibiting each possible combination of @xmath93 and @xmath170 nearest neighbors . we used our supercells to generate these histograms : for each supercell we counted the number of @xmath92 sites with each possible composition of nearest neighbor shell . we first present results for systems consisting of a thin film of species @xmath92 and thickness @xmath172 monolayers embedded in a @xmath93 substrate , i.e. , @xmath171 . we examined such systems with various @xmath172 and @xmath7 , where recall that @xmath7 denotes the degree of interface roughening . following refs . , interface roughening was modeled by convoluting the ` unroughened ' concentration profile with a discrete gaussian function @xcite , where @xmath7 denotes the standard deviation of the function . to elaborate , the concentration profile corresponding to @xmath7 was calculated using the following equation : @xmath173 for all @xmath1 , where @xmath174}{\displaystyle\sum_{d'=-\infty}^{\infty}\exp\big[-(d'/\sigma)^2/2\bigr]}\ ] ] is the discrete analogue of a gaussian function with standard deviation @xmath7 , and @xmath175 denotes the unroughened concentration profile . the quantities @xmath176 , @xmath177 , @xmath178 and @xmath179 for various @xmath171 systems are presented in figs . [ fig : mean_bab ] and [ fig : fwhm_bab ] . in calculating these values , we used the set of @xmath41 described in the previous section . furthermore , we set @xmath180 , @xmath181 , and @xmath182 . results ( in ev ) obtained using these parameters can be generalized to any choice of @xmath42 ( in bohrs ) , @xmath183 , @xmath184 ( both in units of @xmath185 ) and @xmath186 ( in ev ) at the corresponding value of @xmath41 by multiplying all means and fwhms by @xmath187 , and additionally adding @xmath186 to all mean clss . we will use this fact in a moment . some interesting results are immediately apparent from figs . [ fig : mean_bab ] and [ fig : fwhm_bab ] . firstly , as can be seen from both figures , the model results only depend weakly on @xmath41 . this is convenient because the ` true ' value of @xmath41 for a given system is unclear ; values of @xmath41 are very sensitive to the _ ab initio _ method used to obtain them . the same is true for the quantities @xmath43 , and presumably also for the quantities @xmath188 and @xmath189 . this is illustrated in table [ table : ab_initio_compare ] , where the @xmath41 and @xmath190 for the bcc random alloy cu@xmath191zn@xmath191 obtained using different _ ab initio _ methods are compared . in calculating each @xmath41 in the table we took @xmath33 to be the mean of the _ ab initio _ _ q - v _ relation gradients @xmath192 and @xmath193 for cu and zn . furthermore , we calculated @xmath190 from the _ q - v _ relation intercepts via @xmath194 ( see eqn . ) . .@xmath41 and @xmath190 for cu@xmath191zn@xmath191 obtained using various _ ab initio _ calculations . the abbreviation ` lsms ' refers to the locally self - consistent green s function method ; and ` lapw ' refers to the linearized augmented plane wave method . see refs . and references therein for details regarding these calculations . [ cols="^,^,^",options="header " , ] our second observation relates to fig . [ fig : fwhm_bab ] . one might expect that , at a given composition and parametrization of the model ( i.e. , choice of the quantities @xmath41 , @xmath183 , @xmath184 and @xmath186 ) , the @xmath92 fwhm would be maximized at the random alloy configuration . this configuration corresponds to the largest configurational entropy , and therefore might be expected to exhibit the largest range of environments and hence also the largest fwhm . moreover , since , as can be seen from eqn . , @xmath195 is maximized when @xmath196 , one might therefore also expect that the corresponding fwhm , which is an upper bound for the fwhm of a random alloy , is also an upper bound for _ all _ alloys . however , our results reveal that this is _ not _ the case . in each panel of fig . [ fig : fwhm_bab ] , the random alloy upper bound corresponding to the same model parametrization as we used for our @xmath171 systems is indicated by a dotted line . note that for some systems @xmath179 is significantly larger than the random alloy upper bound . it is even possible for the upper bound to be exceeded in ` unroughened ' ultra - thin films - as is evident from the left - most column in fig . [ fig : fwhm_bab ] . therefore inhomogeneous concentration profiles can yield larger disorder broadenings than is possible in random alloys - a fact which we provide an explanation for later . a similar observation has been made in our earlier study@xcite using the linear charge model@xcite : surface segregation was shown to result in a significantly larger disorder broadening relative to the unsegregated random alloy . however in that study surface effects are implicit in the simulated spectra , and hence can not be discounted as a contributing factor to the very large broadening . by contrast here we have shown that a very large broadening can occur in the absence of surface effects . in the aforementioned study we suggested that segregation could explain the anomalously large disorder broadening observed experimentally by medicherla _ _ et al.__@xcite . our results here provide further evidence for this hypothesis . in contrast to @xmath179 , the monolayer fwhms @xmath197 seem to be constrained to be below the random alloy upper bound . this can be understood by noting that the @xmath92 sites in monolayer @xmath14 experience a local environment which closely resembles a random alloy , and hence @xmath197 ( to a good approximation ) can not exceed the upper bound . considering only nearest neighbors , each @xmath92 site in monolayer @xmath14 has @xmath198 nearest neighbors in monolayer @xmath14 , @xmath199 in monolayer @xmath200 , and @xmath199 in monolayer @xmath201 . since _ within each monolayer _ we have assigned sites species randomly in the required concentrations , it follows that the environment of an @xmath92 site in monolayer @xmath14 is approximately that of a random alloy with concentration @xmath202 systems calculated using the model . each column pertains to a different value of @xmath7 , which is indicated above the column ; each row pertains to a different value of @xmath41 , which is indicated to the right of the row . each circle represents @xmath176 for a particular monolayer @xmath14 ; the size of the symbol is proportional to @xmath203 . only results pertaining to monolayers with @xmath204 are shown . the solid lines connect @xmath177 for each value of @xmath172.,scaledwidth=50.0% ] systems calculated using the model . the details of the figure are the same as in the caption to fig . , except that circles represent @xmath178 , and solid lines connect @xmath179 for each value of @xmath172 . furthermore , the dotted line in each panel corresponds to the upper bound on the fwhm of a random alloy at the corresponding value of @xmath41.,scaledwidth=50.0% ] the @xmath171 systems we considered were deliberately chosen to be identical to the fcc systems considered in ref . , in which olovsson _ et al . _ calculated @xmath205 for ni / cu@xmath206/ni and co / cu@xmath206/co . the @xmath205 and @xmath205 determined by olovsson _ are shown in fig . [ fig : mean_olovsson_bab ] . comparing this figure with fig . [ fig : mean_bab ] we see that the model is in excellent qualitative agreement with those of olovsson _ et al . _ for the ni / cu@xmath206/ni systems . the agreement is especially good at higher values of @xmath41 . for the co / cu@xmath206/co systems , the agreement is reasonable . however , it should be borne in mind that the dispersion of clss in co / cu@xmath206/co is very small , and hence any uncertainties implicit in the _ ab initio _ method used by olovsson _ _ will be larger relative to the size of the dispersion : olovsson _ et al . _ quote clss to a precision of 10mev , which is significant on the scale of the co / cu@xmath206/co dispersion , but not for the ni / cu@xmath206/ni dispersion . the @xmath42 used by olovsson _ _ in their calculations was 2.6 bohrs for both ni / cu@xmath206/ni and co / cu@xmath206/co . with this in mind , and recalling the procedure described above for generalizing our model results to different free parameters , we found that @xmath207 , @xmath208ev and @xmath209 gave excellent agreement with olovsson _ _ for ni / cu@xmath206/ni . for co / cu@xmath206/co we found acceptable agreement when @xmath207 , @xmath210 , and @xmath211ev . these values are similar to analogous quantities obtained from _ ab initio _ calculations for other alloys @xcite . however , as mentioned earlier , the values of these quantities are sensitive to the _ ab initio _ method used to obtain them . hence an interesting prospect is to use the model to determine their values experimentally . olovsson _ et al . _ used the @xmath205 from their calculations to simulate core level xps spectra . however , since their calculations did not provide values for @xmath212 , it was necessary for olovsson _ to make some assumptions regarding these quantities . they assumed that @xmath212 was the same for all @xmath14 . [ fig : fwhm_bab ] reveals that , in fact , the quantities @xmath212 vary widely within any one system . hence the assumption made by olovsson _ is incorrect . however , it is unclear whether the breakdown of this assumption is important from a practical point of view . while accurate knowledge of the quantities @xmath98 - as well as perhaps higher moments of the @xmath1 cls distribution for each monolayer - is necessary to reproduce the fine details of the total @xmath1 cls distribution for the system under consideration , in practice these fine details are ` smeared out ' in the experimental spectrum due to complications such as lifetime and experimental broadening . therefore getting some of the fine details wrong in the spectrum before accounting for the aforementioned complications may be inconsequential with regards to accurately reproducing experimental spectra . /ni and co / cu@xmath206/co systems calculated by olovsson _ et al . _ ( ref . ) . the significance of each column , and the symbols and lines , is the same as in fig . [ fig : mean_bab ] . the top row of graphs corresponds to ni / cu@xmath206/ni , while the bottom row corresponds to co / cu@xmath206/co.,scaledwidth=50.0% ] it should be noted that olovsson _ considered a further system in their study in addition to ni / cu@xmath206/ni and co / cu@xmath206/co : fe / cu@xmath206/fe . we do not perform a thorough comparison between our model calculations and those of olovsson _ et al . _ for fe / cu@xmath206/fe because the underlying lattice for fe / cu@xmath206/fe is bcc , while our model calculations are for an fcc underlying lattice . at @xmath213 and @xmath214 , the qualitative nature of the _ ab initio fe / cu@xmath206/fe results is similar to that for ni / cu@xmath206/ni and co / cu@xmath206/co . in these cases we therefore expect that the model will perform well for fe / cu@xmath206/fe . however , as pointed out by olovsson _ , a well - known interface state exists in ordered , but not in disordered fe / cu@xmath206/fe systems , which causes ` anomalous ' results for @xmath10 . it would be interesting to see whether model calculations utilizing a bcc lattice can reproduce this . in the previous subsection we considered an isolated thin film embedded in an infinite substrate . this was done primarily to allow comparison with analogous _ ab initio _ results . however , experimental studies have focused on _ periodic _ multilayer systems . we now consider such systems ; specifically , those comprised of ni and cu in which the repeating unit consists of @xmath172 monolayers of ni adjacent to @xmath216 monolayers of cu , i.e. , [ ni@xmath206/cu@xmath215 . we considered various @xmath172 , @xmath216 and @xmath7 . as above , we used eqn . to obtain the concentration profile for a given @xmath7 . with regards to the parametrization of the model , we used the ` best - fit parametrization ' described above for the ni / cu@xmath206/ni systems , i.e. , @xmath217 , @xmath218ev , and @xmath219 . the @xmath220 and @xmath221 for various [ ni@xmath206/cu@xmath215 systems as a function of @xmath7 are presented in fig . [ fig : trajectories_cuni ] ; we show the ` trajectory ' of the pair @xmath222 as @xmath7 is varied from 0 to @xmath223 . note that at @xmath224 intermixing between the ni and cu regions is absolute , and [ ni@xmath206/cu@xmath215 becomes a random alloy with @xmath225 . as is evident from fig . [ fig : trajectories_cuni ] , as @xmath7 increases , @xmath220 becomes more negative , and @xmath221 increases initially , before decreasing and finally settling on the random alloy value . the exception is [ ni@xmath226/cu@xmath8 ] , for which @xmath221 monotonically decreases with @xmath7 , i.e. , the system exhibits a ` disorder narrowing ' . we will now explain these trends . , determined using the model . each color pertains to a different system ; each system is labeled . symbols indicate the ( mean , fwhm ) evaluated at @xmath7 from 0 to @xmath223 at intervals of @xmath7 of 0.2 . some of the symbols are annotated with their value of @xmath7 . curves connect the symbols , and trace the ( mean , fwhm ) from @xmath10 to @xmath224 for each system.,scaledwidth=50.0% ] a pleasing feature of the model is that it provides a simple means of rationalizing changes in cls spectra due to configurational changes . consider eqn . . an insightful approximation is to ignore the dependence of @xmath20 on the environment of site @xmath2 beyond its nearest neighbor shell , in which case for cu in nicu systems the number of ni nearest neighbors @xmath11 of site @xmath2 wholly determines @xmath20 . for @xmath227 , where recall that @xmath228 denotes the number of sites in any site s @xmath37th nearest neighbor shell , and @xmath229 denotes the global concentration of ni this would lead to better agreement between the histograms and model spectra in fig . [ fig : figure_cu5ni5])(b ) . however , we do not do this here for the sake of simplicity . ] specifically , @xmath20 is linear in @xmath11 : @xmath230 the same then applies to @xmath6 ( from eqn . ): @xmath231 using the above expression the cu spectrum can be decomposed into , or constructed from , components associated with cu atoms with each value of @xmath11 , which allows us to equate changes in the cu spectrum to changes in the system s ` @xmath11 histogram ' for cu . this is done in fig . [ fig : figure_cu5ni5](b ) for [ ni@xmath8/cu@xmath8 ] , where we have used @xmath232ev in the simulated spectra . at @xmath10 cu sites exhibit only two possible environments , @xmath12 or 4 , with the former corresponding to the 3 ` central ' monolayers of the 5 monolayer cu stack and the latter corresponding to the 2 ` edge ' monolayers . on the whole cu sites exhibit higher values of @xmath11 as cu diffuses into the ni region , leading to a shift in the spectrum to low binding energies as @xmath7 increases . the width of the spectrum increases as the @xmath11 histogram becomes ` flat ' near @xmath214 , and then narrows again as @xmath233 and the histogram tends to that corresponding to the random alloy ni@xmath191cu@xmath191 . note that , as we found earlier to be the case for the @xmath171 systems , the disorder broadening is not maximized at the random alloy configuration . an explanation for this is as follows . the @xmath11 histograms for random alloys - substitutionally disordered systems with homogeneous concentration profiles - are binomial distributions . for inhomogeneous concentration profiles , such as those for [ ni@xmath8/cu@xmath8 ] at @xmath234 , the histograms are not constrained to be binomial distributions - they are free to be ` flatter ' . therefore systems with inhomogeneous concentration profiles can exhibit significantly larger disorder broadenings than random alloys - as is borne out in figs . [ fig : figure_cu5ni5](b ) , [ fig : fwhm_bab ] and [ fig : trajectories_cuni ] . as another example , consider [ ni@xmath226/cu@xmath8 ] , which we earlier pointed out exhibits a disorder narrowing . [ fig : figure_cu5ni1 ] is analogous to fig . [ fig : figure_cu5ni5 ] , but for [ ni@xmath226/cu@xmath8 ] . again , we used @xmath232ev in the simulated spectrum . at @xmath10 the cu sites in the 2 edge monolayers of a 5 monolayer cu stack have @xmath235 , while the cu sites in the 3 central monolayers have @xmath12 . this is in fact exactly the same situation as for [ ni@xmath8/cu@xmath8 ] . as @xmath7 is increased the cu sites begin to exhibit other values of @xmath11 ; some cu sites begin to ` occupy ' the @xmath236 ` states ' which were unoccupied at @xmath10 . the same also occurs in [ ni@xmath8/cu@xmath8 ] . however , comparing figs . [ fig : figure_cu5ni5 ] and [ fig : figure_cu5ni1 ] it can be seen that there is a significant occupation of the @xmath237 states in [ ni@xmath8/cu@xmath8 ] which does not occur in [ ni@xmath226/cu@xmath8 ] on account of the lack of ni . hence in [ ni@xmath226/cu@xmath8 ] , loosely speaking , the cu sites spill from their @xmath12 and @xmath235 states into only the @xmath236 states as @xmath7 is increased , which results in a narrowing of the spectrum in this system . , except that this figure pertains to the alloy system [ ni@xmath226/cu@xmath8].,scaledwidth=50.0% ] experimental cu spectra for [ ni@xmath8/cu@xmath8 ] and [ ni@xmath8/cu@xmath238 at various temperatures were obtained in refs . . however , the spectra were found to depend strongly on the material used to cap the sample , and also on the photon energy used @xcite . this is due to the finite escape depth of the photoelectrons , which leads to cu sites near the surface being ` over - represented ' in the cu spectra . our model results do not account for such surface effects ; they pertain to the deep bulk . hence a quantitative comparison between our model results and most of the experimental spectra is meaningless . the possible exception is the spectra for [ ni@xmath8/cu@xmath8 ] in ref . taken using a photon energy of 6030ev - which corresponds to a relatively low surface sensitivity . these spectra reveal a shift of @xmath239ev in [ ni@xmath8/cu@xmath8 ] upon heating over the temperature range corresponding to the transition from @xmath10 to @xmath224 . this is in excellent agreement with our results : as can be seen from fig . [ fig : trajectories_cuni ] , from @xmath10 to @xmath224 the model predicts a shift in the cu cls of @xmath239ev for [ ni@xmath8/cu@xmath8 ] . in the future it would be interesting to perform model calculations which take into account surface effects , and which therefore can be compared directly with the results of refs . . finally , we consider ternary multilayer systems - consisting of three species @xmath92 , @xmath93 and @xmath170 . recall that the @xmath1 cls distribution reflects the @xmath1 charge distribution ( eqn . ) , which itself is determined by charge transfer between @xmath1 and non-@xmath1 sites ( eqn . ) . in binary systems eqn . gives @xmath240 where we have used the fact that @xmath241 . without loss of generality let @xmath242 , which corresponds to choosing species @xmath92 to be the most electronegative of @xmath92 and @xmath93 . from the above equations it can be seen that the electropositivity difference @xmath184 between species @xmath93 and @xmath92 acts only as a scale factor for the @xmath92 and @xmath93 charge distributions ; altering @xmath184 has no effect on the _ qualitative _ nature of the @xmath92 and @xmath93 charge distributions . ternary systems are more complicated . the analogous equation to the above for species @xmath92 in a ternary system is @xmath243.\ ] ] note that altering @xmath184 affects the charge transfer between site @xmath2 and local @xmath93 sites , while leaving the transfer with local @xmath170 sites unaffected ; and conversely if @xmath244 is altered . hence altering @xmath244 or @xmath184 non - trivially alters the @xmath92 charge distribution . in this sense ternary systems are extremely rich . an exhaustive survey of what can be expected from such systems is beyond the scope of this work . we instead limit ourselves to two ternary systems which we have found to exhibit somewhat counterintuitive behaviors . furthermore , we limit our discussion to the qualitative aspects of these systems , with a focus on understanding these behaviors . for both systems species @xmath92 is the focus of our attention . the first ternary system we consider is @xmath245 $ ] . again , we use eqn . to model the interface roughening ; fig . [ fig : figure_b5a5c5](a ) gives a schematic illustration of the system at the @xmath7 we considered . for our calculations we set the electropositivities of each species to be @xmath246 , @xmath247 and @xmath248 : the electropositivity of species @xmath92 is exactly halfway between those of species @xmath93 and @xmath170 , with species @xmath93 being the most electronegative and species @xmath170 being the most electropositive . with regards to the other model parameters we set @xmath182 , @xmath249 , @xmath180 , @xmath250 , and @xmath251 . fig . [ fig : figure_b5a5c5](c ) shows the simulated @xmath92 spectra at each @xmath7 . as is evident from the figure , the mean of the @xmath92 cls distribution is independent of @xmath7 . hence here one can not use the mean to characterize @xmath7 in this system - one must use the fwhm . this stems from the choice of species electropositivities . charge transfer from @xmath93 to @xmath92 and @xmath170 to @xmath92 is always equal and opposite on account of the electropositivity of @xmath92 being exactly between that of @xmath93 and @xmath170 . this , in conjunction with the symmetry of the system , always yields a global mean @xmath92 charge of 0 regardless of @xmath7 , and hence , from eqn . , a @xmath7-independent mean cls . another interesting feature of this system is that the @xmath10 and @xmath224 spectra are ostensibly indistinguishable . hence unambiguously determining @xmath7 in this system from the @xmath92 spectrum alone is difficult . the method described earlier for rationalizing spectral changes in binary systems can be extended to ternary systems . the analogous equation to eqn . for species @xmath92 in a ternary system is @xmath252 + \phi_a.\ ] ] note that here @xmath6 depends on the environment of @xmath2 through the pair @xmath253 ; there is a mapping from @xmath253 to @xmath6 . hence the @xmath92 spectrum reflects the ` @xmath253 histogram ' of species @xmath92 . this is illustrated in fig . [ fig : figure_b5a5c5](b - c ) . fig . [ fig : figure_b5a5c5](b ) illustrates the frequency of @xmath92 sites with each @xmath253 ( i.e. , the @xmath253 histogram for species @xmath92 ) , as well as the cls for each @xmath253 - contours of constant cls in ` @xmath253-space ' are drawn . if an environment @xmath253 is exhibited by a high frequency of sites , then there is a spike in the cls spectrum at the corresponding cls . this allows rationalization of the evolution of the @xmath92 spectrum with @xmath7 . at @xmath10 there are only 3 possible environments for @xmath92 sites : @xmath254 , which corresponds to a site adjacent to the @xmath93 region ; @xmath255 , which corresponds to a site adjacent to the @xmath170 region ; and @xmath256 , which corresponds to a site in the center of the @xmath92 region - surrounded by @xmath92 sites . at @xmath213 there is a small amount of intermixing at the interfaces . this leads to @xmath92 sites near the @xmath93 region exhibiting environments @xmath257 , @xmath258 , @xmath259 , etc . ; and @xmath92 sites near the @xmath170 region exhibiting environments @xmath260 , @xmath261 , @xmath262 , etc . note that at this point no @xmath92 sites have both @xmath93 _ and _ @xmath170 nearest neighbors , and hence the frequency of any environment @xmath253 is only non - zero if @xmath263 or @xmath264 for any @xmath77 . with regards to the spectrum , the ` spreading out ' of the histogram along the left and bottom edges means that environments are exhibited which correspond to more extreme clss , i.e. , @xmath265 and @xmath266 . hence the spectrum broadens from @xmath10 to 0.75 . at @xmath267 the interface roughening is large enough that there are @xmath92 sites with both @xmath93 and @xmath170 neighbors , and hence there is a non - zero frequency for environments away from the left and lower edges of the @xmath253 histogram . at @xmath224 we have a ternary random alloy , and the @xmath253 histogram corresponds to that of a trinomial distribution . /@xmath268/@xmath269 at various @xmath7 ( a ) ; and the corresponding @xmath270 histograms and environment vs. cls maps for species @xmath92 ( b ) , and model spectra ( c ) . in ( b ) , the green - blue squares represents the frequency of @xmath92 sites with the environment @xmath270 , and yellow - red curves are contours of constant cls determined according to eqn . . in ( c ) the dotted curves are spectra for @xmath10 , and the location of some of the cls contours in ( b ) are also shown.,scaledwidth=50.0% ] the final system we consider is @xmath271 , with @xmath92 constrained to be immiscible in @xmath93 and @xmath170 , and the electropositivities set to @xmath247 , @xmath272 , and @xmath273 . the electropositivities correspond to the following situation : species @xmath92 is the most electronegative ; species @xmath93 is the most electropositive ; and species @xmath170 has an intermediate electropositivity very close to that of species @xmath92 . this system is interesting because , like [ ni@xmath226/cu@xmath8 ] discussed earlier , it exhibits a disorder narrowing . constraining @xmath92 to be immiscible in @xmath93 and @xmath170 renders eqn . we therefore instead modeled the interface roughening for this system using the equation @xmath274c^l_x(\sigma=0)\ ] ] for @xmath275 , where we have used the convention that the interface between species @xmath92 and @xmath93 at @xmath10 is located between monolayers 0 and 1 . the resulting concentration profiles are illustrated in fig . [ fig : figure_a5b2c](a ) . similarly to fig . [ fig : figure_b5a5c5 ] , figs . [ fig : figure_a5b2c](b ) and [ fig : figure_a5b2c](c ) show the @xmath253 histogram and simulated spectra for species @xmath92 at each of the considered @xmath7 . for this system we used the same @xmath186 , @xmath41 , @xmath42 , @xmath183 and @xmath169 as for @xmath245 $ ] . earlier we saw that [ ni@xmath226/cu@xmath8 ] exhibits a disorder narrowing . in general , given that the @xmath92 spectrum of a system reflects the @xmath92 charge distribution - as follows from eqn . - a disorder narrowing for @xmath92 occurs when the introduction of substitutional disorder ` quenches ' the width of the @xmath92 charge distribution . this occurs here . species @xmath93 and @xmath92 transfer a certain amount of charge , which results in an @xmath92 charge distribution with a certain width at @xmath10 . as @xmath7 is increased , more @xmath170 sites come within the charge - transfer range of the @xmath92 region . given the tiny electropositivity difference between species @xmath92 and @xmath170 , there is almost no charge transfer between @xmath92 and @xmath170 sites . hence the influx of @xmath170 sites to the @xmath92 interface acts to reduce the charges of the edge @xmath92 sites , bringing them closer to that of the ` non - edge ' @xmath92 sites . this corresponds to a reduction in the width of the @xmath92 charge distribution , and hence also the core level spectrum . alternatively , one can explain the disorder narrowing in terms of the evolution of the @xmath253 histogram for species @xmath92 ( fig . [ fig : figure_a5b2c](b ) ) . at @xmath10 the environment of the @xmath92 sites on the edge monolayer is @xmath254 , which steadily transitions to @xmath276 as @xmath233 . the latter environment has a cls closer to that of the non - edge @xmath92 sites , i.e. , @xmath277 , and hence the spectrum narrows as @xmath7 is increased . , except that this figure pertains to the alloy system @xmath268/@xmath278/@xmath170,scaledwidth=50.0% ] above we have presented a model for core level shifts in alloys , and have used it to add insight into the relationship between atomic environment , charge transfer and disorder broadening in complex systems . our key result is that the mapping between the distribution of atomic environments and core level spectra is often counterintuitive . for instance , systems with inhomogeneous concentration profiles can exhibit significantly larger disorder broadenings than is possible in random alloys , and even a ` disorder narrowing ' in some cases . for the correct interpretation of experimental spectra for complex systems , it is crucially important to understand such phenomena . the model can be easily adapted to treat other core level spectroscopies , the most prominent of which is auger electron spectroscopy . hence we expect that it should find widespread use as a framework in which to interpret experimental results . however , it is by no means a panacea . recall that the model is underpinned by the nra - cefm - which itself is a particular case of the cefm . implicit in the nra - cefm are a number of approximations which may be problematic . one is the _ spherical approximation _ - that only the monopole moments of the charge distribution in each site are considered for the purposes of evaluating the madelung energy . one must go beyond this approximation in order to obtain a quantitatively accurate description of the electron density within disordered alloys @xcite . this is especially true for regions near surfaces . a generalization of the cefm has been described in ref . which does not rely upon the spherical approximation . while the nra - cefm could be generalized in an analogous manner , it is not clear whether this would be fruitful . the strength of the nra - cefm over the ` general ' cefm is its simplicity , with which comes a small loss in accuracy relative to the cefm . it is not clear whether the gain in accuracy achieved by generalizing the nra - cefm to go beyond the spherical approximation is worth the resulting loss in simplicity . another potentially problematic approximation implicit in the nra - cefm is that the nuclei of the system under consideration form an undistorted crystal lattice . the breakdown of this approximation can have far - reaching consequences . the addition of distortions to the crystal lattice of cuau results in a reversal of the average relationship between a site s cls and its number of unlike nearest neighbors @xcite . the reasons for this are not known , and warrant further investigation . the nra - cefm , suitably modified to treat lattice distortions , may add insight into this phenomenon , though it would be optimistic to expect that anything more than a qualitative understanding could be achieved . it should be borne in mind that the aforementioned approximations , while implicit in the nra - cefm , are also utilized in many _ ab initio _ calculations , and are not expected to preclude the model from making _ at least _ qualitatively accurate predictions . a more problematic limitation of the model is that its free parameters are not known _ a priori _ : they must be obtained from _ ab initio _ calculations or by other means . fortunately these parameters are highly transferable between systems ; for details see refs . . in the future we intend to calculate these parameters for a wide range of alloys . this would enable the model to be readily applied to many systems .

we report extensive radio and x - ray observations of sn2003bg whose spectroscopic evolution shows a transition from a broad - lined type ic to a hydrogen - rich type ii and later to a typical hydrogen - poor type ibc . we show that the extraordinarily luminous radio emission is well described by a self - absorption dominated synchrotron spectrum while the observed x - ray emission at days is adequately fit by inverse compton scattering of the optical photons off of the synchrotron emitting electrons . our radio model implies a sub - relativistic ejecta velocity , , at days after the explosion which emphasizes that broad optical absorption lines do not imply relativistic ejecta . we find that the total energy of the radio emitting region evolves as erg assuming equipartition of energy between relativistic electrons and magnetic fields ( ) . the circumstellar density is well described by a stellar wind profile with modest ( factor of ) episodic density enhancements which produce abrupt achromatic flux variations . we estimate an average mass loss rate of ( assuming a wind velocity of ) for the progenitor , consistent with the observed values for galactic wolf - rayet stars . comparison with other events reveals that of radio supernovae show similar short timescale flux variations attributable to circumstellar density irregularities . specifically , the radio light - curves of sn2003bg are strikingly similar to those of the type iib sn2001ig , suggestive of a common progenitor evolution for these two events . based on the relative intensity of the inferred density enhancements , we conclude that the progenitors of sne 2003bg and 2001ig experienced quasi - periodic mass loss episodes just prior to the sn explosion . finally , this study emphasizes that abrupt radio light - curve variations can not be used as a reliable proxy for an engine - driven explosion , including off - axis gamma - ray bursts .

astro-ph0512413

accounting for @xmath9 of the nearby supernova population , type ibc supernovae ( hereafter sne ibc ) are identified by their lack of spectroscopic hydrogen and silicon features ( see @xcite for a review ) . recognized as a rare subclass of core - collapse supernova @xmath10 years ago @xcite , sne ibc have recently enjoyed a revitalized interest . beginning with the discovery of type ic sn1998bw in temporal and spatial coincidence with gamma - ray burst ( grb)980425 @xcite , we now know that most long - duration grbs ( e.g. @xcite ) and x - ray flashes ( xrfs ; @xcite ) are associated with sne ibc . however , radio observations of local sne ibc indicate that the inverse is _ not _ true ; there is a strict limit of @xmath11 on the fraction of sne ibc that could be accompanied by a grb or xrf @xcite . the lack of hydrogen features in both sne ibc and grb / xrf - associated sne imply that they represent the explosion of massive stripped - core progenitors @xcite . the popular models for sne ibc progenitors include massive wolf - rayet ( wr ) stars that eject their envelopes through strong dense winds , and close binary systems where the progenitor star is stripped of its hydrogen - rich layer by the companion @xcite . despite dedicated archival searches , the progenitors of sne ibc are still poorly constrained by pre - explosion images . the best photometric constraints are currently associated with sn2002ap and sn2004gt which exclude wr progenitors in the top @xmath12 and @xmath8 of the population , respectively @xcite . while intriguing , these archival observations lack the sensitivity to clearly discriminate between the wr and binary progenitor models . by studying the circumstellar medium around the supernova , however , it is possible to place independent constraints on the progenitor . observations show that galactic wolf - rayet stars are embedded in wind stratified media @xcite while binary systems are associated with disrupted circumstellar media , possibly including an outflow during a common envelope phase @xcite . radio observations provide the most direct probe of the circumstellar density structure around supernovae @xcite . by observing the dynamical interaction of the ejecta with the surrounding medium we are able to map out the mass loss history of the progenitor star . the identification of irregular density profiles may therefore distinguish between wolf - rayet and binary progenitor systems . here we present extensive radio observations of sn2003bg , discovered as part of our ongoing radio survey of local type ibc supernovae . the peculiar sn2003bg spectroscopically evolved from a broad - lined type ic to a hydrogen - rich type ii and later to a typical type ibc ( hamuy _ et al . _ , in prep ) , thereby bridging the hydrogen - rich and poor divisions of the core - collapse classification system . our densely - sampled radio light - curves show that the extraordinarily luminous radio emission for a sn ibc is characterized by episodic short - timescale variations . we show that these variations are well described by abrupt density enhancements in the circumstellar medium . comparison with other radio supernovae shows that @xmath8 of all well - studied events similarly show evidence for abrupt light - curve variations . we review the mass loss evolution observed ( and inferred ) for the progenitors of these core - collapse sne . using the observed radio properties for the peculiar sn2003bg , we place constraints on the nature of its progenitor system . the organization of this paper is as follows : observations from the very large array ( vla ) and the _ chandra _ x - ray observatory ( cxo ) are described in [ sec : obs ] . preliminary estimates of the energy , velocity and density of the radio emitting region are presented in [ sec : ep ] . modeling of the radio light - curves are presented in [ sec : model ] and [ sec : ssa ] while a discussion of the radio polarization mechanism follows as [ sec : pol_disc ] . our modeling of the x - ray emission is discussed in [ sec : xray_model ] . in [ sec : bumps ] we present a compilation of radio supernovae with abrupt light - curve variations and review their circumstellar irregularities . finally , in [ sec : prog ] we discuss the possible causes of the density enhancements surrounding sn2003bg and the implications for the progenitor system . sn2003bg was optically discovered on 2003 february 25.7 ut , offset 16.3 w and 24".6 s from the center of host galaxy , mcg -05 - 10 - 15 , at @xmath13 mpc @xcite . early spectroscopy on 2003 february 28 ut indicated that sn2003bg was a peculiar type ic supernova with broad optical absorption lines , indicative of fast photospheric velocities and comparable to those seen in sn1998bw @xcite . soon thereafter , sn2003bg developed strong , broad h@xmath14 emission with a p - cygni absorption component indicating an expansion velocity of @xmath15 @xcite . p - cygni components were additionally observed for h@xmath16 and h@xmath17 , prompting the reclassification of sn2003bg as an unusual type ii supernova . nebular spectra taken several months later revealed that all evidence of hydrogen had disappeared , suggestive that the sn had transitioned back to a type ibc event ( hamuy et al . , in prep ) . the peak time of the optical light - curves suggests that sn2003bg was discovered within just a few days of explosion ( hamuy , private communication ) . throughout this paper we therefore assume an approximate explosion date of 2003 february 22 ut . motivated by the spectroscopic similarity of sn2003bg to the radio luminous sn1998bw , we initiated radio observations of sn2003bg on 2003 march 4.07 ut with the very large array ( vla ) . at 8.46 ghz we detected a radio source coincident with the optical position at @xmath18 ( @xmath190.1 arcsec in each coordinate ) with flux density of @xmath20 mjy ( figure [ fig : rgb ] ) . we subsequently began an intense follow - up campaign to study the temporal and spectral evolution of the radio emission . radio data were collected at 1.43 , 4.86 , 8.46 , 15.0 , 22.5 and 43.3 ghz spanning 2003 march through 2005 october and are summarized in table [ tab : vla ] . all vla observations were taken in standard continuum observing mode with a bandwidth of @xmath21 mhz . at 22.5 and 43.3 ghz we included referenced pointing scans to correct for the systematic 10 - 20 arcsec pointing errors of the vla antennas . we used 3c48 , 3c147 and 3c286 ( j0137 + 331 , j0542 + 498 and j1331 + 305 ) for flux calibration , while phase referencing was performed against calibrators j0407 - 330 and j0453 - 281 . data were reduced using standard packages within the astronomical image processing system ( aips ) . no diffuse radio emission from the host galaxy was detected in any of our observations . flux density measurements were obtained by fitting a gaussian model to the sn . in addition to the _ rms _ noise in each measurement , we include a systematic uncertainty of @xmath22 due to the uncertainty in the absolute flux calibration . the sn2003bg radio light - curves are compiled in figures [ fig : oneplot ] and [ fig : lt_curves ] , spanning @xmath23 to @xmath24 days after the explosion . the radio evolution of sn2003bg is characterized by several achromatic short timescale ( @xmath25 ) variations . as will be discussed in [ sec : bumps ] , light - curve variations have also been observed for @xmath8 of well - sampled radio supernovae including type iil sn1979c @xcite , type ic sn1998bw @xcite , and more recently for type iib sn2001ig @xcite . shown in figure [ fig : spectral_indices ] are the spectral indices between each of the adjacent radio frequencies . a spectral turnover is clearly observed for the lower frequencies at early time ; at later time the spectral slope converges to @xmath26 . the observed spectral shape is overall consistent with a non - thermal synchrotron spectrum with a significant absorption component that suppresses the low frequency emission . in the case of radio supernovae , a low frequency turn - over may be attributed to internal synchrotron self - absorption ( ssa ; @xcite ) or free - free absorption ( ffa ; @xcite ) from the external medium . the observed spectral index of the optically - thick spectrum enables us to distinguish between these two processes . for radio supernovae dominated by ssa , the radio spectrum is approximated by @xmath27 \right),\ ] ] where @xmath28 is the peak spectral frequency , @xmath29 is the flux density at @xmath28 , and @xmath30 is the optically - thin spectral index @xcite . on the other hand , for ffa dominated emission , the optically - thick radio spectrum steepens to @xmath31~~\rm mjy\ ] ] where @xmath32 and @xmath33 are normalization constants @xcite . we fit these functions to two well - sampled radio spectra collected on 2003 march 17 and 2004 february 8 ( @xmath34 and 351 days ) . as shown in figure [ fig : spectrum ] , the optically - thick spectral component in each epoch is best described with a spectral index of @xmath35 , and thus entirely consistent with the index predicted by the ssa dominated model . specifically , we find the following best - fit ssa parameters : @xmath36 mjy ( 50 mjy ) and @xmath37 ghz ( 2.2 ghz ) with @xmath38 ( 4.5 ) for the early ( late ) spectrum . the ffa dominated models produce steeper optically - thick spectra ( @xmath39 ) and thus provide a significantly worse fit to the data . we find the following best - fit ffa parameters : @xmath40 ( 35 ) and @xmath41 ( 0.12 ) with @xmath42 ( 8.2 ) for the early ( late ) spectra . here we have adopted @xmath43 for both models , consistent with the observed optically - thin spectral indices from @xmath44 to 1000 days ( figure [ fig : spectral_indices ] ) . clearly the ssa model fit is preferred over the ffa fit in both epochs . we conclude that internal ssa dominates the absorption on the timescale probed by the radio data . since synchrotron emission is inherently polarized , the magnetic field geometry of radio sne can be directly probed through polarization measurements . in order to measure the polarization for sn2003bg , we carried out a full polarization calibration of our 8.46 ghz vla observations taken on 2003 october 7.4 ut ( @xmath45 days ) . this run was chosen for polarization calibration based on the high dynamic range of the radio image . to correct for the instrumental polarization , we observed calibrator j0403 + 260 over a wide range in parallactic angle and computed the leakage terms with aips task pcal . we used 3c147 to calibrate the absolute polarization angle . we do not find any significant polarization for sn2003bg . in particular , we place a limit on the linear polarization intensity ( @xmath46 , where @xmath47 and @xmath48 are the stokes parameters ) of @xmath49 mjy , corresponding to a fractional polarization limit of @xmath50 ( @xmath51 ) . in comparison with other radio sn linear polarization measurements this is one of the deepest limits obtained to date : a factor of @xmath52 deeper than the limit for sn1993j ( @xmath53 , @xcite ) and comparable to that for sn1979c ( @xmath54 , @xcite ) . sn2003bg was observed with the _ chandra _ acis - s detector on 2003 march 24.5 ut ( @xmath55 days after the explosion ) for 50.5 ksec . @xcite reported the detection of an x - ray source at @xmath56 ( @xmath190.5 arcsec in each coordinate ) , coincident with the optical sn position . the source was re - observed with acis - s on 2003 june 22.3 ut ( @xmath57 days ) for 40.5 ksec to search for variability between the two epochs . we retrieved the sn2003bg data from the _ chandra _ data archive and reduced them following the ciao science threads . the source is clearly detected in both epochs and we note that no diffuse emission is detected from the host galaxy . for spectral extraction of the sn emission , we adopt a source aperture of 4.92 arcsec and a large background region located @xmath10 arcsec from the source . after subtracting off the background emission , we measure count rates of @xmath58 and @xmath59 cps ( 0.3 - 10 kev ) in the first and second epochs , respectively . @xcite report that the first epoch data can be reasonably fit with several different spectral models and propose a mekal hot plasma model for the best fit . given that the small number of counts prevents the model fits to be distinguished , we fit the extracted sn spectra with only two basic models : absorbed power - law and thermal bremsstrahlung . table [ tab : cxo ] lists the parameter values for our resulting spectral fits , where we have adopted both fixed and variable @xmath60 values for comparison . @xcite dust maps give @xmath61 mag for the position of sn2003bg . using the standard conversion of @xcite , we find @xmath62 for the galactic column density along our the line - of - sight . the relatively few x - ray counts enable satisfactory fits ( @xmath63 ) for each of the models . given the significantly fewer counts in the second epoch , we fit the spectrum with a fixed ( galactic ) absorption . moreover , for the power - law fit we adopt the best - fit spectral index from our first epoch . the resulting flux values appear in table [ tab : cxo ] for both power - law and thermal bremsstrahlung fits . as in the case of the first epoch , both models provide an equally good representation of the data . in figure [ fig : chandra ] we show the absorbed power - law model fits for epochs 1 and 2 , assuming a fixed galactic @xmath64 . using the associated spectral parameters in table [ tab : cxo ] , we find unabsorbed flux values of @xmath65 and @xmath66 ( 0.310 kev ) for these fits , respectively . the observed temporal evolution is @xmath67 with @xmath68 and the implied spectral indices between the optically thin radio and x - ray band are nearly constant at @xmath69 and @xmath70 , for the first and second _ chandra _ epochs , respectively . at a distance of 19.6 mpc , the observed x - ray luminosity values are thus @xmath71 and @xmath72 , placing sn2003bg among the most x - ray luminous sne ever detected and a factor of @xmath73 fainter than sn1998bw on a comparable timescale . @xcite showed that there is an upper limit to the brightness temperature , @xmath74 , for ssa dominated radio sources . this limit corresponds to the scenario where the fractions of post - shock energy density in relativistic electrons ( @xmath75 ) and magnetic fields ( @xmath76 ) are in equipartition , and the shocked electrons are accelerated into a power - law distribution , @xmath77 , above a minimum lorentz factor , @xmath78 . under these assumptions , the equipartition brightness temperature , @xmath79 k , defines the upper bound on @xmath74 @xcite . here , @xmath80 is the equipartition size ( radius ) of the radio source . it is noted that @xmath81 depends only weakly on the observed peak frequency and flux density . under the assumption that @xmath82 , we can estimate the radius and energy of the radio emitting material for sn2003bg . our first vla epoch in which the spectral turnover is observed is 2003 march 29 ( @xmath83 days ) . in this epoch we observed a peak flux density of @xmath84 mjy at @xmath85 ghz and an optically thin spectral index of @xmath86 ( figure [ fig : spectral_indices ] ) . as shown in figure [ fig : spectrum ] , on the timescale probed by our observations the sn2003bg radio spectrum is well fit with a ssa model and we therefore identify the spectral peak as the synchrotron self - absorption frequency , @xmath87 ( defined as the frequency where the optical depth due to ssa is unity ; @xmath88 ) . using equations 2 and 3 of @xcite ( hereafter s05 ) the observed source properties ( @xmath89 , @xmath28 , @xmath16 , and @xmath90 ) imply an equipartition radius of @xmath91as ( @xmath92 cm ) and a total energy for the radio emitting material of @xmath93 erg , where we have assumed @xmath4 . the average velocity of the radio shell is therefore roughly @xmath94 and the magnetic field is @xmath95 g. a comparison with the ejecta velocities compiled for core - collapse sne @xcite shows that sn2003bg resembles more closely a type ibc ( typical ejecta velocities , @xmath96 ) rather than a type ii event ( @xmath97 ) . moreover , the inferred ejecta velocity for sn2003bg is a factor of @xmath23 slower than the mildly relativistic speed inferred for sn1998bw . this analysis emphasizes that broad optical absorption lines can not be used as a proxy for relativistic ejecta , consistent with the radio analysis for broad - lined sne 2002ap @xcite and sn2003jd @xcite . our preliminary constraint on the minimum energy for sn2003bg places it among the most energetic radio sne ibc ever observed , second only to sn1998bw and comparable to sn2003l ( s05 ) . we note that additional absorption processes ( e.g. ffa ) and departures from equipartition ( e.g. sn1993j ; @xcite ) increase the energy budget further . equipartition analysis may also be used to roughly constrain the characteristic synchrotron frequency , @xmath98 . equating particle kinetic energy across the shock discontinuity , s05 show that @xmath99 ghz which gives @xmath100 ghz for @xmath101 with the values of @xmath102 and @xmath103 given above . at @xmath83 days , @xmath98 is therefore already below our radio observing band and the synchrotron break frequencies are ordered such that @xmath104 . finally , we use these equipartition results at @xmath105 days to estimate the mass loss rate of the progenitor star . from @xcite , the mass loss rate of the star , @xmath106 , can be derived from the post - shock energy density in magnetic fields : @xmath107 where we have assumed a stellar wind - blown medium ( @xmath108 ) . we find @xmath109 ( assuming a wolf - rayet wind velocity of @xmath7 ) . for comparison , we estimate the mass loss rate assuming ffa dominates the absorption . in this scenario , the shock radius is larger than the equipartition estimate and so our equipartition parameters provide only a lower limit on the mass loss rate of the progenitor star . using equation 2.3 of @xcite we find : @xmath110 where @xmath111 is the optical depth to ffa processes and @xmath112 is the temperature normalized to @xmath113 k. here we have adopted the notation @xmath114 where @xmath106 is given in units of @xmath115 and @xmath116 is given in @xmath117 . setting @xmath118 and @xmath119 we find @xmath120 corresponding to a mass loss rate of @xmath121 ( @xmath7 ) at @xmath105 days . this is three orders of magnitude larger than that derived from the ssa interpretation . moreover , this would imply that the total energy of the radio emitting material ( a proxy for the kinetic energy of the fastest ejecta , @xmath122 ) exceeds @xmath123 erg . since hydrodynamic collapse distributes the ejecta kinetic energy as @xmath124 @xcite , this would imply that the optical data ( with @xmath125 ) have @xmath126 erg , larger than any other core - collapse sn to date . this result , taken together with the fact that a ssa model provides a significantly better fit to the radio spectra , leads us to conclude that the observed spectral turn - over is due to internal synchrotron self - absorption . in modeling the radio light - curves of sn2003bg , we adopt the formalism of s05 where we presented a rigorous formulation of the temporal and spectral evolution of synchrotron emission arising from sub - relativistic supernova ejecta . s05 show how the observed sn radio emission spectrum at any single epoch is determined by three parameters : @xmath127 , @xmath128 and @xmath129 . here , @xmath127 and @xmath128 are normalization constants of the flux density and optical depth with cgs units of @xmath130 and @xmath131 , respectively . the parameters @xmath127 , @xmath128 and @xmath129 are in turn determined by the values of four physical parameters : the magnetic field , @xmath132 , the shock radius , @xmath133 , the minimum electron lorentz factor , @xmath134 , and the ratio @xmath135 . with four physical parameters ( @xmath132 , @xmath133 , @xmath134 , @xmath136 ) and only three constraints ( @xmath127 , @xmath128 , @xmath137 ) , we must assume an additional constraint to find a unique solution . this constraint is obtained by adopting a value for @xmath136 . by inverting the equations for @xmath127 , @xmath128 and @xmath129 , s05 derive the following expressions for @xmath132 , @xmath133 and @xmath134 : @xmath138 here , @xmath90 is the distance to the supernova and @xmath139 characterizes the the thickness of the radiating electron shell as @xmath140 . the number density of the synchrotron emitting electrons then follows directly as @xmath141 where it is assumed that the contribution from electrons in a thermal distribution is negligible . the circumstellar density can be expressed in terms of a progenitor mass loss rate @xmath142 [ eqn : m_dot ] by adopting a value for the wind velocity . the total energy of the radio emitting material at a given epoch is then given by @xmath143 where it should be noted that @xmath144 depends not only on @xmath136 but also on an assumed value of @xmath75 . as discussed by @xcite , several models have been proposed for hydrodynamic evolution of the sub - relativistic supernova ejecta . based on separate assumptions , these analytic models constrain the temporal behavior of the shock radius , magnetic field , electron lorentz factor , and energy partition fractions . here , we adopt the standard ssa model ( model 1 of @xcite ) for the hydrodynamical evolution of the ejecta . this model assumes that the evolution is self - similar across the shock discontinuity , and thus @xmath145 with @xmath146 where @xmath147 characterizes the density profile of the outer sn ejecta ( @xmath148 ) and @xmath149 characterizes the density profile of the radiating electrons within the shocked circumstellar material ( @xmath150 ) . in addition , the standard ssa model assumes that the magnetic energy density ( @xmath151 ) and the relativistic electron energy density ( @xmath152 ) scale as the total post - shock energy density ( @xmath153 ) . in this scenario , the magnetic field is amplified by turbulence near the shock discontinuity , implying fixed energy fractions , @xmath75 and @xmath76 , and thus a constant value of @xmath136 throughout the evolution of the ejecta . the evolution of the magnetic field is determined by the csm density profile and the expansion of the shock radius : @xmath154 with @xmath155 . we note that for a wind stratified medium , @xmath156 and @xmath157 . the minimum lorentz factor evolves as @xmath158 with @xmath159 . from the scalings of equation [ eqn : energy ] it follows that @xmath160 , and therefore the total post - shock energy increases with time until the swept up circumstellar mass is comparable to rest mass of the sn ejecta . as shown in figure [ fig : oneplot ] and briefly discussed in [ sec : obs ] , the radio light - curves for sn2003bg are characterized by achromatic short timescale variability , inconsistent with the strict power - law evolution prescribed by the standard model . in an effort to model the overall evolution of the radio ejecta we therefore divide the multi - frequency light - curves into four subsections , each defined by the observed time of abrupt variation . we then apply the ssa model described in [ sec : model ] to each subsection . figure [ fig : lt_curves ] shows that the first peak observed for each of the radio frequencies is chromatic , and therefore produced by a cascading spectral break . we define the time range for this chromatic evolution phase to be @xmath161 days , extending until just before the first achromatic variation is observed . as discussed in [ sec : obs ] the radio emission is well described by a ssa dominated spectrum and we therefore attribute this chromatic subsection to the passage of @xmath87 through the observed frequencies . the evolution of the spectral indices ( figure [ fig : spectral_indices ] ) indicate that the self - absorption frequency cascades as @xmath162 , consistent with typical values observed for other ssa dominated radio supernovae @xcite . using our multi - frequency radio data collected during the time range of the first peak and adopting @xmath163 , we fit for the constants @xmath127 and @xmath128 as well as the temporal indices @xmath164 and @xmath165 . as discussed in section [ sec : ep ] , @xmath98 is estimated to be below the radio band during our observations and thus is unconstrained by the data . we therefore estimate @xmath166 ghz at our chosen reference time of @xmath167 days , consistent with the preliminary value calculated in [ sec : ep ] . we find a best - fit solution ( @xmath168 ; dominated by interstellar scintillation ) for parameters : @xmath169 , @xmath170 , @xmath171 , @xmath172 and @xmath173 . these values imply @xmath174 , @xmath175 and @xmath176 . as explained within the appendix of s05 , we parameterize the sharpness of the @xmath87 spectral break with @xmath177 $ ] and find @xmath178 for our final solution . this ssa model provides a reasonable fit to the chromatic subsection of the light - curves as shown in figures [ fig : oneplot ] and [ fig : lt_curves ] . with this ssa model fit , the physical parameters of the ejecta and csm are uniquely determined by equations [ eqn : b ] - [ eqn : energy ] . @xcite show that for @xmath174 and @xmath175 , @xmath179 . adopting this value , we find that the expansion of the ejecta is described by @xmath180 cm , and the average velocity is @xmath181 . the radio ejecta therefore expand with a modest sub - relativistic velocity comparable to that observed for sn2002ap @xcite and significantly slower than sn1998bw ( bulk lorentz factor , @xmath182 on a similar timescale @xcite . adopting @xmath4 we find that the total energy of the radio emitting material is given by @xmath183 erg , comparable to that of sn1998bw . the magnetic field evolves radially as @xmath184 g , slightly steeper than that observed for type ic sne 2003l ( s05 ) , 2002ap @xcite and the type iib sn1993j @xcite . the electron number density and associated mass loss rate are thus given by @xmath185 and @xmath186 where we have assumed a wolf - rayet wind velocity of @xmath187 and adopted a nucleon - to - electron density ratio of two ( appropriate for wr winds ) . this mass loss rate is consistent with the values observed for galactic wolf - rayet stars @xcite . shortly after the observed ssa peak cascades through 8.46 ghz , an abrupt rise was observed at the optically thin frequencies . by @xmath188 days the radio emission reached an _ second peak . subsequent achromatic light - curve variations were later observed at @xmath189 and 600 days and are most pronounced at frequencies above @xmath87 . as shown in figure [ fig : spectral_indices ] these variations were associated with abrupt changes in the spectral indices , most notably for those indices which straddle @xmath87 . possible causes for these abrupt light - curve variations include energy injection from a central engine , interaction of the reverse shock with density structures in the ejecta , and interaction of the forward shock with csm density enhancements . @xcite invoke energy injection to explain the achromatic second peak in the radio light - curves of sn1998bw at @xmath190 days . this interpretation was supported by the inferred mildly relativistic ejecta speeds and associated prompt gamma - ray emission , suggestive that sn1998bw was powered by a grb - like central engine . in the case of sn2003bg , however , the inferred ejecta speed is merely sub - relativistic and therefore there is no evidence for an engine - driven explosion . we conclude that the variations observed for sn2003bg can not be attributed to long - lived engine activity . @xcite show that the late - time flattening observed for the sn1993j radio light - curves can be produced through the interaction of the reverse shock with density inhomogeneities within the ejecta . while the 1-d hydrodynamical simulations provide a reasonable fit to the smooth evolution of sn1993j at @xmath191 days , it is not clear that they could produce the abrupt rise at @xmath192 days observed for sn2003bg . moreover , any abrupt features in the ejecta would likely be smoothed by instabilities in a 3-d simulation . in the third scenario , the interaction between the forward shock and strong density variations in the circumstellar material produces abrupt changes in the temporal evolution of the radio light - curves . for a wind - stratified medium with constant progenitor mass loss rate , the circumstellar density is predicted to follow a smooth @xmath193 profile . abrupt deviations from a wind density profile have been inferred for several radio supernovae ( e.g. sn1979c ; @xcite ) and are argued to be the result of a variable mass loss history and/or the effects of a binary companion on the structure of the csm . these density induced variations are more pronounced at optically thin frequencies than for optically thick . here we show that the light - curve variations observed for sn2003bg are similarly due to csm density enhancements . we model the sn2003bg radio light - curves with a simple density enhancement model under the following assumptions : ( a ) the density variations can be approximated by modest abrupt jumps in the csm such that the density profile returns to @xmath194 after each enhancement , ( b ) the hydrodynamical evolution of the ejecta returns to the self - similar solution after each jump , thereby maintaining the same values of @xmath164 , @xmath165 and @xmath195 both before and after each density enhancement , ( c ) the microphysics of the shock are not affected by the density jumps and therefore @xmath75 and @xmath76 remain constant throughout the evolution of the ejecta , ( d ) the bulk motion of the ejecta is not significantly affected by the density jumps since the slower , more massive material behind the shock continues to plow forward , and ( e ) the minimum lorentz factor of the radiating electrons is not affected by the csm enhancements , although the number of electrons participating in the radio emission increases . maintaining a self - similar evolution throughout the density variations requires that the magnetic field is enhanced by each of the csm jumps . moreover , since @xmath75 and @xmath76 are assumed to be fixed constants , the additional thermal energy produced by these enhancements causes an increase in the total energy of the radio emitting region . therefore , in fitting the subsequent peaks in the sn2003bg radio light - curves , we assume that the evolution of @xmath133 and @xmath78 remain effectively unchanged while @xmath132 and @xmath144 vary according to the strength of the csm density enhancements . we recognize that this simple density enhancement model is not fully consistent with the hydrodynamic evolution of the ejecta . in fact , a small change in the radial expansion is expected as the shock wave reaches the higher density region . for the @xmath175 self - similar solution adopted here and the modest density enhancements we infer , we estimate this to be minor effect ( @xmath9 ) . therefore , the variations in ejecta parameters ( @xmath132 and @xmath144 ) are primarily the result of the csm density enhancements rather than the minor variations in radial expansion . however , the series of density jumps that we infer cause the evolution of the radio ejecta to gradually deviate from the initial radial evolution , making our model less accurate at very late time ( after several years ) . in addition , we note that the outflow does not immediately settle on the new self - similar solution following each density variation , but instead settles on a timescale comparable for the radius to double ( i.e. the ratio of the settling time to the age of the blastwave is of order unity ) . thus the sharp features appearing in our light - curve fits are merely an artifact of our model assumptions . as discussed in [ sec : model ] , @xmath127 , @xmath128 and @xmath98 are functions of just three physical parameters : @xmath133 , @xmath78 and @xmath132 . of these , only @xmath132 responds significantly to the csm density variations . from the appendix of s05 we have @xmath196 , @xmath197 and @xmath198 . by fitting for @xmath132 at each density variation , the new values of @xmath127 , @xmath128 and @xmath98 are thus directly determined . the observed timescales and estimated circumstellar radii for each density enhancement are given in table [ tab : peaks ] . we note that we do not fit the short rise time of each phase during which the ejecta are settling to the new self - similar solution . as shown in figure [ fig : lt_curves ] , we find reasonable fits for each of the light - curve phases by invoking modest ( factor of @xmath199 ) density enhancements . in table [ tab : peaks ] we list the values of @xmath127 , @xmath128 and @xmath98 and the associated enhancement factors for the density , magnetic field and ejecta energy for each phase . it is interesting to note that the strength of the enhancements decay with time . as discussed in in [ sec : obs ] , our vla observations constrain the linear polarization of the radio emission to less than @xmath200 . although the synchrotron emission from the radio ejecta is inherently polarized , the observed polarization level can be significantly suppressed for two reasons . first , since the sn2003bg radiosphere is not resolved in the vla data our polarization limit applies to the _ integrated _ polarization of the ejecta emission . as shown for supernova remnant cas a , the integrated polarization can be significantly suppressed due to the approximate circular symmetry of the radio ejecta @xcite . second , the radio emission can be depolarized due to internal faraday dispersion within the ejecta @xcite . for the density and magnetic field values we derive from our ssa model , we predict the integrated linear polarization of the sn2003bg radio emission to be negligible , consistent with the observations . x - ray emission in supernovae can be produced by three processes @xcite : [ 1 ] non - thermal synchrotron emission from radiating electrons , [ 2 ] thermal ( free - free ) bremsstrahlung emission from material in the circumstellar shock and/or the ejecta reverse shock , and [ 3 ] inverse compton scattering of photospheric photons by relativistic electrons . we examine each of these scenarios in the context of the bright x - ray emission observed for sn2003bg . as shown in table [ tab : cxo ] , the spectral index between the optically thin radio and x - ray emission is @xmath201 for both _ chandra _ epochs . assuming a constant injection spectral index from the radio to x - ray bands we extrapolate the observed radio spectral index of @xmath202 to the x - ray and find that the synchrotron emission under - predicts the observed x - ray flux by a factor of @xmath23 in both epochs . this discrepancy grows significantly larger when the synchrotron cooling break , @xmath203 , is included , beyond which the spectrum steepens by @xmath204 . using the magnetic field derived in section [ sec : ssa ] , the cooling frequency at @xmath0 ( 120 ) days is @xmath205 ( @xmath206 ) hz . including the cooling break and extrapolating from @xmath207 to @xmath208 as @xmath209 , the synchrotron flux falls five orders of magnitude below the observed x - ray emission . we therefore conclude that the x - ray emission is not dominated by synchrotron emission . in a scenario dominated by thermal bremsstrahlung processes , the x - ray emission is produced by the forward shock plowing into circumstellar material and/or the reverse shock heating of the ejecta . in this case , the strength of the x - ray emission depends strongly on the density of the csm and unshocked ejecta . the shocked material then cools by free - free emission processes . for a wind stratified density profile , the summed free - free luminosity from the forward and reverse shock is generalized @xcite roughly by @xmath210 where @xmath211 is a constant such that @xmath212 . from our ssa radio model we find @xmath172 and @xmath213 ( @xmath214 ) , thus @xmath215 and @xmath216 . the predicted x - ray luminosity is therefore a factor of @xmath217 fainter than the first epoch observation . on the other hand , if we assume the x - ray emission must be dominated by free - free processes , then this implies a significantly higher mass loss rate of @xmath218 ( @xmath214 ) . however , we reiterate that a high density model dominated by external absorption processes is inconsistent with the observed radio spectrum ( figure [ fig : spectrum ] ) . a possible compromise could be that the x - ray emission is produced by dense csm clumps distributed with a very low filling factor ( @xmath219 ) such that the radio emission is optically thin . this scenario was invoked for the radio luminous type iin sn1986j @xcite . however , the dense csm clumps inferred for type iin s also give rise to bright nebular h@xmath14 emission lines which are not present in the nebular spectra of sn2003bg . moreover , it would be difficult to sustain the ejecta velocities inferred from our radio modeling in such a a high density environment ( chevalier & fransson , _ in prep _ ) . we conclude that the radio and optical emission do not support a scenario in which the observed x - ray flux is dominated by thermal bremsstrahlung emission . x - ray emission from supernovae can also be produced by inverse compton scattering ( ics ) of the relativistic electrons by the optical photons . this process was observed to be important in explaining the observed x - ray emission for sn2002ap @xcite and sn2003l ( s05 ) on timescales comparable to the optical maximum . in this scenario the ratio of flux measured in the radio and x - ray bands , respectively is given by @xmath220 where @xmath221 and @xmath222 are the energy densities in magnetic fields and up - scattered photons , respectively . we note that equation [ eqn : fxfr ] is relatively insensitive to the spectral energy index of the electrons . @xcite show that @xmath223 where @xmath224 is the time in days , @xmath225 is the optical luminosity at @xmath226 normalized to @xmath227 and @xmath228 is the shock velocity . at the time at the first x - ray observation , @xmath229 , the optically thin radio flux was @xmath230 and the unabsorbed x - ray flux was @xmath231 ; therefore @xmath232 . from our ssa radio modeling ( [ sec : ssa ] ) we estimate @xmath233 g and @xmath234 at @xmath0 days . based on the optical peak magnitude ( hamuy et al . , private communication ) we find @xmath235 at @xmath229 , coincident with the time of optical maximum . therefore , @xmath236 , and @xmath237 . thus @xmath238 , comparable to the observed ratio for @xmath239 . we conclude that inverse compton emission can produce the bright x - ray luminosity in the first epoch . the second epoch , however , can not be explained through ics . from our radio model at @xmath240 , @xmath241 g , @xmath242 and @xmath243 . the optical light - curve implies @xmath244 at this time . therefore we estimate @xmath245 and a predicted ic x - ray flux of @xmath246 which is a factor of @xmath247 fainter than that observed . we conclude that the x - ray flux observed at 120 days is dominated by a separate process , possibly due to a cosmic ray dominated sn shock which produces a late - time flattening of x - ray light - curves in sne ibc ( chevalier @xmath248 fransson , _ in prep . _ ) we have shown that the abrupt radio light - curve variations observed for sn2003bg can be attributed to density enhancements in the circumstellar medium . in comparison with well - sampled radio supernovae compiled from previous studies including our ongoing survey of type ibc events ( soderberg _ et al . _ , in prep ) , we find that @xmath8 of radio supernovae similarly show evidence for abrupt achromatic variations . at the same time , there are several examples of well - studied radio supernovae with smooth light - curve evolutions including type iib sn1993j @xcite and type ic sn2003l ( s05 ) among others . in table [ tab : bumps ] , we compile the light - curve properties for radio supernovae with abrupt variability . we emphasize that this compilation contains core - collapse sne from nearly all spectroscopic classifications , including hydrogen - rich type iil , iin and iib in addition to hydrogen - poor type ibc . table [ tab : bumps ] shows that the short timescale variability observed for radio supernovae can be qualitatively divided into three types . the most common effect is an abrupt steepening of the radio light - curves as observed for sn1988z @xcite . on the other hand , an abrupt rise in the optically thin flux has been observed for a couple of events ( e.g. sn1987a , @xcite ) . more intriguingly , episodic ( and perhaps periodic ) light - curve bumps are observed for sne 1979c @xcite , 2001ig @xcite , 1998bw @xcite and 2003bg , all of which are characterized by modest flux deviations ( factor of @xmath199 ) . these three types of radio variability are generally attributed to dynamical interaction of the ejecta with irregularities in the circumstellar medium . abrupt density variations in the media surrounding core - collapse supernovae are known to be produced in a variety of ways , including : [ 1 ] fluctuations in the progenitor mass loss rate and/or wind velocity . this scenario is supported through observations of local supernova remnants and massive stars which collectively suggest a complex picture for mass loss during the final stage of stellar evolution . [ 2 ] termination shocks between stellar winds ejected during the evolution of the progenitor star may produce abrupt density enhancements at radii of @xmath249 pc @xcite . case 2 is observationally motivated by the circumstellar nebulae associated with local wolf - rayet stars , which are produced by interacting stellar winds @xcite . [ 3 ] a clumpy stellar wind perhaps resulting from fragmentation of circumstellar shells by rayleigh - taylor instabilities @xcite . this scenario is motivated by narrow h@xmath14 line profiles observed for type iin supernovae . in the case of sn1988z , @xcite propose a shocked dense wind component at @xmath250 cm to explain the low velocity line emission . [ 4 ] the effects of a binary companion on the circumstellar wind profile . this case is observationally motivated by the binary companion for type iib sn1993j @xcite and the dense circumstellar ring around sn1987a at @xmath251 cm @xcite commonly attributed to the effects of a companion star @xcite . in this scenario , the binary system parameters determine the circumstellar radii at which the progenitor wind is most affected . [ 5 ] a close binary progenitor system embedded in a common envelope @xcite . given that the rate of sn progenitors in common envelope binary systems is predicted to be low ( @xmath252 of sne ibc ; @xcite ) this scenario is expected to be rare . as an example , a common envelope system has been argued to explain the large inferred density enhancement at @xmath253 cm responsible for strong radio @xcite , x - ray @xcite and optical ( h@xmath14 ; @xcite ) emission in the unusual type ic sn2001em at @xmath254 years @xcite . to summarize , the strength and location of circumstellar density variations depends on the method by which the density profile was disturbed . we note that cases 3 and 5 are generally associated with hydrogen - rich events whereas cases 1 , 2 and 4 are applicable to all classes of core - collapse supernovae . finally , in light of this compilation it is interesting to note that an abrupt rise in the optically thin radio emission of type ibc supernovae can also be attributed to the presence of an off - axis gamma - ray burst jet @xcite . as the jet sweeps up circumstellar material and decelerates , it eventually undergoes a dynamical transition to sub - relativistic expansion on a timescale of @xmath255 yr @xcite . once sub - relativistic , the jet spreads sideways and the ejecta rapidly approach spherical symmetry . the observational signature of an off - axis grb is therefore an abrupt optically thin rise as the jet sweeps through our line - of - sight . due to the location of the synchrotron spectral peak , this signature is is most easily detected in the radio band on a timescale of @xmath255 to 10 years @xcite . yet as we have shown here , many radio supernovae show evidence for density induced variability on this same timescale . we therefore emphasize that abrupt flux variations in radio supernovae may not be used as a reliable proxy for an off - axis grb jet and/or energy injection from a central engine . a comparison of the radio light - curves represented in table [ tab : bumps ] shows that the variations are generally unique to each supernova . however , as figure [ fig : lt_curve_comparison ] shows , the separation and intensity of the observed variations in sn2003bg and the type iib sn2001ig are strikingly similar . by simply scaling our sn2003bg light - curve fits by a factor of @xmath256 we obtain a reasonable match to the sn2001ig radio evolution . with this scaling we derive the following physical parameters for the sn2001ig radio emitting ejecta : @xmath257 cm , @xmath258 , @xmath259 g and @xmath260 erg for @xmath4 . the electron number density is thus given by @xmath261 with successive peaks of relative intensity comparable to those observed in sn2003bg ( table [ tab : peaks ] ) and an associated mass loss rate of @xmath262 ( @xmath187 ) . we note that this mass loss estimate is comparable to that reported by @xcite , but they adopt free - free absorption for the dominant absorption process and a progenitor wind velocity of @xmath263 , typical of red supergiant type ii sn progenitors . our ssa modeling , however , shows the ejecta velocity of sn2001ig is consistent with those of sne ibc and the mass loss rate is in fact comparable to galactic wolf - rayet stars , similar to the case of sn2003bg . figure [ fig : lt_curve_comparison ] shows that strong light - curve deviations are observed for both sne on timescales of @xmath264 and 300 days with evidence for a weaker feature at @xmath265 days . making the reasonable approximation that the ejecta evolve as @xmath266 cm , these variations correspond to density enhancements at radial distances of @xmath267 and @xmath268 cm . assuming a progenitor wind velocity of @xmath187 , we infer ejection timescales of @xmath269 and 24 years before the sn explosion . moreover , as shown in the figure [ fig : lt_curve_comparison ] the strength of the light - curve variations decreases with time implying that the largest enhancements are closest to the explosion site . in addition to their radio light - curves , the optical evolution for sne 2003bg and 2001ig are also similar . in both cases there was spectroscopic evidence for a thin layer of hydrogen near the time of optical maximum @xcite ; however nebular spectra revealed no hydrogen emission in either event ( @xcite ; hamuy , private communication ) days past explosion . ] . this evolution is characteristic of type iib supernovae ( e.g. sn1993j ; @xcite ) . however , their inferred ejecta velocities resemble more closely those of sne ibc and therefore imply compact ( wolf - rayet ) progenitors . we conclude that these two events are intermediate between types iib and ibc , and show more overall similarities to hydrogen - poor events . moreover , the likeness of their observed radio and optical evolution may indicate similar environments and progenitor evolutions . the overall resemblance of sn2003bg and sn2001ig as outlined in the previous section suggests that a similar process is responsible for the observed density variations . based on the location ( @xmath270 cm ) and relative intensity ( factor of @xmath199 ) of the density enhancements , we find that cases 1 and 4 ( a variable mass loss rate and the tidal effects of a binary companion , respectively ) provide the most likely explanations for the inferred circumstellar structure . noting the abrupt fluctuations in the sn2001ig radio light - curves , @xcite recently proposed a binary induced pinwheel nebula for the inferred episodic density enhancements . in this scenario , the colliding stellar winds produce a smooth dusty plume that spirals in the plane the binary orbit . a highly eccentric orbit is required to produce structure at well - defined intervals and a favorable viewing angle is necessary to reproduce the observed flux variations . here the location and intensity of the pinwheel structure is determined by the binary parameters . @xcite argue further support for this model based on the discovery of a late - b - late - f supergiant coincident ( rms uncertainty of 0.03 arcsec , corresponding to 1.7 pc at @xmath271 mpc ) with the sn2001ig optical position which they interpret as the putative binary companion . however , in cases of observed pinwheel nebulae ( e.g. @xcite ) , the dust plumes are attributed to the strong wind - wind interaction between a wolf - rayet and o- or early - b star companion . taken together with the substantial uncertainty in the position of the putative companion , this suggests that a binary induced pinwheel nebula for sn2001ig is not clearly substantiated . adopting a pinwheel model for both sne 2003bg and 2001ig would imply that the geometry of their circumstellar nebulae are nearly identical and thus they share remarkably similar binary systems . moreover , this model would imply that our viewing angle with respect to their circumstellar nebulae must also be very similar . while this scenario can not be ruled out , it is clearly unlikely . assuming a single star progenitor model , the mass loss rates derived through our ssa modeling for sne 2003bg and 2001ig imply massive wolf - rayet progenitors . broadband observations of galactic wr stars show that the circumstellar media for these events are often significantly disturbed . in particular , x - ray observations show evidence for colliding stellar winds and dense clumps ( e.g. @xcite ) . if the progenitors of sne 2003bg and 2001ig were single wr stars then the observed density variations could be produced by a series of enhanced mass loss episodes . to be consistent with the radio data , the mass loss ejections must ramp up as the star nears explosion . in this scenario , the observed variation timescale and the lack of csm structure at radii inward of @xmath272 cm may be explained by a quasi - periodic mechanism which drives enhanced mass loss episodes every @xmath269 years . if accompanied by wind velocity variations , these episodes could produce shells of circumstellar material , overall consistent with the circularly symmetric density structures we infer for sne 2003bg and 2001ig . the observed similarity between sne 2003bg and 2001ig implies that these mass loss episodes may be common among some fraction of hydrogen - poor events . these clues suggest that episodic variations in the progenitor mass loss rate and/or wind velocity provide a reasonable explanation to the radio light - curve variations observed for both sne 2003bg and 2001ig . given the predicted timescale for the mass ejections , these two sne provide some of the best evidence for unusual mass loss evolution immediately preceding the supernova explosion . while the presence of a binary companion can not be ruled out for either of these hydrogen - poor events , additional studies of short timescale variability in radio supernovae will enable discrimination between single and binary progenitor models . we report extensive radio and x - ray observations of sn2003bg . the spectroscopic evolution of this unusual supernova shows a transition from a broad - lined type ic to a hydrogen - rich type ii and later to a typical hydrogen - poor type ibc . this evolution strengthens the connection between type ii and type ibc ( including broad - lined ) events . we show that the extraordinarily luminous radio emission for a sn ibc is well described by a self - absorption dominated synchrotron spectrum while the observed x - ray emission at @xmath0 days is adequately fit by inverse compton scattering of the optical photons off of the synchrotron emitting electrons . our radio model implies a sub - relativistic ejecta velocity , @xmath1 , and a size of @xmath273 cm at @xmath2 days . this analysis emphasizes that broad optical absorption lines do not imply relativistic ejecta . we find that the total energy of the radio emitting region evolves as @xmath3 erg assuming equipartition of energy between relativistic electrons and magnetic fields ( @xmath4 ) . the circumstellar density is well described by stellar wind profile , @xmath274 , with modest ( factor of @xmath5 ) episodic enhancements which produce abrupt achromatic flux variations . we show that free - free absorption does not contribute significantly to the radio spectrum and estimate an average mass loss rate of @xmath275 , consistent with observed values for local wolf - rayet stars . comparison with other events reveals that @xmath8 of radio supernovae show similar short timescale light - curve variations which are attributable to circumstellar density irregularities . this compilation emphasizes that abrupt radio light - curve variations can not be used as a reliable proxy for an engine - driven explosion , including off - axis gamma - ray bursts . finally , the radio light - curves and spectroscopic evolution for sn2003bg are strikingly similar to those of sn2001ig , suggestive of a common progenitor evolution for these two events . the overall similarity of sne 2003bg and 2001ig to radio sne ibc is suggestive of a compact wolf - rayet progenitor model . based on the relative intensity of the inferred density enhancements , we conclude that the progenitors of hydrogen - poor sne 2003bg and 2001ig experienced quasi - periodic mass loss episodes just prior to the sn explosion . we are indebted to barry clark for his generous scheduling of the vla for this project . we thank m. hamuy and a. filippenko for making their optical sn2003bg data available to us during this study . the authors thank reem sari , doug leonard , edo berger , michael rupen , paolo mazzali , avishay gal - yam , p. brian cameron , derek fox , ehud nakar and kurt weiler for useful discussions . caltech sn research is supported by nsf and nasa grants . ams acknowledges support by the national radio astronomy observatory graduate summer student research assistantship program and the nasa graduate student research program . rac acknowledges nsf grant ast-0307366 . , r. n. , gaensler , b. m. , wheaton , v. c. , staveley - smith , l. , tzioumis , a. k. , bizunok , n. s. , kesteven , m. j. , and reynolds , j. e. 2002 , publications of the astronomical society of australia , 19 , 207 . lrrrrrrrc [ tab : vla ] 2003 mar 4 & 10 & & & @xmath276 & & & & d + 2003 mar 6 & 12 & @xmath277 & & @xmath278 & & @xmath279 & & d + 2003 mar 17 & 23 & @xmath280 & @xmath281 & @xmath282 & @xmath283 & @xmath284 & & d + 2003 mar 29 & 35 & & @xmath285 & @xmath286 & @xmath287 & @xmath288 & @xmath289 & d + 2003 apr 11 & 48 & & @xmath290 & @xmath291 & @xmath292 & @xmath293 & @xmath294 & d + 2003 apr 21 & 58 & & @xmath295 & @xmath296 & @xmath297 & @xmath298 & @xmath299 & d + 2003 apr 26 & 63 & & @xmath300 & @xmath301 & @xmath302 & @xmath303 & @xmath304 & d + 2003 may 6 & 73 & & @xmath305 & @xmath306 & @xmath307 & @xmath308 & @xmath309 & d + 2003 may 18 & 85 & & @xmath310 & @xmath311 & @xmath312 & @xmath313 & & dna + 2003 may 24 & 91 & & @xmath314 & @xmath315 & @xmath316 & @xmath317 & & dna + 2003 jun 17 & 115 & & @xmath318 & @xmath319 & @xmath320 & @xmath321 & & a + 2003 jun 31 & 129 & @xmath322 & @xmath323 & @xmath324 & @xmath325 & @xmath326 & & a + 2003 jul 4 & 132 & @xmath327 & @xmath328 & @xmath329 & @xmath330 & & & a + 2003 jul 14 & 142 & @xmath331 & @xmath332 & @xmath333 & @xmath334 & & & a + 2003 jul 29 & 157 & @xmath335 & @xmath336 & @xmath337 & @xmath338 & & & a + 2003 aug 2 & 161 & @xmath339 & @xmath340 & @xmath341 & @xmath342 & & & a + 2003 aug 22 & 181 & @xmath343 & @xmath344 & @xmath345 & @xmath346 & @xmath347 & & a + 2003 sep 11 & 201 & @xmath348 & @xmath349 & @xmath350 & @xmath351 & @xmath352 & & bna + 2003 sep 24 & 214 & @xmath353 & @xmath354 & @xmath355 & @xmath356 & @xmath357 & & bna + 2003 oct 7 & 227 & @xmath358 & @xmath359 & @xmath360 & @xmath361 & @xmath362 & & bna + 2003 oct 22 & 242 & @xmath363 & @xmath364 & @xmath365 & & @xmath366 & & b + 2003 nov 4 & 255 & @xmath367 & @xmath368 & @xmath369 & @xmath370 & @xmath371 & & b + 2003 nov 15 & 266 & @xmath372 & @xmath373 & @xmath374 & @xmath375 & @xmath376 & & b + 2003 dec 4 & 285 & @xmath377 & @xmath378 & @xmath379 & @xmath380 & @xmath381 & & b + 2003 dec 19 & 300 & @xmath382 & @xmath383 & @xmath384 & @xmath385 & @xmath386 & & b + 2004 jan 14 & 326 & @xmath387 & @xmath388 & @xmath389 & @xmath390 & @xmath391 & & b + 2004 jan 25 & 337 & @xmath392 & @xmath393 & @xmath394 & @xmath395 & @xmath396 & & cnb + 2004 feb 8 & 351 & @xmath397 & @xmath398 & @xmath399 & @xmath400 & @xmath401 & @xmath402 & cnb + 2004 feb 25 & 368 & @xmath403 & @xmath404 & @xmath405 & @xmath406 & @xmath407 & & c + 2004 apr 2 & 405 & & & @xmath408 & & @xmath409 & & c + 2004 apr 7 & 410 & @xmath410 & @xmath411 & @xmath412 & & & & c + 2004 apr 21 & 424 & @xmath413 & @xmath414 & @xmath415 & @xmath416 & @xmath417 & @xmath418 & c + 2004 apr 31 & 434 & @xmath419 & @xmath420 & @xmath421 & @xmath422 & @xmath423 & & c + 2004 may 2 & 435 & & & @xmath424 & & & @xmath425 & c + 2004 jun 29 & 493 & @xmath426 & @xmath427 & @xmath428 & @xmath429 & @xmath430 & @xmath431 & d + 2004 aug 8 & 533 & @xmath432 & @xmath433 & @xmath434 & @xmath435 & @xmath436 & & d + 2004 nov 15 & 632 & & & @xmath437 & @xmath438 & @xmath439 & & a + 2004 nov 29 & 646 & @xmath440 & @xmath441 & & & & & a + 2005 jan 24 & 702 & @xmath442 & @xmath443 & @xmath444 & @xmath445 & & & bna + 2005 jan 31 & 709 & & & & & @xmath446 & & bna + 2005 mar 19 & 756 & & & @xmath447 & @xmath448 & & & b + 2005 apr 11 & 779 & @xmath449 & @xmath450 & & & & & b + 2005 may 22 & 820 & @xmath451 & @xmath452 & @xmath453 & @xmath454 & @xmath455 & & b + 2005 aug 12 & 902 & @xmath456 & @xmath457 & @xmath458 & @xmath459 & @xmath460 & @xmath461 & c + 2005 oct 27 & 978 & @xmath462 & @xmath463 & @xmath464 & & & & dnc + [ tab : vla ] lrlrrrrrr 2003 march 24.5 & 30 & power law & 0.013 & @xmath465 & & 24/20 & @xmath466 & -0.94 + & & & @xmath467 & @xmath468 & & 24/19 & @xmath469 & -0.94 + & & th . & 0.013 & & @xmath470 & 26/20 & @xmath471 & -0.95 + & & & @xmath472 & & @xmath473 & 25/19 & @xmath474 & -0.95 + 2003 june 22.3 & 120 & power law & 0.013 & @xmath475 & & 5/7 & @xmath476 & -0.89 + & & th . & 0.013 & & @xmath477 & 4/7 & @xmath478 & -0.97 + [ tab : cxo ] lrrrrrrrr 1 & @xmath479 & @xmath480 & 3.1 & 6.2 & 0.11 & & & + 2 & 120 - 250 & 6.7 - 12 & 2.7 & 23 & 0.15 & 1.8 & 1.3 & 1.8 + 3 & 300 - 500 & 14 - 21 & 2.5 & 46 & 0.18 & 1.4 & 1.2 & 1.4 + 4 & @xmath481 & @xmath482 & 2.4 & 67 & 0.19 & 1.2 & 1.1 & 1.2 + [ tab : peaks ] lcrrrr 1957d & ii & @xmath483 & 4.9 & abrupt steepening & 1,2,3 + 1978k & iin & 2.6 & 4.9 & abrupt steepening & 4 + 1979c & iil & 1.2 & 5.0 & episodic bumps & 1,5,6 + 1980k & iil & 0.4 & 5.0 & abrupt steepening & 1,7 + 1986j & iin & 3.8 & 4.9 & abrupt steepening & 8 + 1987a & ii & 0.0093 & 0.84 & abrupt rise & 9,10 + 1988z & iin & 2.5 & 4.9 & abrupt steepening & 11 + 1998bw & ic & 0.030 & 4.9 & episodic bumps & 12,13,14 + 2001em & ic / ii & @xmath484 & 4.9 & abrupt rise & 15,16 + 2001ig & iib & 0.20 & 5.0 & episodic bumps & 17 + 2003bg & ic / ii & 0.27 & 4.9 & episodic bumps & 18 + 2004c & ic & @xmath485 & 4.9 & abrupt rise & 19 + 2004cc & ic & 0.068 & 8.5 & abrupt steepening & 19 + [ tab : bumps ]

we have developed a relativistic coupled - cluster theory to incorporate nuclear spin - dependent interaction hamiltonians perturbatively . in this theory , the coupled - cluster operators in the electronic sector are defined as tensor operators of rank one and we introduce suitable diagrammatic representations . for properties calculations , the electronic part is first calculated and later coupled to the nuclear spin part . the method is ideal to calculate parity violating nuclear spin - dependent electric dipole transition amplitudes , , of heavy atoms . to validate the proposed method the of the transition incs is calculated for selected mbpt diagrams and compared with the results from our theory .

1105.3447

experimental observation of nuclear spin dependent ( nsd ) parity non - conservation ( pnc ) is a signature of nuclear anapole moment ( nam ) @xcite . it is the most dominant nsd - pnc effect among three possible sources . the other two are : combination of hyperfine and nuclear spin - independent pnc ; and spin - dependent @xmath3 exchange between electrons and nucleus . to date , the most precise atomic pnc measurement , of atomic cs , has provided the only experimental evidence of nam @xcite . one major hurdle to a clear observation of nuclear anapole moment is the large nuclear spin - independent ( nsi ) signal , which overwhelms the nsd signature . however , proposed experiments with single ba@xmath4 ion @xcite could probe pnc in the @xmath5 transition , where the nsi component is zero . this could then provide an unambiguous observation of nsd - pnc and nam in particular . the ongoing experiments with atomic ytterbium @xcite is another possibility , the @xmath6 transition , to observe nsd - pnc with minimal mixture from the nsi component . one crucial input , which is also the source of large uncertainty , to extract the value of nam and nuclear weak charge in nsi - pnc is the input from atomic theory calculations . in the case of isotope chain measurements , like in yb , the pnc observable is a ratio . atomic theory contributions then cancel and parameters can be extracted without atomic theory calculations . however , results from atomic theory calculations are important in estimating the expected value of pnc transition amplitudes and extracting nam does require results from atomic theory . for these reasons , it is important to employ reliable and accurate many - body theory in the atomic theory calculations . recently , atomic theory calculations have investigated the nsd - pnc of ba@xmath4 and ra@xmath4 using the atomic many - body perturbation theory ( mbpt ) @xcite . another recent work reports the theoretical estimate of the nsd - pnc observable of yb @xcite and the calculations are based on the ci - mbpt method @xcite . there is also an earlier work on nsd - pnc of yb using the same method @xcite . very recently , the ci - mbpt method is used to to calculate the nsd - pnc observables of ba@xmath4 , yb@xmath4 and ra@xmath4 @xcite . besides the methods mentioned , earlier works on ba@xmath4 @xcite and yb @xcite used configuration interaction ( ci ) . to date , coupled - cluster theory ( cct ) considered as one of the most reliable and accurate many - body theory has not been used in the nsd - pnc calculations . the difficulty of developing a suitable theory stems from the complications of dealing with the nuclear spin dependent interaction . in a previous paper of ours we reported the development of a relativistic coupled - cluster ( rcc ) theory to incorporate nuclear spin - dependent interaction in a consistent scheme @xcite . in this work we provide elaborate details of the theory and touch upon related subtle issues . the coupled - cluster ( cc ) theory@xcite is one of the most reliable many - body theory to incorporate electron correlation in atomic calculations . in atomic physics , the relativistic coupled - cluster ( rcc ) theory has been used extensively in atomic properties calculations , for example , hyperfine structure constants @xcite , electric dipole moments @xcite , and electromagnetic transition properties @xcite . in atomic pnc calculations too , rcc is the preferred theory and several groups have used it to calculate nsi - pnc of atoms @xcite . however , the calculations in ref . @xcite are entirely based on rcc with a variation we refer to as perturbed rcc ( prcc ) , where as the calculations in ref . @xcite are based on sum over states with cc wave functions . naturally , the former incorporates electron correlation more precisely than the later approach . in this work we provide a detailed explanation on the approach we have adopted to incorporate nuclear spin - dependent interaction hamiltonian as a perturbation in rcc theory . in section . [ review_rcc ] , we give a very short account of rcc theory to serve as a quick reference . we then give the details of our formulation to represent a nsd interaction perturbed cc operator in section . [ pert_cc_fn ] . prcc equations of closed - shell and one - valence systems is derived in this section . to simplify angular factor calculations , the calculations in the electronic sector is separated out . based on this the calculation of nsd - pnc observable , in the electronic sector , is described in section . [ e1pnc_elec ] . next section , section . [ couple_nsd ] , describes the coupling of results from the electronic sector with the nuclear spin . this completes the theoretical development and then , the correctness of theory is verified in section . [ validation_prcc ] . for which , we use mbpt results and calculate the nsd - pnc of the @xmath7 transition in cs . we then conclude and the angular factors of all the diagrams in the linearized prcc are given in the appendix . for @xmath8-electron atoms or ions the dirac - coulomb hamiltonian , appropriate to account for relativistic effects , is @xmath9 + \sum_{i < j}\frac{1}{r_{ij } } , \label{dchamil}\ ] ] where @xmath10 and @xmath11 are the dirac matrices , @xmath12 is the linear momentum , @xmath13 is the nuclear coulomb potential and last term is the electron - electron coulomb interactions . for one - valence systems it satisfies the eigen value equation @xmath14 where @xmath15 and @xmath16 are the atomic state and energy respectively . in the cc method , the atomic state is expressed in terms of @xmath17 and @xmath18 , the closed - shell and one - valence cluster operators respectively , as @xmath19 |\phi_v\rangle , \label{cceqn_1v}\ ] ] where @xmath20 is the one - valence dirac - fock reference state . it is obtained by adding an electron to the closed - shell reference state , @xmath21 . in the coupled - cluster singles doubles ( ccsd ) approximation @xmath22 and are the solutions of the nonlinear coupled equations [ t_eqn ] @xmath23 where @xmath24 is the similarity transformed hamiltonian and the normal order hamiltonian @xmath25 . the states @xmath26 and @xmath27 are the singly and doubly excited determinants , respectively and @xmath28 ( @xmath29 ) denote occupied ( virtual ) orbitals . the details of the derivation are given in ref . like in @xmath30 , the one - valence cluster operator @xmath31 in the ccsd approximation is @xmath32 . and these are the solutions of [ cc_sin_dou ] @xmath33 { \bar}{h}{_n}{s } \bar h_n s^{(0)}\ } |\phi_v\rangle & = & e_v^{\rm att}\langle\phi_v^p|s^{(0)}_1|\phi_v\rangle , \label{ccsingles } \\ \langle \phi_{va}^{pq}|\bar h_n + \{\contraction[0.5ex ] { \bar}{h}{_n}{s}\bar h_n s^{(0)}\ } |\phi_v\rangle & = & e_v^{\rm att}\langle\phi_{va}^{pq}|s^{(0)}_2|\phi_v\rangle , \label{ccdoubles}\end{aligned}\ ] ] where @xmath34 is the attachment energy of the valence electron . in our previous work @xcite , we provide details of the derivation . time independent perturbation theory is the standard procedure to incorporate external perturbations or additional interactions in atomic many - body calculations . however , a basic requirement of perturbative calculations is a complete set of intermediate atomic states , which is non - trivial to generate . the perturbed cc method @xcite , on the other hand , implicitly accounts for all the possible intermediate states . in this work , we consider the perturbation as the nuclear spin - dependent parity non - conserving ( pnc ) interaction @xmath35 where @xmath36 is the weak nuclear moment of the nucleus and @xmath37 is the nuclear density . the weak nuclear moment is expressed in terms of the neutron and proton numbers @xmath38 , where @xmath39 and @xmath40 are respectively the vector electron and the axial vector nucleon coupling coefficients . there are two complications arising from the nuclear spin operator @xmath41 in @xmath42 . first , the cluster operators in the electron space are rank one operators , and second , the atomic states in the one - valence sector are eigenstates of total angular momentum @xmath43 . both of these are relatively simple to incorporate at lower order mbpt calculations , however , implementing these in non - perturbative theory like rcc is nontrivial . accordingly , the method we have developed and implemented in the current work are very different from our previous works @xcite . the other simplifying feature of the previous works is , the methods developed were for closed - shell systems where the total electronic angular momentum @xmath44 . with the pnc interaction , the total atomic hamiltonian is @xmath45 where @xmath46 is the perturbation parameter . as @xmath42 mixes atomic states of opposite parities , the eigenvalue equation is modified to @xmath47 note that the first order energy correction @xmath48 as @xmath42 is an odd parity interaction hamiltonian . this is taken into account while writing the above eigenvalue equation . here @xmath49 are the mixed parity atomic states and to first order in perturbation @xmath50 if @xmath15 and @xmath51 are atomic states of same parity , then the @xmath42 induced electric dipole ( @xmath52 ) transition amplitude is @xmath53 where @xmath54 is the dipole operator . the perturbation expression in eq . ( [ psi_pert ] ) require a complete set of intermediate atomic states @xmath55 . this , as mentioned at the beginning of the section , is non - trivial to obtain in atomic many - body calculations . summation over intermediate states is circumvented when @xmath0 is calculated with cc atomic states . for this define a new set of cluster operators @xmath56 , which unlike @xmath30 connects the reference state to opposite parity states . this is the result of incorporating one order of @xmath42 and for this reason we refer to @xmath57 as the perturbed cluster operators . although hyperfine states are natural to @xmath42 , cluster operator @xmath57 is defined to operate only in the electronic space and is a rank one operator . so , the mixed parity atomic state in rcc is @xmath58 the scalar product with the nuclear spin @xmath59 in the exponent restores @xmath57 to the correct form of the wave operator . at this point it is convenient to separate out the electronic part of the interaction hamiltonian @xmath60 which operates only in the electronic space , so that @xmath61 the remaining part of this section describe how to arrive at a consistent representation of @xmath56 and extension of the method to one - valence systems , where @xmath62 are the perturbed cluster operators . to define the multipole and parity selection rules of @xmath63 , we examine the second order mbpt wave function with @xmath42 as one of the perturbations . from the generalized bloch equation @xcite , the total wave operator is @xcite @xmath64 here @xmath65 has @xmath66 orders of residual coulomb interaction , where as @xmath67 has one order of @xmath42 and @xmath66 orders of residual coulomb interaction . following which the mixed parity state @xmath68 where @xmath69 is the opposite parity component arising from the @xmath42 . the first order wave function is then , @xmath70 |\phi_v \rangle + \omega^{(1)}_{v , 1 } |\phi_v \rangle . \label{pert_state_mbpt}\ ] ] although , notation wise @xmath71 seem first order , it is second order in perturbation : one order each in residual coulomb interaction and @xmath42 . for model space consisting of same parity states @xmath72 p = qh_{\rm pnc}^{\rm nsd } \omega_{v,0 } p + qv_{\rm res}\omega_{v,1}^{(0 ) } p. \label{omega_11}\ ] ] here , and @xmath73 and @xmath74 are the projection operators of the model and complementary spaces , respectively , @xmath75 is the dirac - fock hamiltonian , @xmath76 is the residual coulomb interaction and the zeroth order mixed parity wave operator @xmath77 using @xmath78 and @xmath79 , we may compute @xmath80 to third order in perturbation , the details of which are discussed in ref . @xcite . the expression of @xmath81 and associated selection rules are what we need to arrive at a consistent description of @xmath82 . in fig [ pcc_mbpt](a ) we show an mbpt wave function diagram of @xmath71 arising from the second term in eq . ( [ omega_v1 ] ) . the algebraic expression of the diagram is @xmath83 where @xmath84 are the orbital energies . the states @xmath85 represent uncoupled atomic states comprising of electronic and nuclear states . on closer examination , the component of @xmath42 which operates in the nuclear subspace is diagonal . so , for further calculations , we separate out the matrix elements in the electronic subspace and combine with the nuclear part at the end of the calculations . for the mbpt diagram considered earlier fig . [ pcc_mbpt](a ) , the electronic part is as shown in fig . [ pcc_mbpt](b ) . the corresponding algebraic expression is @xmath86 details of the angular reduction when electronic state is coupled with the nuclear state are discussed in the later sections of the paper . like wise , the perturbed cluster operators @xmath56 , as defined earlier operate only in the electronic . for this consider the angular part of the matrix elements in eq . ( [ mbpt_2body ] ) , the diagrammatic representation is shown in fig . [ pcc_mbpt1 ] . the angular diagrams are based on the conventions used in lindgren and morrison @xcite and the same is followed in the remaining of the paper while referring to angular momentum diagrams . in the figure , the angular diagram on the right hand side indicates the multipole structure of the @xmath87 and like the electron - electron coulomb interaction @xmath88 where @xmath89 are c - tensor operators and @xmath90 indicates the two c - tensor operators couple to a rank one tensor operator . following general rules of coupling tensor operators , the rank of the tensor operators must satisfy the triangular conditions @xmath91 , @xmath92 and @xmath93 . effectively , @xmath56 is a rank one operator in the electronic subspace and forms a scalar operator after coupling with @xmath59 . the other important difference from the @xmath94 is the parity selection rule at the vertices . the combined parities at the vertices are opposite @xmath95 . after a similar analysis , the singles operator is @xmath96 based on the multipole structures , the perturbed cluster operators are diagrammatically represented as shown in fig . [ pert_cc_op ] . for the doubles @xmath97 , to indicate the multipole structure , an additional line is added to the interaction line . the mixed parity states , eigenstates of @xmath98 , in the perturbed relativistic coupled - cluster ( prcc ) is as given in eq . ( [ prcc_cl ] ) . however , as @xmath42 is considered to first order only , it is sufficient to consider up to linear terms in the expansion of @xmath99 . the prcc atomic state is then @xmath100 |\phi_0\rangle . \label{pccwave2}\ ] ] the cluster operator @xmath56 , as described earlier , incorporates @xmath101 to first order and residual coulomb interaction to all order . the closed - shell equivalent of eq . ( [ pnc_eigen ] ) , the mixed parity eigenvalue equation , is @xmath102 from the definition of the coupled - cluster mixed parity state in eq . ( [ pccwave2 ] ) , the eigen value equation in terms of the coupled - cluster wavefunction is @xmath103 |\phi_0 \rangle \nonumber \\ = e_0 e^{t^{(0)}}\left [ 1 + \lambda \pt\cdot\mathbf{i } \right]|\phi_0\rangle.\end{aligned}\ ] ] in terms of @xmath104 , defined earlier , the eigenvalue equation is simplified to @xmath105 |\phi_0 \rangle \nonumber \\ = \delta e_0 e^{t^{(0 ) } } \left [ 1 + \lambda \pt\cdot\mathbf{i } \right ] \phi_0 \rangle,\end{aligned}\ ] ] where @xmath106 is the closed - shell correlation energy . retaining terms which are first - order in @xmath46 , we get @xmath107 e^{t^{(0 ) } } |\phi_0 \rangle = \delta e_0 \pt\cdot\mathbf{i } e^{t^{(0 ) } } |\phi_0 \rangle.\ ] ] from here on , for simplicity , we drop the nuclear spin operator @xmath59 and work only in the electronic space . at a later stage of the calculations the electronic part is coupled with the nuclear spin . operating with @xmath108 and projecting on singly and doubly excited states @xmath109 and @xmath110 , respectively , we get the cc equations for singles and doubles perturbed cluster amplitudes as [ pcceq ] @xmath111 where , @xmath112 , is the similarity transformed pnc interaction hamiltonian in the electronic subspace . the perturbed cluster operators are solutions of these coupled linear equations . however , the equations are nonlinear in the unperturbed cluster operators @xmath30 . the advantage of separating the cluster operators into two categories @xmath94 and @xmath56 is , the two sets of equations can be solved sequentially . as the @xmath113 has no dependence on @xmath56 , the @xmath113 equations are solved first and then the @xmath56 equations are solved . an approximate form of eq . ( [ pcceq1 ] ) and ( [ pcceq2 ] ) , but which contains all the important many - body effects , are the linearized cluster equations . this is obtained by considering [ lpcc_term ] @xmath114 the diagrams in the singles and doubles equation arising from @xmath115 are shown in fig . [ psingles ] and [ pdoubles ] , respectively . the other form of eq . ( [ lpcc_term ] ) is to write the equation in terms of specific cluster amplitudes . the singles equation is then , @xmath116 where @xmath117 and @xmath118 , are the matrix elements @xmath119 and @xmath120 , respectively , and @xmath121 and @xmath122 are the unperturbed single and double excitation rcc amplitudes , respectively . for compact notation , we have defined @xmath123 . similarly , @xmath124 and @xmath125 are the antisymmetized unperturbed and perturbed cc amplitudes respectively . the equation of the double excitation perturbed cluster amplitudes is @xmath126 where @xmath127 represents terms similar to those within parenthesis but with the combined permutations @xmath128 and @xmath129 . these equations are similar to the unperturbed cluster equations in ref . there is , however , a major difference from the all - order equations of unperturbed cluster operators used in other works @xcite which use antisymmetized cluster operators . here , all order is the same as the linearized coupled - cluster theory and in the antisymmetized representation @xmath130 . for our work , which is based on diagrammatic evaluation of angular factors , the representation without anti - symmetrization is preferable as it follows directly from the diagrams . to solve eq . ( [ lin_s ] ) and ( [ lin_d ] ) , the expressions of @xmath131 are separated into multipole components and the angular factor of each terms in the two equations are given in the appendix . from eq . ( [ pnc_grnd ] ) , we may write the perturbed eigenvalue equation of the one - valence system as @xmath132 as defined earlier , @xmath16 is the energy of the one - valence system . similar to the closed - shell case , the perturbed wavefunction in cc theory is @xmath133 \left [ 1 + s^{(0 ) } + \lambda \ps \cdot\mathbf{i } \right ] |\phi_v \rangle , \label{psiptrb1v}\ ] ] where @xmath62 is the perturbed cc operator of the one - valence part . the diagrammatic representation of the valence single and double perturbed cluster operators @xmath134 and @xmath135 , respectively , are shown in fig . [ pert_cc_op ] . these are topologically equivalent to the diagrams of the closed - shell operators @xmath136 and @xmath137 , however , with one of the core lines rotated . ( [ psiptrb1v ] ) in terms of the prcc wavefunction is @xmath138 \left [ 1 + s^{(0 ) } \right . \nonumber \\ & & \left . + \lambda\ps\cdot\mathbf{i } \right ] |\phi_v \rangle = e_v e^{t^{(0 ) } } \left [ 1 + \lambda \pt\cdot\mathbf{i } \right ] \left [ 1 + s^{(0 ) } \right . \nonumber \\ & & \left . + \lambda \ps\cdot\mathbf{i } \right ] |\phi_v \rangle.\end{aligned}\ ] ] to derive the @xmath62 equations , like in the closed - shell case , project the above equation on @xmath139 and retain the terms linear in @xmath46 . for further simplification , use normal - ordered form of the hamiltonian , which for the one - valence system is @xmath140 . after these sequence of operations , and retaining only the electronic part like in the closed - shell case , the eigen value equation is modified to @xmath141 |\phi_v \rangle \nonumber \\ = \left [ \delta e_v \ps + \delta e_v \pt ( 1 + s ) \right]|\phi_v \rangle , \label{deltae1v}\end{aligned}\ ] ] where @xmath142 , is the correlation energy of the one - valence system . projecting eq . ( [ deltae1v ] ) with the excited determinants @xmath143 and @xmath144 , we get the perturbed cc equations of the singles and doubles respectively , in the form @xmath145{}{v}{_{\rm n}t^{(1)}}{s}\bar{h}_{\rm n } \mathbf{t}^{(1)}s^{(0)}\ } + \bar{\mathbf{h}}_{\rm elec}^{\rm nsd } \nonumber \\ & & + \ { \contraction{}{h}{_{\rm elec}^{\rm nsd}}{s } \bar{\mathbf{h}}_{\rm elec}^{\rm nsd}{s}^{(0 ) } \}|\phi_v \rangle = \delta e_v \langle \phi^p_v | \pso|\phi_v \rangle , \label{ccsptrb1v1}\end{aligned}\ ] ] @xmath146{}{v}{_{\rm n}t^{(1)}}{s}\bar{h}_{\rm n } \mathbf{t}^{(1)}s^{(0)}\ } + \bar{\mathbf{h}}_{\rm elec}^{\rm nsd } \nonumber \\ & & + \ { \contraction{}{h}{_{\rm elec}^{\rm nsd}}{s } \bar{\mathbf{h}}_{\rm elec}^{\rm nsd}{s}^{(0 ) } \}|\phi_v \rangle = \delta e_v \langle \phi^{pq}_{vb } | \pst|\phi_v \rangle . \label{ccsptrb1v2}\end{aligned}\ ] ] [ prcc_eqn ] where , we have used the relations @xmath147 as @xmath56 , being the cluster operator of closed - shell sector , does not contribute to the cc equation of @xmath134 and @xmath135 . in terms of specific components , as in eq . ( [ lin_s ] ) , the single excitation perturbed cluster amplitude equation is @xmath148 this is eq . ( [ lin_s ] ) with two important modifications . first , the valence orbital @xmath149 replaces the orbital @xmath150 and second , the single particle energy of @xmath149 is modified to include the correlation energy . following similar modifications , the equation of the double excitation cluster amplitudes is @xmath151 angular factors obtained from the closed - shell sector , with replacement of @xmath150 by @xmath149 , can be used to rewrite the cluster equations in terms specific multipole components from these equations . with these modifications , the same methods used to solve the @xmath63 are used to calculate the @xmath152 cluster amplitudes . the fore going derivations shows that prcc equations of one - valence sector are not very different from the closed - shell equations . but the same rationale does not apply to two - valence prcc equations . for two - valence systems , even at the level of rcc , require due considerations on the nature of model space . properties calculations from the rcc wave functions is another level of sophistication , these and related issues on rcc calculations of two - valence systems is explored in one of our recent works @xcite . in this section , as prelude to the calculation of @xmath153 from the prcc states , the details of @xmath154 calculation with mbpt is discussed at the beginning . the mbpt calculation , however , is in the hyperfine states and is equivalent to the prcc expressions of @xmath153 after coupling with the nuclear states . to calculate the @xmath155 induced electric dipole transition amplitude @xmath0 with mbpt , we use the wavefunction in eq . ( [ pert_state_mbpt ] ) . for the purpose of using mbpt calculations as the basics of analyzing the prcc , it is sufficient to consider the total wave operator @xmath156 the pnc induced @xmath52 transition amplitude is then @xmath157^\dagger\mathbf{d } \nonumber \\ & & \left [ 1 + \omega^{(1)}_{v,0 } + \lambda \omega^{(0)}_{v,1 } \right ] \red\phi_v \rangle.\end{aligned}\ ] ] the expressions arising from these terms contain all the symmetry information to describe the properties of @xmath63 . consider terms linear in @xmath46 , @xmath158 this is the mbpt expression of @xmath154 , which has one order each of @xmath42 and residual coulomb interactions . the diagrams which arises from @xmath159 and @xmath160 are shown in fig . [ e1pnc_mbpt ] . a similar set of diagrams , which arise from @xmath161 and @xmath162 , are obtained from interchanging @xmath163 and @xmath54 vertices in the diagrams . besides providing insights on the nature of perturbed cluster operators , detailed mbpt calculations serve another valuable purpose . these serve as reference calculations to validate the prcc implementations . like in mbpt , from the expression of the prcc wave function in eq . ( [ psiptrb1v ] ) , the @xmath155 induced @xmath52 transition amplitude is @xmath164^\dagger \left [ 1 + s + \lambda \ps\cdot\mathbf{i } \right]^\dagger \mathbf{d } \nonumber \\ & & \times e^t \left [ 1 + \lambda \pt\cdot\mathbf{i } \right ] \left [ 1 + s + \lambda \ps\cdot\mathbf{i}\right ] \red \phi_v \rangle.\end{aligned}\ ] ] retain terms linear in @xmath46 as the remaining are zero from parity selection rule . considering only the electronic component , like in the calculation of the cluster amplitudes @xmath56 and @xmath62 , define @xmath165 as the @xmath166 induced @xmath52 amplitude in the electronic space . from the commutation relation of the cluster - operators , the expression is reduced to @xmath167 + \left [ \pt + \ps \right . \nonumber \\ & & \left . + \pt s \right]^\dagger \bar{\mathbf{d } } + s^\dagger\bar{\mathbf{d } } \left [ \pt + \ps + \pt s \right ] \nonumber \\ & & + \left [ \pt + \ps + \pt s \right]^\dagger \bar{\mathbf{d } } s \red \phi_v \rangle,\end{aligned}\ ] ] where @xmath168 is the dressed electric dipole operator . it is evident that @xmath169 is a non - terminating series of the closed - shell cluster operators . it is non - trivial to incorporate @xmath17 to all orders in numerical computations . for this reason @xmath169 approximated as @xmath170 this captures all the important contributions arising from the core - polarization and pair - correlation effects . terms not included in this approximation are terms which are third and higher order in @xmath30 . the expression used in our calculations is then @xmath171 the eq . ( [ e1pnccc1v ] ) , as mentioned earlier , includes term up to second order in cluster amplitudes . from our previous study of properties calculations @xcite , we conclude that the contributions from the higher order are negligible . selected diagrams from the leading - order and next to leading order terms of eq . ( [ e1pnccc1v ] ) are shown in the fig . [ e1pnc_cc ] . the prcc diagrams of @xmath153 in fig . [ e1pnc_cc ] are , as explained earlier , defined in the electronic subspace . to calculate @xmath172 , the nuclear spin part of the operator is coupled with @xmath153 . the @xmath0 diagrams must then include a vertex which operates on the nuclear spin states . this is evident from the example mbpt diagram in fig . [ pcc_mbpt]@xmath173 . although complicated , the coupling with the nuclear part of the operator require only angular integration , where as the evaluation in the electronic space involves radial integrals . in our calculations , we use diagrammatic to carry out the angular integrals and as mentioned earlier , we follow the conventions of lindgren and morrison @xcite . to elaborate on the coupling schemes , we discuss specific cases involving single and double prcc operators . for the former , we select the topologically simplest , and for the later , one of the more complicated one . so that , these represent the range of topology and complexity of the diagrams . in each of these the main objective is to reduce the electronic part to a form common to a set of diagrams . and then , combine with the nuclear part . to define the coupling of singles diagrams , either @xmath174 or @xmath134 , with nuclear spin part consider the diagram in fig . [ e1pnc_cc](a ) . it represents the dirac - fock contribution and is perhaps the simplest diagram . but it is the most dominant and naturally the most important . it is one of the diagrams arising from @xmath175 in the prcc calculations . the angular diagram for the electronic part of this diagram has same topology as in fig . [ e1pnc_cc](a ) . however , the orbitals lines are replaced by angular momentum lines and arrows on appropriate lines to represent phase factors . the angular momentum diagram , which includes coupling with the nuclear spin , is shown in fig . [ e1pnc_d5 ] . the electronic part is the portion enclosed with the dashed line rectangle . in the diagram , @xmath176 and @xmath177 are the total angular momentum of the single particle states of the initial and final one - valance atomic states , respectively . the line @xmath178 represents a rank one multipole line and denotes the angular part of the dipole operator . the multipole line @xmath179 is the all important representation of @xmath42 , it is rank one multipole in electronic part as well as the nuclear part . in the electronic part , it represents the angular part of operator @xmath180 and produces a transition from @xmath176 to @xmath181 . here , @xmath181 is the total angular momentum of the intermediate single particle state . on the other hand , in the nuclear sector , @xmath179 represents the operator @xmath59 . since it is diagonal operator of the nuclear spin state @xmath182 , the @xmath179 line does not change the nuclear spin . the last step in the angular momentum diagram representation is to coupled the electron and nuclear momenta to form the total angular momentum of the hyperfine states . the initial and final hyperfine states are @xmath183 and @xmath184 , respectively . the lines marked as @xmath185 and @xmath186 represent the angular momenta of the hyperfine states in the diagram . using wigner - eckert theorem , we can write algebraic equivalent of the diagram as @xmath187 where @xmath188 is a rank one operator , @xmath189 are the hyperfine magnetic quantum numbers and @xmath190 is the component of @xmath191 . the phase factor and @xmath192-symbol in the above expression are the free lines in the right hand side of the diagrammatic equation in fig . [ e1pnc_d5 ] . remaining expression on the right hand side , phase factor and @xmath193-symbols , is the @xmath194 independent angular component of @xmath195 . for the coupling with nuclear spin involving either @xmath136 or @xmath134 , consider a more complicated diagram as shown in fig . [ e1pnc_d1](a ) . it is the exchange of diagram in fig . [ e1pnc_cc](j ) and arises from the term @xmath196 in the prcc expression of @xmath197 . the angular momentum diagram is shown in fig . [ e1pnc_d1](b ) and note that the angular momentum representation of @xmath135 is as described in section . [ pert_cc_fn ] . the labels of the lines are different from the one in fig . [ pcc_mbpt1 ] , however , these are dummy labels and same selection rules apply to @xmath198 and @xmath199 , and @xmath179 represents a rank one multipole line . it has the same role in the electronic sector as @xmath179 of the diagram in fig . [ e1pnc_d5 ] . to demonstrate the non - trivial angular integration in the calculations with hyperfine atomic states , the angular momentum diagram of @xmath200 , of the diagram in fig . [ e1pnc_d1 ] , is shown in fig . [ e1pnc_d2 ] . the portion of the diagram within the rectangle in dashed - line is the angular momentum part of the electronic sector and except for the topological rotation of the ( @xmath198 , @xmath199 , @xmath179 ) vertex , it is identical to the diagram in fig . [ e1pnc_d1](b ) . the evaluation of the angular integral of the electronic sector , for the example considered , is shown in fig . [ e1pnc_d3 ] . important point to be observed is the structure of the free part in the right hand side of the diagrammatic equation in fig . [ e1pnc_d3 ] . although the multipole lines of @xmath201 and @xmath202 , @xmath203 and @xmath204 respectively , are coupled to an effective multipole @xmath205 , the @xmath203 and @xmath204 are present as free lines . the effective multipole @xmath205 operates on @xmath176 and transforms it to @xmath177 . in terms of wigner - eckert theorem , the diagram is equivalent to @xmath206 where @xmath207 represents coupling of rank one tensor operators @xmath208 and @xmath62 to an operator of rank @xmath205 . this coupling is a structure common to any prcc term of @xmath153 . that is , for any term , the angular integral in the electronic sector is reducible to a form where the free lines is similar to the one on the right hand side of fig . [ e1pnc_d3 ] . from the triangular condition , @xmath209 are the allowed values , however , what values of @xmath205 contribute depends on @xmath176 and @xmath177 . for example , @xmath210 contribute in the pnc @xmath7 transition of atomic cs @xcite , where as only @xmath211 contributes to the proposed pnc @xmath212 transition in ba@xmath4 @xcite . the form of the free lines in the fig . [ e1pnc_d3 ] require one due consideration while combining with the nuclear part . it is the multipole line @xmath179 , inherited from @xmath56 or @xmath213 , which couples with the operator @xmath59 to form a scalar operator . the diagrammatic representation is shown on the left side of fig . [ e1pnc_d4 ] . after evaluation , it reduces to a @xmath214-symbol and free line part . from wigner - eckert theorem , the matrix element in the hyperfine states is @xmath215^{l_4 } \mathbf{i}\right \}^1 |f_vm_v\rangle = ( -1)^{f_w - m_w } \nonumber \\ & & \times \left ( \begin{array}{ccc } f_w & 1 & f_v \\ -m_w & q & m_v \end{array } \right ) \langle f_w\red d_{\rm eff}\red f_v\rangle , \label{deff_dbl}\end{aligned}\ ] ] where @xmath216^{l_4 } \mathbf{i}\right \}^1 , \ ] ] is the effective dipole operator in the hyperfine states . it is of the same form as the effective operator in eq . ( [ deff_sing ] ) . however , the coupling sequence in eq . ( [ deff_dbl ] ) is general and applies to all the terms in the @xmath0 . the rcc method described so far involves intricate but tractable angular momentum coupling and most of the calculations are in the configuration space of reduced dimension , namely of the electrons . considering the complexity of the method , it is desirable to validate the method with few selected terms before a full scale implementation . here , we present a method of validation by comparing with dominant third order mbpt diagrams . this is possible as we solve the prcc equations iteratively with the first order mbpt wave functions as the initial guess . in particular , we evaluate the @xmath217 of the transition @xmath218 in @xmath2cs . to check the prcc equation and angular momentum coupling in the calculations of @xmath219 , consider the term @xmath220 in @xmath221 . two of the third order mbpt diagrams , which are equivalent to the @xmath220 at the first iteration of the prcc equation are shown in fig . [ validation_s ] . in the figure , the portion of the diagrams within the rectangle of dashed - line arises from the second order wave operator @xmath222 define in eq . ( [ omega_11 ] ) . as for the solutions of linearized prcc equations , the initial guess is the first order mbpt wave function @xmath223 . single excitation cluster amplitudes @xmath224 are then calculated iteratively from eq . ( [ lin_s ] ) . at the first iteration of the linearized prcc equation , the diagram within the dashed rectangle in fig . [ validation_s](a ) is equal to the one valence version of the diagram in fig . [ psingles](c ) . and , the diagram fig . [ validation_s](b ) is the exchange counterpart . these mbpt diagrams are the dominant ones after the df and other theories like the prcc must be able to reproduce matching results . .validation of singles prcc amplitudes . the values listed are in units of @xmath225 . numbers in the square bracket represent the power of 10 . [ cols="^,^,^",options="header " , ] the results from the mbpt and prcc calculations are listed in table . [ table_s2 ] . here too , the results from the two calculations are in excellent agreement . perhaps , it must be mentioned that , the angular momentum factor calculations of the mbpt and prcc diagrams are done in very different steps . electronic part of the portion within the rectangle in dashed - line in fig . [ validation_d ] are evaluated to reduce to the representation of @xmath226 as discussed in section . [ pert_cl_rep ] . mbpt diagrams , on the other hand , may be evaluated without the need to associate with an effective operator of specific form . the relativistic coupled - cluster method we have developed to incorporate a nuclear spin - dependent perturbation is an apt one to calculate nuclear spin - dependent parity non - conservation in atoms and ions . the representation of the cluster amplitude within the electronic sector as a tensor operator of rank one and coupling with the nuclear spin part at a later stage of property calculation leads to simplification of the calculations . otherwise , the entire calculation must be done with hyperfine states , which involves complicated angular momentum couplings at all stages of the calculations . the proposed scheme , on the other hand , introduces the nuclear spin coupling is with a simplified effective operator in the electronic sector . the validity of the representation is explicitly tested and verified with selected diagrams . although limited in number , the diagrams and terms selected in the sample test calculations are varied enough to account for complex as well as subtle issues related to the method . based on the results presented , we conclude that the method works and in future publications we shall report results of sophisticated and large scale calculations using prcc . we wish to thank s. chattopadhyay , s. gautam , k. v. p. latha , b. sahoo and s. a. silotri for useful discussions . the results presented in the paper are based on computations using the hpc cluster at physical research laboratory , ahmedabad . here , we give the linearized prcc equations of the closed - shell sector , eqs . ( [ lin_s ] ) and ( [ lin_d ] ) , after angular integration . the one - valence cluster amplitude equations can be obtained after suitable modifications . the angular reduction of each diagram is such that the free parts are reduced to the form in fig . [ pert_cc_op](a ) . the free part is common to all the diagrams and is avoided in the computational implementation . so , solutions of the prcc equations are the cluster amplitudes in reduced matrix form . for @xmath227 and @xmath228 , the rank of the operator , and hence the free part , is one . for convenience , the representation of @xmath229 is redefined as @xmath230 like the prcc operators , we use @xmath231 and @xmath232 to represent the unperturbed single and double excitation amplitudes , respectively . the tensor operator structure of @xmath233 is @xmath234 for the slater integrals , the reduced matrix element is @xmath235 where , @xmath236 is the radial part of the slater integral . with these definitions , the eq . ( [ lin_s ] ) is written in terms of reduced matrix elements and appropriate angular factors . @xmath237 + \sum_{bqk_2 } \bm{\tau}^q_b\bigg [ \frac{\delta{(k_1 , k_2 } ) } { \sqrt{(2 k_1 + 1 ) } } ( -1)^{j_q - j_b + k_1 } x_{k_2}(bpqa ) \nonumber \\ & & - ( -1)^{j_b + j_q + k_1 } \left \ { \begin{array}{ccc } j_a & j_b & k_2 \\ j_q & j_p & k_1 \end{array}\right\ } x_{k_2}(bpaq ) \bigg ] + \sum_{bqrk_2 } \sum_{l_1 l_2}\bm{\tau}^{qr}_{ba}(l_1,l_2 ) \bigg [ \frac{\delta{(k_2 , l_2})}{\sqrt{(2 k_2 + 1 ) } } ( -1)^{j_q - j_b + j_a + j_p + l_1 } \nonumber \\ & & \times\;\left \ { \begin{array}{ccc } j_r & j_a & l_1 \\ k_1 & k_2 & j_p \end{array}\right\ } x_{k_2}(bpqr ) - ( -1)^{j_a + j_p + j_b + j_q + l_1 } \left \ { \begin{array}{ccc } j_b & k_2 & j_r \\ j_p & l_2 & j_q \end{array}\right\ } \left \ { \begin{array}{ccc } j_r & j_a & l_1 \\ k_1 & l_2 & j_p \end{array}\right\ } x_{k_2}(bprq ) \bigg ] \nonumber \\ & & - \sum_{bcqk_2 } \sum_{l_1 l_2 } x_{k_2}(bcqa)\bigg [ \frac{\delta{(k_2 , l_2 } ) } { \sqrt{(2 k_2 + 1 ) } } ( -1)^{j_a - j_b + j_p + j_q + k_1 + k_2 } \left \ { \begin{array}{ccc } j_p & l_1 & j_c \\ k_2 & j_a & k_1 \end{array}\right\ } \bm{\tau}^{qp}_{bc}(l_1,l_2 ) \nonumber \\ & & - ( -1)^{j_q + j_c + j_a + j_p + k_1 + l_2 } \left \ { \begin{array}{ccc } k_2 & j_a & j_c \\ l_2 & j_q & j_b \end{array}\right\ } \left \ { \begin{array}{ccc } j_b & j_p & l_1 \\ k_1 & l_2 & j_a \end{array}\right\ } \bm{\tau}^{qp}_{cb}(l_1,l_2 ) \bigg ] .\end{aligned}\ ] ] for the double excitation prcc operator @xmath238 , the common free part of the angular factors in eq . ( [ lin_d ] ) is as shown in fig . [ pert_cc_op](b ) . in the angular factors , the multipoles @xmath178 , @xmath179 , @xmath198 and @xmath199 have the same interpretations as in the singles . multipoles @xmath239 and @xmath205 are , however , arise from coupling @xmath240 and @xmath241 , respectively . the double excitation cluster equation obtained from the projection on @xmath242 along with appropriate angular factors is @xmath243 + \bigg [ \begin{array}{c } p\leftrightarrow q \\ a\leftrightarrow b \end{array } \bigg ] \nonumber \\ & & + \sum_{rsk_2}\sum_{l_3 l_4}(2 l_1 + 1)(2 l_2 + 1 ) ( -1)^{j_a + j_p + j_b + j_q + k_1 + k_2 + l_4 + l_2 } \left \ { \begin{array}{ccc } k_2 & j_p & j_r \\ j_a & l_3 & l_1 \end{array}\right\ } \left \ { \begin{array}{ccc } l_1 & l_3 & k_2 \\ l_4 & l_2 & k_1 \end{array}\right\ } \left \ { \begin{array}{ccc } k_2 & j_s & j_q \\ j_b & l_2 & l_4 \end{array}\right\ } \nonumber \\ & & \times x_{k_2}(pqrs ) \bm{\tau}^{rs}_{ab}(l_3,l_4 ) + \sum_{cdk_2 } \sum_{l_3 l_4}(2 l_1 + 1)(2 l_2 + 1 ) ( -1)^{j_a + j_p + j_b + j_q + k_1 + k_2 + l_3 + l_1 } \left \ { \begin{array}{ccc } k_2 & j_a & j_c \\ j_p & l_3 & l_1 \end{array}\right\ } \nonumber \\ & & \times \left \ { \begin{array}{ccc } l_2 & l_1 & k_1 \\ l_3 & l_4 & k_2 \end{array}\right\ } \left \ { \begin{array}{ccc } k_2 & j_d & j_b \\ j_q & l_2 & l_4 \end{array}\right\ } x_{k_2}(cdab ) \bm{\tau}^{pq}_{cd}(l_3,l_4).\end{aligned}\ ] ] 36 y. zeldovich , jetp * 6 * , 1184 ( 1958 ) . c. s. wood , et al . science * 275 * , 1759 ( 1997 ) . n. fortson , phys . lett . * 70 * , 2383 ( 1993 ) . k. tsigutkin , d. dounas - frazer , a. family , j. e. stalnaker , v. v. yashchuk , and d. budker , phys . lett . * 103 * , 071601 ( 2009 ) ; phys . a * 81 * , 032114 ( 2010 ) . b. k. sahoo , p. mandal and m. mukherjee , phys . a * 83 * , 030502 ( 2011 ) . v. a. dzuba and v. v. flambaum , phys . a * 83 * , 042514 ( 2011 ) . v. a. dzuba , v. v. flambaum , and m. g. kozlov , phys . a * 54 * , 3948 ( 1996 ) . s. g. porsev , m. g. kozlov , and yu . g. rakhlina , hyperfine interact . * 127 * , 395 ( 2000 ) . v. a. dzuba , v. v. flambaum , arxiv:1104.0086 . k. p. geetha , a. d. singh , b. p. das , and c. s. unnikrishnan , phys . a * 58 * , r16 ( 1998 ) . a. d. singh and b. p. das , j. phys . b * 32 * , 4905 ( 1999 ) . b. k. mani and d. angom , arxiv:1104.3473v1 . f. coester , nucl . phys . * 7 * , 421 ( 1958 ) . f. coester and h. kmmel , nucl * 17 * , 477 ( 1960 ) . r. pal , m. s. safronova , w. r. johnson , a. derevianko , and s. g. porsev , phys . rev . a * 75 * , 042515 ( 2007 ) . b. k. sahoo , l. w. wansbeek , k. jungmann , and r. g. e. timmermans , phys . a * 79 * , 052512 ( 2009 ) . h. s. nataraj , b. k. sahoo , b. p. das , and d. mukherjee , phys . * 101 * , 033002 ( 2008 ) . k. v. p. latha , d. angom , b. p. das , and d. mukherjee , phys . lett . * 103 * , 083001 ( 2009 ) . c. thierfelder and p. schwerdtfeger , phys . a * 79 * , 032512 ( 2009 ) . b. k. sahoo , b. p. das , and d. mukherjee , phys . a * 79 * , 052511 ( 2009 ) . l. w. wansbeek , b. k. sahoo , r. g. e. timmermans , k. jungmann , b. p. das , and d. mukherjee , phys . a * 78 * , 050501(r ) ( 2008 ) . r. pal , d. jiang , m. s. safronova , and u. i. safronova , phys . rev . a * 79 * , 062505 ( 2009 ) . s. g. porsev , k. beloy , and a. derevianko , phys . d * 82 * , 036008 ( 2010 ) . b. k. mani , k. v. p. latha , and d. angom , phys . a * 80 * , 062505 ( 2009 ) . b. k. mani and d. angom , phys . a * 81 * , 042514 ( 2010 ) . b. k. sahoo and b. p. das , phys . a * 77 * , 062516 ( 2008 ) . k. v. p. latha , d. angom , r. k. chaudhari , b. p. das and d. mukherjee , j. phys . b. * 41 * , 035005 ( 2008 ) . i. lindgren and j. morrison , _ atomic many - body theory _ , edited by g. ecker , p. lambropoulos , and h. walther ( springer - verlag , 1985 ) . s. a. blundell , w. r. johnson , z. w. liu and j. sapirstein , phys . a * 39 * , 3768 ( 1989 ) . s. a. blundell , w. r. johnson , z. w. liu and j. sapirstein , phys . a * 40 * , 2233 ( 1989 ) . b. k. mani and d. angom , phys . a * 83 * , 012501 ( 2011 ) .

parity - odd domains , corresponding to non - trivial topological solutions of the qcd vacuum , might be created during relativistic heavy - ion collisions . these domains are predicted to lead to charge separation of quarks along the orbital momentum of the system created in non - central collisions . to study this effect , we investigate a three particle mixed harmonics azimuthal correlator which is a -even observable , but directly sensitive to the charge separation effect . we report measurements of this observable using the star detector in au+au and cu+cu collisions at=200 and 62 gev . the results are presented as a function of collision centrality , particle separation in rapidity , and particle transverse momentum . a signal consistent with several of the theoretical expectations is detected in all four data sets . we compare our results to the predictions of existing event generators , and discuss in detail possible contributions from other effects that are not related to parity violation .

0909.1717

quantum chromodynamics ( qcd ) is widely accepted as the theory of the strong interaction . the perturbative regime , applying to processes with large momentum transfer , is theoretically calculable and has been extensively tested @xcite . on the other hand , the regime in which quarks and gluons interact with modest momenta and with an effective coupling constant that is too large for perturbation theory to apply , can not be reliably calculated by analytic methods . lattice gauge theory is one first principle approach which can be used . it predicts the existence of a new state of strongly - interacting matter at high energy density . this state has now been observed in high - energy heavy - ion collisions at the relativistic heavy ion collider ( rhic ) at brookhaven national laboratory @xcite . many interesting features of this new state of matter produced in these collisions have been observed . in this paper we focus on a new phenomenon which we refer to as local strong parity ( ) violation . it is well known that the strong interaction conserves parity meaning that strong interactions do not lead to reactions which produce a finite expectation value for any parity odd ( changing sign under parity transformation ) observable . the best evidence for this comes from experiments which set limits on the electric dipole moment of the neutron @xcite . these experiments show that the parameter @xmath1 whose nonzero value would describe parity violation in qcd must have magnitude less than @xmath2 . this limit effectively makes direct global parity violation unobservable in heavy - ion reactions . our measurement of a similar -odd observable is consistent with zero at the level of the experimental precision of @xmath3 ( see section [ sec : results ] ) . the concept of local parity violation at high temperatures or in high - energy heavy - ion collisions was discussed by lee and wick @xcite and morley and schmidt @xcite and elaborated by kharzeev et al . @xcite . in a dense highly excited state , gluon fields can produce configurations , local in space and time , which cause , time reversal , and , via the cpt theorem , violating effects . these field configurations form in different ways in different events and averaged over many events they would not yield a finite expectation value for a -odd observable . each space - time region , occupied by such a configuration , is spontaneously produced with a random sign of -violation , which in the theory is determined by the gluonic field topological charge . field configurations with non - zero topological charge have a finite expectation value for @xmath4 , where @xmath5 and @xmath6 are the chromoelectric and chromomagnetic fields , and the average is taken over the region occupied by the configuration . since the space time symmetries of chromodynamic fields are the same as those of electromagnetic fields , with electric field being a vector and magnetic field being a pseudovector , this region is not invariant under ( and ) transformations . quark interactions with such topological gluonic configurations change the quark chirality leading to asymmetry in the number of left and right handed quarks : @xmath7 , where @xmath8 is the number of light quark flavors and @xmath9 is the topological charge of the gluonic configuration . thus , the gluonic field configurations with non - zero topological charge induce the local -violating effects . different aspects of an experimental detection of this phenomenon were discussed in @xcite . in non - central collisions such a domain can manifest itself via preferential same charge particle emission along the system angular momentum @xcite ( see fig . 1 ) . opposite charge quarks would tend to be emitted in opposite directions relative to the system angular momentum . this asymmetry in the emission of quarks would be reflected in , for example , an analogous asymmetry between positive and negative pion emission directions . this phenomenon is driven by the large ( electro- ) magnetic field produced in non - central heavy - ion collisions @xcite . peak magnetic field strengths can reach levels of the order of @xmath10 t. the combined effect of this magnetic field ( which tends to align the magnetic moments of the quarks with the field ) and the difference in the number of quarks with positive and negative chiralities ( which is induced by their presence in a `` -violating bubble '' ) gives rise to the `` chiral magnetic effect '' . the same phenomenon can also be described in terms of induction of electric field by the ( quasi ) static magnetic field , which occurs in the presence of these topologically non - trivial vacuum solutions @xcite . the induced electric field is parallel to the magnetic field , and leads to the charge separation in that direction . thus the charge separation can be viewed as a non - zero electric dipole moment of the system ( see fig . [ fig : charge_sep ] ) . , and that of the magnetic field @xmath11 , is indicated by an arrow . ] depending on the sign of the domain s topological charge , positively charged particles will be preferentially emitted either along , or in the direction opposite to , the system orbital angular momentum , with negative particles flowing oppositely to the positive particles . the magnetic field and the angular momentum are normal to the plane containing the trajectories of the two colliding ions . this plane , called the reaction plane , can be found experimentally in each collision by observation of the azimuthal distribution of produced particles in that event . when two heavy ions collide with a finite impact parameter , the probability for particles to be emitted in a given azimuthal direction is often described with a fourier decomposition @xcite : & & 1 + 2 v_1 , ( ) + 2 v_2 , ( 2)+ ... , [ eq : expansionflow ] where @xmath12 is the particle azimuthal direction relative to the reaction plane , as shown in fig . 2 . @xmath13 and @xmath14 are coefficients accounting for the so - called directed and elliptic flow , respectively , and @xmath15 indicates the particle type . they depend on the impact parameter of the colliding nuclei , the particle type ( @xmath16 , @xmath17 , @xmath18 , ... ) , transverse momentum ( @xmath19 ) , and pseudorapidity ( @xmath20 ) of the produced particles . for collisions of identical nuclei , symmetry requires @xmath13 to be an odd function of rapidity and @xmath14 to be an even function of rapidity . measurements ( for a review and references , see @xcite ) have found that , at rhic , @xmath13 is quite small at mid - rapidity ; typically , @xmath21 for @xmath22 . in contrast , @xmath14 is found to be sizable and positive . in au+au collisions at = 200 gev , for unidentified charged hadrons , @xmath14 reaches 0.25 for @xmath23 gev / c , and 0.06 when integrated over all @xmath19 . phenomenologically , the charge separation due to a domain with a given sign of the topological charge can be described by adding -odd sine terms to the fourier decomposition eq . [ eq : expansionflow ] @xcite : & & 1 + 2 v_1 , ( ) + 2 v_2 , ( 2)+ ... + & + & 2 a_1 , ( ) + 2 a_2 , ( 2 ) + ... , [ eq : expansion ] where the @xmath24 parameters describe the -violating effect . equation [ eq : expansion ] describes the azimuthal distribution of particles of a given transverse momentum and rapidity and , like the flow coefficients , @xmath24 coefficients depend on transverse momentum and rapidity of the particles . in addition , they depend also on the rapidity ( position ) of the domain . one expects that only particles close in rapidity to the domain position are affected . according to the theory , the signs of @xmath24 coefficients vary following the fluctuations in the domain s topological charge . if the particle distributions are averaged over many events , then these coefficients will vanish because the distributions are averaged over several domains with different signs of the topological charge . however , the effect of these domains on charged particle correlations will not vanish in this average , as discussed below . in this analysis we consider only the first harmonic coefficient @xmath25 , which is expected to account for most of the effect although higher harmonics determine the exact shape of the distribution . for brevity we will omit the harmonic number , and write @xmath26 . the index @xmath15 takes only two values , @xmath27 and @xmath28 , for positively and negatively charged particles respectively . the effects of local parity violation can not be significantly observed in a single event because of the statistical fluctuations in the large number of particles , which are not affected by the -violating fields . the average of @xmath29 over many events , @xmath30 , must be zero . the observation of the effect is possible only via correlations , e.g. measuring @xmath31 with the average taken over all events in a given event sample . the correlator @xmath31 is , however , a -even quantity , and an experimental measurement of this quantity may contain contributions from effects unrelated to parity violation . the correlator @xmath31 can be in principle evaluated via measuring @xmath32 with the average in the last expression taken over all pairs of particles of a given type from a kinematic region under study and then over all events . the problem is that this form of correlator contains also a large contribution from correlations not related to the reaction plane orientation ( such correlations are not accounted for by eq . [ eq : expansion ] , which is a single particle distribution ) . for example , a pair of particles originating from a single jet will typically be emitted with a small azimuthal separation . these particle pairs will make a positive contribution to @xmath33 , even if the parent jets are emitted isotropically relative to the reaction plane . therefore , we separate = + b_out , [ eq : bout ] where @xmath31 is caused by parity violation and @xmath34 ( defined by this expression ) includes all other correlations projected onto the direction perpendicular to the reaction plane ( `` out of plane '' ) . the effects contributing to @xmath34 may be large and are difficult to estimate reliably . for this reason , a different correlator was proposed @xcite : & = [ eq : obs1 ] + & -= [ eq : cossin ] + & [ + b_in ] - [ + b_out ] , [ eq : v - a ] where , similarly to eq . [ eq : bout ] , @xmath35 is defined via : = + b_in . [ eq : bin ] the correlator eq . [ eq : obs1 ] represents the difference between correlations of the projections of the particle transverse momentum unit vectors onto an axis in the reaction plane and the correlations of the projections onto an axis that is out - of - plane or perpendicular to the reaction plane . the key advantage of using eq . [ eq : v - a ] is that it removes all the correlations among particles @xmath15 and @xmath36 that are not related to the reaction plane orientation @xcite . the contribution given by the term @xmath37 can be neglected because directed flow averages to zero in a rapidity region symmetric with respect to mid - rapidity , as used in this analysis and the contribution due to directed flow fluctuations is very small ( see section [ sec : simulations ] for a quantitative estimate ) . equation [ eq : v - a ] then implies that by using @xmath38 instead of @xmath39 , the background to our measurement of @xmath31 is now not @xmath34 , but @xmath40 $ ] , where @xmath35 is the contribution of the in - plane correlations which are analogous to @xmath34 . only the parts of such correlations that depend on azimuthal orientation with respect to the reaction plane remain as backgrounds . studies of the various physics contributions to @xmath40 $ ] are discussed in detail in section [ sec : simulations ] . based on the current theoretical understanding of the chiral magnetic effect one might expect the following features of the correlator @xmath31 : * _ magnitude _ : the first estimates @xcite predicted a signal of the order of @xmath41 , where @xmath42 is the net topological charge and @xmath43 is the positive pion multiplicity in one unit of rapidity the expected rapidity scale for correlations due to topological domains , see below . more accurate estimates @xcite including the strength of the magnetic field and topological domains production rates , were found to be close to the same number . it corresponds to values of @xmath44 of the order of @xmath45 for mid - central collisions , and to @xmath3 for the correlator @xmath31 . * _ charge combinations _ : if the particles , after leaving the domain experience no medium effects ( re - interaction with other particles in the system ) , one would expect @xmath46 . thus , in the absence of medium effects , one expects @xmath47 . if the process occurs in a dense medium one needs to account for correlation modifications due to particle interaction with the medium @xcite . the effect of these modifications is similar to the modification of the jet - like two - particle correlations which experience strong suppression of the back - to - back correlations : @xmath48 . the effect of strong radial flow can further modify this relation such that the opposite charge correlations can even become positive . * _ centrality dependence : _ under the assumption that the average size of the -violating domain does not change with centrality , the correlator should follow a @xmath49 dependence ( typical for any kind of correlations due to clusters ; @xmath50 is the multiplicity ) multiplied by a factor accounting for the variation of the magnetic field . the latter is difficult to predict reliably at present , other than that it should be zero in perfectly central collisions . thus at large centralities the effect should decrease with centrality somewhat faster than @xmath49 . * _ rapidity dependence : _ the correlated particles are produced in a domain of the order of 1 fm , and it is expected that the correlations should have a width in @xmath51 of the order unity , as is typical for hadronic production from clusters @xcite . * _ transverse momentum dependence : _ local parity violation is non - perturbative in nature and the main contribution to the signal should `` come from particles which have transverse momentum smaller than 1 gev / c '' @xcite . the actual limits might be affected by the radial flow . * _ beam species dependence : _ the effect should be proportional to the square @xmath52 of the nuclear charge , but the atomic number @xmath53 dependence is not well understood . one qualitative prediction is that the suppression of the back - to - back correlations should be smaller in collisions of lighter nuclei . * _ collision energy dependence : _ the effect might be stronger at lower energies , as the time integral of the magnetic field is larger . at the same time , the charge separation effect is expected to depend strongly on deconfinement and chiral symmetry restoration @xcite , and the signal might be greatly suppressed or completely absent at an energy below that at which a quark - gluon plasma can be formed . the main systematic uncertainty in application of the correlator eq . [ eq : obs1 ] to measurements of anisotropies in particle production with respect to the reaction plane , is due to processes when particles @xmath15 and @xmath36 are products of a cluster ( e.g. resonance , jet ) decay , and the cluster itself exhibits elliptic flow @xcite . detailed discussion of this and other effects which could mimic the effect of local strong parity violation in experimental measurements is presented in section [ sec : simulations ] . in this paper , we report our measurements of correlators shown in eqs . [ eq : obs1]-[eq : v - a ] and present systematic studies of the background effects that affect the measurements . section [ sec : setup ] discusses the experimental setup , while section [ sec : method ] discusses the observables and the methods for estimating the reaction plane angle and corrections for finite reaction plane resolution . sections [ sec : detector_effects ] and [ sec : syst ] present the data and a discussion of systematic effects that can affect the measurements . our main results , and how they systematically change with system size , centrality , particle transverse momentum , and separation in rapidity , are presented in section [ sec : results ] . physics backgrounds that can mimic the -violating effect are discussed in section [ sec : simulations ] . the data were collected with the star detector at brookhaven national laboratory during the 2004 and 2005 runs . collisions of au+au and cu+cu beams were recorded at @xmath54 and 62 gev incident energies ; for a total of four beam - energy combinations . charged particle tracks were reconstructed in a cylindrical time projection chamber ( tpc ) @xcite . the tpc is a 4.2 m long barrel with a 2 m radius which was operated in a solenoidal magnetic field of 0.5 t. the tpc detects charged tracks with pseudorapidity @xmath55 and @xmath56 mev / c with an absolute efficiency that ranges from 80 to 90% . the tpc is nearly azimuthally symmetric and records tracks at all azimuthal angles ; however , sector boundaries and other regular detector features are responsible for an approximately 10% loss of particles due to the finite acceptance of the detector . track merging , and other tracking artifacts that depend on track density , can cause an additional 0 - 10% loss of reconstructed tracks ; so the overall efficiency is typically 85% per event . the tpc s pseudorapidity coverage of an event is supplemented by two cylindrical and azimuthally symmetric forward time projection chambers ( ftpc ) . the are placed in the forward and backward direction relative to the main tpc and cover pseudorapidity intervals @xmath57 @xcite . in the most forward direction , star has two zero degree calorimeter - shower maximum detectors ( zdc - smd ) @xcite which are sensitive to the directed flow of neutrons in the beam rapidity regions . a minimum bias trigger was used during data - taking . events with a primary vertex within 30 cm along the beam line from the center of the main tpc were selected for the analysis . standard star software cuts were applied to suppress pile - up and other malformed events or tracks . the results presented here are based on 14.7 m au+au and 13.9 m cu+cu events at the center of mass energy of a nucleon pair = 200 gev , and 2.4 m au+au and 6.3 m cu+cu events at = 62 gev . the data were taken with the magnetic field in the full field ( ff ) , and reverse full field ( rff ) configurations , with the strength of the magnetic field at 0.5 t. the centrality of the collision is determined according to the reference multiplicity ( refmult ) , which is the recorded multiplicity of charged particles in @xmath58 that satisfy specific track quality cuts . the correlations are reported in the pseudorapidity region @xmath59 covered by the main tpc . for this analysis , the tracks in the tpc are required to have @xmath60 gev / c . for the results integrated over transverse momentum we also impose an upper cut of @xmath61 gev / c . standard star track quality cuts are applied : a minimum of 15 tracking points are required for a track to be considered good . the ratio of the number of hit points on a track to the maximum possible given the track geometry is required to be greater than 0.52 to avoid the effects of track splitting . the data with reverse magnetic field were used to assess systematic effects , as the biases for positive and negative charged particles interchange . the final results reported here are averaged over both field polarities . we use particle identification via specific energy loss ( @xmath62 ) in the volume of the tpc to reject electrons as a check that the signal we present is determined by hadron production . in practice the reaction plane angle for a given collision is not known . in order to evaluate the correlator defined in eq . [ eq : obs1 ] , one estimates the reaction plane with the so - called event plane reconstructed from particle azimuthal distributions @xcite . for the event plane determination one can use particles found in the same detector that is used to detect particles @xmath15 and @xmath36 ( in our case star s main tpc ) or different detectors ( we have used the star ftpcs and the zdc - smd ) . the second order event plane ( determined by the second harmonic modulation in particle distribution ) is sufficient for this study . we make use of the large elliptic flow measured at rhic @xcite to determine the event plane from particle distributions in the main and forward tpcs . when using the zdc - smd for event plane reconstruction , the first - order event plane can be determined through the measured directed flow of spectator neutrons . in the three - particle correlation technique , the explicit determination of the event plane is not required ; instead , the role of the event plane is played by the third particle that enters the correlator with double the azimuth @xcite . under the assumption that particle @xmath63 is correlated with particles @xmath15 and @xmath36 only via common correlation to the reaction plane , we have : ( _ a + _ -2_c ) = ( _ a + _ -2 ) v_2,c , [ e3p ] where @xmath64 is the elliptic flow value of the particle @xmath63 . we check this assumption by using particles @xmath63 from different detectors and exhibiting different elliptic flow . we also study the effect of using only positive or only negative particles to determine the event plane and compare the results obtained with different field polarities in our estimates of the systematic uncertainties . all the correlators presented in this paper have been calculated by first averaging over all particles under study in a given event and subsequently averaging the results over all events in a given event sample . the star tpcs have quite uniform azimuthal acceptance . nevertheless , tpc sector boundaries , malfunctioning electronics , etc . , may introduce biases in the analysis , in particular as the acceptance for positive and negative particles is different . to avoid these effects we use a recentering procedure @xcite in which we substitute : @xmath65 and @xmath66 and similarly for the second harmonic . the typical values of @xmath67 and @xmath68 for the tracks in the main tpc are @xmath69 , but for high @xmath19 particles and the most central collisions could go as high as 1.5% . in the ftpc region , the typical correction is of the order of a few percent . the validity of the recentering method can be verified by calculating three - particle cumulants @xcite : & = + & = \ { > + + + & + -2 } , [ eq : cumulant ] where we use notations @xmath70 and @xmath71 . @xmath72 denotes the real part , and double angle brackets denote cumulants . in the case of perfect acceptance , the cumulant @xmath73 coincides with the correlator @xmath74 . as can be seen from eq . [ eq : cumulant ] , to account for the acceptance effect it is sufficient to perform a recentering procedure . all results presented in this paper have been corrected for acceptance effects , where applicable , by this method . the cumulant eq . [ eq : cumulant ] can be calculated directly by correcting the three - particle correlator with the corresponding products of two and single particle averages . we have compared the results obtained by directly calculating cumulants with the results obtained by the recentering method and found them to be consistent . because the detector acceptance varied during the period of data taking , we perform the corresponding correction run - by - run , separately for positive and negative particles , and for each centrality bin . we also account for the acceptance dependence on particle pseudorapidity and transverse momentum . we consider separately the east ( @xmath75 ) and west ( @xmath76 ) ftpcs . we have found that the corrections do not depend significantly on the collision vertex position along the beam line . figure [ fig : no_recenter ] shows the three - particle correlator , eq . [ e3p ] , as a function of reference multiplicity in au+au collisions at = 200 gev for two field polarities before the recentering procedure . all three particles are from main tpc region , @xmath77 . figure [ fig : recenter ] shows results for the same correlator after correction . the correlator has been scaled by the reference multiplicity for clarity at high centralities , where the absolute values of the signal are small . these figures are intended only to illustrate the effect of the recentering ; for that reason and also to have finer binning in centrality , we plot the correlator directly versus reference multiplicity . all other results are presented as a function of the fraction of the total interaction cross section ( which is calculated taking into account the track , event vertex reconstruction , and trigger inefficiencies ) . the acceptance effects are most noticeable for central collisions , where the signal is small ; there is a slight difference in results depending on whether the third ( @xmath63 ) particle is positive or negative and the difference changes sign depending on the polarity of the magnetic field . this difference disappears after the acceptance correction . results for particles @xmath15 and @xmath36 being both positive or both negative are consistent within statistical errors , and later we combine them as same charge correlations . as expected for the case when particles @xmath15 and @xmath36 are correlated to the particle @xmath63 only via common correlation to the reaction plane , the results do not depend on the charge of the particle @xmath63 . as a function of reference multiplicity for different charge combinations , before corrections for acceptance effects . in the legend the signs indicate the charge of particles @xmath15 , @xmath36 , and @xmath63 . the results shown are for au+au collisions at 200 gev obtained in ( a ) the reversed full field , and ( b ) the full field configurations . , scaledwidth=50.0% ] ( a ) as a function of reference multiplicity for different charge combinations , before corrections for acceptance effects . in the legend the signs indicate the charge of particles @xmath15 , @xmath36 , and @xmath63 . the results shown are for au+au collisions at 200 gev obtained in ( a ) the reversed full field , and ( b ) the full field configurations . , scaledwidth=50.0% ] ( b ) after correction for acceptance effects . , scaledwidth=50.0% ] ( a ) after correction for acceptance effects . , scaledwidth=50.0% ] ( b ) the acceptance effects are larger in the average correlation , @xmath78 , than in the correlator @xmath79 , because the latter represents the difference in correlations projected onto the reaction plane and to the direction normal to the reaction plane . as the reaction plane is uniformly distributed in azimuth , many of the possible acceptance effects average out to zero . figure [ fig : uvau200 ] presents the correlator @xmath78 for different charge combinations from the au+au 200 gev data obtained with ff and rff magnetic field settings as a function of collision centrality . in this figure and later in the paper , the centrality is quantified by the fraction of the total interaction cross section , with the centrality bins corresponding to ( ordered from most to least central ) 0 - 5% , 5 - 10% , 10 - 20% , ... , 70 - 80% of the most central collisions . the points are plotted at the middle of the bin , not reflecting possible small biases due to higher weight of events with larger multiplicity within the bin . before acceptance corrections are applied , @xmath80 correlations are slightly different from @xmath81 correlations , with the difference changing sign in different field orientations . after the correction , the results from different field polarities coincide with each other . as a function of centrality for different charge combinations and ff and rff configurations . the data points corresponding to different charge and field configurations are slightly shifted in the horizontal direction with respect to each other for clarity . the error bars are statistical . also shown are model predictions described in section [ sec : simulations ] . ] we have performed several additional checks to ensure that the signal is not due to detector effects . high accelerator luminosity leads to significant charge buildup in the tpc , which leads to distortions in the recorded track positions , affecting the reconstructed momenta . we have compared the results obtained from the 2002 rhic run ( a low luminosity run ) , with results from 2004 - 2005 divided into high and low luminosity events ( selection is based on zdc coincidence rate ) . all three data samples yield the same signal within statistical uncertainties . the acceptance of the detector depends weakly on the position of the event vertex relative to the center of the tpc . we applied the acceptance corrections differentially according to the event vertex position , and explicitly checked the dependence of the signal on the vertex position . no dependence has been found . the main tpc consists of two parts which are separated by a central membrane . a particle track will occasionally cross the central membrane , and be separately reconstructed in each half - barrel of the tpc . these two track parts can be displaced one with respect to the other . in order to check that this effect does not contribute to the signal , we calculated the correlator using only tracks that do not cross the membrane . taking into account the signal dependence on the track separation in pseudorapidity , the observed signal was found to be consistent with the signal obtained without such a requirement . tracks in the tpc are characterized by the distance of closest approach ( @xmath82 ) , the distance between the projection of the track , and the event vertex . particles originating from weak decays ( @xmath83 , @xmath84 , etc . ) can have larger @xmath82s than the direct primary particles we are studying . we compared the results obtained with a cut @xmath85 cm to those of @xmath86 cm , and found only negligible differences with a somewhat larger signal ( of the order of the statistical error ) for tracks with @xmath85 cm . the correlator used in this analysis is the difference between the correlations projected onto the reaction plane and the correlations projected onto the direction normal to the reaction plane . the correlator calculated by projecting onto an axis rotated by @xmath87 relative to the reaction plane should only be non - zero due to detector effects . we have explicitly calculated the correlator in this rotated frame and found it to be zero within statistical error . figure [ fig : au200f](a ) compares the three - particle correlations obtained for different charge combinations , as a function of centrality , when the third particle is selected from the main tpc with when it is selected from the forward tpcs . assuming that the second harmonic of the third particle is correlated with the first harmonic of the first two particles via a common correlation to the reaction plane , the correlator should then be proportional to the elliptic flow of the third particle . on average , the elliptic flow in the ftpc region is significantly smaller than that in the tpc region @xcite , explaining the different magnitudes of the three - particle correlations shown in fig . [ fig : au200f](a ) . figure [ fig : au200f](b ) shows the three - particle correlator after it has been divided by @xmath14 of the third particle according to eq . [ e3p ] . resulting signals are in very good agreement in the two cases . in this and subsequent plots , for the elliptic flow of particle @xmath63 in the main tpc region we use estimates obtained from the correlations of particles in the main tpc region , @xmath88 , with particles in the ftpc , @xmath89 . these estimates are less affected by non - flow effects , compared to elliptic flow derived from two - particle correlations with both particles taken from the main tpc . the shaded band in fig . [ fig : au200f](b ) and the subsequent figures illustrate the systematic change in the results that occur when different estimates of the elliptic flow are used . the upper ( in magnitude ) limit is obtained with flow from four - particle correlations and the lower limit from the two - particle cumulant method . all elliptic flow data have been taken from ref . @xcite , obtained from correlations of particles in this region with those in ftpcs was denoted as @xmath90 ; an estimate from two - particle correlations with both particles in the main tpc as @xmath91 . elliptic flow from four - particle correlations , denoted as @xmath92 , is considered to be least affected by non - flow effects . for a review of flow measurements , see @xcite . ] . four - particle cumulant values are not available for all collision systems and energies studied here . therefore in figs . [ fig : uuv2_200][fig : uuv2_aunpart ] , we plot systematic upper limits obtained with extrapolation of available data assuming that the measurements with ftpc suppress only 50% of the non - flow contribution . the magnitude of the elliptic flow in the ftpc region was estimated from correlations between particles in the east and west ftpcs . section [ sec : syst ] has further details on the systematic uncertainties associated with different @xmath14 estimates . scaling ( see text for details ) . in both panels , the tpc and ftpc points are shifted horizontally relative to one another for clarity purposes . the error bars are statistical . , scaledwidth=50.0% ] ( a ) scaling ( see text for details ) . in both panels , the tpc and ftpc points are shifted horizontally relative to one another for clarity purposes . the error bars are statistical . , scaledwidth=50.0% ] ( b ) results obtained with the event plane reconstructed with zdc - smd are consistent with those shown in fig . [ fig : au200f](b ) , though the statistical errors on zdc - smd results are about 5 times larger because the ( second order ) reaction plane resolution from zdc - smd is worse . figure [ fig : au200f](b ) shows very good agreement between the same charge correlations obtained with the third particle in the tpc and ftpc regions , which supports for such correlations the assumption @xmath93 . the opposite charge correlations are small in magnitude and it is difficult to conclude on validity of the assumption for such correlations based only on results presented in fig . [ fig : au200f](b ) . similarly , in the most peripheral collisions , the statistical errors are large , which also prohibits making a definite conclusion . there is one class of uncertainties , related to the question of factorization of eq . [ e3p ] , which would arise if the events contained a large number of correlated groups of particles such as minijets . even if these `` clusters '' were produced isotropically in azimuth , they might contribute to our observable through correlations between the particles used to determine the reaction plane ( particle @xmath63 in eq . [ e3p ] ) and the particles ( @xmath94 ) used to measure the signal . we consider this effect in detail in section [ sec : simulations ] . as will be shown there , in cu+cu and peripheral au+au collisions this effect could cause opposite charge correlations of the sign and magnitude we observe , but does not produce the same charge correlations . we proceed with discussion of the results assuming @xmath95 but indicate in all plots the hijing @xcite ( default , quenching - off settings ) three - particle correlation results . the latter can be considered as an estimate of the systematic uncertainty from correlations not related to the reaction plane . in future high statistic measurements such uncertainty can be decreased by taking particle @xmath63 from a rapidity region separated from particles @xmath15 and @xmath36 . one dominant systematic uncertainty in the correlator @xmath38 is due to uncertainty in the elliptic flow measurements of the particle used to determine the reaction plane . this contributes a fractional uncertainty , on average of the order of 15% and somewhat larger in most peripheral and most central collisions @xcite . from comparison of the results obtained in different field configurations , and other studies presented in section [ sec : detector_effects ] we conclude that after acceptance corrections are performed , the remaining systematic uncertainties in three - particle correlations due to detector effects are comparable to or smaller than the statistical errors . we have performed an additional study to estimate the size of possible error caused by acceptance effects before and after the recentering correction is applied : we have run simulations in which tracks were generated using realistic single particle distributions but having no correlation except due to elliptic flow . an efficiency loss is introduced similar to that of the star detector as a function of azimuth , transverse momentum , and particle charge . we then study the effect of distorting the efficiency in additional and more extreme ways . in all of these cases , after the recentering correction is applied , the value of @xmath96 is zero for all centralities within the statistical precision of the study , which is about @xmath97 for the most peripheral bin and decreases to less than @xmath98 for the most central bin . this is many times smaller than the measured signal for all centralities in all cases . errors in measuring the magnitude of particle momenta make negligible contributions to the correlator used in this analysis , which uses only measured azimuthal angles . it is therefore robust against many systematic errors which are commonly encountered in the analyses of the star data ( space charge distortion errors leading to momentum biases , etc . ) . theoretical treatments of the correlator defined in eq . [ eq : obs1 ] were developed with charged hadrons in mind . by using cuts ( based on specific energy loss ) to suppress the presence of electrons in our sample , we have verified that this bias is also smaller than the statistical errors . final results presented in this section have been obtained with three - particle correlation using eq . [ e3p ] with all three particles from the pseudorapidity region @xmath77 . figure [ fig : uuv2_200 ] presents the correlator @xmath38 for au+au and cu+cu collisions at = 200 gev . positive - positive and negative - negative correlations are found to be the same within statistical errors , see fig . [ fig : recenter](b ) , and are combined together as same - charge correlations . opposite - charge correlations are relatively smaller than same - charge correlations , in agreement with possible suppression of the back - to - back correlations discussed in the introduction . the correlations in cu+cu collisions , shown by open symbols , appear to be larger than the correlations in au+au for the same centrality of the collision . one reason for this difference may be the difference in number of participants ( or charge multiplicity ) in au+au and cu+cu collisions at the same centrality . the signal is expected to have a @xmath49 dependence , and at the same centrality of the collision the multiplicity is smaller in cu+cu collisions than in au+au . the difference in magnitude between same and opposite charge correlations is considerably smaller in cu+cu than in au+au , qualitatively in agreement with the scenario of stronger suppression of the back - to - back correlations in au+au collisions . in fig . [ fig : uuv2_200 ] and below , error bars indicate statistical uncertainties . the shaded bands show the systematic uncertainty associated with measurements of elliptic flow which are used to rescale the three - particle correlator . in this section we assume the factorization of correlator eq . [ e3p ] . the possible error due to this assumption which may be large for peripheral bins in the opposite charge correlation is denoted by the thick lines in fig . [ fig : uuv2_200 ] and subsequent figures and is explained in section [ sec : simulations ] . other systematic uncertainties have been discussed in section [ sec : syst ] . figure [ fig : uuv2_62 ] shows results for collisions at = 62.4 gev . the signal is similar in magnitude , with slightly more pronounced opposite - charge correlations compared to those at = 200 gev . this is consistent with weaker suppression of opposite - charge correlations in the less dense 62 gev system . in au+au and cu+cu collisions at @xmath54 gev calculated using eq . the error - bars show the statistical errors . the shaded area reflects the uncertainty in the elliptic flow values used in calculations , with lower ( in magnitude ) limit obtained with elliptic flow from two - particle correlations and upper limit from four - particle cumulants . for details , see section [ sec : detector_effects ] . thick solid ( au+au ) and dashed ( cu+cu ) lines represent possible non - reaction - plane dependent contribution from many - particle clusters as estimated by hijing , see section [ sec : backgrounds_sub1 ] . ] in au+au and cu+cu collisions at @xmath99 gev calculated using eq . the error - bars indicate the statistical errors . the shaded area reflects the uncertainty in the elliptic flow values used in calculations . for details , see section [ sec : detector_effects ] . thick solid ( au+au ) and dashed ( cu+cu ) lines represent possible non - reaction - plane dependent contribution from many - particle clusters as estimated by hijing , see section [ sec : backgrounds_sub1 ] . , scaledwidth=50.0% ] . thick solid ( au+au ) and dashed ( cu+cu ) lines represent possible non - reaction - plane dependent contribution from many - particle clusters as estimated by hijing , see section [ sec : backgrounds_sub1 ] . , scaledwidth=50.0% ] ( a ) . thick solid ( au+au ) and dashed ( cu+cu ) lines represent possible non - reaction - plane dependent contribution from many - particle clusters as estimated by hijing , see section [ sec : backgrounds_sub1 ] . , scaledwidth=50.0% ] ( b ) the correlations are weaker in more central collisions compared to more peripheral collisions , which partially can be attributed to dilution of correlations which occurs in the case of particle production from multiple sources . to compensate for this effect and to present a more complete picture of the centrality dependence , we show in fig . [ fig : uuv2_aunpart ] results multiplied by the number of participants . the number of nucleon participants is estimated from a monte - carlo glauber model @xcite . figure [ fig : uuv2_aunpart](a ) presents the results as a function of centrality , and fig . [ fig : uuv2_aunpart](b ) as a function of @xmath100 . smaller correlations in most central collisions are expected in the parity violation picture as the magnetic field weakens . the same and opposite charge correlations clearly exhibit very different behavior . figure [ fig : uuv2_aunpart](a ) demonstrates that the same - charge correlations show similar centrality dependencies , as would be expected if the geometry of the collision is important . the opposite - charge correlations in au+au and cu+cu collisions are found to be close at similar values of @xmath100 in rough qualitative agreement with the picture in which their values are mostly determined by the suppression of back - to - back correlations . figure [ fig : uuv2eta ] shows the dependence of the signal on the difference in pseudorapidities of two particles , @xmath101 , for 30 - 50% and 10 - 30% centralities . the signal has a typical hadronic width of about one unit of pseudorapidity . the dependence on @xmath102 has been calculated for all charged tracks with @xmath103 gev / c . figure [ fig : uuv2pt ] shows the dependence of the signal on the sum of the transverse momentum ( magnitudes ) of the two particles for these same centralities . results presented is this figure have no upper @xmath19 cut . we do not observe the signal concentration in the low @xmath19 region as naively might be expected for -violation effects . for ( a ) centrality 30 - 50% , and ( b ) centrality 10 - 30% . the shaded band indicates uncertainty associated with @xmath14 measurements and has been calculated using two- and four - particle cumulant results as the limits . , scaledwidth=50.0% ] ( a ) for ( a ) centrality 30 - 50% , and ( b ) centrality 10 - 30% . the shaded band indicates uncertainty associated with @xmath14 measurements and has been calculated using two- and four - particle cumulant results as the limits . , scaledwidth=50.0% ] ( b ) for ( a ) centrality 30 - 50% , and ( b ) centrality 10 - 30% . the shaded band has the same meaning as that in fig . [ fig : uuv2eta ] . , scaledwidth=50.0% ] ( a ) for ( a ) centrality 30 - 50% , and ( b ) centrality 10 - 30% . the shaded band has the same meaning as that in fig . [ fig : uuv2eta ] . , scaledwidth=50.0% ] ( b ) for ( a ) centrality 30 - 50% , and ( b ) centrality 10 - 30% . the shaded band has the same meaning as that in fig . [ fig : uuv2eta ] . , scaledwidth=50.0% ] ( a ) for ( a ) centrality 30 - 50% , and ( b ) centrality 10 - 30% . the shaded band has the same meaning as that in fig . [ fig : uuv2eta ] . , scaledwidth=50.0% ] ( b ) figure [ fig : uuv2ptdiff ] shows the dependence of the signal on the difference in the magnitudes of the two particle transverse momenta . we find that the correlation depends very weakly on @xmath104 . this excludes quantum interference ( hbt ) or coulomb effects as possible explanations for the signal . there are no specific theoretical predictions on this dependence for the chiral magnetic effect , though naively one expects that the signal should not extend to large values of @xmath104 . finally , the zdc - smd detector has good first - order reaction - plane resolution . for mid - central collisions the resolution , @xmath105 , is of the order of 0.350.4 . the zdc - smd allows us to test the first - order ( -odd ) effect of the charge separation along the system orbital momentum , which would correspond to @xmath106 . in theory this is possible only if the vacuum @xmath107 . our measurements are consistent with zero , averaged over all centralities @xmath108 and @xmath109 . reaction - plane independent background is caused by three ( or more ) particle clusters which affect the factorization of eq . [ e3p ] . with future high statistics data sets , it will be possible to reduce such backgrounds significantly by determining the reaction plane using particles far remote in rapidity from the signal particles . in order to estimate possible contribution to the three - particle correlator of effects not related to the reaction plane orientation we use the hijing @xcite event generator which is based on the minijet picture of heavy - ion collisions . for all hijing results presented in this paper we use default , quenching - off setting . figure [ fig:200hj ] presents the results for the three - particle correlator , @xmath110 , measured in au+au and cu+cu collisions as a function of centrality together with hijing results for the correlations among three particles from many particle clusters . in this figure the most central points correspond to centrality 05% and the most peripheral to 6070% for au+au collisions and 5060% for cu+cu collisions . the correlator is scaled with number of participants for clarity at large centralities , where the signal is small in magnitude . the correlations are shown as a function of the number of participants because this gives very similar hijing results for au+au and cu+cu collisions , implying a dependence only on the charged particle rapidity density . we have separately checked that hijing results scale as @xmath111 as expected for contributions from many particle clusters . figure [ fig:200hj ] shows that if this minijet picture in hijing is correct , in peripheral collisions the entire opposite charge signal may be dominated by contributions from clusters not related to the reaction plane orientation . the same charge correlations in hijing are significantly smaller in magnitude than in data and have opposite sign . hijing results for three - particle correlations among three particles all of the same charge are consistent with zero in sharp contrast to the data shown in fig . [ fig : recenter ] . we have also studied such reaction - plane independent backgrounds using the event generator urqmd @xcite and found that the predicted contributions to both opposite - charge and same - charge correlations are at least a factor of two lower than those predicted by hijing . . the correlator has been scaled with number of participants and are plotted versus number of participants . in this representation hijing results for au+au and cu+cu collisions coincide in the region of overlap . , scaledwidth=50.0% ] unlike those discussed in sec . [ sec : backgrounds_sub1 ] , reaction - plane dependent physics backgrounds can not be suppressed by better methods of determining the reaction plane . the correlator @xmath79 is a -even observable and can exhibit a non - zero signal for effects not related to -violation . among those are processes in which particles @xmath15 and @xmath36 are products of a cluster ( e.g. resonance , jet , di - jets ) decay , and the cluster itself exhibits elliptic flow @xcite or decays ( fragments ) differently when emitted in - plane compared to out - of - plane . if `` flowing clusters '' are the only contribution to the correlator , we can write : & ( _ + _ -2 ) = + & a_clust ( ( _ + _ -2_clust ) + 2(_clust- ) ) _ clust + & = a_clust ( _ + _ -2_clust ) _ clust v_2,clust , [ eq : resonance ] where @xmath112 indicates that the average is performed only over pairs consisting of two daughters from the same cluster and the resulting normalization factor is @xmath113 . equation [ eq : resonance ] assumes that there is no reaction plane dependence of @xmath114 . the term @xmath115 is a measure of the azimuthal correlations of decay products with respect to the cluster azimuth , while @xmath116 is cluster elliptic flow . in the case of resonance decays , @xmath117 is zero if the resonance is at rest , and becomes non - zero only due to resonance motion . estimates of the contribution of `` flowing resonances '' , based on eq . [ eq : resonance ] and reasonable values of resonance abundances and values of elliptic flow , indicate that they should not produce a fake signal . given the relative scarcity of parents decaying to two same - charge daughters , a much smaller magnitude is expected for same charge than opposite charge correlations from this source . kinematic studies demonstrate that it is very difficult for the sign of the correlations observed in the data to be created in the same - charge correlations without postulating a negative value of @xmath14 for the resonances or particles from cluster decays . to study the contribution from resonances in greater detail we have carried out simulations using the mevsim event generator @xcite . mevsim generates particles according to the single particle momentum distributions measured at rhic . the only correlations included are an overall bulk elliptic flow pattern and correlations between daughters of the same resonance decay ( resonances included are @xmath118 , @xmath119 , @xmath120 , @xmath121 , and @xmath122 ) . mevsim simulation results are shown as solid squares in fig . [ fig : auausimulations ] ; the opposite charge correlations are larger than what is seen in the data , while the same charge correlations are far smaller in magnitude and of the wrong sign to match parity violation correlations . we conclude that resonances are not responsible for the observed signal . in addition to contributing to reaction - plane independent background as discussed in section [ sec : backgrounds_sub1 ] , jets are another potential source of reaction - plane dependent background since their properties may vary with respect to the reaction plane . for those jets in a heavy - ion event which include a charged particle of sufficiently high @xmath19 to act as a trigger particle for a jet analysis , we may estimate the contribution using the results of previous star jet studies @xcite . with trigger transverse momentum values that allow such analysis ( @xmath123 gev / c ) the contribution to @xmath38 is roughly two orders of magnitude below the same charge signal shown in fig . [ fig : uuv2_200 ] . to extend the study of jet contributions to lower momentum , we rely on event generators ( in particular , hijing ) calculations . several correlation measurements from rhic @xcite and earlier measurements at isr ( see review @xcite ) indicate that cluster formation plays an important role in multiparticle production at high energies . these clusters , with a size inferred in @xcite to be 2.53 charged particles per cluster , may account for production of a significant fraction of all particles . because we have limited information about the nature of these clusters we do not make an estimate of their contribution to the observed correlations . our studies indicate that in order to fake the same - charge correlations observed in the data , there should be several types of clusters with some of them having negative values of elliptic flow . it is hoped that with a better understanding of the cause and properties ( including charge dependence and @xmath14 ) of such clusters , a clearer statement can be made regarding their contributions . we have also run simulations with several p+p and au+au event generators . with pythia @xcite p+p events we find that the correlations in @xmath124 are significantly smaller than those seen in au+au data when scaled by @xmath49 , and are similar for all charge combinations . we add modulation with respect to the reaction plane by adding @xmath14 through angular correlations or strong ( elliptically modulated ) radial flow . this way we create non - zero values for @xmath38 , albeit with correlations different from the data , being always positive and similar in magnitude for all charge combinations . figure [ fig : auausimulations ] shows results for ( reaction - plane dependent ) physics backgrounds to @xmath38 calculated with 200 gev au+au events from the event generators urqmd @xcite and hijing @xcite . because the modulation of @xmath125 with respect to the reaction plane is smaller in hijing than seen in rhic data , we also run hijing with an added `` afterburner '' which adds elliptic flow using as input @xmath14 values consistent with star measurements at the given centrality . elliptic flow is introduced by the `` shifting '' method @xcite , which preserves other correlations that exist in the model . figure [ fig : auausimulations ] shows that no generator gives qualitative agreement with the data ; the model values of @xmath38 are significantly smaller in magnitude than what is seen in the data , and the correlations calculated in these models tend to be very similar for same and opposite charge correlations . these models do not match the correlations for @xmath124 that are seen in the data either , as shown in fig . [ fig : uvau200 ] . hijing predicts very similar same and opposite charge correlations that are much smaller in magnitude than seen in the data . urqmd overestimates the same charge correlations . it predicts opposite charge correlations that are much smaller in magnitude and opposite in sign from the data . this points to the need for better modeling of two - particle correlations to give quantitatively meaningful comparisons for @xmath38 . in fig . [ fig : auausimulations ] we connect urqmd points by dashed lines to illustrate that the `` reference line '' for strong parity correlations might be not at zero . in this particular case of urqmd , both same and opposite - charge correlations have values below zero . note that the same - charge correlations sit somewhat above the opposite - charge correlations , opposite to the expectation from local parity violation . calculated for 200 gev au+au events with event generators hijing ( with and without an `` elliptic flow afterburner '' ) , urqmd , and mevsim . blue symbols mark opposite - charge correlations , and red are same - charge . solid stars represent the values from the data to facilitate comparison . acceptance cuts of @xmath126 gev / c and @xmath77 were used in all cases . for mevsim , hijing , and urqmd points the true reaction plane from the generated event was used for @xmath127 . thick solid lighter colored lines represent possible non - reaction - plane dependent contribution from many - particle clusters as estimated by hijing and discussed in section [ sec : backgrounds_sub1 ] . corresponding estimates from urqmd are about factor of two smaller . , scaledwidth=50.0% ] directed flow , which on average is zero in a symmetric pseudorapidity interval , can contribute to the correlator @xmath38 via flow fluctuations . this effect is of the opposite sign , see eq . [ eq : v - a ] , and is similar for different charge combinations unlike the signal . if one assumes that the amplitude of the fluctuations is of the same order of magnitude as the maximum directed flow in the pseudorapidity interval under study , then the flow fluctuation contribution is no more than 10@xmath128 for centrality 30 - 60% , significantly smaller than the observed signal . global polarization of hyperons @xcite , the phenomenon of the polarization of secondary produced particles along the direction of the system s angular momentum , may also contribute to the correlator eq . [ eq : obs1 ] via -odd weak decays . this effect could lead to a charge asymmetry with respect to the reaction plane , which is always pointing in the same direction relative to the orientation of the angular momentum . our main analysis based on the reaction plane reconstructed from the elliptic flow does not distinguish the direction of the angular momentum , and is susceptible to this effect . but as we pointed out in section [ sec : results ] , our measurement of charge separation along the system orbital angular momentum is zero based on the first - order reaction plane reconstructed in the zdc - smd . global polarization has also been found to be consistent with zero , @xmath129 @xcite . an analysis using three - particle correlations that are directly sensitive to the -violation effects in heavy - ion collisions has been presented for au+au and cu+cu collisions at @xmath0=200 and 62 gev . the results are reported for different particle charge combinations as a function of collision centrality , particle separation in pseudorapidity , and particle transverse momentum . qualitatively the results agree with the magnitude and gross features of the theoretical predictions for local -violation in heavy - ion collisions , except that the signal persists to higher transverse momenta than expected @xcite . the particular observable used in our analysis is -even and might be sensitive to non - parity - violating effects . so far , with the systematics checks discussed in this paper , we have not identified effects that would explain the observed same - charge correlations . the observed signal can not be described by the background models that we have studied ( hijing , hijing+@xmath14 , urqmd , mevsim ) , which span a broad range of hadronic physics . a number of future experiments and analyses are naturally suggested by these results . one of them is the study of the correlation dependence on the energy of the colliding ions . the charge separation effect is expected to depend strongly on the formation of a quark - gluon plasma @xcite , and the signal might be greatly suppressed or completely absent at an energy below that at which a quark - gluon plasma can be formed . improved theoretical calculations of the expected signal and potential physics backgrounds in high energy heavy ion collisions are essential to understand whether or not the observed signal is due to local strong parity violation , and to further experimental study of this phenomenon . we thank d. kharzeev for discussions of the local strong parity violation phenomenon and its experimental signatures . we thank the rhic operations group and rcf at bnl , and the nersc center at lbnl and the resources provided by the open science grid consortium for their support . this work was supported in part by the offices of np and hep within the u.s . doe office of science , the u.s . nsf , the sloan foundation , the dfg cluster of excellence ` origin and structure of the universe ' , cnrs / in2p3 , ra , rpl , and emn of france , stfc and epsrc of the united kingdom , fapesp of brazil , the russian ministry of sci . and tech . , the nnsfc , cas , most , and moe of china , irp and ga of the czech republic , fom of the netherlands , dae , dst , and csir of the government of india , the polish state committee for scientific research , and the korea sci . & eng . foundation . c. amsler _ et al . _ [ particle data group ] , phys . b * 667 * , 1 ( 2008 ) . i. arsene _ [ brahms collaboration ] , nucl . phys . a * 757 * , 1 ( 2005 ) ; b.b . back _ et al . _ [ phobos collaboration ] , nucl . phys . a * 757 * , 28 ( 2005 ) ; j. adams _ et al . _ [ star collaboration ] , nucl . phys . a * 757 * , 102 ( 2005 ) ; k. adcox _ [ phenix collaboration ] , nucl . phys . a * 757 * , 184 ( 2005 ) . m. pospelov and a. ritz , phys . lett . * 83 * , 2526 ( 1999 ) [ arxiv : hep - ph/9904483 ] . c. a. baker _ et al . _ , lett . * 97 * , 131801 ( 2006 ) [ arxiv : hep - ex/0602020 ] . t. d. lee , phys . d * 8 * , 1226 ( 1973 ) . t. d. lee and g. c. wick , phys . d * 9 * , 2291 ( 1974 ) . d. morley and i. a. schmidt , z. phys . c * 26 * , 627 ( 1985 ) . d. kharzeev , r. d. pisarski and m. h. g. tytgat , phys . lett . * 81 * , 512 ( 1998 ) . d. kharzeev and r.d . pisarski , phys . d * 61 * , 111901 ( 2000 ) . e. v. shuryak and t. schafer , ann . nucl . part . sci . * 47 * , 359 ( 1997 ) . d. diakonov , prog . part . phys . * 51 * , 173 ( 2003 ) . s. a. voloshin , phys . c * 62 * , 044901 ( 2000 ) . l. e. finch , a. chikanian , r. s. longacre , j. sandweiss and j. h. thomas , phys . c * 65 * , 014908 ( 2001 ) . d. kharzeev , phys . lett . b * 633 * , 260 ( 2006 ) . d. kharzeev and a. zhitnitsky , nucl . phys . a * 797 * , 67 ( 2007 ) . d. e. kharzeev , l. d. mclerran and h. j. warringa , nucl . a * 803 * , 227 ( 2008 ) . k. fukushima , d. e. kharzeev and h. j. warringa , phys . d * 78 * , 074033 ( 2008 ) . s. voloshin and y. zhang , z. phys . c * 70 * , 665 ( 1996 ) . s. a. voloshin , a. m. poskanzer and r. snellings , arxiv:0809.2949 [ nucl - ex ] . s. a. voloshin , phys . c * 70 * , 057901 ( 2004 ) . n. borghini , p. m. dinh and j. y. ollitrault , phys . c * 66 * , 014905 ( 2002 ) . l. foa , phys . * 22 * , 1 ( 1975 ) . j. adams _ et al . _ [ star collaboration ] , phys . * 92 * , 062301 ( 2004 ) . n. borghini , p. m. dinh and j. y. ollitrault , phys . c * 64 * , 054901 ( 2001 ) . k. h. ackermann _ et al . _ [ star collaboration ] , nucl . instrum . a * 499 * , 624 ( 2003 ) . m. anderson _ _ , nucl . instrum . a * 499 * , 659 ( 2003 ) . k. h. ackermann _ et al . _ , nucl . instrum . meth . a * 499 * , 713 ( 2003 ) . j. adams _ et al . _ [ star collaboration ] , phys . rev . c * 73 * , 034903 ( 2006 ) . c. adler , h. strobele , a. denisov , e. garcia , m. murray and s. white , nucl . instrum . a * 461 * , 337 ( 2001 ) ; the star zdc - smd has the same structure as the star eemc smd : c. e. allgower _ et al . _ [ star collaboration ] , nucl . instrum . a * 499 * , 740 ( 2003 ) ; star zdc - smd proposal , star note sn-0448 ( 2003 ) . a. m. poskanzer and s. a. voloshin , phys . c * 58 * , 1671 ( 1998 ) . k. h. ackermann _ et al . _ [ star collaboration ] , phys . lett . * 86 * , 402 ( 2001 ) . i. selyuzhenkov and s. voloshin , phys . c * 77 * , 034904 ( 2008 ) . j. adams _ et al . _ [ star collaboration ] , phys . rev . c * 72 * , 014904 ( 2005 ) . s. a. voloshin [ star collaboration ] , j. phys . g * 34 * , s883 ( 2007 ) . m. gyulassy and x .- wang , comput . phys . commun . * 83 * , 307 ( 1994 ) ; x.n . wang and m. gyulassy , phys . d * 44 * , 3501 ( 1991 ) . b. i. abelev _ et al . _ [ star collaboration ] , phys . c * 79 * , 064903 ( 2009 ) . r. l. ray and r. s. longacre , arxiv : nucl - ex/0008009 . j. adams _ et al . _ [ star collaboration ] , phys . rev . lett . * 95 * , 152301 ( 2005 ) . j. adams _ et al . _ [ star collaboration ] , phys . lett . * 91 * , 172302 ( 2003 ) . a. feng [ star collaboration ] , j. phys . g * 35 * , 104082 ( 2008 ) . b. alver _ et al . _ [ phobos collaboration ] , phys . c * 75 * , 054913 ( 2007 ) . m. daugherity [ star collaboration ] , j. phys . g * 35 * , 104090 ( 2008 ) . t. sjostrand , s. mrenna and p. skands , jhep * 0605 * , 026 ( 2006 ) . s. a. bass _ et al . . part . phys . * 41 * , 255 ( 1998 ) . z. t. liang and x. n. wang , phys . lett . * 94 * , 102301 ( 2005 ) . s. a. voloshin , arxiv : nucl - th/0410089 . b. i. abelev _ et al . _ [ star collaboration ] , phys . c * 76 * , 024915 ( 2007 ) .

we present the concentration ()-virial mass ( ) relation of 39 galaxy systems ranging in mass from individual early - type galaxies up to the most massive galaxy clusters , . we selected for analysis the most relaxed systems possessing the highest quality data currently available in the _ chandra _ and _ xmm _ public data archives . a power - law model fitted to the x - ray relation requires at high significance ( ) that decreases with increasing , which is a general feature of cdm models . the median and scatter of the relation produced by the flat , concordance model ( , ) agrees with the x - ray data provided the sample is comprised of the most relaxed , early forming systems , which is consistent with our selection criteria . holding the rest of the cosmological parameters fixed to those in the concordance model the relation requires ( 99% conf . ) , assuming a 10% upward bias in the concentrations for early forming systems . the tilted , low- model suggested by a new wmap analysis is rejected at confidence , but a model with the same tilt and normalization can be reconciled with the x - ray data by increasing the dark energy equation of state parameter to . when imposing the additional constraint of the tight relation between and from studies of cluster abundances , the x - ray relation excludes ( conf . ) both open cdm models and flat cdm models with . this result provides novel evidence for a flat , low- universe with dark energy using observations only in the local ( ) universe . possible systematic errors in the x - ray mass measurements of a magnitude suggested by cdm simulations do not change our conclusions . we discuss other sources of systematic error in the measurements and theoretical predictions that need to be addressed for future precision cosmological studies using the relation .

astro-ph0610135

the current cosmological paradigm of a flat universe with low matter density ( ) consisting of cold dark matter ( cdm ) and dark energy has become established recently through a variety of astronomical observations , especially the cosmic microwave background ( cmb ) , supernova hubble diagram , and statistics of large galaxy surveys ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? these key techniques probe the large - scale , high - redshift universe where the development of cosmic structure may be studied in the linear clustering regime ( @xmath16 ) . although the standard model appears to have withstood every test thus far ( e.g. , * ? ? ? * ) , testing the model on small scales , fully into the non - linear regime ( @xmath17 ) , will further our understanding of galaxy formation and evolution and may also refine the fundamental parameters of the world model itself . the dark matter halos of galaxies and clusters are the largest virialized manifestations of small - scale , non - linear clustering . an important probe of the structure of dark matter halos is the relationship between concentration and virial mass . the concentration parameter is defined as , @xmath18 , where @xmath19 the `` virial radius '' is usually taken to be the radius within which the average density equals @xmath20 , where @xmath21 is the critical density of the universe , and @xmath22 is a number typically between @xmath23 . the quantity @xmath24 is the scale radius of the nfw profile @xcite , but it is replaced by @xmath25 , the radius where the logarithmic density slope equals -2 , for more general profiles @xcite . the virial mass ( @xmath1 ) is the mass enclosed within @xmath19 . n - body simulations of cdm models find that @xmath0 declines slowly with increasing @xmath1 but with substantial intrinsic scatter , independent of @xmath1 ( @xcite , hereafter b01 ; @xcite , hereafter d04 ; @xcite ) . for a given @xmath1 the value of @xmath0 in cdm models varies significantly with changes in cosmological parameters , particularly the normalization , , and @xmath26 , the parameter representing the dark energy equation of state ( @xcite ; d04 ; @xcite ) . the amount of intrinsic scatter in the @xmath3 relation is a robust prediction of cdm models , though preferentially selecting relaxed , early forming systems yields smaller scatter than for the entire halo population @xcite . the expected decrease of @xmath0 with increasing @xmath1 in cdm models has yet to be confirmed by observations . optical studies of groups and clusters ( @xmath27 ) using several different techniques galaxy dynamics @xcite , redshift - space caustics @xcite , and weak gravitational lensing @xcite all are consistent with no variation in @xmath0 . however , the @xmath3 relations obtained by these studies are consistent with the concordance model . x - ray studies of the @xmath3 relation using data from the _ asca_satellite @xcite found , @xmath28 , much steeper than produced in cdm models ( b01 ; d04 ; @xcite ) . however , it has only become possible with data from the _ chandra _ and _ xmm _ satellites to obtain spatially resolved temperature measurements of sufficient quality for reliable x - ray constraints on the mass profiles of elliptical galaxies , galaxy groups , and clusters ( e.g. , see reviews by * ? ? ? * ; * ? ? ? * and references therein ) . indeed , our recent study of the mass profiles of seven early - type galaxies with _ chandra _ indicates @xmath3 values consistent with @xcite . the @xmath3 relations for clusters ( @xmath29 ) obtained using _ @xcite and _ chandra _ @xcite each are consistent with no variation in @xmath0 and with the gentle decline with increasing @xmath1 expected for cdm models . in each case the scatter in @xmath30 about the mean relation is not well constrained and is consistent with the cdm prediction . our present investigation aims to improve significantly the constraints on the @xmath3 relation by analyzing a wider mass range with many more systems than the previous _ chandra _ and _ xmm _ studies . for this purpose we undertook a program to obtain accurate mass constraints on relaxed systems with @xmath31 , representing 24 individual early - type galaxies and galaxy groups / clusters , each of which possesses high - quality _ chandra _ and/or _ xmm _ data @xcite . to these systems we added results for 15 relaxed , massive clusters from @xcite and @xcite . using this combined sample , optimized for x - ray mass measurements , we obtain empirical constraints on the local @xmath3 relation with a simple power - law model and compare this model to the predictions of a suite of cdm models . we note that the primary purpose of this paper is to assemble accurate @xmath3 measurements from _ chandra _ and _ xmm _ observations from galaxy to cluster scales and to test whether @xmath0 decreases with increasing @xmath1 similar to that predicted by cdm models . however , we also make an initial attempt to constrain the parameters of cdm models to examine the efficacy of the @xmath3 relation in this regard and to highlight sources of systematic error both in the models and data that need to be addressed with future studies . the paper is organized as follows . we describe the x - ray @xmath0 and @xmath1 measurements in [ obs ] . our procedure for fitting a power - law model to the @xmath3 data is discussed in [ method ] . we define the cosmological models that are compared to the x - ray data in [ models ] . we present our results in [ results ] and discuss sources of systematic error in [ sys ] . in [ conc ] we give our conclusions . all distance - related quantities , unless stated otherwise , are computed assuming a flat universe with @xmath5 , @xmath32 , and @xmath33 km s@xmath34 mpc@xmath34 with @xmath35 . lcc|ccc|ccc|ccc + & & & & & & @xmath1 & @xmath36 & & & + name & @xmath37 & @xmath22 & @xmath38 & @xmath39 & @xmath40 & ( @xmath41 ) & ( @xmath41 ) & @xmath42 & corr & corr@xmath43 & ref . + ( 1 ) & ( 2 ) & ( 3 ) & ( 4 ) & ( 5 ) & ( 6 ) & ( 7 ) & ( 8) & ( 9 ) & ( 10 ) & ( 11 ) & ( 12 ) + + ngc4125 & 0.0045 & 101.5 & 10.19 & 1.78 & 7.52e-02 & 6.24e+12 & 8.04e+11 & 6.24e-02 & -7.87e-01 & -7.61e-01 & a + ngc720 & 0.0058 & 101.6 & 18.54 & 7.25 & 1.65e-01 & 6.65e+12 & 1.42e+12 & 9.63e-02 & -7.88e-01 & -7.83e-01 & a + ngc6482 & 0.0131 & 102.3 & 18.40 & 5.39 & 1.35e-01 & 7.15e+12 & 1.72e+12 & 8.63e-02 & -8.46e-01 & -8.87e-01 & a + ngc5129 & 0.0230 & 103.1 & 14.95 & 2.46 & 7.71e-02 & 1.54e+13 & 2.18e+12 & 5.96e-02 & -8.99e-01 & -8.94e-01 & b + ngc1407 & 0.0059 & 101.7 & 17.88 & 4.93 & 1.09e-01 & 1.59e+13 & 3.12e+12 & 9.35e-02 & -8.41e-01 & -8.35e-01 & a + ngc533 & 0.0185 & 102.7 & 17.32 & 1.21 & 3.14e-02 & 2.29e+13 & 1.11e+12 & 2.05e-02 & -9.03e-01 & -9.12e-01 & b + ngc2563 & 0.0149 & 102.4 & 10.16 & 3.43 & 1.16e-01 & 2.61e+13 & 5.46e+12 & 9.78e-02 & -9.01e-01 & -9.74e-01 & b + rgh80 & 0.0379 & 104.4 & 10.28 & 1.01 & 4.15e-02 & 2.81e+13 & 1.60e+12 & 2.55e-02 & -9.23e-01 & -9.28e-01 & b + ngc1550 & 0.0124 & 102.2 & 17.16 & 1.12 & 2.80e-02 & 3.12e+13 & 1.49e+12 & 2.08e-02 & -9.53e-01 & -9.58e-01 & b + ngc4472 & 0.0033 & 101.4 & 13.07 & 1.64 & 5.09e-02 & 3.27e+13 & 3.99e+12 & 5.97e-02 & -8.90e-01 & -8.89e-01 & a + ngc4325 & 0.0257 & 103.3 & 11.46 & 1.28 & 5.61e-02 & 3.50e+13 & 8.10e+12 & 8.34e-02 & -9.71e-01 & -9.83e-01 & b + ngc4649 & 0.0037 & 101.5 & 20.80 & 2.45 & 5.17e-02 & 3.50e+13 & 4.62e+12 & 6.29e-02 & -8.88e-01 & -8.63e-01 & a + ngc5044 & 0.0090 & 101.9 & 11.32 & 0.29 & 1.09e-02 & 3.73e+13 & 1.17e+12 & 1.40e-02 & -9.50e-01 & -9.53e-01 & b + ic1860 & 0.0223 & 103.1 & 9.59 & 0.88 & 3.91e-02 & 5.10e+13 & 4.62e+12 & 3.98e-02 & -9.66e-01 & -9.72e-01 & b + mkw4 & 0.0200 & 102.9 & 12.72 & 0.86 & 2.96e-02 & 6.22e+13 & 3.38e+12 & 2.39e-02 & -9.43e-01 & -9.46e-01 & b + ngc4261 & 0.0075 & 101.8 & 3.72 & 0.86 & 1.09e-01 & 6.69e+13 & 1.03e+13 & 6.35e-02 & -8.29e-01 & -7.95e-01 & a + rxj1159.8 + 5531 & 0.0810 & 108.0 & 11.48 & 2.86 & 1.28e-01 & 9.10e+13 & 6.89e+13 & 1.73e-01 & -8.12e-01 & -9.79e-01 & b + a262 & 0.0163 & 102.5 & 8.87 & 0.67 & 3.46e-02 & 1.11e+14 & 1.06e+13 & 3.99e-02 & -9.77e-01 & -9.83e-01 & b + ms0116.3 - 0115 & 0.0452 & 105.0 & 6.57 & 2.21 & 1.41e-01 & 1.28e+14 & 8.04e+13 & 2.14e-01 & -8.35e-01 & -9.72e-01 & b + eso5520200 & 0.0314 & 103.8 & 7.79 & 0.75 & 4.46e-02 & 1.31e+14 & 1.96e+13 & 6.06e-02 & -9.53e-01 & -9.57e-01 & b + mkw9 & 0.0498 & 105.4 & 7.48 & 0.93 & 5.41e-02 & 1.44e+14 & 3.61e+13 & 1.11e-01 & -6.49e-01 & -9.36e-01 & d + awm4 & 0.0317 & 103.9 & 9.27 & 0.84 & 3.72e-02 & 1.62e+14 & 1.82e+13 & 5.10e-02 & -9.18e-01 & -9.33e-01 & b + eso3060170 & 0.0358 & 104.2 & 9.17 & 1.04 & 6.12e-02 & 1.81e+14 & 7.56e+13 & 1.09e-01 & -8.84e-01 & -9.76e-01 & b + a2717 & 0.0490 & 105.3 & 6.45 & 0.36 & 2.46e-02 & 1.83e+14 & 1.20e+13 & 2.84e-02 & -9.79e-01 & -9.83e-01 & b + a1991 & 0.0592 & 106.2 & 8.91 & 0.66 & 3.22e-02 & 1.93e+14 & 2.67e+13 & 6.04e-02 & -1.00e+00 & -1.00e+00 & e + a1983 & 0.0442 & 104.9 & 5.34 & 0.99 & 8.15e-02 & 1.99e+14 & 7.62e+13 & 1.76e-01 & -2.87e-01 & -3.92e-01 & d + a2589 & 0.0414 & 104.7 & 6.36 & 0.32 & 2.33e-02 & 3.24e+14 & 2.37e+13 & 3.00e-02 & -9.56e-01 & -9.66e-01 & c + a2597 & 0.0852 & 108.4 & 8.26 & 0.70 & 3.71e-02 & 3.56e+14 & 3.91e+13 & 4.80e-02 & -7.87e-01 & -1.00e+00 & d + a383 & 0.1883 & 116.8 & 9.61 & 0.89 & 4.03e-02 & 4.70e+14 & 4.76e+13 & 4.41e-02 & -5.12e-01 & -1.00e+00 & e + a133 & 0.0569 & 106.0 & 6.72 & 0.61 & 3.97e-02 & 5.37e+14 & 6.44e+13 & 5.23e-02 & -5.50e-01 & -1.00e+00 & e + a1068 & 0.1375 & 112.7 & 5.44 & 0.38 & 3.07e-02 & 6.96e+14 & 6.01e+13 & 3.76e-02 & -9.42e-01 & -1.00e+00 & d + a907 & 0.1603 & 114.5 & 7.75 & 0.94 & 5.27e-02 & 7.39e+14 & 5.99e+13 & 3.53e-02 & -3.87e-01 & -1.00e+00 & e + a1795 & 0.0622 & 106.5 & 6.80 & 0.38 & 2.44e-02 & 1.02e+15 & 8.78e+13 & 3.75e-02 & -6.48e-01 & -1.00e+00 & e + pks0745 & 0.1028 & 109.9 & 7.32 & 0.57 & 3.40e-02 & 1.20e+15 & 1.44e+14 & 5.24e-02 & -2.64e-01 & -1.00e+00 & d + a478 & 0.0881 & 108.6 & 7.61 & 0.58 & 3.29e-02 & 1.25e+15 & 1.65e+14 & 5.74e-02 & -2.28e-01 & -1.00e+00 & e + a2029 & 0.0779 & 107.8 & 8.47 & 0.44 & 2.26e-02 & 1.27e+15 & 1.18e+14 & 4.02e-02 & -4.19e-01 & -1.00e+00 & e + a1413 & 0.1429 & 113.1 & 6.56 & 0.38 & 2.52e-02 & 1.29e+15 & 1.29e+14 & 4.37e-02 & -4.41e-01 & -1.00e+00 & e + a2204 & 0.1523 & 113.9 & 6.77 & 0.55 & 3.51e-02 & 1.41e+15 & 1.55e+14 & 4.80e-02 & -2.55e-01 & -1.00e+00 & d + a2390 & 0.2302 & 120.0 & 4.13 & 0.32 & 3.41e-02 & 2.13e+15 & 2.14e+14 & 4.38e-02 & -3.13e-01 & -1.00e+00 & e + + the measurements of concentrations and virial masses of 23 early - type galaxies and galaxy groups @xcite form the core of our sample . in these papers we considered only those systems with the highest quality _ chandra _ and/or _ xmm _ data . moreover , to insure that hydrostatic equilibrium is a good approximation , we visually inspected all the early - type galaxies and groups in the public data archives and selected those systems that possessed the most regularly shaped x - ray images devoid of strong asymmetries . to these we added several clusters to populate the high - mass portion of the @xmath3 diagram . firstly , we include our analysis of the radio - quiet ( and very symmetrical ) cluster a2589 @xcite . secondly , we added clusters from the _ xmm_study by @xcite and the _ chandra _ study by @xcite , each of whom also focused on the highest quality observations of the most relaxed systems . for clusters that are common to both studies , we used results from @xcite because of more accurate background subtraction and temperature constraints , the latter resulting from the much smaller point spread function of _ chandra _ ; i.e. , less biased temperature measurements in the presence of strong temperature gradients . for those few ( lowest mass ) systems in the @xcite and @xcite studies that overlap with our work cited above , we use our values . the final sample ( see table [ tab.data ] ) consists of 39 systems spanning the mass range @xmath44 , with @xmath0 and @xmath1 values inferred using the nfw dark matter profile . we follow the convention that defines the virial radius ( ) so that the mean density within equals @xmath20 , where @xmath45 is the critical density of the universe at redshift @xmath37 , and @xmath22 is obtained from the solution to the top - hat spherical collapse model . to evaluate @xmath22 we use the approximation obtained by @xcite . for simplicity , when considering dark energy ( de ) models with @xmath46 , we adopt the value of @xmath22 appropriate for @xmath47 . although technically the virial overdensity changes with @xmath48 @xcite , this choice has no effect on our conclusions . the values of @xmath22 for the concordance cosmology at the appropriate redshift for each system are listed in table [ tab.data ] . for results obtained using @xmath22 values different from those in table [ tab.data ] , we converted @xmath0 and @xmath1 to our adopted @xmath22 using the formula of @xcite , which assumes an nfw mass profile . we converted the standard deviations for @xmath0 and @xmath1 in one of two ways . for those systems taken from our previous work , we have at least 20 monte carlo error simulations . hence , we convert each simulated value of @xmath0 and @xmath1 to the new @xmath22 and then compute the standard deviation of the converted values . this procedure provides a self - consistent conversion of the standard deviations . for systems taken from the literature , we simply convert the lower limit ( e.g. , @xmath49 ) and upper limit ( e.g. , @xmath50 ) each to the new @xmath22 . the new standard deviation is then set to one - half the difference of the converted upper and lower limits . ( by performing this procedure on our own data and comparing to the rigorous method using the error simulations , we find that this procedure is adequate for our present study . ) since we perform our analysis of the @xmath3 relation in log space ( see below ) , we also require the standard deviations of @xmath51 and @xmath52 . as above , for systems analyzed in our previous studies , we compute self - consistently the log errors using monte carlo error simulations . the other systems use the simple prescription noted above , but using the log values . the log errors on @xmath0 and @xmath1 are listed in table [ tab.data ] as , respectively , @xmath53 and @xmath54 . for the systems with monte carlo error simulations we also computed the covariance between @xmath0 and @xmath1 ( and @xmath51 and @xmath55 ) for the different systems ; e.g. , @xmath56 , where the brackets represent the mean quantity , and the sum is over the simulations . in table [ tab.data ] we report these as the correlation coefficient , which is the covariance divided by @xmath57 ( or @xmath58 ) . for the objects from the literature we used the average covariance obtained from the others to evaluate the correlation coefficient , setting the coefficient to -1 if a value @xmath59 was obtained . most of the correlation coefficients have values near -1 indicating that @xmath0 and @xmath1 are anti - correlated . finally , the results listed in table [ tab.data ] for the concentration parameter actually refer to the quantity , @xmath60 , since this quantity is what is fitted in our analysis below . note that @xmath38 is also used when computing the covariance and correlation coefficient . we focus our analysis on a simple power - law representation of the @xmath3 data , @xmath61 where @xmath37 is the redshift , and both @xmath62 and @xmath63 are constants independent of @xmath1 . we set the reference mass to , @xmath64 , which lies close to the midpoint ( in log space ) of the mass range of our sample . cdm models generally predict that @xmath0 decreases with increasing @xmath1 ( e.g. , b01 ; d04 ; @xcite ) . a key goal of our study , therefore , is to determine whether @xmath65 . for the most massive systems ( @xmath66 ) in cdm models d04 find that @xmath62 is much more sensitive to cosmological parameter variations than @xmath63 ( see [ models ] ) . to constrain the parameters of eqn . [ [ eqn.cm ] ] it is necessary to account for the error estimates on both @xmath0 and @xmath1 . consequently , we employ the bces method ( i.e. , bivariate correlated errors with intrinsic scatter ) described by @xcite to estimate @xmath63 , @xmath62 , and the intrinsic scatter about the best relation . we performed the bces fitting using software kindly provided by m. bershadymab / archive / stats / stats.html ] . since bces is a linear regression method we transform eqn.to the form , @xmath67 , where @xmath68 , @xmath69 , and @xmath70 . the parameter @xmath62 is derived from @xmath63 and @xmath71 : @xmath72 . since the fractional error @xmath73 is typically smaller by more than an order of magnitude of the fractional error , @xmath74 ( see table [ tab.data ] ) , we always use @xmath75 as the independent variable . by default we also use the corr@xmath43 values in table [ tab.data ] for the bces method . despite the strong anti - correlation between @xmath0 and @xmath1 , we find that including the corr@xmath43 values in the analysis has an insignificant impact on the estimated @xmath63 and @xmath62 values . we determine the best estimate of @xmath63 and @xmath62 by performing the bces method on @xmath76 bootstrap resamplings of the data . we take the mean of the bootstrap simulations to be the best estimates and construct error ellipses about these best values . the confidence contour spacings computed from the bootstrap simulations correspond quite closely to the @xmath77 values for two parameters assuming normally distributed errors . we estimate the intrinsic scatter on the concentration as , @xmath78 where @xmath79 is the number of data points @xmath80 corresponding to the entries in table [ tab.data ] ; @xmath81 , where @xmath63 and @xmath71 are the best ( mean ) estimates from the bootstrap simulations ; and @xmath82 in table [ tab.data ] . lccccccc + name & & & @xmath83 & @xmath84 & @xmath85 & @xmath86 & -w + + & 0.30 & 0.70 & 0.022 & 0.7 & 0.90 & 1.00 & 1.0 + & 0.24 & 0.76 & 0.022 & 0.73 & 0.76 & 0.96 & 1.0 + decdm & 0.30 & 0.70 & 0.022 & 0.7 & 0.90 & 1.00 & 0.6 + qcdm & 0.30 & 0.70 & 0.022 & 0.7 & 0.82 & 1.00 & @xmath87 + ocdm & 0.30 & 0.00 & 0.022 & 0.7 & 0.90 & 1.00 & @xmath88 + + in table [ tab.models ] we define the cdm - based cosmological models to be compared with the observations . is the standard concordance model , which essentially reflects the combined constraints from the first year of wmap cmb observations , the supernovae hubble diagram , and the large - scale clustering of galaxies ( e.g. , * ? ? ? . effectively updates using the third year of wmap @xcite . each of these models assumes a constant dark energy equation of state with @xmath89 ( i.e. , a cosmological constant ) . in addition , we consider a dark energy model with @xmath90 ( decdm ) and a quintessence model ( qcdm ) with a ratra - peebles potential ( * ? ? ? * ; * ? ? ? * and references therein ) as implemented in d04 . the qcdm model has a nearly constant @xmath91 ( see fig . 1 of d04 ) and a lower . finally , we include an open model ( ocdm ) with the same parameters as except with @xmath92 . the median relation between concentration and virial mass as a function of redshift for cdm halos is described well by the semi - analytic model proposed by b01 , @xmath93 where @xmath94 is the normalization constant , @xmath37 is the redshift of the halo , and @xmath95 is the redshift when the halo `` collapsed '' . this collapse redshift is defined implicitly by the equation , @xmath96 where @xmath97 is the equivalent linear overdensity for spherical collapse at @xmath95 , @xmath98 is the @xmath99 linear rms density fluctuation corresponding to a mass @xmath100 , and @xmath101 is a constant . ( note that eqn.is not a power - law , though over small mass ranges within a factor of 5 - 10 the b01 model can be well approximated by one . ) the constants @xmath101 and @xmath94 must be specified by comparison with numerical simulations . for halos with @xmath102 it is found that @xmath103 and @xmath104 provide the best description of the results from n - body simulations , although @xmath105 and @xmath106 is acceptable ( b01 ; @xcite ) . larger box simulations that focus on higher mass ranges clearly prefer @xmath105 . one of us ( j. bullock ) finds that @xmath105 and @xmath107 matches the @xmath3 results for cdm halos ( @xmath108 ) simulated by @xcite . for simulations of even more massive clusters ( @xmath109 ) , d04 find higher concentrations and that @xmath110 is required for the same @xmath101 . consequently , when comparing cdm models to the full mass range of the x - ray data in table [ tab.data ] we shall consider the @xmath3 relations obtained with the b01 model using @xmath105 with both @xmath107 and @xmath110 . ( in [ fk ] we discuss further the different @xmath101 and @xmath94 values obtained by different investigators . ) lcc + name & @xmath63 & @xmath62 + + & -0.104 & 7.63 + decdm & -0.094 & 9.05 + qcdm & -0.111 & 7.39 + ocdm & -0.091 & 11.47 + @xmath111 & -0.142 & 8.43 + @xmath111 & -0.155 & 6.23 + + in the high - mass `` cluster '' regime d04 find that the @xmath3 relation for cdm models is adequately parameterized by the simple power - law model given by eqn . [ [ eqn.cm ] ] . we list the power - law parameters obtained by d04 in table [ tab.models.cm ] but converted to our definition of virial radius . ( d04 use @xmath112 for @xmath5 . ) that is , initially we compute @xmath113 using @xmath62 and @xmath63 from d04 ( at @xmath99 ) . then we convert the resulting @xmath0 and @xmath1 values to @xmath114 using the approximation of @xcite . these values are used to compute the slope and @xmath62 , which are listed in table [ tab.models.cm ] . the parameter @xmath62 decreases by @xmath115 while @xmath63 remains nearly constant under this transformation . for comparison , in table [ tab.models.cm ] we also list @xmath63 and @xmath62 computed from the tangent line to the @xmath3 profiles located at @xmath116 of the and models obtained using the b01 approach with @xmath105 and @xmath110 . these results are listed as @xmath111 and @xmath111 respectively . ( we note that the virial quantities for the x - ray data in table [ tab.data ] all refer to the cosmology . for consistent comparison , we also use the cosmology when converting all the d04 models mentioned above . ) the concentration is sensitive to the formation time and dynamical state of the halo ( e.g. , @xcite ; b01 ; @xcite ) . for example , @xcite and @xcite find that their `` relaxed '' dark halo samples have concentrations that are systematically larger by @xmath117 compared to the whole population . @xcite found that the halo concentration at fixed mass is set almost exclusively by the `` formation epoch '' of the halo ( the time when the mass accretion rate of the halos is slowed below a critical value ) . indeed , when @xcite focused on halos with no major mergers since @xmath118 they found @xmath119 higher concentrations for that population . therefore , a proper comparison between the theoretical @xmath3 relation with observations requires the observed and simulated halos be selected in a consistent manner . the x - ray data presented in table [ tab.data ] represent objects that were selected to be the most relaxed , x - ray bright , systems . but the theoretical @xmath3 models discussed above ( table [ tab.models.cm ] ) were obtained using all available halos in the n - body simulations . consequently , we consider this source of systematic error in our comparisons of the cdm models with the x - ray data . finally , the magnitude of the intrinsic scatter about the median theoretical @xmath3 relation does not vary over a large class of cdm models and is independent of @xmath1 . the value of @xmath120 ( @xmath121 ) obtained by b01 has been found by several independent investigations ( @xcite ; d04 ; @xcite ) ; note this consistency is observed after accounting for the fact that @xmath122 , rather than , is used in these other studies . early forming , relaxed , halos tend to exhibit smaller scatter @xmath123 @xcite . in [ sys ] we discuss possible sources of systematic error both in the measurements from the x - ray data and the theoretical predictions . possible systematic errors affecting the x - ray mass measurements of our sample of relaxed galaxies and clusters are expected to be small and do not change the conclusions of the analysis presented here ( see beginning of [ sys ] and [ he ] ) . when fitting the power - law relation ( eqn . [ eqn.cm ] ) to the entire sample we obtain @xmath124 and @xmath125 ( quoted errors @xmath126 ) . the estimated intrinsic scatter in @xmath127 is @xmath128 ( eqn . [ eqn.scat ] ) , with a total scatter of 0.12 . in figure we plot the best - fitting model with intrinsic scatter . inspection of figure [ fig.cm ] reveals that the power - law with intrinsic scatter is a good approximate representation of the x - ray results , especially for @xmath129 . for lower masses the x - ray measurements tend to lie above the best - fitting model , though most are consistent with lying within the @xmath126 range covered by the intrinsic scatter . using equation [ 3 ] of @xcite we compute a reduced @xmath130 value of 1.07 ( 37 dof ) for the fit , confirming the visual impression that the power law is a good , but not perfect , representation of the x - ray data . the slope @xmath63 is constrained to be negative at the @xmath4 level , demonstrating at high significance that @xmath0 decreases with increasing @xmath1 , as expected in cdm models . in figure [ fig.cm ] we also plot the theoretical @xmath3 relation for the model for both @xmath107 and @xmath110 as discussed in [ models ] . for @xmath131 the @xmath110 model recommended by d04 for clusters lies very close to , but just below , the empirical power - law fit . allowing for a systematic @xmath15 increase in concentration for relaxed , early forming halos @xcite lifts the @xmath110 model slightly over the power - law , with a marginal improvement in the agreement between the models . a recent theoretical study @xcite finds that the higher mass halos experience less of bias associated with relaxation and formation time . in fact , systematically increasing the concentrations of the @xmath110 model by @xmath132 would provide a closer match to the empirical power - law fit to the x - ray data . for @xmath133 the @xmath110 model must quickly transform into the @xmath107 model as @xmath1 decreases from the cluster to the galaxy / group regime as discussed in [ models ] . but in this mass range the @xmath107 model lies consistently below the power - law model and is similar to the @xmath126 lower limit on the power - law given by the intrinsic scatter . increasing the concentrations of the @xmath107 model by @xmath15 for relaxed halos improves the agreement with the power - law , and has the effect of essentially pushing the @xmath107 model into the @xmath110 model . with this accounting for the systematic bias arising from relaxed , early forming halos the model is an equally good representation of the x - ray data for masses @xmath134 . for the lowest masses , the model still lies below the power - law ( within the @xmath126 intrinsic scatter ) and may represent a real discrepancy . we summarize this qualitative discussion with a quantitative comparison of the model with the data as represented by the results of the empirical power - law fit . ( we do not fit the b01 model to the data for the following reasons . first , it is non - trivial to account for the error bars in both coordinates when fitting a general model ; the akritas & bershady method only applies to linear regression . second , the prediction of the b01 model possesses significant systematic uncertainty in the parameter @xmath94 ( at least for masses @xmath135 ) as discussed above . by showing the b01 model predictions separately for the range of interesting k values , without fitting , we provide a clear demonstration of the importance of reducing this systematic uncertainty with future theoretical studies . when that happens , formal fitting of the b01 model will be investigated . ) in figure [ fig.errors ] we show the error contours estimated for @xmath63 and @xmath62 from the bootstrap simulations ( [ method ] ) . we calculated the slope and concentration of the @xmath3 relation of the model at @xmath136 for both the @xmath107 and @xmath110 cases . ( the values for @xmath110 are listed in table [ tab.models.cm ] . ) these results are plotted in figure [ fig.errors ] as `` 1 '' for @xmath107 and @xmath137 for @xmath110 , the latter representing the `` cluster '' regime . the right panel of figure [ fig.errors ] increases the concentrations by 10% to represent relaxed , early forming systems , and shows the model lies near the @xmath138 confidence contour . this level of agreement applies for masses @xmath129 . as discussed above , the agreement is worse at the lowest masses . however , performing a similar comparison at @xmath139 we find the local slope and normalization lie just within the 95% confidence contour of the power - law model ( both with and without the 10% correction for relaxed systems ) ; i.e. , the disagreement is not very significant even at the low - mass end . if real , the discrepancy may signify very early forming fossil groups in the sample @xcite . the intrinsic scatter of 0.10 in @xmath127 obtained for the power - law fit is smaller than the value of 0.14 for all dark matter halos in cdm simulations but agrees extremely well with the value expected for the most relaxed , early forming systems @xcite . consideration of this result for the scatter and the average @xmath3 relation above , we conclude that the model is consistent with the x - ray data , provided the x - ray sample reflects the most relaxed , early forming systems in the population . ( this corroborates our selection criteria discussed in [ obs ] . ) now we perform the analogous comparison of the model with the x - ray data and associated power - law fit . the @xmath3 relation is displayed in the right panel of figure [ fig.cm ] . the model lies well - below the power - law at all masses , even when allowing for the expected 10% increase in concentration for relaxed halos . in figure [ fig.errors ] we compare the model at @xmath136 with the empirical power - law . even when considering the 10% increase in concentrations for relaxed halos , the model lies on the 99.99% contour . systematic errors associated with the x - ray measurements can not explain this level of disagreement ( see [ sys ] ) . the key parameter responsible for the poor performance of the model with respect to is the low value of ( 0.76 ) . in order to bring the model within the 99% contour in the right panel of figure [ fig.errors ] requires @xmath140 where we have kept the other cosmological parameters fixed to their values in table [ tab.models ] . this limit is conservative since ( 1 ) we use @xmath110 , ( 2 ) we assume a full 10% upward shift for the bias from relaxed , early forming systems , which may be less for massive clusters , and ( 3 ) we have approximated the b01 models as power - laws using their predictions only near @xmath136 . the sensitivity of the concentrations to results from the impact that has on the average halo formation times ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? other cosmological parameters , however , contribute to the large discrepancy of the model . if we remove the tilt of the power spectrum ( i.e. , set @xmath141 ) then we obtain @xmath142 at the 99% confidence level . finally , if we further set @xmath5 and @xmath143 ( so the model is the same as except with variable ) , then the 99% constraint falls to @xmath144 , the value associated with the model . ( we reiterate this limit is conservative as noted above . similarly , we obtain @xmath145 at 99% confidence , where the @xmath107 model is used here to be conservative . ) hence , although the lower value of is the primary cause of the poor performance of the model , the combined action of the power - spectrum tilt with the lower value of exacerbate the discrepancy with the x - ray @xmath3 relation . if the parameters are correct , particularly the low value of , then a fundamental modification of the model is required to increase the concentration values to match the x - ray results . both d04 and @xcite have shown that changing the dark energy equation of state parameter ( @xmath26 ) has the effect of systematically raising ( larger @xmath26 ) or lowering ( smaller @xmath26 ) halo concentrations . as we show below , a model with @xmath10 and @xmath146 can describe the x - ray data in the cluster regime . results of fitting eqn . [ eqn.cm ] in log space only for halos with @xmath29 . displayed are the best - fitting model ( solid blue line ) and the @xmath126 intrinsic scatter ( dotted blue lines ) . also shown is the predicted relation for the model from cosmological simulations . the ( solid red ) line is the power - law fit obtained by d04 while the ( dashed red ) line is the b01 model with @xmath105 and @xmath110 found by d04 to best match the simulated clusters . note the d04 model is converted to our definition of the virial radius ( [ models ] ) . ] as discussed in [ models ] , d04 provide results of fitting eqn.to a suite of cdm models for massive clusters , including an open model and multiple dark energy models . we may compare these theoretical predictions directly to the power - law fit to the x - ray data with the following considerations . firstly , we convert d04 s results to our definition of the virial radius ( see [ models ] ) . secondly , since d04 analyze only massive clusters ( @xmath147 , @xmath35 and converted to @xmath114 ) , we restrict our analysis to high - mass systems as well . in order to allow more precise constraints on the power - law fit we consider clusters down to a somewhat smaller mass limit , @xmath148 . ( as shown below , this choice is justified since there is no obvious trend in the @xmath3 relation over this mass range . ) finally , we need to consider the biases ( @xmath15 higher @xmath0 , smaller scatter in @xmath30 ) for relaxed , early forming systems as done for the whole sample in the previous section . however , the systematic increase in @xmath0 may be less since we are considering the most massive systems @xcite . in figure [ fig.cm.clusters ] we show the result of fitting eqn.to the high - mass sample , for which we obtain the following parameter constraints , @xmath149 and @xmath150 . these parameters are consistent with a constant @xmath3 relation , as well as profiles that both decrease and increase with increasing @xmath1 , and agree well with similar fits obtained in this mass range from previous x - ray studies with either _ xmm _ @xcite or _ chandra _ @xcite . since the power - law fit is quite consistent with @xmath151 , it follows that it is the lower mass systems ( @xmath135 ) which require @xmath152 found when fitting the whole sample in the previous section ( see * ? ? ? ( note that the power - law parameters inferred in the previous section for the full sample lies on the 99% confidence contour obtained for the fit to the high - mass subsample see figure [ fig.errors.clusters ] . ) we measure an intrinsic scatter in @xmath127 of @xmath153 which is somewhat less than the value of 0.10 obtained for the entire sample . d04 also obtain 0.10 for the intrinsic scatter ( after accounting for a factor of 2.3 converting @xmath122 to ) for all of their cdm clusters . the smaller scatter we measure presumably can be explained because the x - ray sample preferentially contains relaxed , early forming systems . for comparison we plot two versions of the model in figure [ fig.cm.clusters ] . the first version we show is the power - law fit obtained by d04 converted to our definition of the virial radius ( see [ models ] ) . the second version arises from using the b01 model with @xmath110 , which d04 found to best represent their high - mass cluster simulations . both models are very similar and provide a reasonable description of the x - ray data , though the b01 representation is slightly steeper . in the left panel of figure [ fig.errors.clusters ] we plot the power - laws obtained by d04 for the cdm models as well as the b01 representation of the model with @xmath110 . the values of @xmath63 and @xmath62 for all of these models ( table [ tab.models.cm ] ) are plotted in the right panel of figure [ fig.errors.clusters ] along with the error contours derived from the power - law fit to the x - ray data . as done in the previous section , the @xmath63 and @xmath62 values for the b01 model were obtained using the @xmath3 slope near @xmath154 . we emphasize that the d04 parameters may be compared directly to the results of our power - law fit to the x - ray data , since the same model is fitted over nearly the same mass range . since the b01 representation of the model is very nearly a power - law over the range being investigated , a similarly direct comparison is appropriate . the results for the and models obtained using only the high - mass subsample corroborate those obtained from the full sample . as noted above , the model is an acceptable match to the x - ray data ; the parameters lie near the 68% confidence contour in figure [ fig.errors.clusters ] , even if @xmath0 is increased to correct for early forming halos . in contrast , the model is very inconsistent with the high mass data , with parameters lying outside the 99.99% contour . even if we apply the full 10% correction to @xmath62 appropriate for the most relaxed , early forming systems , then the model merely moves on top of the 99.99% contour . however , as noted above , it is expected that this correction is less for the highest mass systems . hence , the model is also rejected using only the most massive clusters . the @xmath3 data for high - mass clusters clearly exclude the ocdm model . the concentrations lie systematically above the data and the model . the latter is expected because structures form earlier in the ocdm model . in principle , the ocdm model may be brought into acceptable agreement with the x - ray data by lowering . we estimate the effect of lowering using the b01 model . for example , lowering @xmath8 from 0.90 to 0.70 lowers @xmath62 by 25% . this would move the ocdm model to the right in figure [ fig.errors.clusters ] on to the 99% confidence contour . considering any of the 10% systematic bias expected for early forming systems would increase the discrepancy . ( note that for @xmath155 , appropriate for the model , the ocdm model would be rejected at the @xmath156 confidence level without additional consideration of early formation bias . ) studies of the abundances of galaxy clusters using weak gravitational lensing typically find consistent with 0.9 ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? the optical study of @xcite uses the caustic technique to measure cluster abundances and also finds @xmath157 with a lower limit of 0.72 ( 95% conf . ) . however , x - ray studies of cluster abundances show a large variation in , with values ranging from 0.7 - 0.9 ( see * ? ? ? * and references therein ) . the x - ray studies use a simple conversion between mass and global temperature or mass and global x - ray luminosity ( @xmath158 ) . a recent study by @xcite argues that considering reasonable errors in the @xmath159 conversion allows agreement of cluster abundances with the recent wmap parameters with @xmath160 and @xmath161 though see @xcite . we take @xmath162 to be a conservative lower limit established by the abundances of galaxy clusters . therefore , an open cdm model with @xmath163 is ruled out ( @xmath11 confidence ) from joint consideration of only the x - ray @xmath3 relation and cluster abundances . since @xmath8 and @xmath13 are tightly coupled from cluster abundance studies , and all the studies cited above give @xmath164 for @xmath12 , using the x - ray @xmath3 data we find that we can also exclude cdm models with @xmath12 at high significance ( @xmath165 confidence ) . for this comparison we have employed the entire sample of 39 systems and have set @xmath166 for the @xmath12 models to better satisfy cosmic age constraints from stellar populations in globular clusters @xcite . consequently , the combination of the x - ray @xmath3 relation and cluster abundances , assisted by constraints on the cosmic age , provides novel evidence for a flat , low-@xmath13 universe with dark energy using observations only in the local ( @xmath14 ) universe . finally , we consider the alternative dark energy models , decdm and qcdm . the decdm model lies systematically above the model and deviates from the power - law fit to the x - ray data at the 99.9% confidence level . this behavior is expected since @xmath90 implies halos form earlier compared to the @xmath167 model ( see also * ? ? ? * ) . by lowering it should be possible to bring the decdm model into good agreement with the data . to illustrate this trade - off between @xmath26 and , we consider the qcdm model which has @xmath10 and @xmath168 . as shown in figure [ fig.errors.clusters ] the qcdm model matches the x - ray results nearly as well as the model . in fact , a mild increase in @xmath62 owing to effect from selecting relaxed , early forming halos would produce even better agreement . our analysis indicates that a model similar to qcdm is able to satisfy the most recent constraints from other cosmological observations since @xmath10 is marginally consistent with the constraints imposed by supernovas ( @xmath169 , * ? ? ? * ) , and @xmath168 is also marginally consistent with the 3-yr wmap results . put another way , if the low value of is correct , then the x - ray @xmath3 data imply @xmath170 . when comparing to theoretical predictions of the @xmath3 relation we have considered the small biases in the mean concentration level and scatter expected for the highly relaxed , early forming systems selected for our study . we found that the model is a good representation of the x - ray data provided this bias for relaxed systems applies , especially for the low - mass range of our sample ( @xmath171 ) . however , the model does not agree with the @xmath3 data , even considering this bias . could other sources of systematic error seriously affect these conclusions ? for 24 of 39 systems in table [ tab.data ] we have performed a detailed investigation of systematic errors on @xmath0 and @xmath1 @xcite . generally , we find that the estimated systematic error is less than the statistical error , particularly for the systems of lowest mass . for the higher mass systems , the errors are usually comparable . but from consideration of all of these systems we recognize no obvious trend that would systematically shift the observed @xmath3 relation in one direction . although we do not believe systematic errors associated with the x - ray data analysis compromise the conclusions of our present investigation , here we list the most important issues to be resolved in future studies that seek to use the x - ray @xmath3 relation for precision constraints on cosmological models . [ he ] we also consider the impact on our present analysis of systematic errors resulting from the hydrostatic equilibrium approximation . ( note we verified that our analysis is insensitive to the definition of the virial radius . an interesting example is to compare to results obtained for @xmath172 , because within this radius all of the systems in our study possess good x - ray constraints . when defining the virial radius to correspond to @xmath172 for every system , we arrive at the same conclusions obtained in [ results ] but with correspondingly different @xmath63 and @xmath62 values . the predictions of the cosmological models were also converted to @xmath172 for this comparison . ) accounting for the biases associated with preferentially observing the most relaxed , early forming systems is required for precision constraints on cosmology . while @xcite found a nearly one - to - one correlation between halo `` formation epoch '' and concentration , it is not obvious how this theoretically motivated parameter should connect with the dynamical state of a real cluster . more direct quantifiers ( e.g. rejecting cases with recent major mergers and disturbed profiles ) suggest that `` relaxed '' halos should have @xmath15 larger average @xmath0 and smaller scatter . for massive systems @xmath173 ) , the outlook for analyzing a sample with a well - determined , observational selection function is excellent , because there exist several well - defined catalogs of the brightest , most massive clusters in x - rays ( e.g. , * ? ? ? * ) , of which many of the systems have been observed either with _ chandra _ or _ xmm_. for lower masses the number of x - ray catalogs of complete samples is small , and the number with good coverage with _ chandra _ or _ xmm _ observations is even smaller . for this purpose we have scheduled observations of a complete , x - ray flux - limited sample of 15 systems in the approximate mass range @xmath174 with _ chandra_. the determination of the mass distribution from x - ray observations requires that hydrostatic equilibrium is a suitable approximation . this approximation has been tested for massive clusters by comparing to results obtained from gravitational lensing ( e.g. , * ? ? ? * and references therein ) . there is very good agreement between the methods , especially outside the inner cores . this agreement is especially encouraging since some of the clusters are manifestly not completely relaxed ( e.g. , a2390 , * ? ? ? . at the low - mass end , good agreement between x - ray mass measurements with stellar dynamics in elliptical galaxies provides further indication that the hydrostatic equilibrium approximation is quite accurate for obviously relaxed systems @xcite . for over ten years cosmological hydrodynamical simulations have found that the hydrostatic equilibrium approximation is quite accurate in massive galaxy clusters ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? the most recent studies conclude that x - ray mass estimates of the most massive , relaxed clusters should typically underestimate the mass by a small amount ( @xmath15 ) because of turbulent pressure in the hot gas , with less of an effect in lower mass systems ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? the systems analyzed in our present investigation were selected to be highly relaxed as indicated by regularly shaped x - ray image morphology . this is reinforced by our analysis of the x - ray @xmath3 relation ( [ results ] ) , especially by the small intrinsic scatter in the concentrations which is a robust prediction of cdm models @xcite . underestimating the virial masses of our relaxed clusters by @xmath175 , as suggested by cdm simulations , would indicate that our measurement of @xmath62 should be raised by a factor @xmath176 , using @xmath177 . in addition , since @xmath178 , where @xmath24 is the nfw scale radius , the total increase in @xmath62 should be @xmath179 , provided the estimate of @xmath24 is unaffected by the presence of turbulent pressure . this systematic error , if real , would have the effect of increasing the discrepancies between the x - ray @xmath3 relation and ( most ) cdm models . however , essentially all of conclusions in [ results ] remain unchanged if we consider the bias for relaxed early forming systems to be 15% rather than 10% , within the range obtained by current simulations @xcite . the exceptions are the ocdm and decdm models ( [ clusters ] , see fig.[fig.errors.clusters ] ) , for which the @xmath179 boost in @xmath62 would bring the models into slightly better agreement with the data . however , the ocdm model is still rejected at @xmath180 confidence , and the @xmath163 open cdm models are still excluded ( @xmath11 conf . ) , the latter provided that at least a 5% bias ( of the expected @xmath15 ) for relaxed , early forming systems applies . we conclude , therefore , that the level of systematic error suggested by cdm simulations to affect the x - ray mass measurements does not change the conclusions of our present study . in order to use the x - ray @xmath3 relation for precision constraints on cosmological parameters , it is necessary to be able to predict halo concentrations produced in different cosmological models with high precision . this must be achieved with a semi - analytic procedure , because it is not feasible to resort to n - body simulations to fully investigate parameter space for obtaining confidence regions . the procedure proposed by b01 is currently one of the most promising models of this kind . with just two parameters @xmath101 and @xmath94 ( see [ models ] ) it can reproduce the results of cdm n - body simulations with ranges of power spectrum shapes , @xmath8 normalizations , matter content , and dark energy variables @xmath26 ( b01 , @xcite ) . recently , @xcite show that the normalizations of the b01 models ( i.e. , @xmath94 values ) obtained by different investigators analyzing halos covering approximately the same mass range ( @xmath181 ) can differ by 10 - 20% . for halos with @xmath182 , the halos of @xcite prefer @xmath105 and @xmath183 while those @xcite prefer @xmath107 . the `` cluster '' halos ( @xmath184 ) studied by d04 suggest a higher normalization @xmath110 . because so few studies have investigated the high - mass halo regime , it is unclear whether these differences reflect numerical details in the simulations or demand a revision of the b01 model . more studies of the concentrations of the highest mass halos are very much needed . in this context it is important to note that @xcite have recently studied the @xmath3 relation of massive halos , where most of their halos have @xmath185 . they obtain power - law fit parameters @xmath186 and @xmath187 for the model but with @xmath188 . converting their results to @xmath189 yields a 6% reduction in the concentration , @xmath190 , which is just 2% larger than the d04 value quoted in table [ tab.models.cm ] . we consider this quite good agreement , considering the differences between theoretical studies noted above for lower mass halos . finally , we mention that if we adopt the lowest normalization quoted in the literature ( @xmath105 , @xmath183 , @xcite ) for our entire mass range , then in order for the model to match the x - ray data , it is necessary to boost the concentrations produced in the model by another 16% over the @xmath175 attributed to the preferential selection of relaxed , early forming systems . this 16% increase can be achieved by increasing either ( @xmath191 ) , @xmath26 ( @xmath192 ) , or both . we note that the 99% upper limit derived in [ all ] would increase from 1.07 to 1.15 if @xmath183 . the published cdm predictions for the @xmath3 relation we have considered in this paper are all derived from dissipationless n - body simulations containing only dark matter . while the details of gasdynamics should be unimportant for massive systems ( @xmath193 ) , the effects of dissipation and feedback from star formation and agn likely influence the dark matter profile inferred from observations of lower mass halos . it is noteworthy that the systems with @xmath194 show the largest deviations from the model ( [ all ] ) . for most of these the _ chandra _ data require a significant contribution of stellar mass from the central galaxy @xcite . however , allowing for adiabatic contraction ( e.g. , * ? ? ? * ; * ? ? ? * ) of the dark matter profile in most cases degrades the fits , suggesting a more complex interplay between the baryons and the dark matter ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? some cosmological simulations with gas predict a small , but significant , systematic increase ( @xmath195 ) in the concentration of the total mass @xcite , but the inability of cdm simulations to reproduce observed x - ray temperature and density profiles within 50 - 100 kpc of cluster centers ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) makes it difficult to interpret the reliability of such results . future precision cosmology studies which aim to use the x - ray @xmath3 relation for low - mass systems ( @xmath196 ) will require better understanding of the influence of the central galaxy on the inferred dark matter distribution . we present the concentration ( @xmath0)-virial mass ( @xmath1 ) relation of 39 galaxy systems ranging in mass from individual early - type galaxies up to the most massive galaxy clusters , @xmath2 . we selected for analysis the most relaxed systems possessing the highest quality data currently available in the _ chandra _ and _ xmm _ public data archives . measurements for 24 systems were taken from our recent work @xcite which populate the lower mass portion of the sample ( @xmath197 ) . we obtained results for 15 massive galaxy clusters in our sample from the studies by @xcite and @xcite . our principal objective is to measure the @xmath3 relation accurately from galaxy to cluster scales and determine whether @xmath0 decreases with increasing @xmath1 as generally predicted by cdm models . however , we also use the @xmath3 relation to provide an initial demonstration of the ability of the @xmath3 relation to constrain cosmological parameters , which also serves to highlight key sources of systematic error both in the theoretical models and the observations . we parameterize the x - ray @xmath3 relation using a simple power - law model . the best estimates of the parameters the slope ( @xmath63 ) and normalization ( @xmath62 ) evaluated at @xmath198 were obtained via linear regression in log space taking into account the uncertainties on both @xmath0 and @xmath1 . we employed the bces method of @xcite for this analysis . fitting the power - law model to the entire sample yields @xmath124 and @xmath125 ( quoted errors @xmath126 ) . the slope @xmath63 is negative and inconsistent with 0 at @xmath4 . the previous studies of galaxy clusters ( @xmath193 ) with _ chandra _ and _ xmm _ by @xcite and @xcite did not place strong constraints on @xmath63 , and were quite consistent with @xmath151 ; i.e. , it is the lower mass galaxy groups that require @xmath152 ( see * ? ? ? recent optical studies of the @xmath3 relation in the group - cluster regime ( see [ intro ] ) also do not place strong constraints on the slope and are very consistent with @xmath199 . our analysis , therefore , provides crucial evidence that @xmath0 decreases with increasing @xmath1 , as expected in cdm models ( @xcite ; b01 ; d04 ; @xcite ) . we compare the x - ray data to the model ( with @xmath5 , @xmath6 , @xmath141 ) , the `` concordance model '' effectively representing the combined constraints from the first year of wmap cmb data , supernovae , and galaxy surveys ( e.g. , * ? ? ? we judge the median @xmath3 relation of the model to be consistent with the empirical power - law fit provided the x - ray sample consists of the most relaxed , early forming systems , for which a systematic increase in the concentrations of @xmath15 is expected @xcite . we measure an intrinsic scatter in @xmath127 of @xmath128 in excellent agreement with the prediction of cdm simulations for the most relaxed , early forming systems . the amount of scatter is a robust prediction of cdm variants and provides additional evidence that our sample comprises the most relaxed systems that are best suited for x - ray mass measurements requiring approximate hydrostatic equilibrium . the x - ray @xmath3 relation places interesting constraints on . within the context of the concordance model noted above , the @xmath3 relation requires @xmath7 ( 99% conf . ) , assuming a 10% upward bias in the concentrations for early forming systems . this confidence range is conservative as explained in [ all ] . next we compare the x - ray data to the model ( with @xmath161 , @xmath143 , @xmath155 , @xmath200 ) which effectively updates the model using the analysis of the third year of wmap data @xcite . the x - ray @xmath3 relation rejects the model at the 99.99% confidence level . the primary reason for the poor performance of the model is the low value of , but the lower value of and the tilt of the power spectrum also contribute to the poor fit . for this comparison we have assumed a uniform bias for relaxed , early forming halos of @xmath201 , as suggested by numerical simulations . this bias would have to be @xmath202 to bring the model into agreement with the x - ray data . while the early - type galaxy and group - cluster mass halos @xmath203 studied here apparently prefer slightly higher concentrations than predicted for typical halos in the concordance @xmath204cdm model , the opposite is true for late - type galaxies ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? this may indicate that a selection / formation - time bias operates across the galaxy type spectrum , with late - type galaxies inhabiting the low - concentration tail of the distribution ( though see * ? ? ? * ; * ? ? ? since d04 provide results of power - law fits to the @xmath3 relations of massive clusters formed in a variety of cdm models , including an open model and several dark energy models , we analyze separately the x - ray @xmath3 relation for the 22 systems in our sample with @xmath29 . in this mass range we obtain @xmath62 and @xmath205 values consistent with those inferred from previous _ chandra _ and _ xmm _ studies in the cluster mass regime noted above @xcite . as also found for the entire sample , the @xmath3 relation of the high - mass subsample is consistent with the model and very inconsistent with the model . however , an open model with @xmath163 is ruled out ( @xmath11 confidence ) from joint consideration of only the x - ray @xmath3 relation and published constraints on from the analysis of the abundances of galaxy clusters ( @xmath206 , see [ clusters ] ) . since cluster abundance studies also find @xmath164 if @xmath12 , using the x - ray @xmath3 data we find that we can also reject cdm models with @xmath12 at a high significance level ( @xmath165 confidence ) . consequently , the combination of the x - ray @xmath3 relation and cluster abundances ( and local constraints on the age of the universe ) provides novel evidence for a flat , low-@xmath13 universe with dark energy using observations only in the local ( @xmath14 ) universe . if the values of , , and @xmath86 of the model are correct , agreement with the x - ray @xmath3 relation may be achieved by increasing the dark energy equation of state parameter @xmath26 . we find that a quintessence model with @xmath10 and @xmath168 performs nearly as well as the model , and the larger value of @xmath26 remains marginally consistent with supernova constraints @xcite . finally , we discuss key sources of systematic error associated with both the x - ray measurements and theoretical models that need to be addressed before the x - ray @xmath3 relation is suitable for precision cosmology . in particular , if the virial masses are systematically underestimated by @xmath117 , as suggested by cdm simulations , then we estimate that @xmath62 is increased by @xmath207 , less than the @xmath15 increase expected for selecting relaxed , early forming systems . this level of systematic error does not change the conclusions of our present study ( see [ he ] ) , but it will be important for future studies of precision cosmology . we thank a. cooray for discussions and comments on the manuscript . , p.j.h . , and . gratefully acknowledge partial support from nasa grants nng04ge76 g , issued through the office of space sciences long - term space astrophysics program , and nag5 - 13059 , issued through the office of space science astrophysics data program . partial support for this work was also provided by nasa through chandra award numbers go4 - 5139x and go6 - 7118x issued by the chandra x - ray observatory center , which is operated by the smithsonian astrophysical observatory for and on behalf of nasa under contract nas8 - 03060 . we also are grateful for partial support from nasa - xmm grants , nag5 - 13643 , nag5 - 13693 , nng04gl06 g , and nng05gl02 g .

the thermal conductivity of the ( 2 + 1)-dimensional njl model in the presence of a constant magnetic field is calculated in the mean - field approximation and its different asymptotic regimes are analyzed . taking into account the dynamical generation of a fermion mass due to the magnetic catalysis phenomenon , it is shown that for certain relations among the theory s parameters ( particle width , temperature and magnetic field ) , the profile of the thermal conductivity versus the applied field exhibits kink- and plateau - like behaviors . we point out possible applications to planar condensed matter . pacs numbers : 11.10kk , 11.30qc , 11.10wx , 12.20.ds

cond-mat0203217

it is now well established , from the study of many relativistic theories of massless fermions in the presence of an external magnetic field , that a magnetic field can be a strong catalyst for chiral symmetry breaking with the consequent generation of a fermion dynamical mass even at the weakest attractive interaction among fermions @xcite . this magnetic catalysis ( mc ) of chiral symmetry breaking has proven to be universal , its main features being independent of the model under consideration ( see refs . @xcite@xmath0@xcite for various aspects of this phenomenon ) . the universality character of the magnetic catalysis has motivated many recent works @xcite-@xcite aimed to apply it to diverse areas of quantum physics . the essence of the mc effect lies in the dimensional reduction of the fermion pairing dynamics due to the confinement of these particles to their lowest landau level ( lll ) , when the pairing energy is much less than the landau gap @xmath1 ( @xmath2 is the magnitude of the magnetic field induction ) . under these circumstances , any attraction between fermions , whenever small it might be , is strengthened by the effective dimensional reduction in the presence of the magnetic field , and therefore , a condensate of fermion - antifermion is formed with the subsequent generation of a fermion mass . the lowest ll plays in this case a role similar to that of the fermi surface in bcs superconductivity @xcite . it is known that several quasiplanar systems have low - energy excitation spectrum of quasiparticles ( qp ) , characterized by a linear dispersion , around the fermi surface consisting of isolated points . the dynamics of these qp can be described by a `` relativistic '' quantum field theory of massless fermions . when such dirac - like qp are electrically charged , they can couple to an externally applied magnetic field which can catalyze the condensation of qp - antiqp pairs . then , one would expect the realization of mc in such a kind of condensed matter systems . as a matter of fact , the mc was suggested as the possible explanation @xcite@xmath0@xcite for the profile of the thermal conductivity in an applied magnetic field observed in recent experiments in planar high-@xmath3 cuprates @xcite@xmath0@xcite . the mc has also been proposed @xcite as the source of the semimetal - insulator phase transition observed in the so - called highly oriented pyrolitic graphites ( hopg ) @xcite in the presence of a magnetic field . given that the heat transport is a convenient probe to understand many basic properties of quasi - planar condensed matter systems , as gap structure , qp density , scattering rate , etc , the study of heat transport in a quasiplanar system subjected to mc may turn out physically revealing . it is the goal of the present paper to study how the mc affects the thermal transfer properties of a ( 2 + 1)-dimensional fermion system under an applied constant magnetic field . in particular , we calculate the thermal conductivity of a system of qp described by a ( 2 + 1)-dimensional nambu - jona - lasinio ( njl ) model which exhibits a mass ( gap ) generation for fermions in the presence of a magnetic field , and discuss possible applications to planar condensed matter systems . to obtain the results here reported , we used the same approximation of a constant magnetic field that was already explored in refs.@xcite . however , our calculations deviate considerably from what was done in these papers . not only we take into account the contribution of all landau levels , but the definition we use for the heat current itself is different . in our formulation , when the gap induced by the magnetic field is opened , the thermal conductivity exhibits a new term proportional to @xmath4 ( @xmath5 is the gap ) . near the phase transition point , the gap behaves like @xmath6 in the mean - field approximation . hence , the term proportional to @xmath4 yields a positive contribution in the slope of the thermal conductivity , leading to a jump in the slope of @xmath7 at @xmath8 ( kink - like behavior ) . notice that the magnetic catalysis is mainly responsible for the kink effect , being the critical behavior of the gap an essential factor for this result . we underline that to obtain such a kink - like behavior , it is crucial to go beyond the lll contribution in the calculation of the thermal conductivity , since , as we show , the heat transfer takes place due to transitions between neighboring lls . physically , this is easy to understand keeping in mind that the spatial momentum of the qp lies in the system plane . in the presence of a magnetic field perpendicularly applied to the two - dimensional sample , the qp spatial momentum is purely transverse and hence quantized into discrete landau levels . therefore , the transfer of kinetic energy can only occur by means of transitions between landau levels . even though our results were obtained by using a particular model , we point out that the main outcome of the present work is of a more general and theoretical character , as we show that the mc phenomenon can be responsible for a kink - like effect in the thermal conductivity of a whole class of ( 2 + 1)-dimensional relativistic fermion systems . that is , we show that the kink effect is essentially model independent , since it is determined by the critical behavior of the dynamically generated mass near the phase transition point . this fact makes the basic outcome of our investigation relevant beyond the particular model under consideration , linking it to the universality class of theories with such a critical behavior . in connection with this we conjecture that , since the hopg may be described @xcite by a model that belongs to the same universality class as that of the model used here , the hopg thermal conductivity in the presence of a magnetic field should display similar kink - like behavior . the plan of the paper is as follows . in sec . [ thcond - general ] we derive the expression for the thermal conductivity in the ( 2 + 1)-dimensional njl model in the presence of a constant magnetic field and analytically study its different asymptotics , underlying the possible application of each result . in sec . [ numerics ] we obtain the thermal conductivity vs magnetic field profile using numerical calculations . in the reported graph , the change of slope in the thermal conductivity profile is shown to occur at the critical magnetic field where the fermion dynamical mass is generated at the given temperature . the conclusions and discussion of potential applications of our results are presented in sec . [ concl ] . in appendix [ app - a ] we derive the critical curve in the @xmath9 plane which separates the symmetric and the symmetry - broken phases in the ( 2 + 1)-dimensional njl model , and study the scaling of the mass near the critical curve . a derivation of the kubo formula in the framework of matsubara formalism is given in appendix b. we start from the ( 2 + 1)-dimensional njl lagrangian density in an external magnetic field @xmath10 + \frac{g}{2n}\left ( \bar{\psi}_{i}\psi _ { i}\right ) ^{2 } , \label{lagrangian}\ ] ] where @xmath11 is the covariant derivative and the vector potential for the external magnetic field is taken in the symmetric gauge @xmath12 we assume that the fermions carry an additional flavor index @xmath13 ( @xmath14 for realistic @xmath15-wave superconductors ) . the dirac @xmath16-matrices are taken in the reducible four - component representation . in the absence of the bare mass term @xmath17 , the lagrangian density ( [ lagrangian ] ) is invariant under discrete chiral symmetry @xmath18 which forbids the fermion mass generation in perturbation theory . the appearance of the mass ( energy gap ) is due to the spontaneous breaking of the above discrete symmetry that leads to a neutral condensate of fermion - antifermion pairs . in condensed matter physics this could correspond to the condensation of excitons ( electron - hole bound states ) . introducing the composite field @xmath19 , the lagrangian ( [ lagrangian ] ) can be written in the form @xmath10 -\sigma\bar\psi_i\psi_i-\frac{n\sigma^2}{2g}. \label{lagrangian1}\ ] ] one readily verifies the equivalence of the lagrangians ( [ lagrangian ] ) and ( [ lagrangian1 ] ) by making use of the euler - lagrange equations ( or performing the integration over the field @xmath20 in the functional integral ) . the field @xmath20 has no dynamics at the tree level , however , it acquires a kinetic term due to fermion loops . the vacuum expectation value of @xmath20 gives a dynamical mass ( gap ) to fermions . the effective action for the composite field @xmath21 can be obtained by integrating over fermions in the path integral ( see appendix [ app - a ] ) . it is well known that in the absence of a magnetic field and at zero temperature the mass generation occurs only if the coupling constant exceeds some critical value @xcite . this can be seen from the stationary equation for the effective potential corresponding to the lagrangian density ( [ lagrangian ] ) , which in leading order of @xmath22-expansion is given by @xmath23 = 0 \label{stat - eq}\ ] ] ( it follows from eq . ( [ gapequation ] ) in the limit @xmath24 ) . in ( [ stat - eq ] ) we introduced the dimensionless coupling constant @xmath25 and @xmath26 is the ultraviolet cutoff parameter . from eq . ( [ stat - eq ] ) it is easy to see that there exists a critical value for the coupling constant @xmath27 such that , if @xmath28 eq . ( [ stat - eq ] ) has only the trivial solution ( @xmath29 ) , while for the strong coupling limit @xmath30 a nontrivial solution ( @xmath31 @xmath32 ) is reached which leads to the generation of a fermion mass . hence , the critical coupling @xmath33 separates two phases : the weakly coupled massless phase at @xmath34 and the strongly coupled massive one ( @xmath35 ) . an applied magnetic field changes the situation dramatically , so that the mass generation now takes place at all @xmath36 @xcite , hence the name magnetic catalysis . at finite @xmath37 and @xmath38 there is a critical curve in the @xmath9 plane separating the symmetric and the symmetry broken phases ( the derivation and analysis of the critical curve in the @xmath9 plane are given in appendix [ app - a ] ) . to derive an expression for the static thermal conductivity in an isotropic system we follow the familiar linear response method and apply kubo s formula @xcite @xmath39 where @xmath40 is the volume of the system , @xmath41 is the temperature , and @xmath42 is the heat - current density operator . the brackets denote averaging in the canonical ensemble with the density matrix @xmath43 . physically the thermal conductivity @xmath44 appears as a coefficient in the equation relating the heat current to the temperature gradient @xmath45 under the condition of absence of particle flow . if we neglect the chemical potential the heat density coincides with the energy density , hence the quantity that satisfies the continuity equation @xmath46 can be interpreted as the heat current density . equation ( [ continuity ] ) defines @xmath47 to within a divergenceless vector , which is sufficient for calculating the conductivity . the vector @xmath47 is obtained automatically from the lagrangian density ( [ lagrangian ] ) as @xmath48 when using equations of motion the last expression can be represented in the form @xmath49 at this point it is useful to underline that our definition of the heat current does not coincide with the one used in refs . @xcite , @xmath50 . operator @xmath51 can not be obtained from equation ( [ l - derivative ] ) by using the equations of motion , unless the external field is zero , so it does not lead to the correct thermal conductivity in the presence of a magnetic field . note also , that our quantity @xmath52 is explicitly gauge invariant in contrast to @xmath53 . the correlator of the heat currents ( or polarization function ) in ( [ heat_cur_cor ] ) is evaluated in the following way ( see appendix b for details on the formalism ) . first , it is computed in the matsubara finite temperature formalism replacing the time @xmath54 by the imaginary time @xmath55 ( @xmath56 ) : @xmath57 the thermal conductivity is then given by the discontinuity of the retarded function @xmath58 which is obtained by analytic continuation from imaginary discrete frequencies @xmath59 : @xmath60 .\ ] ] neglecting vertex corrections , the calculation of the thermal conductivity reduces to the evaluation of the bubble diagram @xmath61 , \label{polar_oper}\ ] ] where @xmath62 is the fourier transform of the translation invariant part @xmath63 of the fermion propagator in an external magnetic field : @xmath64 note , that the translation non - invariant phase of the fermion green s function cancels in the computation of @xmath65 . defining a spectral representation for @xmath66 , we can write @xmath67 the spectral representation allows one to make analytic continuation and find the retarded , @xmath68 , and advanced , @xmath69 , green functions according to the rule @xmath70 and @xmath71 . the spectral function @xmath72 is given by @xmath73 = -\frac{1}{\pi } { \rm im}% s^{r}(\omega + i\epsilon , { \vec{k } } ) . \label{spectrfunction}\ ] ] plugging the spectral representation ( [ spectr_repr ] ) into eq.([polar_oper ] ) the sum over matsubara frequencies can be performed . ) . this divergence disappears when the problem is treated more carefully . the prescription is simply to ignore it . ] after this has been done , we can continue the external frequencies to the real axis to get @xmath74 . finally , we arrive at the following expression for the thermal conductivity @xmath75 . \label{kappathrougha}\ ] ] in order to compute @xmath44 we have to specify now the fermion propagator . since the fermion mass is generated in the ( 2 + 1)-dimensional njl model in a magnetic field already at weak coupling , we take the standard expression for the massive fermion propagator in a magnetic field , decomposed over the landau level poles @xcite @xmath76 where @xmath77 + 4{\vec{k}}{\vec{\gamma}}l_{n-1}^{1}\left ( \frac{2{\vec{k}}% ^{2}}{eb } \right)\ ] ] with @xmath78 being projectors and @xmath79 laguerre s polynomials ( @xmath80 ) . here @xmath20 is the fermion dynamical mass obtained from the finite temperature gap equation in a constant magnetic field ( see appendix [ app - a ] ) . the spectral function according to ( [ spectrfunction ] ) is found to be @xmath81 , \label{spect - fun - magfield}\ ] ] where @xmath82 and @xmath83 , \quad f_{2}({\ \vec{k}})=4{\vec{k}}{\vec{\gamma}}l_{n-1}^{1}\left ( \frac{2{\vec{k}}^{2}}{eb } \right ) . \label{expr : f1-f2}\ ] ] here , we introduced the width @xmath84 of the quasiparticles , which is due to interaction processes , in particular , scattering on impurities , having replaced @xmath85 in ( [ spectrfunction ] ) by a finite @xmath84 . in general , the scattering rate @xmath86 , which is defined through the fermion self - energy , @xmath87 , is a frequency - dependent quantity ( as well as temperature and field dependent ) . it must be determined , like the dynamical mass , self - consistently from the schwinger - dyson equations . at low temperatures we are interested in its value at @xmath88 , so we will consider it as a phenomenological parameter . in the absence of a magnetic field ( @xmath89 ) it can be shown that the spectral function ( [ spect - fun - magfield ] ) reduces to @xmath90 , \quad e=\sqrt{{\vec k}% ^2+\sigma^2},\end{aligned}\ ] ] hence , the retarded fermion green s function is @xmath91 straightforward calculation of the trace in ( [ kappathrougha ] ) , with @xmath92 from eqs.([spect - fun - magfield]),([expr : f1-f2 ] ) gives @xmath93 = \nonumber \\ & & -\frac{16\gamma ^{2}n}{\pi ^{2}}e^{-2{\vec{k}}^{2}/eb}\sum \limits_{n , m=0}^{\infty } ( -1)^{m+n+1}\frac{(\omega ^{2}+m_{n}^{2}+\gamma ^{2})(\omega ^{2}+m_{m}^{2}+\gamma ^{2})-4\omega ^{2}\sigma ^{2}}{[(\omega ^{2}+m_{n}^{2}+\gamma ^{2})^{2}-4\omega ^{2}m_{n}^{2}][(\omega ^{2}+m_{m}^{2}+\gamma ^{2})^{2}-4\omega ^{2}m_{m}^{2 } ] } \nonumber \\ & & \times \left ( l_{n}\left ( \frac{2{\vec{k}}^{2}}{eb}\right ) l_{m-1}\left ( \frac{2{\vec{k}}^{2}}{eb}\right ) + l_{n-1}\left ( \frac{2{\vec{k}}^{2}}{eb}% \right ) l_{m}\left ( \frac{2{\vec{k}}^{2}}{eb}\right ) \right ) .\end{aligned}\ ] ] performing now the integration over momenta in eq . ( [ kappathrougha ] ) produces kronnecker s delta symbols @xmath94 due to the orthogonality of laguerre s polynomials , thus we get @xmath95[(\omega ^{2}+m_{n+1}^{2}+\gamma ^{2})^{2}-4\omega ^{2}m_{n+1}^{2}]}. \label{kappa}\ ] ] note that the factor @xmath96 in front of the right hand side of ( [ kappa ] ) originated from integrating over transverse momenta and gives the degeneracy of landau levels ( more exactly , the degeneracy is @xmath97 for the lowest ll , @xmath98 , and @xmath99 for levels with @xmath100 ) . we stress the important result that because of the appearance of the mentioned kronnecker deltas only transitions between neighboring landau levels contribute into the heat transfer . had we restrict ourselves to the lll as in @xcite , we would have gotten zero result for @xmath44 . further summation over @xmath101 in eq . ( [ kappa ] ) can be performed expanding the integrand in terms of partial fractions . the resulting sums are expressed through digamma functions by means of the formula @xmath102=-\left[a\psi(a)+b\psi(b)+c\psi(c)+d\psi(d)\right],\ ] ] where for convergency @xmath103 . after some algebraic manipulations the final expression for @xmath104 is written as follows @xmath105 this formula is the main result of our paper . note that it is independent of the particular model used to describe the qp interactions unless we specify the dependence of the dynamical mass on @xmath106 for a concrete model . another representation of eq.([kappamaggeneral ] ) which is particularly convenient for studying the small width limit @xmath107 is @xmath108 \right\ } . \label{kappa_small_gamma}\end{aligned}\ ] ] it is obtained from ( [ kappamaggeneral ] ) if one takes the series representation of the @xmath109-function in the integrand of ( [ kappamaggeneral ] ) and writes the expression in the curved brackets in fractions of @xmath110 . we are now in the position to study different asymptotic regimes defined by different relations among the dimensional parameters @xmath111 . first , we consider the limit of vanishing magnetic field ( @xmath89 ) . for that , one can use the formula for the asymptotic of @xmath109-function at large values of the argument @xmath112 to get @xmath113 . \label{thermcondfree}\ ] ] this expression describes the behavior of the thermal conductivity as a function of temperature for massive dirac s particles and is relevant for the supercritical phase of the njl model , where the mass is generated spontaneously even at zero magnetic field . the last expression can be evaluated analytically in two regimes , @xmath114 : @xmath115 and @xmath116 : @xmath117 . \label{thcond - t0-mass}\end{aligned}\ ] ] accordingly , in the weak coupling phase of the njl model where the dynamical mass is not generated we obtain : @xmath118 , \quad \gamma \ll t , \label{kappagamma < t}\ ] ] and @xmath119,\quad \gamma \gg t . \label{universal_kappa}\ ] ] eqs . ( [ kappagamma < t ] ) , ( [ universal_kappa ] ) up to an overall factor coincide with the corresponding expressions obtained in ref.@xcite for the vortex state of nodal quasiparticles in the @xmath120wave superconducting phase of high-@xmath121 cuprates . the overall factor there equals @xmath122 where @xmath123 are respectively the velocities perpendicular and tangential to the fermi surface . they originate from the quasiparticle excitation spectrum in the vicinity of the gap nodes which takes the form of an anisotropic dirac cone @xmath124 . with the overall factor replacing @xmath125 ( @xmath126 in real @xmath120wave superconductor ) , the first term in eq . ( [ universal_kappa ] ) reproduces the universal ( or residual ) thermal conductivity at low @xmath37 in the so - called `` dirty '' limit , @xmath127 , since it is independent of the impurity density , thus it will not depend on the specific characteristics of the scattering processes in the sample . the residual conductivity was recently observed in experiments @xcite confirming the existence of gapless quasiparticles in @xmath15-wave cuprates at @xmath128 . note , however , that in contrast to what was claimed in ref.@xcite , the low - temperature thermal conductivity for massive quasiparticles ( eq.([thcond - t0-mass ] ) ) does not exhibit a universal behavior when @xmath129 . this peculiarity of the low - temperature thermal conductivity can be used to find out experimentally the second gap in cuprates . expressions ( [ kappagamma < t ] ) , ( [ universal_kappa ] ) were recently used to propose a scenario for the arising of a plateaux at high magnetic fields @xcite in krishana s experiment . in that scenario the width @xmath84 of qps becomes dependent on the field due to scattering on disordered vortices , thus @xmath84 becomes @xmath130 where the field induced width @xmath131 is calculated to be @xmath132 . all the information regarding the magnetic field is encoded now in the total width @xmath133 hence the magnetic field is not explicitly present . if we start with a weak magnetic field , when @xmath134 , the thermal conductivity follows first the expression ( [ kappagamma < t ] ) ( weak field regime ) decreasing with the field . at some point , when @xmath84 becomes of order @xmath37 a crossover takes place to the high field regime ( [ universal_kappa ] ) with plateaux . physically , such a scenario is applicable only if there is a small number of vortices with large distances between them and the magnetic field is basically confined in tubes . however , for the field range of interest , @xmath135 , where @xmath136 are the lower and upper critical magnetic fields of the high @xmath3 superconductor respectively , the vortices are dense enough to overlap strongly giving rise to an effective uniform magnetic field in the whole plane @xcite , so the above scenario is not perhaps the most appropriate for this field range . we shall analyze now the thermal conductivity in the presence of a uniform magnetic field in the whole plane . the analysis in all cases will be made at fixed @xmath37 and @xmath84 , and we do not assume the last one to be dependent on the field . from the phase transition analysis of the ( 2 + 1)-dimensional njl model ( see appendix [ app - a ] ) it follows that at finite temperature there exists a critical value of the magnetic field @xmath137 , above which the magnetic catalysis phenomenon occurs generating a dynamical fermion mass even at weak coupling ( in what follows we consider only the weak coupling case @xmath138 ) . for magnetic fields less than the critical one ( @xmath139 ) the dynamical mass is zero ( @xmath29 ) . to study the narrow width limit @xmath140 , we replace the fractions @xmath141 in eq.([kappamaggeneral ] ) by @xmath142 then , after integrating over @xmath143 , what is equivalent to evaluating in the mass shell for the different ll , we obtain @xmath144 in ( [ kappagammato0 ] ) we kept @xmath145 in the denominators in order to be able to reproduce a smooth behavior of @xmath7 in the limit @xmath146 . the origin of the first term in ( [ kappagammato0 ] ) , whose essential role in the kink - like behavior is discussed below , can be traced back to the leading contribution of the zeroth to first ll transitions . note that it contributes only when the dynamical mass is present , i.e. if @xmath147 . that is , because the ratio @xmath148 appearing in eq.([kappamaggeneral ] ) becomes @xmath149 in this limit , it does not contribute to ( [ kappagammato0 ] ) unless @xmath150 due to the presence of the factor @xmath151 in the integrand of eq.([kappamaggeneral ] ) . this means that in the narrow width limit the magnetic catalysis is not only connected to the generation of the mass , but it is responsible also for the enhancement of the transitions between zeroth and first lls . the fact that the thermal conductivity is proportional in this limit to the scattering rate ( width ) @xmath84 means that it results from transitions of quasiparticles between cyclotron orbits mediated by scattering of qps on impurities . [ [ weak - field - limit ] ] weak field limit , + + + + + + + + + + + + + + + + + @xmath152 . taking into account that at weak coupling no dynamical mass is generated for fields below the critical value ( @xmath153 ) , we take @xmath29 in the calculation that follows . making use of the euler - maclaurin formula @xmath154 expression ( [ kappagammato0 ] ) with zero dynamical mass can be recast for subcritical fields @xmath155 in the form @xmath156 eq.([kappa_sub_crit ] ) shows monotonic decreasing of @xmath44 with increasing magnetic field @xmath2 ( at @xmath89 it reproduces the leading term in eq.([kappagamma < t ] ) ) . note that the scale @xmath157 marks the crossover point where the transition from superweak ( @xmath158 ) to weak ( @xmath159 ) fields takes place . [ [ strong - field - limit ] ] strong field limit , + + + + + + + + + + + + + + + + + + + @xmath160 . we shall consider now the strong field regime , @xmath160 , where a nonzero dynamical fermion mass is generated in the weakly interacting system ( we are interested mainly in the region of coupling constants @xmath161 where the scaling @xmath162 is achieved ) . let us start , however , analyzing the case of free massless fermions ( @xmath163 ) . in this case , one can use eq . ( [ kappagammato0 ] ) , after evaluating it in @xmath29 , to describe the very large field ( @xmath164 ) behavior of @xmath104 in the narrow width case , what yields an exponential fall of the conductivity @xmath165 coming back to the interacting case and after dropping the term depending on @xmath86 in the denominators of eq.([kappagammato0 ] ) , we obtain @xmath166 in the limit of large fields ( @xmath167 the first term in ( [ kappa_large_b ] ) is the leading one . such a term would produce a sharp plateau , were the fermion mass a constant . since in the model under consideration the mass is dynamical and it behaves as @xmath168 at @xmath169 the first term gives rise to an exponential decrease as in the case of free massless fermions ( [ kappa_free_large_b ] ) for asymptotically large fields . [ [ near - the - phase - transition - point ] ] near the phase transition point , + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + @xmath170 . in this case , the thermal conductivity can still be approximated by eq.([kappa_large_b ] ) . as shown in appendix a , near the mean field phase transition point @xmath171 so if the field lies in the interval @xmath172 , the dynamical mass @xmath173 and hence the @xmath174 appearing in the first term of eq.([kappa_large_b ] ) is of order one . in this field region the first term of eq.([kappa_large_b ] ) gives a positive contribution to the derivative of @xmath44 close to the critical point . that positive contribution leads to a jump in the slope of @xmath44 at @xmath175 therefore showing a kink - like behavior for @xmath44 in the narrow - width case . let us consider now the case where the width @xmath84 is small but finite . we are particularly interested in the behavior of @xmath44 near the phase transition point , where @xmath176 and therefore @xmath177 , @xmath178 from eq.([kappa ] ) one can see that for these fields the contribution of transitions between landau levels with @xmath179 in the integrand behaves as @xmath180while the transitions between zeroth and first ll decrease as @xmath181 . however , the ll degeneracy is also proportional to @xmath182 what implies that the transitions between the zeroth and first ll are not suppressed despite the fact that the gap between levels grows with the field . the leading in @xmath183 behavior is easy to obtain from eq.([kappa_small_gamma ] ) . it is given by the expression @xmath184 as discussed in the previous subsection , near the transition the dynamical mass behaves as @xmath171 so we can expand eq.([leading_b_behavior ] ) around @xmath29 to obtain @xmath185 \right\}. \label{kappa_sigma expansion}\ ] ] clearly , for @xmath186 the term proportional to @xmath4 gives a positive contribution to the derivative of @xmath44 with respect to@xmath187 near the transition point , so there is a jump in the slope of @xmath44 at the critical point : a kink - like effect . notice that if @xmath21 were zero or constant , the derivative of the thermal conductivity would satisfy @xmath188with c positive , so no kink - like effect would be present . on the other hand , since @xmath5 is dynamical , near the transition point @xmath189 @xmath190 with @xmath191 and @xmath192 positive , hence one can see that the dynamical mass not only allows for a jump in the slope , but it flattens the profile after the critical field , allowing at least for certain region of the parameter space a behavior of almost zero slope : a plateau - like profile . this kink and plateau -like behavior is corroborated by numerical calculations in the next section . we highlight that the dynamical mass is needed to obtain the kink - like effect in both narrow and finite width cases . hence , in our model the kink of the thermal conductivity is directly linked to the magnetic catalysis phenomenon . moreover , any model with the same critical behavior for @xmath21 would lead to a similar effect . this means that our results are indeed model independent , since any relativistic theory of interacting fermions that belongs to the universality class determined by the critical behavior here considered would yield a similar kink - like feature . one should note that the mean field behavior of the dynamical mass @xmath193 may change if higher order corrections ( fluctuations ) are taken into account in the gap equation . the fluctuations could either change the phase transition to a first order one , with a discontinuity in @xmath20 at the phase transition point ( this was a suggestion made by laughlin in @xcite ) , or to a non - mean - field order phase transition , with the scaling law @xmath194 where @xmath195 . while in the former case a discontinuity will appear in the thermal conductivity , in the latter case the conductivity will be a smooth function of the magnetic field , and a singularity will move to its higher derivatives . finally , we give an expression for the thermal conductivity when the temperature is much less than both @xmath86 and @xmath96 . at low @xmath37 the function @xmath196 in eq.([kappamaggeneral ] ) is very sharply peaked at @xmath88 , thus , expanding the rest of the integrand over @xmath197 and performing the integration we get @xmath198\big\}. \label{kappa - low - t}\end{aligned}\ ] ] it is easy to see , using the asymptotic of the @xmath199function at large values of its argument , that when @xmath200 the expression ( [ kappa - low - t ] ) goes to eq.([thcond - t0-mass ] ) in spite of the fact that we approach @xmath201 from the side @xmath202 . for large fields we get @xmath203\big\}. \label{small - t - large - b}\end{aligned}\ ] ] note that at large @xmath96 the thermal conductivity would approach some constant value in the case of constant mass , if @xmath204 , @xmath205 approaches that asymptotical value from below . this resembles the low @xmath37 behavior of the thermal conductivity in @xmath120wave cuprates @xcite . on the other hand , if the mass @xmath20 is a dynamical one with asymptotic behavior @xmath162 as in our four - fermion model , then @xmath205 goes to zero as @xmath183 at large fields . this is different from the thermal conductivity behavior for massless particles which tends to the universal constant @xmath206 for @xmath207 at large @xmath2 ( see , eq.([small - t - large - b ] ) ) . in this section we do a numerical study of the profile of the thermal conductivity versus the applied magnetic field , taking into account the generation of the dynamical mass at a critical field that depends on the temperature . the field - dependence of the finite - temperature dynamical mass is obtained from the solution of the gap equation ( [ gapequation ] ) derived in appendix [ app - a ] . to numerically investigate the behavior of the thermal conductivity ( [ kappamaggeneral ] ) within a parameter range that can be of interest for condensed matter applications , we need to restore all the model parameters , like @xmath208 . following ref.@xcite we write the lagrangian density as @xmath209+\frac{gv}{2n}(\bar{\psi}_{i}\psi _ { i})^{2 } , \label{n1}\ ] ] where @xmath210 and @xmath211 entering in @xmath212 were defined in the previous section , and the external potential is given by eq.([ext_potential ] ) . as known , this lagrangian is equivalent to @xmath213-\sigma v\bar{\psi}_{i}\psi _ { i}-\frac { n\sigma ^{2}v}{2 g } , \label{n2}\ ] ] since the euler - lagrange equation for the auxiliary scalar field @xmath21 obeys the constraint @xmath214 so that the lagrangian density ( [ n2 ] ) reproduces eq . ( [ n1 ] ) upon application of this constraint . the effective action for the composite field @xmath21 can be obtained by integrating over fermions in the path integral . from the minimum condition of the effective potential @xmath215 one finds that , at fixed @xmath37 , there is a critical value of the magnetic field @xmath216 such that for subcritical fields @xmath217 the gap is zero , while for @xmath218 it is given by the non - trivial solution of the gap equation ( [ gapequation ] ) ( see appendix [ app - a ] ) . notice that @xmath219 has dimension of energy and plays the role of @xmath220 in the dirac lagrangian density . to generate a plot of @xmath221 ( @xmath222 is the thermal conductivity at zero field ) versus the magnetic field , we need to substitute @xmath223 , where @xmath2 is measured in gauss . it is convenient to measure all energetic quantities in degrees of @xmath224 , what leads to the replacement @xmath225 , where the magnetic field is measured now in tesla s . thus , using the approximated value of the characteristic velocity @xmath226 @xcite,@xcite , @xmath227 , and we obtain the critical curve @xmath228 . let us numerically find the profile of @xmath44 versus the magnetic field at a fixed @xmath37 . in fig.[fig:1 ] @xmath229 has been plotted for two different temperatures taking into account the generation of the dynamical mass for @xmath218 . here we corroborate what we had already argued in section 2 based on the analytic result found for @xmath44 : due to the appearance of @xmath21 at a critical field that depends on the temperature , @xmath44 exhibits a kink behavior in its profile with the magnetic field . moreover , for @xmath230 the kink is followed by a region where @xmath44 is only weakly dependent on the field ( plateau - like region ) . with decreasing temperature , the position of the kink moves to the left in accordance with the critical line @xmath231 . the numerical calculations revealed the sensitivity of the kink - plateau feature of the thermal conductivity to the relation between @xmath86 and @xmath37 . only when @xmath86 was not much smaller than @xmath37 the thermal conductivity showed a kink - plateau profile ( in fig.[fig:1 ] the curves shown correspond to the ratios @xmath232 and @xmath233 ) ) . in the present paper we study the thermal conductivity of relativistic fermions in a ( 2 + 1)-dimensional four - fermion interaction model as a function of the applied magnetic field , the temperature and the particle width . we have shown that , for certain relations among these parameters , the profile of the thermal conductivity versus the applied field exhibits a kink - like behavior at @xmath234 , where @xmath235 is the critical field for the generation of a fermion dynamical mass @xmath21 , followed by a plateau - like region at @xmath236 we point out that the kink effect is the consequence of two main features : the generation of a fermion gap in the presence of the magnetic field ( mc phenomenon ) , and the enhancement of the zeroth - to - first lls transitions . a main outcome of our investigation is that the relevant properties of the thermal conductivity of the ( 2 + 1)-dimensional relativistic qp system around the critical point are model independent . indeed , the essential ingredient of the effective model required to produce the kink - like effect in the thermal conductivity is the critical behavior of the dynamical mass induced by mc near the phase transition , so not much depends on the concrete form of the effective lagrangian . this fact makes our result relevant beyond the particular model under consideration , linking it to the universality class of theories with such a critical behavior . such an universal character opens a window for possible applications . from a quantum field theory viewpoint , condensed matter systems whose fermi surface are only characterized by nodal points are especially interesting for us , since at low energies they can be described by relativistic quantum field theory models of massless fermions @xcite . along that direction , a feasible possibility for the application of our results is the heat transport properties of graphite in the presence of a magnetic field . let us recall that hopg materials @xcite have layered structure with two isolated points in the brillouin zone where the dispersion is linear . their electronic states can be thus described in terms of relativistic charged particles @xcite . this graphite could exhibit the phenomenon of mc as suggested in ref.@xcite to explain the semimetal - insulator phase transition observed in hopg in the presence of a magnetic field perpendicular to the layers . as the quasiparticles in the graphite are subjected to landau quantization under a perpendicularly applied magnetic field , our results should have full strength there and we anticipate that the thermal conductivity of these systems will show a behavior similar to the one reported in the present work ( for computation of the electric conductivity in graphite along the lines followed in the present work , see recent paper @xcite ) . on the other hand , the characteristic feature of @xmath15-wave superconductors is also the existence of nodal points ( four in this case ) where the order parameter vanishes , thus the fermi surface consists of four isolated points with excitations around them being well - defined gapless quasiparticles ( qp ) . the kinetic part of the qps effective lagrangian is nothing but the dirac lagrangian for two species of massless four - component spinors @xcite . at low temperatures such qps give the main contribution to thermodynamic and transport properties . there is now considerable experimental evidence for the existence of well - defined qps in the superconducting state of cuprates ( see ref.@xcite and references therein ) . as was mentioned in the introduction , the mc has been suggested @xcite@xmath237@xcite to be behind the odd behavior of the thermal conductivity of high - tc superconducting cuprates in a magnetic field observed in the experiment @xcite , @xcite,@xcite , although alternative solutions , not based on mc , has also been proposed @xcite,@xcite . according to the experiment done by krishana et al . @xcite , and later reproduced by other authors @xcite,@xcite , at temperatures significantly lower than @xmath3 of superconductivity , the thermal conductivity @xmath205 , as a function of a magnetic field perpendicularly applied to the cuprate planes of the samples , displays a sharp break in its slope ( kink - like behavior ) at a transition field @xmath238 , followed by a plateau region in which it ceases to change with increasing field up to the highest attainable fields @xmath239 . the critical temperature for the appearance of the kink - like behavior scales with the magnetic field as @xmath240 similar to the scaling of the critical temperature with the field found in njl models @xcite . it is worthy to mention here that the reliability of the thermal conductivity experiments of krishana et al . @xcite has been a matter of debate in the last years @xcite , although it seems that finally the contradictory results have been clarified and understood @xcite . we emphasize that none of the previously mentioned @xcite@xmath0@xcite,@xcite,@xcite attempted explanations of krishana s experiment were able to obtain on a theoretical basis the kink - like behavior observed in the experiment @xcite . in the case of our results , although it is striking that , as discussed in our previous paper @xcite and shown in fig.[fig:1 ] , our numerical curves exhibit a profile of the thermal conductivity with the applied magnetic field with similar qualitative characteristics the crossing of the curves occurs in such a way that the lower @xmath37 curve reaches the higher value at large fields . ] to those observed in the experiment @xcite , we can not claim that our results are valid in the vortex state of the superconductor since our calculations are based on the landau level quantization , whose applicability to the d - wave superconductor in the presence of vortices has been recently subjected to intense criticism @xcite . nevertheless , the mechanism of generating a kink - like effect in the thermal conductivity via the mc phenomenon , as studied in the previous sections , should be relevant for condensed matter systems with dirac - like charged qp on which landau level quantization is feasible . in this direction , we would like to point out that there is still some chance for the realization of mc in cuprate systems . as it has been recently argued in refs . @xcite,@xcite , the physical picture underlying the description of superconducting cuprates may involve the interplay of two different phases with their corresponding order parameters : one @xmath241 superconducting ( dsc ) and one of density wave order ( ddw ) . the ddw state , unlike the dsc state , does not break gauge invariance , so the quasiparticles of the ddw state can form landau levels under an applied magnetic field @xcite . although the above scenario is still speculative and more experimental confirmation is required , one can venture that given that the charged qp excitations around the nodes of the ddw order parameter have landau levels , they can be subjected to a mc phase transition at some critical magnetic field , and hence our theoretical results could be relevant in that case for the description of the heat transport properties of the ddw - state in the magnetic field . we would like to acknowledge v.a . miransky for useful and stimulating discussions and to z. tesanovic for calling our attention to ref.@xcite . we are grateful to a. hams and i. shovkovy for help in the numerical calculation . this research has been supported in part by the national science foundation under grant no . the work of v.p.g . is supported also by the scopes - projects 7 ip 062607 and 7ukpj062150.00/1 of the swiss national science foundation and grant - in - aid of japan society for the promotion of science ( jsps ) no . he wishes to acknowledge the jsps for financial support . the effective potential of the ( 2 + 1)-dimensional njl model in a constant external magnetic field at finite temperature was computed in @xcite . the integration over fermion fields in the functional integral with the lagrangian ( [ lagrangian1 ] ) can be performed to give the effective potential in the form @xmath242 where the temperature independent part of the potential is given by @xmath243 , \label{eq : poten41}\end{aligned}\ ] ] ( the mass scale parameter @xmath244 and @xmath26 is the ultraviolet cutoff which is taken much bigger than all other parameters in the model ) and the part depending on temperature is @xmath245 .\end{aligned}\ ] ] here @xmath246 is the jacobi s elliptic function . the gap equation @xmath247 follows from ( [ eq : vbt ] ) @xmath248\hspace{-1mm}\right]\hspace{% -1mm}=0 . \label{gapequation}\ ] ] the solution of this equation defines the fermion dynamical mass . at @xmath249 the gap equation admits a nontrivial solution only if the coupling @xmath250 is supercritical , @xmath251 ( @xmath252 ) . the gap equation at @xmath253 was also studied in the literature ( see , for example , refs.@xcite ) . it was shown that it always has a nontrivial solution at all @xmath254 no matter how small the magnetic field @xmath2 might be . in the weak coupling phase the field - induced dynamical mass at zero temperature behaves as @xmath255 in weak fields ( @xmath256 ) , while at high fields ( @xmath257 ) it is @xmath258 . at finite temperature the critical line in the @xmath259 plane separating the massless and massive phases was calculated numerically in @xcite ( see also @xcite ) . we shall derive here an analytical solution of the gap equation near such a critical line . we start , actually , with the derivation of the landau - ginzburg - like potential by expanding @xmath215 in powers of @xmath21 , since near the phase transition point @xmath21 is small , and write @xmath260 as it is accustomed for second - order phase transitions , the equality of the coefficient @xmath261 to zero defines the phase transition curve . the region of parameters @xmath262 where @xmath263 corresponds to the spontaneously broken phase with fermions acquiring the mass , whereas the region @xmath264 corresponds to the massless phase . our goal is to obtain the coefficients @xmath265 and @xmath266 . when it is done the solution of the gap equation is given by @xmath267 from ( [ sigmma2 ] ) we obtain the behavior of the dynamical mass near the phase transition point ( @xmath268 ) . in particular , at fixed temperature and @xmath269 we find @xmath270 is easily expanded in powers of @xmath271 . let us turn to @xmath272 which we write as the sum of two terms @xmath273 \nonumber \\ & & + \frac{neb}{4\pi ^{3/2}}\int_{0}^{\infty } \frac{dt}{t^{3/2}}e^{-t\sigma ^{2}}\left ( \coth { ebt}-1\right ) \left [ \theta _ { 4}\big(0|\frac{i}{4\pi t^{2}t}\big)-1\right ] .\end{aligned}\ ] ] the second term in the last expression can be expanded in a series in @xmath274 since the two brackets ( with cotangent and @xmath275function ) regularize the behavior of the integrand at infinity and zero , respectively . thus we need to calculate the first term @xmath276 = 2\sum\limits_{n=1}^{\infty } ( -1)^{n}\int_{0}^{\infty } \frac{dt}{t^{3/2}}e^{-t\sigma ^{2}-\frac{n^{2}}{% 4t^{2}t } } \nonumber \\ & = & 4\sqrt{2t\sigma } \sum\limits_{n=1}^{\infty } \frac{(-1)^{n}}{\sqrt{n}}% k_{1/2}\left ( \frac{n\sigma } { t}\right ) = -4\sqrt{\pi } t\log \left ( 1+e^{-% \frac{\sigma } { t}}\right ) , \end{aligned}\ ] ] where @xmath277 is a modified bessel function ( @xmath278 ) . finally , we obtain the following expressions for the coefficients @xmath279 \big\ } , \label{m_function } \\ \lambda ( t , b ) & = & \frac{n}{\pi \sqrt{eb}}\big\{\frac{\zeta ( 3/2)}{4\sqrt{2}}+ \frac{1}{48}\left ( \frac{\sqrt{eb}}{t}\right ) ^{3}+\frac{(eb)^{3/2}}{2\pi } \int_{0}^{\infty } dtt^{1/2}\left ( \coth { ebt}-1\right ) \nonumber \\ & \times & \left [ \theta _ { 4}\big(0|\frac{i}{4\pi t^{2}t}\big)-1\right ] \big\}. \label{lambda_func}\end{aligned}\ ] ] the critical curve @xmath280 can be analyzed analytically at @xmath281 where the integral in eq.([m_function ] ) is exponentially small and for @xmath3 we get the equation @xmath282 for the solution to exist , it must satisfy that @xmath283 , what gives the critical temperature @xmath284 where @xmath285 is the dynamical mass at zero temperature . one can convince oneself that there is no solution of the equation @xmath280 when @xmath286 . indeed , for that let us write the integral in ( [ m_function ] ) as @xmath287 \nonumber \\ & = & \frac{1}{\sqrt{eb}}\int_{0}^{\infty } \frac{dt}{t^{1/2}}\left ( \coth { t } -1\right ) \left [ \theta _ { 4}\big(0|\frac{ieb}{4\pi t^{2}t}\big)-1\right ] . \label{i_integral}\end{aligned}\ ] ] we further divide the integrand into three pieces @xmath288 -\int_{0}^{\infty } \frac{dt}{t^{1/2}}\left ( \coth { t}-\frac{1}{t}-1\right ) \right . \nonumber \\ & + & \left . \int_{0}^{\infty } \frac{dt}{t^{1/2}}\left ( \coth { t}-\frac{1}{t } -1\right ) \theta _ { 4}\big(0|\frac{ieb}{4\pi t^{2}t}\big)\right\ } .\end{aligned}\ ] ] the first and second integrals in the last expression can be evaluated exactly ( after changing the variable @xmath289 in the first integral ) with the help of the formulas @xcite @xmath290 = -4\sqrt{\pi } \log { 2}\frac{t}{\sqrt{eb } } ; \\ & & \int_{0}^{\infty } \frac{dt}{t^{1/2}}\left ( \coth { t}-\frac{1}{t}-1\right ) = \sqrt{2\pi } \zeta ( \frac{1}{2}).\end{aligned}\ ] ] the integral with @xmath291function is calculated using the jacobi imaginary transformation to @xmath292function and keeping in it only the first term in the series when @xmath293 : @xmath294 this reduces the third integral to @xmath295 ( when proceeding to the last equality we changed the variable @xmath54 to @xmath296 and then expanded over @xmath96 ) . thus , combining all formulas we get the following expression for the @xmath297 function : @xmath298 as seen , there is no solution as @xmath293 for @xmath299 . ( in case @xmath300 and @xmath301 we get the standard expression for the critical temperature @xmath302 @xcite ) . hence , we arrive at the conclusion that the only remaining possibility is that the root of the equation @xmath280 is of the order of @xmath303 . however , in this case we can not expand the integral in eq.([m_function ] ) and should turn to a numerical calculation . the root of the function @xmath304 when the parameter @xmath305 is found to be @xmath306 what defines the critical line . for the critical temperature this gives @xmath307 in agreement with the result of ref.@xcite . we calculated numerically the coefficient before @xmath308 in eq . ( [ near_crit_scaling ] ) which is found to be @xmath309 , thus the scaling of the dynamical mass near the critical line is given by the formula @xmath310 for the sake of completeness we derive here the expression for the thermal conductivity in the two - dimensional case used in section ii . we start from the kubo s formula for the thermal conductivity tensor @xcite integrating over @xmath54 by parts in eq.([1a ] ) , and taking into account that the currents go to zero at @xmath315 , we obtain @xmath316\ } , \label{6a}\end{aligned}\ ] ] where we used also the fact that the quantity under @xmath317 is a function only of @xmath318 . we have @xmath322\}={\rm tr}\{\rho u_{i}(t)u_{j}(0)\}- { \rm % tr}\{\rho u_{i}(t)u_{j}(0)\}^{\dagger } = 2i{\rm im}{\rm tr}\{\rho u_{i}(t)u_{j}(0)\}. \label{11a}\end{aligned}\ ] ] then , we can write ( [ 8a ] ) as the thermal conductivity for an isotropic system is given by @xmath324 where the summation over repeated indices is understood ( @xmath15 is the number of space dimensions , in our case @xmath325 ) . equation ( [ 12a ] ) then takes the form where the hermiticity of the heat current operators was used . similarly @xmath330 now using eqs.([16a ] ) and ( [ 18a ] ) and the symmetry of the matrix elements under the interchange @xmath331 we can write the correlator function in the form the retarded fourier transform of @xmath333 is given by @xmath334 to obtain the spectral representation of @xmath335 we define @xmath336 \nonumber \\ & = & \sum_{n , m}\frac{1}{z}e^{-\beta e_{n}}\left ( 1-e^{-\beta ( e_{m}-e_{n})}\right ) \mid < n\mid u_{i}\mid m>\mid ^{2}\delta ( e_{m}-e_{n}-\omega ) . \label{22a}\end{aligned}\ ] ] from ( [ 21a ] ) and ( [ 22a ] ) we can write @xmath337 we can use this representation in order to express the thermal conductivity in terms of the spectral density @xmath338 , which is an important step for the calculations in the green s function formalism . using the inverse fourier transform , we can write to express @xmath44 in terms of the spectral function we substitute with ( [ 26a ] ) in ( [ 14aa ] ) , so we get @xmath342 taking into account that the spectral function is an odd function , @xmath343 we can see that only the imaginary part of @xmath344 remains , what leads to @xmath345 or equivalently , from the representation ( [ 22a ] ) for the spectral function it can be shown that the thermal conductivity ( [ 30a ] ) can be expressed in terms of imaginary time green s functions . indeed , let us introduce the following thermal green function @xmath347 and its fourier transform @xmath348 inserting the complete set of energy eigenstates @xmath349 we can perform the integration over @xmath55 as indicated in ( [ 33a ] ) , to find @xmath350 if we now define an analytical function @xmath351 in such a way that at discrete points @xmath352 it coincides with @xmath353 and has a branch cut along the real axis , then the spectral function ( [ 22a ] ) is related to the discontinuity of @xmath351 across the cut @xmath354 . \label{32a}\ ] ] substituting ( [ 32a ] ) in ( [ 30a ] ) we arrive to eq . ( 10 ) of section ii . v. p. gusynin , v. a. miransky , and i. a. shovkovy , , 3499 ( 1994 ) . v. p. gusynin , v. a. miransky , and i. a. shovkovy , , 4718 ( 1995 ) . gusynin , v. a. miransky , and i. a. shovkovy , b * 349 * , 477 ( 1995 ) . k. g. klimenko , z. phys . c*54 * , 323 ( 1992 ) ; teor . mat . fiz . * 90 * , 3 ( 1992 ) ; i. v. krive and s. a. naftulin , phys . d * 46 * , 2737 ( 1992 ) . gusynin , v. a. miransky , and i. a. shovkovy , , 4747 ( 1995 ) ; nucl . b * 462 * , 249 ( 1996 ) ; , 1291 ( 1999 ) ; nucl . b * 563 * , 361 ( 1999 ) ; v. elias _ et al . d * 54 * , 7884 ( 1996 ) ; v. p. gusynin and a. v. smilga , phys.lett . b * 450 * , 267 ( 1999 ) . c. n. leung , y. j. ng , and a. w. ackley , , 4181 ( 1996 ) ; d. k. hong , y. kim , and s .- j . sin , , 7879 ( 1996 ) ; a. v. shpagin , hep - ph/9611412 ; d. s. lee , c. n. leung , and y. j. ng , phys . d * 55 * , 6504 ( 1997 ) ; v. p. gusynin and i. a. shovkovy , phys . d * 56 * , 5251 ( 1997 ) . i. a. shushpanov and a. v. smilga , b * 402 * , 351 ( 1997 ) ; s. kanemura , h. -t . sato and h. tochimura , nucl . b * 517 * , 567 ( 1998 ) ; a. yu . babansky , e. v. gorbar , and g. v. shchepanyuk , b * 419 * , 272 ( 1998 ) ; t. itoh and h. kato , , 30 ( 1998 ) ; e. j. ferrer and v. de la incera , phys . lett . b * 481*,287 ( 2000 ) ; v. p. gusynin , ukrainian j. phys . * 45 * , 603 ( 2000 ) ; v. ch . zhukovsky _ et al . _ , jetp lett . * 73 * , 121 ( 2001 ) . k. farakos and n.e . mavromatos , cond - mat/9710288 . k. farakos , g. koutsoumbas , and n.e . mavromatos , int . b*12 * , 2475 ( 1998 ) . g. w. semenoff , i. a. shovkovy , and l. c. r. wijewardhana , mod . a*13 * , 1143 ( 1998 ) . w. v. liu , nucl . b * 556 * , 563 ( 1999 ) . e. j. ferrer , v. p. gusynin , and v. de la incera , mod . b*16 * , 107 ( 2002 ) . e. j. ferrer , v. p. gusynin , and v. de la incera , phys . b * 455 * , 217 ( 1999 ) ; g. w. semenoff , i. a. shovkovy , and l. c. r. wijewardhana , phys . d * 60 * , 105024 ( 1999 ) . k. krishana _ et al . _ , science , * 277 * , 83 ( 1997 ) , n. p. ong et al . , cond - mat/9904160 . d. v. khveshchenko , phys . 87 * , 206401 ( 2001 ) . m. s. sercheli et al . , cond - mat/0106232 . b. rosenstein and b. warr , phys . repts . * 205 * , 59 ( 1991 ) . v. ambegaokar and a. griffin , phys . a * 137 * , 1151 ( 1965 ) . a. c. durst and p. a. lee , phys . b * 62 * , 1270 ( 2000 ) . a. chodos , k. everding , and d. a. owen , phys . d * 42 * , 2881 ( 1990 ) . m. franz , phys . lett . * 82 * , 1760 ( 1999 ) . l. taillefer et al . , phys . lett . * 79 * , 483 ( 1997 ) ; k. behnia et al . , j. low temp . phys . * 117 * , 1089 ( 1999 ) . m. franz and o. vafek , , 220501(r ) ( 2001 ) . p. w. anderson , cond - mat/9812063 . i. vekhter and a. houghton , phys . lett . * 83 * , 4626 ( 1999 ) . g. semenoff , phys . * 53 * , 2449 ( 1984 ) ; f. d. m. haldane , _ ibid _ * 61 * , 2015 ( 1988 ) ; j. gonzales , f. guinea , and m. a. h. vozmediano , phys . lett . * 77 * , 3589 ( 1996 ) . g. e. volovik , proc . sci . * 96 * , 6042 ( 1999 ) ; phys . rept . * 351 * 195 ( 2001 ) . gorbar , v.p . gusynin , v.a . miransky , and i.a . shovkovy , cond - mat/0202422 . to be published in prb . g. w. semenoff and l. c. r. wijewardhana , , 2633 ( 1989 ) ; j. b. marston , phys . . lett . * 64 * , 1166 ( 1990 ) ; n. dorey and n. e. mavromatos , nucl . b * 368 * , 614 ( 1992 ) ; s. h. simon and p. a. lee , , 1548 ( 1997 ) . a. paramekanti _ et al . _ , , 6786 ( 2000 ) . r. b. laughlin , , 5188 ( 1998 ) ; t. v. ramakrishnan , j. phys . chem . solids , * 59 * , 1750 ( 1998 ) . a. s. melnikov j. phys . matter * 11 * , 4219 ( 1999 ) . l. marinelli et al . , phys . b * 62 * , 3499 ( 2000 ) ; j. ye , phys . , * 86 * , 316 ( 2001 ) ; o. vafek et al , phys . rev . b * 63 * , 134509 ( 2001 ) ; j. ye and a. millis , cond - mat/0101032 ; d. knapp , c. kallin and a. j. berlinsky , phys . rev . b , 64 , 014502 ( 2001 ) ; a. vishwanath , phys . 87 , 217004 ( 2001 ) .

we present optical spectroscopic identifications of hard x - ray ( 5 - 10 kev ) selected sources belonging to the hellas sample obtained with _ bepposax _ down to a 5 - 10 kev flux limit of erg s . the sample consists of 118 sources . 25 sources have been identified trough correlations with catalogues of known sources . 49 have been searched for spectroscopic identification at the telescope . 13 fields resulted empty down to r=21 . 37 sources have been identified with type 1 agn and 9 with type 2 agn . the remaining are : 5 narrow emission line galaxies , 6 clusters , 2 bl lac , 1 radio galaxy and 1 star . combining these objects with other hard x - ray selected agns from _ asca _ and _ heao1 _ , we find that the local luminosity function of type 1 agn in the 2 - 10 kev band is fairly well represented by a double - power - law - function . there is evidence for significant cosmological evolution according to a pure luminosity evolution ( ple ) model , with=2.12 and=2.22 ( ) in a (,)=(1.0,0.0 ) and in a (,)=(0.3,0.7 ) cosmology , respectively . the data show an excess of faint high redshift type 1 agn which is well modeled by a luminosity dependent density evolution ( ldde ) , similarly to what observed in the soft x - rays . however , in both cosmologies , the statistic is not significant enough to distinguish between the ple and ldde models . the fitted models imply a contribution of agn1 to the 2 - 10 kev x - ray background from 35% up to 60% .

astro-ph0112455

agn have first been discovered in the radio and soon after searched in the optical band . consequently , they have been classified using their optical characteristics and mainly divided into two categories : type 1 ( agn1 ) and 2 ( agn2 ) according to the presence or not of broad emission lines in their optical spectra ( we will keep this definition of agn1 throughout this paper ) . before the advent of the last generation of hard x - ray telescopes , agn samples where predominantly based on agn1 selected either in the optical or , later on , in the soft x - rays by _ einstein _ and _ rosat_. in these bands the evolution of agn1 has been well measured ( see e.g. della ceca et al . 1992 ; boyle et al . 2000 ; miyaji , hasinger , & schmidt 2000 ) . on the contrary the production of samples of agn2 has been difficult at any wavelength and limited to few local surveys . the general picture was in favor of a model in which agn1 objects were associated to agn with low absorption in the hard x - rays while agn2 to obscured sources with large column densities and spectra strongly depressed in the soft x - rays , as expected in the unification models ( e.g. antonucci 1993 ) . in the last decade the advent of the _ asca _ and _ bepposax _ satellites has allowed for the first time the detection and identification of agn as the main counterparts of hard ( 2 - 10 kev ) x - ray sources down to fluxes @xmath14 erg @xmath3 s@xmath4 , more than 2 orders of magnitude fainter than _ heao1 _ ( wood et al . 1984 ) . these identifications accounted for about 30% of the 2 - 10 kev hard x - ray background ( ueda et al . 1998 ; fiore et al . recently the new generation of x - ray satellites such as _ chandra _ and _ xmm - newton _ , have reached fluxes 100 times fainter , identifying hundreds of sources and almost resolving the hard ( 2 - 10 kev ) x - ray background ( e.g. mushotzky et al . 2000 ; fiore et al . 2000 ; giacconi et al . 2001 ; hornschemeier et al . 2001 ; hasinger et al . 2001 ; tozzi et al . 2001 ; baldi et al . 2001 ) . thanks to their excellent angular resolution ( @xmath11 - 5@xmath15 ) , the first spectroscopic identifications projects have been able to observe faint ( i@xmath123 ) optical counterparts . at variance with the classical type-1/type-2 model in the optical , a significant number of the counterparts ( @xmath130% ) resulted to be apparently optical normal galaxies , with x - ray luminosities @xmath16@xmath10@xmath17 erg s@xmath4 typical of agn activity , and moreover part of the optical type 1 agns resulted to be absorbed in the hard x - rays ( see e.g. fiore et al . 2000 ; barger et al . 2001 ; tozzi et al . 2001 ; hornschemeier et al . 2001 ; comastri et al . 2002 ) . these observations have complicated the picture of the agn model . in this framework the computation of the density of agn has become an even more difficult task . in fact , it is not clear how to classify the sources and to take into account the selection biases introduced by the observation in the 2 - 10 kev range , where the absorption still play a relevant role . these recent deep surveys with _ chandra _ and _ xmm - newton _ have reached fluxes @xmath1 @xmath18 erg @xmath3 s@xmath4(2 - 10 kev ) in quite small areas ( less than 1 deg@xmath19 ) . as a consequence these surveys are not able to provide statistical significant samples at brighter fluxes ( @xmath20 erg @xmath3 s@xmath4 ; 5 - 10 kev ) where the density of sources is about 5/deg@xmath19 ( fiore et al . 2001 ) and tens of square degrees are to be covered . such data are necessary to provide large numbers of spectroscopic identified sources in a wide range of x - ray fluxes in order to cover as much as possible the @xmath21 plane and hence to derive their x - ray luminosity function ( lf ) . in this paper we report the results of the spectroscopic identifications of one of such brighter samples . the x - ray sources have been detected by the _ bepposax_-mecs instruments in the 5 - 10 kev band in the framework of the high energy llarge area survey ( hellas ) . preliminary results have been presented in fiore et al . ( 1999 ) and la franca et al . the whole survey and the catalogue is described by fiore et al . the data have been analyzed in the framework of the synthesis models for the x - ray background by comastri et al . ( 2001 ) , and the correlation with the soft x - rays has been investigated by vignali et al . ( 2001 ) . in section 2 we describe our x - ray and optical observations . in section 3 we present an analysis of the evolution of agn in the 2 - 10 kev band . because of the reasons previously described , the selection and definition of type 2/absorbed sources is still not clear , and thus we restricted our evolutionary studies to type 1 agn only . the results are discussed in section 4 . the spectroscopic follow up of the hellas sources has been carried out in a subsample enclosed in a region with @xmath22 , and outside @xmath23 and @xmath24 . in this region the number of sources is 118 out of a total of 147 . their flux distribution is shown in figure 1 and the sky coverage is shown in figure 2 and listed in table 1 . the _ bepposax _ x - ray positions have an uncertainty of about 1 - 1.5 arcmin , the larger at larger off - axis distances . we have thus searched for optical counterparts having r magnitude brighter than 21.0 in a circular region of 1 - 1.5 arcmin of radius around the hellas positions ( see below and section 3.1.1 for a discussion on the choice of this optical limit ) . in the case of large off - axis distances , the larger error - boxes ( 1.5@xmath25 ) have been used . 25 sources have been identified with cross - correlation with existing catalogues ( ned ) , and 49 have been investigated at the telescope . the total resulting sample of 74 sources has been built up in such a way that to a ) randomly sample the flux distribution of the parent catalogue of 118 sources ( see figure 1 ) , and b ) to avoid possible biases introduced by the cross - correlation with existing catalogues ( see appendix a for a detailed discussion ) . cl = 0.2 cm 1998 dec 28 : 3 nights & eso/3.6 m efosc2 1999 jun 8 : 3 nights & eso/3.6 m efosc2 2000 jan 3 : 3 nights & eso/3.6 m efosc2 2000 jul29 : 3 nights & eso/3.6 m efosc2 [ tab2 ] the follow - up observations have been carried out using the efosc2 at the 3.6 m eso telescope . previously published identifications ( fiore et al . 1999 ) have been carried out with the rc spectrograph ( rcsp ) at the kitt peak 4 m telescope , the fast spectrograph at the whipple 60@xmath15 telescope , and the hawaii 88@xmath15 telescope . long slit and multi object spectroscopy have been carried with resolution between 7 and 16 . the reduction process used standard midas facilities ( banse et al . the raw data were sky - subtracted and corrected for pixel - to - pixel variations by division with a suitably normalized exposure of the spectrum of an incandescent source ( flat - field ) . wavelength calibrations were carried out by comparison with exposures of he - ar , he , ar and ne lamps . relative flux calibration was carried out by observations of spectrophotometric standard stars ( oke 1990 ) . the campaign dates of the observations carried out at eso are listed in table 2 . we have divided the identified sources in agn1 , agn2 , galaxies with narrow emission lines showing moderate - to - high degree of excitation ( elg ) , bl lac objects , clusters of galaxies and stars . agn2 includes agn1.5 , agn1.8 and agn1.9 . no distinction has been made between radio - loud and radio quite objects . the radio properties of the hellas sources will be discussed in a forthcoming paper ( p. ciliegi , in preparation ) . the r - band magnitudes of these identifications have been pushed down to the limit of obtaining about less than 10@xmath26 probability of having a spurious identification , due to chance coincidences , for each of the hellas sources ( i.e. 90@xmath26 reliability for the whole sample ) . for agn1 we have chosen a limit of r=21 where the surface density is @xmath160 - 70/deg@xmath19 ( zitelli et al . tresse et al . ( 1996 ) found that 17@xmath26 of galaxies with @xmath27 host an emission line spectrum not typical of excitation from starburst ( but typical of agn2 activity ) , this percentage reduces to at minimum 8% if the maximum likely effect of absorption under the balmer lines is taken into account . they also found that the galaxies having forbidden emission lines without being agn2 are at least 36@xmath26 . these objects are mainly starburst galaxies and low ionization narrow emission line galaxies ( liner ) . we will call here all these sources simply emission line galaxies ( elg ) . we have assumed roughly constant these percentages and thus chosen a limit of r=19 for agn2 where the surface density of all galaxies is about 500/deg@xmath19 and consequently the expected density of agn2 is @xmath170 - 80/deg@xmath19 . for elg we have chosen a limit of r=17.5 where the surface density of all galaxies is about 160/deg@xmath19 , and therefore the expected density of elg is @xmath160/deg@xmath19 . the classification of the narrow emission line galaxies has been carried out using standard line ratio diagnostics ( tresse et al . some of the spectra showed h@xmath28 in emission and h@xmath29 having a ) very small emission , or b ) absent , or c ) small absorption . such spectra show however spectral features typical of strong - emission - line spectra such as [ oii]@xmath30 , [ oiii]@xmath31 , and [ sii]@xmath32 . in such cases the [ oiii]/h@xmath29 push the diagnostic ratio in the locus of agn2 , but stellar absorption at h@xmath29 could be significant affecting the classification . following tresse et al . ( 1996 ) , we have tested the classification under the assumption of a possible absorption of ew=2 over h@xmath29 . we have spectroscopically identified 36 out of the 49 hellas sources whose field has been investigated at the telescopes . for 13 sources we have found no optical counterparts brighter than the chosen magnitude limits ( see discussion above ) . the astrometric accuracy of the new generation x - ray telescopes ( _ chandra _ and _ xmm - newton _ ) would probably allow an unambiguous identification of the counterparts of these hellas sources , but in the framework of this work these fields have been declared _ empty_. these sources have harder x - ray spectra than the total sample . they have an average softness ratio of @xmath33 in comparison with the average value @xmath34 of the total sample of 118 sources , where s and h are the 1.3 - 4.5 kev the 4.5 - 10 kev counts , respectively ( see fiore et al . the absence of bright counterparts and the average harder x - ray spectrum favor the hypothesis that most of these sources preferably harbor absorbed agn . if these sources correspond to optical normal galaxies , such as those observed by _ chandra _ and _ xmm - newton _ ( see e.g. fiore et al . 2000 ; barger et al . 2001 ; tozzi et al . 2001 ; hornschemeier et al . 2001 ; comastri et al . 2002 ) , we would have not been able to properly identify them inside our error - boxes . however , our sample is statistically well defined , and from the number of empty fields we can directly derive an upper limit of @xmath118% ( 13/74 ) for the presence of these optical normal galaxies in our survey . the list of the observed hellas sources with their most probable spectroscopic identification is shown in table 3 . 80% of our identifications are inside an error - box of 60 arcsecs . 2 bright galaxies ( r@xmath115 - 16 ) are at a distance of about 110@xmath15 . the 13 empty fields are listed in table 4 . the optical spectra are shown in figure 3 ; as a reference , a list of the most typical emission lines for agn are over - plotted with the corresponding redshift . in table 5 the ew , fwhm , fwzi in the observed frame of the most relevant emission lines are listed . in total , 63@xmath26 ( 74/118 ) of our hellas subsample has been searched for spectroscopic identification . 61 have been identified : 37 agn1 , 9 agn2 , 5 elg , 6 clusters , 2 bl lac , 1 radio galaxy and 1 star . ccrrrllll = 0.2 cm 00 26 36.5 @xmath3519 44 14 & 00 26 36.1 @xmath3519 44 16 & 7 & 13 & 3.4 & 18.1 & agn2 & 0.238 + 00 27 09.9 @xmath3519 26 31 & 00 27 09.8 @xmath3519 26 14 & 18 & 6 & 1.8 & 17.7 & agn1 & 0.227@xmath36 + 00 45 49.6 @xmath3525 15 14 & 00 45 46.3 @xmath3525 15 50 & 59 & 24 & 3.3 & 17.5 & agn2 & 0.111 + 01 21 56.8 @xmath3558 44 05 & 01 21 56.9 @xmath3558 44 42 & 37 & 16 & 2.6 & 16.8 & agn2 & 0.118 + 01 34 33.3 @xmath3529 58 38 & 01 34 33.8 @xmath3529 58 16 & 24 & 6 & 0.9 & 18.0 & agn1 & 2.217 + 01 35 30.2 @xmath3529 51 22 & 01 35 32.7 @xmath3529 52 02 & 53 & 8 & 0.9 & 17.7 & agn1 & 1.344 + 01 40 08.9 @xmath3567 48 13 & 01 40 14.7 @xmath3567 48 54 & 53 & 8 & 2.8 & 12.4 & star & 0.000 + 02 42 01.8 @xmath3700 00 46 & 02 42 00.9 @xmath3700 00 22 & 29 & 10 & 1.5 & 18.6 & agn1 & 1.112@xmath38 + 03 15 45.0 @xmath3555 29 26 & 03 15 47.5 @xmath3555 29 04 & 31 & 15 & 2.7 & 17.9 & agn1 & 0.464@xmath38 + 03 17 32.4 @xmath3555 20 12 & 03 17 32.7 @xmath3555 20 26 & 14 & 21 & 4.1 & 17.5 & agn1 & 0.406@xmath38 + 03 33 09.6 @xmath3536 19 40 & 03 33 12.2 @xmath3536 19 48 & 33 & 15 & 4.0 & 17.5 & blac & 0.308@xmath38 + 03 34 07.4 @xmath3536 04 22 & 03 34 07.5 @xmath3536 04 04 & 19 & 5 & 1.9 & 20.1 & agn1 & 0.904 + 03 36 51.3 @xmath3536 15 57 & 03 36 54.0 @xmath3536 16 06 & 34 & 15 & 3.7 & 17.7 & agn1 & 1.537@xmath38 + 04 37 14.5 @xmath3547 30 58 & 04 37 11.8 @xmath3547 31 48 & 57 & 16 & 2.7 & 17.3 & agn1 & 0.142 + 04 38 47.9 @xmath3547 29 07 & 04 38 47.0 @xmath3547 28 02 & 67 & 20 & 4.7 & 20.5 & agn1 & 1.453 + 06 46 39.3 @xmath3544 15 35 & 06 46 37.6 @xmath3544 15 34 & 19 & 17 & 4.3 & 16.6 & agn1 & 0.153 + 07 21 29.6 @xmath3771 14 04 & 07 21 36.2 @xmath3771 13 24 & 52 & 8 & 0.8 & 17.7 & agn1 & 0.232@xmath38 + 07 41 40.3 @xmath3774 14 58 & 07 41 50.0 @xmath3774 14 48 & 41 & 23 & 30.7 & .... & clust . & 0.216@xmath38 + 07 43 09.1 @xmath3774 29 20 & 07 43 12.6 @xmath3774 29 36 & 22 & 7 & 6.0 & 16.4 & agn1 & 0.312@xmath38 + 08 37 37.2 @xmath3725 47 49 & 08 37 37.1 @xmath3725 47 52 & 4 & 12 & 2.6 & 16.9 & agn1 & 0.077 + 08 38 59.9 @xmath3726 08 14 & 08 38 59.2 @xmath3726 08 14 & 10 & 23 & 16.4 & 15.3 & elg & 0.048 + 10 32 15.8 @xmath3750 51 04 & 10 32 16.2 @xmath3750 51 20 & 18 & 10 & 3.1 & 15.9 & agn1 & 0.174 + 10 54 19.8 @xmath3757 25 09 & 10 54 21.2 @xmath3757 25 44 & 38 & 13 & 2.6 & 18.5 & agn2 & 0.205@xmath38 + 11 01 46.4 @xmath3772 26 11 & 11 01 48.8 @xmath3772 25 38 & 36 & 22 & 7.3 & 16.7 & agn1 & 1.460@xmath38 + 11 02 37.2 @xmath3772 46 38 & 11 02 36.8 @xmath3772 46 40 & 3 & 21 & 7.9 & 15.1 & agn1 & 0.089@xmath38 + 11 06 14.0 @xmath3772 43 16 & 11 06 16.1 @xmath3772 44 14 & 59 & 9 & 1.7 & 18.5 & agn1 & 0.680@xmath38 + 11 18 11.9 @xmath3740 28 33 & 11 18 13.8 @xmath3740 28 38 & 23 & 4 & 0.8 & 18.7 & agn1 & 0.387 + 11 18 46.2 @xmath3740 27 39 & 11 18 48.7 @xmath3740 26 48 & 59 & 5 & 1.4 & 18.5 & agn1 & 1.129 + 12 04 07.6 @xmath3728 08 31 & 12 04 04.0 @xmath3728 07 24 & 82 & 16 & 5.1 & .... & clust . & 0.167@xmath38 + 12 17 50.3 @xmath3730 07 08 & 12 17 52.1 @xmath3730 07 02 & 25 & 20 & 3.5 & 14.0 & bllac & 0.237@xmath38 + 12 18 55.0 @xmath3729 58 12 & 12 18 52.5 @xmath3729 59 02 & 59 & 13 & 2.0 & 18.6 & agn2 & 0.176 + 12 19 45.7 @xmath3747 20 43 & 12 19 52.3 @xmath3747 20 58 & 69 & 8 & 1.2 & 19.3 & agn1 & 0.654@xmath38 + 12 22 06.8 @xmath3775 26 17 & 12 22 07.0 @xmath3775 26 18 & 2 & 7 & 2.5 & .... & clust . & 0.240@xmath38 + 12 40 26.0 @xmath3505 13 20 & 12 40 27.8 @xmath3505 14 02 & 50 & 12 & 3.1 & 18.8 & agn1 & 0.300 + 12 40 29.6 @xmath3505 07 46 & 12 40 36.4 @xmath3505 07 52 & 102 & 17 & 1.9 & 15.2 & elg & 0.008 + 13 04 38.2 @xmath3510 15 47 & 13 04 35.6 @xmath3510 15 48 & 39 & 6 & 1.4 & 20.1 & agn1 & 2.386 + 13 05 32.3 @xmath3510 32 36 & 13 05 33.0 @xmath3510 33 20 & 45 & 22 & 19.3 & 14.9 & agn1 & 0.278@xmath38 + 13 36 34.3 @xmath3533 57 48 & 13 36 39.0 @xmath3533 57 58 & 60 & 22 & 3.2 & 10.5 & radiog . & 0.013@xmath38 + 13 42 59.3 @xmath3700 01 38 & 13 42 56.5 @xmath3700 00 58 & 59 & 21 & 3.2 & 18.7 & agn1 & 0.804@xmath38 + 13 48 20.8 @xmath3530 11 06 & 13 48 19.5 @xmath3530 11 56 & 52 & 14 & 2.2 & 15.3 & agn2 & 0.128 + 13 48 45.4 @xmath3530 29 37 & 13 48 44.7 @xmath3530 29 46 & 13 & 14 & 5.1 & 17.1 & agn1 & 0.330 + 13 50 09.4 @xmath3530 19 55 & 13 50 15.4 @xmath3530 20 10 & 80 & 11 & 5.1 & 16.5 & elg & 0.074 + 13 53 54.6 @xmath3718 20 33 & 13 53 54.4 @xmath3718 20 16 & 18 & 18 & 6.8 & 17.3 & agn1 & 0.217 + 14 17 12.5 @xmath3724 59 29 & 14 17 18.8 @xmath3724 59 30 & 86 & 13 & 0.7 & 19.5 & agn1 & 1.057@xmath38 + 14 18 31.1 @xmath3725 11 07 & 14 18 31.2 @xmath3725 10 50 & 17 & 9 & 6.1 & .... & clust . & 0.240@xmath38 + 15 19 39.9 @xmath3765 35 46 & 15 19 33.7 @xmath3765 35 58 & 41 & 14 & 9.4 & 14.4 & agn2 & 0.044 + 15 28 47.3 @xmath3719 39 10 & 15 28 47.7 @xmath3719 38 54 & 18 & 5 & 1.6 & 20.3 & agn1 & 0.657@xmath39 + 16 26 59.9 @xmath3755 28 21 & 16 26 59.0 @xmath3755 27 24 & 57 & 11 & 12.1 & .... & clust . & 0.130@xmath38 + 16 34 11.8 @xmath3759 45 29 & 16 34 18.5 @xmath3759 45 44 & 53 & 3 & 0.8 & 19.0 & agn2 & 0.341@xmath40 + 16 49 57.9 @xmath3704 53 33 & 16 50 00.0 @xmath3704 54 00 & 42 & 20 & 9.5 & .... & clust . & 0.154@xmath38 + 16 50 40.1 @xmath3704 37 17 & 16 50 42.7 @xmath3704 36 18 & 71 & 25 & 12.2 & 14.6 & agn2 & 0.031 + 16 52 38.0 @xmath3702 22 18 & 16 52 37.5 @xmath3702 22 06 & 14 & 5 & 0.7 & 20.7 & agn1 & 0.395 + 20 42 47.6 @xmath3510 38 31 & 20 42 53.0 @xmath3510 38 26 & 80 & 21 & 5.3 & 17.9 & agn1 & 0.363 + 20 44 34.8 @xmath3510 27 34 & 20 44 34.8 @xmath3510 28 08 & 35 & 15 & 2.0 & 17.7 & agn1 & 2.755 + 22 26 30.3 @xmath3721 11 57 & 22 26 31.5 @xmath3721 11 38 & 26 & 14 & 3.9 & 17.6 & agn1 & 0.260 + 23 15 36.4 @xmath3559 03 40 & 23 15 46.8 @xmath3559 03 14 & 85 & 8 & 2.0 & 11.2 & elg & 0.044@xmath38 + 23 19 22.1 @xmath3542 41 50 & 23 19 32.0 @xmath3542 42 28 & 116 & 21 & 5.7 & 16.5 & elg@xmath41 & 0.101 + 23 27 28.7 @xmath3708 49 30 & 23 27 28.7 @xmath3708 49 26 & 4 & 7 & 0.5 & 18.5 & agn1@xmath42 & 0.154 + 23 29 02.4 @xmath3708 34 39 & 23 29 05.8 @xmath3708 34 16 & 56 & 20 & 2.9 & 20.3 & agn1 & 0.953 + 23 31 55.6 @xmath3719 38 34 & 23 31 54.3 @xmath3719 38 36 & 19 & 17 & 3.7 & 18.8 & agn1 & 0.475 + 23 55 53.3 @xmath3728 36 06 & 23 55 54.3 @xmath3728 35 58 & 16 & 12 & 4.2 & 17.9 & agn1 & 0.731@xmath38 + [ tab3 ] cc = 0.2 cm 01 34 28.6 @xmath3530 06 04 & 1.33 + 01 34 49.6 @xmath3530 02 34 & 0.71 + 09 46 05.3 @xmath3514 02 59 & 2.82 + 09 46 32.8 @xmath3514 06 16 & 3.22 + 13 04 24.3 @xmath3510 23 53 & 1.28 + 13 04 45.1 @xmath3505 33 37 & 1.26 + 13 48 24.3 @xmath3530 25 47 & 3.15 + 14 11 58.7 @xmath3503 07 02 & 3.93 + 22 03 00.5 @xmath3532 04 18 & 2.83 + 22 31 49.6 @xmath3711 32 08 & 1.94 + 23 02 30.1 @xmath3708 37 06 & 2.67 + 23 02 36.2 @xmath3708 56 42 & 3.17 + 23 16 09.8 @xmath3559 11 24 & 1.32 + [ tab4 ] because of the difficulties previously described in quantifying all the selection biases in building up samples of type 2 ( and/or absorbed ) agn , we have decided to limit our evolutionary analysis to agn1 only . see appendix a for a discussion on the absence of substantial biases on our sample of agn1 . we have assumed h@xmath43=50 km / s / mpc and either the ( @xmath12,@xmath13)=(1.0,0.0 ) or the ( @xmath12,@xmath13)=(0.3,0.7 ) cosmologies . we have computed the 2 - 10 kev luminosity for agn1 assuming a typical spectral energy distribution as computed by pompilio , la franca , & matt ( 2000 ) , which assumes a power slope in energy with index @xmath44=0.9 ( @xmath45 ) and takes into account the reflection . this spectrum is roughly approximated by a single power @xmath44=0.6 in the range 2 - 50 kev ( i.e. up to @xmath46=4 ) . this is in agreement with the average slope of the spectra of agn1 from the hellas sample ( vignali et al . 2001 ) . as our spectroscopic identifications are limited to magnitudes brighter than @xmath47 , in order to estimate the density of agn1 , we have evaluated which is their completeness in our sample . we have assumed that all agn1 follow the same distribution of the ratio of the 5 kev to r - band optical luminosity log(l@xmath48/l@xmath49 ) without dependencies on luminosity and/or redshift . we have estimated the relationship between optical and hard x - ray luminosity for agn1 as follow . we selected a mixed sample of both 25 optical and 25 hard x - ray selected type 1 agn , mostly in the redshift range 0@xmath50@xmath46@xmath502 , from mineo et al . ( 2000 ) , george et al . ( 2000 ) and akiyama et al . the 25 optically selected agn1 have an average log(l@xmath48/l@xmath49 ) ratio of -4.78 with a standard deviation of 0.59 , while the 25 x - ray selected agn1 have an average log(l@xmath48/l@xmath49 ) ratio of -4.38 with a standard deviation of 0.58 . in figure 4 the histogram of the ratio of 5 kev and r - band optical luminosity of the total sample of 50 agn1 at @xmath46=0 is shown . we fitted the distribution with a gaussian having mean -4.58 and standard deviation 0.61 . we have estimated that with this distribution , a limit of @xmath47 corresponds to only 1 agn1 lost from our hellas survey , and consequently we can say that a limit of @xmath47 assure a high level of completeness for the identification of agn1 at the fluxes of our hellas survey . at variance , as already discussed , our hellas identification programme is biased against absorbed type 2 sources which probably harbor in the 13 empty fields found . although the incompleteness for agn1 in the hellas sample is quite low , during the computation of the lf of agn1 , we have estimated the correct area coverage for our survey by multiplying the area coverage corresponding to each point of the @xmath21 plane for : a ) the spectroscopic completeness of the hellas sample ( 74/118 ) , and b ) the fraction of found agn1 as a function of @xmath46 , as described before . in order to increase the statistical significance of our analysis , the hellas data have been combined to other hard x - ray samples of agn1 identified by grossan ( 1992 ) , boyle et al . ( 1998 ) , and akiyama et al . ( 2000 ) . the sample of grossan ( 1992 ) consists of 84 agn1 and 12 agn2 detected by _ predominantly at low redshift @xmath27 . the sample covers an area of 26919 deg@xmath19 down to a flux limit of @xmath51 erg @xmath3 s@xmath4 . the quoted fluxes at 5 kev have been converted to fluxes in the 2 - 10 kev assuming @xmath52 . the sample of boyle et al . ( 1998 ) consists of 12 agn1 and 6 agn2 detected by _ asca_. 2 - 10 kev counts have been converted to fluxes with a count - rate - to - flux conversion factor of @xmath53 erg @xmath3 s@xmath4/count , assuming @xmath52 ( georgantopoulos et al . , 1997 ) . the sky coverage has been taken from georgantopoulos et al . ( 1997 , see their figure 1 ) . the sample of akiyama et al . ( 2000 ) consists of 25 agn1 , 5 agn2 , 2 clusters , 1 star and 1 unidentified source . the sources have been detected by _ asca_. in order to convert the counts of the sky coverage presented in their table 1 , a conversion has been applied assuming that 1 count corresponds to @xmath54 erg @xmath3 s@xmath4 in the 2 - 10 kev band for an x - ray source with a slope @xmath44=0.6 . a total of 158 agn1 have been used in the redshift range 0@xmath50@xmath46@xmath503 . their luminosity - redshift distribution is shown in figure 5 . in this distribution a gap between the sample of grossan from _ hea01 _ and the other fainter samples from _ asca _ and _ bepposax _ is evident . this is because surveys at fluxes in the range @xmath55@xmath56 erg @xmath3 s@xmath4 in the 2 - 10 kev energy band where some hundreds of deg@xmath19 need to be covered are still missing . these surveys are in program with _ xmm - newton _ ( see e.g. barcons et al . 2001 ) . the best - fit parameters for the 2 - 10 kev lf and its cosmological evolution have been derived by minimization of the @xmath57 computed by comparison of the observed and expected number of agn1 in 6@xmath583 bins in the @xmath21 plane . the adopted luminosity / redshift grid was : 6 log@xmath16 bins , with @xmath59log@xmath16=1 in the range 42.2 - 48.2 , and 3 redshift bins ( 0.0 - 0.2 , 0.2 - 1.0 , 1.0 - 3.0 ) . table 6 summarizes the results for a number of different models . the errors quoted for the parameters are 68@xmath26 ( 1@xmath60 ) confidence intervals . they correspond to variations of @xmath61=1.0 , obtained perturbing each parameter in turn with respect to its best - fit value , and looking for a minimization with the remaining parameters free to float . the fitted lf have been tested with the 2d kolmogorov - smirnov ( 2dks ) test . two different functional forms have been used . following ceballos and barcons ( 1996 ) and boyle et al . ( 1998 ) , we used the pure luminosity evolution ( ple ) for the agn1 lf . we also used the luminosity dependent density evolution ( ldde ) model similar to the one fitted in the soft x - rays by miyaji et al . ( 2000 ) . the local ( @xmath46=0 ) agn1 lf used for the ple model has been represented by a two - power - law : @xmath62 where @xmath63 is expressed in units of @xmath64 . a standard power - law luminosity evolution model was used to parameterize the cosmological evolution of this lf : @xmath65 . in the case of the ldde model , as an analytical expression of the present day ( @xmath46=0 ) luminosity function , we used the smoothly connected two power - law form : @xmath66^{-1}.\ ] ] the description of the evolution is given by a factor @xmath67 such that : @xmath68 where @xmath69 } ) } & ( z \leq z_{\rm c } ; l_{\rm x}<l_{\rm a } ) \\ ( 1+z)^{p1 } & ( z \leq { z_{\rm c } } ; l_{\rm x}\ge l_{\rm a})\nonumber \\ e(l_{\rm x},{z_{\rm c } } ) \left[(1+z)/(1+{z_{\rm c } } ) \right]^{p2 } & ( z>{z_{\rm c } } ) . \\ \end{array } \right . \ ] ] lclllllllr = 0.2 cm 0 ) ple from boyle & 1.0,0.0 & 1.73 & 2.96 & 44.16 & 2.00 & ... & @xmath70 & @xmath71 & 24 + 1 ) ple & 1.0,0.0 & 1.83 & 3.00 & 44.17 & 2.12 & ... & @xmath72 & @xmath73 & 35 + 2 ) ple with @xmath74 & 1.0,0.0 & 1.87 & 3.03 & 44.17 & 2.52 & 1.39 & @xmath75 & @xmath76 & 34 + 3 ) ldde & 1.0,0.0 & 0.68 & 2.02 & 44.03 & 4.66 & 1.55@xmath40 & @xmath77 @xmath78 & @xmath79 & 40 + 4 ) ple & 0.3,0.7 & 1.93 & 2.97 & 44.24 & 2.22 & ... & @xmath80 & @xmath79 & 56 + 5 ) ple with @xmath74 & 0.3,0.7 & 1.94 & 2.97 & 44.24 & 2.26 & 2.38 & @xmath80 & @xmath81 & 54 + 6 ) ldde & 0.3,0.7 & 0.69 & 1.95 & 43.96 & 4.56 & 1.58@xmath40 & @xmath82 @xmath41 & @xmath83 & 60 + + 7 ) ldde softx & 1.0,0.0 & 0.62@xmath40 & 2.25@xmath40 & 44.13 & 5.40@xmath40 & 1.55@xmath40 & @xmath84 @xmath85 & @xmath86 & 48 + + 68@xmath26 confidence errors & & @xmath87 & @xmath88 & @xmath89 & @xmath90 & @xmath91 & 8@xmath26 + [ tab6 ] the parameters are defined as in miyaji et al . in particular , @xmath44 is proportional to the decrease of the density evolution of the faint objects . in this parameterization of the lf @xmath16 is expressed in units of @xmath92 . in both models the lf has been computed for luminosities brighter than logl@xmath93=42.2 . using this parameterization , for ( @xmath12,@xmath13)=(1.0,0.0 ) , a fit was done by keeping fixed the parameters @xmath94=@xmath95 ( the redshift at which the evolution stops ) , @xmath44=@xmath96 , and the ratio @xmath97 to the value found by miyaji et al . ( 2000 ) for agn1 only ( their ldde1 model ) in the soft x - rays , and leaving all the remaining parameters ( the two slopes @xmath98 and @xmath99 , the break luminosity @xmath100 , and the speed of the density evolution @xmath101 ) free to vary . in the case of an ( @xmath12,@xmath13)=(0.3,0.7 ) universe , as the fit for agn1 only from miyaji et al . ( 2000 ) was not available , their parameters from the fit for all agns ( agn1+agn2 ) have been used : @xmath94=@xmath102 , @xmath44=@xmath103 , @xmath104 . we started our computation of the lf in an ( @xmath12,@xmath105)=(1.0,0.0 ) cosmology . previous estimates of the shape and evolution of the lf of agn1 in the 2 - 10 kev range are from ceballos and barcons ( 1996 ) , and boyle et al . boyle et al . ( 1998 ) combined the local sample of 84 agn1 observed by _ heao1 _ from grossan ( 1992 ) with a fainter sample of 12 agn1 observed by _ asca _ ( see their distribution in the l@xmath106-@xmath46 plane in figure 4 ) . they found that the 2 - 10 kev agn x - ray lf is best represented by a two - power - law function evolving according to a pure luminosity evolution ( ple ) model : @xmath5@xmath6@xmath107 ( see model 0 in table 6 ) . in figure 6 the lf from only the 37 agn1 from hellas in three redshift intervals ( 0.0@xmath50@[email protected] , 0.2@xmath50@[email protected] , and 1.0@xmath50@[email protected] ) is shown . for the sake of comparison , the best - fit lf from boyle et al . ( 1998 ) is also shown . the data have been represented by correcting for evolution within the redshift bins as explained in la franca & cristiani ( 1997 ) . the error bars in the figures are based on poisson statistics at the 68@xmath26 confidence level . the data are in rough agreement with the previous estimate of boyle et al . ( 1998 ) , but show an excess of faint agn1 at redshift larger than @[email protected] . this feature is still present after combining our data with the other agn1 samples from grossan ( 1992 ) , boyle et al . ( 1998 ) , and akyiama et al . ( 2000 ) ( figure 7 ) . as already discussed , these samples all together collect 158 agn1 . these data have a @xmath57 probability of 0.04 to be drawn from a ple model such as that computed by boyle et al . ( 1998 ) ( model 0 in table 6 ) . we did not use the 2dks as this test is not appropriate in this case : it uses the cumulative distributions in the @xmath21 plane regardless of the normalization of the lf . our best - fit to the data with a ple model found a lf with similar slopes and break luminosity as that one of boyle et al . ( 1998 ) but with a slightly larger evolution ( @xmath108 instead of @xmath8=@xmath109 ) , and a significantly 20(@xmath110)@xmath26 larger normalization ( the normalization has a poisson uncertainty of 8@xmath26 . see model 1 in table 6 ) . our larger values of the evolution parameter @xmath8 and normalization @xmath111 are originated by the necessity of better fitting the observed higher density of faint agn1 at high redshift . the 2dks test gives a probability of 0.22 for this fit ( see table 6 ) . a even better probability of 0.31 is obtained if a stop in the evolution is applied at redshift @xmath74=@xmath112 ( model 2 in table 6 ) and a larger evolution ( @xmath8=@xmath113 ) is used . although our fits are already statistically adequate , our ple models do not fully describe an over - density of faint agn1 which is observed at high redshift . the data are not sufficiently faint to properly probe this part of the lf , but as this feature is similar to what observed in the soft x - rays ( miyaji et al . 2000 ) , we tried to obtain a even better fit of the data , by using the luminosity dependent density evolution ( ldde ) model similar to the one fitted in the soft x - rays , as described in the previous section . this model differentiates from the ple especially in the faint part of the lf , at luminosities lower than @xmath100 . in this part of the lf the density evolution decreases in proportion to the faintness of the objects . as our data just start to probe the part of the lf which is fainter than @xmath100 , we can not expect to find directly a fit to all the parameters of the ldde model to our data , and we limited our analysis to a check of the compatibility ( with limited changes ) of the ldde parameters found by miyaji et al . ( 2000 ) to our data . with this model we found the best - fit to our data ( model 3 : ldde , figure 8) by keeping fixed the parameters describing the stop in the evolution ( @xmath94=@xmath95 ) , and the dependence on luminosity of the density evolution ( @xmath44=@xmath96 ) to the value found by miyaji et al . ( 2000 ) for agn1 only , and leaving all the remaining parameters free to vary ( see section 3.2 ) . the 2dks test gives , for this model , a probability of 0.47 . if an ( @xmath12,@xmath13)=(0.3,0.7 ) cosmology is assumed , the ple models provide an ever better representation of the data in comparison with what found in the ( @xmath12,@xmath13)=(1.0,0.0 ) universe ( see figure 9 ) . the 2dks probability is 0.70 and 0.47 , with and without the introduction of the @xmath74 parameter , respectively . in this case even the simple ple model obtain a quite good fit of the data , and the introduction of the @xmath74 parameter is necessary to stop the evolution only at redshifts larger than 2.4 . however our data contain not enough agn1 at redshift larger than 2 in order to obtain an accurate measure of the @xmath74 parameter ( see figure 4 ) , therefore the errors in this parameter are quite large . also the ldde model obtain a satisfactory fit of the data ( see figure 10 ) . thanks to the identification of 61 sources of the hellas sample we have been able to double the number of hard x - ray agn1 available for statistical analysis at fluxes in the range @xmath114@xmath55 erg @xmath3 s@xmath4 . in total we can use 74 agn1 at these fluxes ( 37 from hellas ) , which combined with the local sample of grossan have allowed to show directly the shape of the lf of agn1 as function of redshift and measure its evolution . the ple models provide satisfactory fits of the data both in the ( @xmath12,@xmath13)=(1.0,0.0 ) and in the ( @xmath12,@xmath13)=(0.3,0.7 ) cosmologies . our estimate of the lf in the ( @xmath12,@xmath13)=(1.0,0.0 ) has a significantly larger normalization in comparison to the previous measure from boyle et al . ( 1998 ) . the data start to probe in the hard x - rays the faint part of the lf where the excess of density of agn1 has been observed in the soft x - rays , justifying the implementation of the ldde models . however , in both cosmologies , the statistic is not significant enough to distinguish between the ple and ldde models ( see table 6 ) . in fact , in the prediction of the differential counts of agn1 shown in figure 11 , the models differentiates at fluxes fainter than @xmath115 erg @xmath3 s@xmath4 , where the statistic is still poor . the new upcoming fainter surveys from _ chandra _ and _ xmm - newton _ will easily test which model is correct . in table 6 the percentages of the contribution to the 2 - 10 kev x - ray background are shown . the x - ray background has been computed integrating the lf up to @xmath116 for @xmath16@xmath11710@xmath118 @xmath119 . at variance with the results of 24% obtained from the evolution of the lf derived by boyle et al . ( 1998 ) , our models reproduces from @xmath135% up to @xmath160% of the xrb . we used @xmath120 erg @xmath3 s@xmath4 deg@xmath121 from chen , fabian and gendrau ( 1997 ) . the highest percentages would probably imply that part of the absorbed population necessary to reproduce the xrb is already included in the agn1 at high redshift . however , more detailed analysis of this issue are beyond the scope of this paper . the percentages would decreas if a value f @xmath122 erg @xmath3 s@xmath4 deg@xmath121 from vecchi et al . ( 1999 ) is assumed . it is interesting to notice that agn1 in the 2 - 10 kev range show an evolution up to @xmath123 which is fairly well compatible with what observed in the soft x - rays . in fact we found a good fit of the data in the ( @xmath12,@xmath13)=(1.0,0.0 ) universe if we assume exactly the same parameters found in the soft x - rays by miyaji et al . ( 2000 ) for agn1 , by only looking for a fit with the break luminosity @xmath100 ( model 7 in table 6 ) . miyaji et al . ( 2000 ) found log@xmath100=43.78 . the value @xmath100 found by our fit in the 2 - 10 kev band is log@xmath100=44.13 . this luminosity difference implies an x - ray spectrum for agn1 with slope @xmath44=0.6 , which is the same slope used in our computation of the lf . we note however that miyaji et al . ( 2000 ) used a slope @xmath44=1.0 in their computations of the soft x - ray lf . therefore , taken at a face value , the match between the soft and hard x - rays lfs implies that agn1 have a broad band concave spectrum getting steeper going toward lower energies . as already discussed in the previous section , the ( @xmath12,@xmath13)=(0.3,0.7 ) cosmology for agn1 alone has not been analyzed by miyaji et al . ( 2000 ) . the agreement of the evolution measured in the soft and hard x - rays is also shown in figure 12 , where the evolution of the density of agn1 brighter than logl@xmath93=44.8 is shown . the soft x - ray data from miyaji et al . ( 2000 ) have been over - plotted assuming a slope @xmath52 , which corresponds to a limit logl@xmath106(0.5 - 2 kev)@xmath11744.5 . the continuous line are the predictions of our ldde model for ( @xmath12,@xmath13)=(1.0,0.0 ) . based on observations collected at the european southern observatory , chile , eso n@xmath124 : 62.p-0783 , 63.o-0117(a ) , 64.o-0595(a ) , 65.o-0541(a ) . this research has made use of the nasa / ipac extragalactic database ( ned ) which is operated by the jet propulsion laboratory , california institute of technology , under contract with the national aeronautics and space administration . this research has been partially supported by asi contract ars-99 - 75 , murst grants cofin-98 - 02 - 32 , cofin-99 - 034 , cofin-00 - 02 - 36 and a 1999 cnaa grant . as already described , 25 sources have been identified by cross - correlation with existing catalogues . however , as the sample is not fully spectroscopic identified , the cross - correlation with the existing catalogues could alter the average characteristics ( fluxes , percentages of classes of counterparts , etc . ) of the subsample of identified sources . namely , the subsample could not be representative of the whole sample . let s explain this with an example . in an ideal sample of 50 agn1 and 50 agn2 a random identification of a subsample of 40 sources will identify about 50@xmath26 agn1 and 50@xmath26 agn2 according to sampling errors . but if a cross - correlation is first made with a catalogue of only agn1 ( let say 30 ) and later only 10 sources are randomly identified , the fraction of agn1 in the total identified subsample of 40 sources will be artificially increased ( they will be at minimum 75@xmath26 ) . our subsample is risking the same sort of bias , and we wish to quantify it . a kolmogorov - smirnov ( ks ) two - sample test gives a 9@xmath26 probability that the x - ray flux distribution of the 25 sources identified through the cross - correlation with catalogues belong to the same parent population of the total sample of 118 sources . although not significant , this low probability is due to the average slightly brighter fluxes of the cross - correlated subsample ( @xmath125 ) in comparison to the total sample ( @xmath126 ) . we have thus tried to populate the faintest bins during the observing runs at the telescope , and , indeed , the ks test gives a 66@xmath26 probability that the whole subsample of the 74 identified sources ( included the empty fields ) belong to the same parent population of the total sample of 118 sources ( see figure 1 ) . in this way , our observing runs at the telescope have recovered the possible alteration on the fraction of classes of sources which are identified . as the catalogues used for the cross correlation are mainly populated by agn1 ( we used ned ) , their fraction in our sample , as explained before , could be artificially increased . agn1 are 14 out of the 25 cross - correlated sources ( 56@xmath26 ) . the fraction of agn1 is , as expected ( but not significantly ) , lower for the sources observed at the telescope : 23 out of the 49 ( 47@xmath26 ) . the two values are not statistically distinguishable from the observed fraction 50@xmath26 of agn1 in the whole sample , which we thus consider representative of the whole hellas sample . la franca f. , fiore f. , vignali c. , comastri a. , pompilio f. , 2001 , in asp conf . 232 , the new era of wide field astronomy , ed . clowes , a.j . adamson & g.e . bromage ( san francisco : asp ) , 96 , preprint(astro - ph/00111008 ) = 0.2 cm -10.34 & 55.51 -11.85 & 55.51 -11.89 & 55.48 -11.93 & 55.39 -11.97 & 55.28 -12.00 & 55.05 -12.04 & 54.66 -12.08 & 54.04 -12.11 & 53.16 -12.15 & 51.97 -12.19 & 50.63 -12.22 & 49.04 -12.26 & 47.13 -12.30 & 44.88 -12.34 & 42.44 -12.37 & 39.93 -12.41 & 37.49 -12.45 & 34.88 -12.48 & 31.93 -12.52 & 28.83 -12.56 & 25.67 -12.59 & 22.75 -12.63 & 20.03 -12.67 & 17.53 -12.70 & 15.23 -12.74 & 13.16 -12.78 & 11.32 -12.82 & 9.73 -12.85 & 8.30 -12.89 & 6.96 -12.93 & 5.67 -12.96 & 4.56 -13.00 & 3.60 -13.04 & 2.88 -13.07 & 2.35 -13.11 & 1.96 -13.15 & 1.60 -13.19 & 1.27 -13.22 & 0.94 -13.26 & 0.68 -13.30 & 0.47 -13.33 & 0.33 -13.37 & 0.22 -13.41 & 0.14 -13.44 & 0.08 -13.48 & 0.05 -13.52 & 0.03 -13.56 & 0.01 -13.59 & 0.00 [ tab1 ] = 0.2 cm h002636@xmath35194416 & 0.238 & fwzi & ... & 62.1 & 38.1 & 62.1 & ... & ... + & & fwhm & ... & 19.7 & 15.4 & 18.5 & ... & ... + & & ew & ... & 82.4 & 14.6 & 133.1 & ... & ... + + h004546@xmath35251550 & 0.111 & fwzi & ... & ... & ... & 40.1 & 104.2 & 44.1 + & & fwhm & ... & ... & ... & 14.6 & 69.7 & 50.4 + & & ew & ... & -1.3 & -0.2 & 5.6 & 22.2 & 2.5 + + h012157@xmath35584442 & 0.118 & fwzi & ... & 44.1 & 38.1 & 64.1 & 120.2 & 48.1 + & & fwhm & ... & 19.7 & 13.1 & 16.8 & 45.0 & 80.0 + & & ew & ... & 30.4 & 2.9 & 21.4 & 53.5 & 20.8 + + h013434@xmath35295816 & 2.217 & fwzi & @xmath127452.7 & ... & ... & ... & ... & ... + & & fwhm & @xmath127123.6 & ... & ... & ... & ... & ... + & & ew & @xmath127156.5 & ... & ... & ... & ... & ... + + h013533@xmath35295202 & 1.344 & fwzi & 282.5 & ... & ... & ... & ... & ... + & & fwhm & 67.6 & ... & ... & ... & ... & ... + & & ew & 61.2 & ... & ... & ... & ... & ... + + h033408@xmath35360403 & 0.904 & fwzi & 412.6 & 40.1 & ... & ... & ... & ... + & & fwhm & 174.0 & 26.5 & ... & ... & ... & ... + & & ew & 120.5 & 6.1 & ... & ... & ... & ... + + h043712@xmath35473148 & 0.142 & fwzi & ... & 36.0 & 122.2 & 64.1 & 162.3 & 70.1 + & & fwhm & ... & 16.4 & 36.4 & 20.9 & 39.7 & 73.0 + & & ew & ... & 5.2 & 65.4 & 23.6 & 241.2 & 17.0 + + h043847@xmath35472802 & 1.453 & 372.6 & ... & ... & ... & ... & ... + & & 64.9 & ... & ... & ... & ... & ... + & & 61.4 & ... & ... & ... & ... & ... + + h064638@xmath35441534 & 0.153 & fwzi & ... & 60.1 & 128.2 & 60.1 & 122.2 & 46.1 + & & fwhm & ... & 18.3 & 47.2 & 21.3 & 36.1 & 29.0 + & & ew & ... & 7.9 & 61.0 & 78.9 & 168.4 & 10.0 + + h083737@xmath37254752 & 0.077 & fwzi & ... & 52.1 & 69.4 & 56.8 & 327.2 & 60.1 + & & fwhm & ... & 40.3 & 28.8 & 15.2 & 71.8 & 43.0 + & & ew & ... & 51.8 & 16.0 & 11.3 & 291.6 & 17.0 + + h083859@xmath37260814 & 0.048 & fwzi & ... & ... & 36.7 & 40.1 & 110.2 & 70.1 + & & fwhm & ... & ... & 11.7 & 23.2 & 42.1 & 32.2 + & & ew & ... & -8.9 & 1.9 & 4.1 & 46.0 & 6.7 + + h103216@xmath37505120 & 0.174 & fwzi & ... & 30.0 & 203.7 & 36.7 & 150.2 & 33.4 + & & fwhm & ... & 12.0 & 68.1 & 13.2 & 79.9 & 15.0 + & & ew & ... & 5.8 & 65.1 & 13.1 & 163.2 & 6.5 + + h111814@xmath37402838 & 0.387 & fwzi & 340.5 & 50.1 & ... & 40.1 & ... & ... + & & fwhm & 184.0 & 46.7 & ... & 10.3 & ... & ... + & & ew & 393.8 & 6.7 & -0.7 & 6.7 & ... & ... + + h111849@xmath37402648 & 1.129 & fwzi & @xmath127200.3 & ... & ... & ... & ... & ... + & & fwhm & @xmath12765.6 & ... & ... & ... & ... & ... + & & ew & @xmath12759.9 & ... & ... & ... & ... & ... + + h121853@xmath37295902 & 0.176 & fwzi & ... & 43.4 & 36.1 & 56.8 & 153.6 & 50.1 + & & fwhm & ... & 12.4 & 12.1 & 12.5 & 26.8 & 20.9 + & & ew & ... & 25.4 & 17.8 & 199.6 & 186.1 & 22.6 + + h124028@xmath35051402 & 0.300 & fwzi & ... & 80.1 & 166.9 & 63.4 & ... & ... + & & fwhm & ... & 25.6 & 61.3 & 23.9 & ... & ... + & & ew & ... & 36.1 & 90.8 & 67.1 & ... & ... + [ tab5 ] = 0.18 cm h124036@xmath35050752 & 0.008 & fwzi & ... & ... & 20.0 & -10.0 & 90.1 & 76.8 + & & fwhm & ... & ... & 13.9 & -7.0 & 34.0 & 26.0 + & & ew & ... & ... & 0.8 & -0.4 & 26.3 & 6.8 + + h130436@xmath35101549 & 2.386 & fwzi & @xmath36468.7 & ... & ... & ... & ... & ... + & & fwhm & @xmath127128.4 & ... & ... & ... & ... & ... + & & ew & @xmath127537.0 & ... & ... & ... & ... & ... + + h134820@xmath35301156 & 0.128 & fwzi & ... & 52.1 & ... & 56.1 & 112.2 & ... + & & fwhm & ... & 21.5 & ... & 22.0 & 53.9 & ... + & & ew & ... & 19.0 & -0.4 & 12.7 & 63.0 & ... + + h134845@xmath35302946 & 0.330 & fwzi & ... & 36.0 & 220.4 & 72.1 & ... & ... + & & fwhm & ... & 15.8 & 83.0 & 20.5 & ... & ... + & & ew & ... & 2.3 & 84.2 & 55.8 & ... & ... + + h135015@xmath35302010 & 0.074 & fwzi & ... & 60.1 & ... & 52.1 & 168.3 & 56.1 + & & fwhm & ... & 23.1 & ... & 24.5 & 64.4 & 31.7 + & & ew & ... & 100.5 & -0.6 & 16.0 & 48.8 & 8.0 + + h135354@xmath37182016 & 0.217 & fwzi & ... & 52.1 & 24.0 & 64.1 & 252.4 & 40.1 + & & fwhm & ... & 13.7 & 9.3 & 19.7 & 95.0 & 47.9 + & & ew & ... & 47.3 & 2.1 & 22.8 & 154.0 & 12.4 + + h151934@xmath37653558 & 0.044 & fwzi & ... & 36.0 & ... & 44.1 & 92.2 & 44.1 + & & fwhm & ... & 13.6 & ... & 9.7 & 9.0 & 14.5 + & & ew & ... & 43.6 & -1.3 & 22.6 & 26.4 & 6.1 + + h163419@xmath37594504 & 0.341 & fwzi & 144.2 & 48.1 & 72.1 & 56.1 & ... & ... + & & fwhm & 19.9 & 12.9 & 12.9 & 13.7 & ... & ... + & & ew & 33.1 & 19.7 & 30.8 & 267.2 & ... & ... + + h165043@xmath37043618 & 0.031 & fwzi & ... & 48.1 & 36.1 & 52.1 & 124.2 & 56.1 + & & fwhm & ... & 40.1 & 27.2 & 21.0 & 49.1 & 32.6 + & & ew & ... & 45.9 & 2.8 & 26.6 & 43.0 & 12.3 + + h165238@xmath37022206 & 0.395 & fwzi & 184.3 & 56.1 & 164.3 & 64.1 & ... & ... + & & fwhm & 70.0 & 24.1 & 129.5 & 21.5 & ... & ... + & & ew & 66.6 & 22.5 & 37.5 & 68.5 & ... & ... + + h204253@xmath35103826 & 0.363 & fwzi & ... & 72.1 & 184.3 & 76.1 & ... & ... + & & fwhm & ... & 22.5 & 48.5 & 22.3 & ... & ... + & & ew & ... & 15.4 & 73.0 & 201.8 & ... & ... + + h204435@xmath35102808 & 2.755 & fwzi & @xmath127248.4 & ... & ... & ... & ... & ... + & & fwhm & @xmath127101.9 & ... & ... & ... & ... & ... + & & ew & @xmath12756.6 & ... & ... & ... & ... & ... + + h222632@xmath37211138 & 0.260 & fwzi & ... & ... & 125.5 & 26.7 & ... & ... + & & fwhm & ... & ... & 41.6 & 7.9 & ... & ... + & & ew & ... & ... & 36.4 & 9.3 & ... & ... + + h231932@xmath35424228 & 0.101 & fwzi & ... & 68.1 & 40.1 & 56.1 & 172.3 & 64.1 + & & fwhm & ... & 23.1 & 26.3 & 21.8 & 46.9 & 38.4 + & & ew & ... & 20.4 & 5.0 & 24.2 & 77.6 & 7.5 + + h232729@xmath37084926 & 0.154 & fwzi & ... & 28.0 & 92.2 & 60.1 & 116.2 & 56.1 + & & fwhm & ... & 15.9 & 80.3 & 18.7 & 34.2 & 37.5 + & & ew & ... & 13.0 & 16.8 & 19.1 & 60.9 & 16.2 + + h232906@xmath37083416 & 0.953 & fwzi & 368.6 & ... & ... & ... & ... & ... + & & fwhm & 76.1 & ... & ... & ... & ... & ... + & & ew & 133.6 & ... & ... & ... & ... & ... + + h233154@xmath37193836 & 0.475 & fwzi & 344.5 & ... & ... & ... & ... & ... + & & fwhm & 75.0 & ... & ... & ... & ... & ... + & & ew & 93.5 & ... & ... & ... & ... & ... + [ tab5 ]

we present our temporal and spectral analyses of 29 bursts from , detected with the gamma - ray burst monitor onboard the _ fermi _ gamma - ray space telescope during the 13 days of the source activation in 2008 ( august 22 to september 3 ) . we find that the durations of the bursts can be fit with a log - normal distribution with a mean value ofms . we also estimate for the first time event durations of soft gamma repeater ( sgr ) bursts in photon space ( i.e. , using their deconvolved spectra ) and find that these are very similar to thes estimated in count space ( following a log - normal distribution with a mean value ofms ) . we fit the time - integrated spectra for each burst and the time - resolved spectra of the five brightest bursts with several models . we find that a single power law with an exponential cutoff model fits all 29 bursts well , while 18 of the events can also be fit with two black body functions . we expand on the physical interpretation of these two models and we compare their parameters and discuss their evolution . we show that the time - integrated and time - resolved spectra reveal that decreases with energy flux ( and fluence ) to a minimum ofkev at erg s , increasing steadily afterwards . two more sources exhibit a similar trend : sgrs j and . the isotropic luminosity , , corresponding to these flux values is roughly similar for all sources ( erg s ) .

1107.2121

magnetars are slowly rotating neutron stars associated with extreme magnetic fields ( @xmath12 g ) . several obscure neutron star subpopulations haven been claimed as magnetar candidates , in particular soft gamma repeaters ( sgrs ) and anomalous x - ray pulsars ( axps ) ; for reviews see @xcite . most magnetars have been discovered either from their persistent x - ray emission properties or when they enter into randomly occurring outbursts , during which they emit a multitude of short ( @xmath13 milliseconds ) , soft @xmath14/hard x@xmath15ray bursts . thus far , approximately twenty magnetar sources are known and most of them reside in the galactic plane with a higher concentration close to the center ; two are located in the magellanic clouds . was discovered with _ swift _ on 2008 august 22 , when it emitted a series of bright , soft , short bursts @xcite . the first burst also triggered the gamma - ray burst monitor ( gbm ) onboard the _ fermi _ gamma - ray space telescope ( _ fgst _ ; hereafter _ fermi _ ) . soon after , our target of opportunity ( too ) observations with the _ rossi x - ray timing explorer _ ( _ rxte _ ) established a period of @xmath16 5.76 s in the persistent x - ray emission of the source @xcite . further observations with _ rxte _ and the _ swift_/x - ray telescope ( xrt ) revealed a spin - down rate of @xmath17 s s@xmath7 , indicating a dipole magnetic field of @xmath18 g @xcite . our subsequent @xmath19 too observations established an accurate location of the source at @xmath20 , @xmath21 , with an 1@xmath22 uncertainty of @xmath23 @xcite . this is the first magnetar location at roughly the galactic anticenter direction , placing most likely at the perseus arm of our galaxy at @xmath24 kpc @xcite . the outburst lasted approximately 2 weeks , during which several bright bursts were detected with _ swift _ , gbm , _ rxte _ , _ konus_-wind , and _ suzaku _ ( enoto et al . 2009 ; aptekar et al . 2009 ; kumar et al . 2010 ; see also table [ obs ] ) . after the burst activity ceased , the source flux decreased exponentially with an e - folding time of @xmath25 days @xcite . during the entire outburst interval , gbm triggered on 26 bursts ; in addition , an untriggered event search in the daily data sets revealed seven more bursts . we present here our analyses of the 29 gbm bursts from for which we have high - resolution data ; the properties of the x - ray persistent source emission have been published by @xcite and @xcite . in section [ data ] , we describe the instrument , observations and data selection . in sections [ tempo ] and [ spec ] , we present our temporal and spectral results , respectively . we discuss the interpretation of our results in section [ discussion ] . the _ fermi_/gbm has a wide field of view ( 8 sr ; un - occulted ) and a continuous broad - band energy coverage ( 8 kev 40 mev ) . it consists of 12 nai detectors ( 8 1000 kev ) arranged in 4 clusters of three each and 2 bgo detectors ( 0.2 40 mev ) placed at opposite sides of the spacecraft @xcite . in trigger mode gbm provides three types of science data : cspec with continuous high spectral resolution ( 1024 ms and 128 energy channels ) , ctime for continuous high time resolution ( 64 ms and 8 energy channels ) , and time - tagged photon event ( tte ) data ( 2 @xmath26s and 128 energy channels ) . for a detailed description of the instrument and data types , see @xcite . when a gbm trigger occurs , tte data are provided from @xmath27 s pre - trigger to @xmath28 s post - trigger . with its very high temporal resolution the tte datatype is most suitable for the detailed temporal and spectral analyses of very short events like sgr bursts and is , therefore , the only datatype used throughout this paper . during the burst active period of the source ( 2008 august 22 2008 september 3 ) , gbm triggered on 26 bursts . we further implemented the algorithm described in @xcite to search for any untriggered events in the daily data sets of 2008 august 21 through 2008 september 14 , and found seven individual short bursts located in the same direction as . among these only three had tte data and were , therefore , included in our analyses . two of the 29 bursts were very bright , causing saturation of the high speed science data bus ; the saturated parts of these bursts were excluded from any spectral analysis . for all spectra and durations we selected the nai detectors with an angle to the source smaller than @xmath29 to avoid attenuation effects . we also excluded any detectors blocked by the _ fermi_/large area telescope ( lat ) , or by the spacecraft radiators or solar panels . the bgo detectors were not used as there was no obvious emission in the nais above 200 kev . in columns two through four of table [ obs ] we list for each of the 29 gbm events their trigger numbers , times , and the selected nai detectors used for the following analyses . figure [ lc ] exhibits the light curves of four representative bursts . their profiles vary from a single short pulse ( figure [ lc]a ) to a multi - pulse event ( figure [ lc]c ) . we note that the flat top in the event of figure [ lc]d is caused by saturation . we have used the gbm tte data to estimate the event durations in both count and photon space . although the former process has been used exclusively in the past to estimate sgr and axp burst durations @xcite , the latter is used for magnetar bursts for the first time here . we describe the methods and the results below ; table [ durations ] contains the mean values and widths of the distributions of all temporal parameters . + for the @xmath0 ( @xmath30 ) is the duration during which the background - subtracted cumulative counts increase from @xmath31 ( @xmath32 ) to @xmath33 ( @xmath34 ) of the total counts . ] estimate we used the algorithm originally developed by @xcite for gamma ray bursts ( grbs ) and later adapted for sgr bursts @xcite , modified slightly to accommodate the gbm tte data . each duration was calculated between @xmath35kev with 2ms bins as follows . first we fit the burst background using two time intervals before and after each burst ( [ -2 s , -0.5 s ] and [ 0.5 s , 2 s ] ) with a first order polynomial ; these intervals were kept mostly the same for all bursts unless a precursor and/or a tail were detected . then we fit the background - subtracted , cumulative burst counts , with a linear plus a step function simultaneously ; the linear part was user - selected ( before and after the burst ) , while the step part was determined from the fit linear trend . the height of the step function was then used to represent the net total counts ( @xmath36 ) of the burst , subsequently used for the @xmath0 determination . panels a and b in figure [ duration_dis ] show the distributions of @xmath0 and @xmath30 for all 29 bursts ; all individual @xmath0 values can be found in table [ obs ] ( column 4 ) . for comparison reasons with other magnetar duration distributions ( see section 3.4 ) we fit each distribution with log - normal functions and obtained @xmath37 ms ( @xmath38 , where @xmath22 is the width of the distribution in the log - frame ) and @xmath39 ms ( @xmath40 ) . the average values of the raw data weighted by their errors are @xmath41 ms , and @xmath42 ms . + + the distributions above did not account for the uncertainties in the durations . we estimate the ( max , min ) errors ( @xmath43 , @xmath44 ) for each @xmath0 ( similarly @xmath30 ) as : @xmath45 where @xmath46 is the time when the cumulative light curve reaches _ n _ counts . we then created probability distribution functions ( pdfs ) for each @xmath0 and @xmath30 , as described by @xcite and @xcite . each pdf was constructed using a two - sided normal distribution , where the width of each half ( @xmath22 ) was set to the uncertainty of the duration ( @xmath47 , @xmath44 , @xmath48 , and @xmath49 ) : @xmath50 each pdf describes the likelihood to obtain _ x _ ( i.e. , @xmath0 or @xmath30 ) given its measured value @xmath51 . finally , we averaged all sample pdfs to create the total pdf of @xmath0 ( @xmath30 ) shown in figure [ duration_dis ] panels ( a ) and ( b ) . note that the shorter ( fainter ) events have larger errors , resulting in `` pulling '' the pdf towards shorter durations . the photon - based durations , @xmath52 , are estimated with an algorithm similar to the one used above over each burst cumulative fluence in erg @xmath6 . we used the same time resolution ( 2ms ) and energy range ( @xmath35kev ) as in the count durations . the essential difference here , is that these measurements utilize the intrinsic ( deconvolved ) burst spectra instead of the detector recorded counts to define the burst intrinsic durations independent of different instruments . to perform these estimates , we used the gbm public software tool rmfit v3.3 ( for a description of this tool see also * ? ? ? * ) and the new datatype ctte specially created to facilitate analyses of short events . this datatype simply bins the 128 tte energy channels into the same 8 bins as the ctime data . the errors in the duration estimates are taken from @xcite and @xcite . a detailed description of the photon - based durations can be found in the first two years grb catalog of the _ fermi_/gbm @xcite . in short , an adequate background interval is selected before and after each burst and fit with the lowest acceptable order of a polynomial to determine the background model parameters . next the entire burst interval is fit to determine the default set of photon model parameters . the model used in these fits is a power law with an exponential cutoff ( compt ; described in detail in section 4 ) . when all background and source model selections are determined for each 2ms time bin , we subtract the background , fit its spectrum using the compt model , and calculate its photon flux . these values are then used as inputs for the @xmath52 ( @xmath53 ) estimates , performed with the same algorithm described above . figure [ duration_photon ] shows the distributions of @xmath52 ( @xmath53 ) fit with a log - normal function ( panels a and b ) , obtaining @xmath54 ms ( @xmath55 , where @xmath22 is the width of the distribution in the log - frame ) and @xmath56 ms ( @xmath57 ) . the average values of the raw data weighted by their errors are @xmath58 ms , and @xmath59 ms . the individual @xmath52 values can be found in table [ obs ] ( column 5 ) . the solid curves in figure [ duration_photon ] exhibit the pdfs for @xmath52 and @xmath53 after taking into account the symmetrized errors in the values . this plot is an adaptive kernel density estimation : it was made by adding up a set of normalized gaussian functions , one at each of the data points and with a 1@xmath22 width given by the corresponding error estimate . the multiple narrow spikes in this figure , mostly at the long - duration end of the plot , are largely due to the fact that the corresponding errors ( in contrast to those of shorter and fainter durations ) are relatively small , yielding tall and narrow gaussian components . this is a rather high - variance estimate of the distribution function . as an alternative we show in figure [ duration_bb ] a completely different density estimate based on an adaptation of the bayesian block algorithm for histogramming of the @xmath52 and @xmath53 data , taking into account measurement errors in the independent variable @xcite . this algorithm finds the optimal piece - wise constant model to represent the data ; the optimization corresponds to the maximum likelihood for a constant - rate poisson model for the data in each of the bins . the sizes and locations of the bins are all determined by this optimization , not pre - defined as in ordinary histograms . the solid curve is the bayesian block representation superimposed on such an ordinary histogram with bins chosen so that the value plotted is more than one only if there are duplicate values . this analysis suggests that the spiky structure of the pdfs in figure [ duration_photon ] is mostly due to noise fluctuations ( the spikes corresponding to a small number of points with small formal measurement errors ) and does not support a quantized or a multimodal duration distribution for the bursts from . we compare the two duration distributions ( in count and photon space ) in section [ comp_durations ] . ( a ) and @xmath53 ( b ) . the histograms show the raw data and the solid curves show the pdfs . the vertical dot - dashed lines indicate the mean value of the log - normal fits of the histograms . [ duration_photon],title="fig : " ] ( a ) and @xmath53 ( b ) . the histograms show the raw data and the solid curves show the pdfs . the vertical dot - dashed lines indicate the mean value of the log - normal fits of the histograms . [ duration_photon],title="fig : " ] ( _ left _ ) , and @xmath53 ( _ right _ ) , taking into account measurement errors . the histograms show the raw data and the solid curves show the bayesian blocks . [ duration_bb ] ] the emission time @xmath60 ( @xmath61 ) for each burst was determined by adding the time bins ( 2ms each ) of high fluence in decreasing fluence rank until 90% ( 50% ) of the fluence was reached . the emission time interval , thus , characterizes the duration of high - fluence emission . this parameter was also first introduced for grbs as complementary to their @xmath0 ( @xmath30 ) duration measures @xcite . panels c and d in figure [ duration_dis ] show the distributions of @xmath60 and @xmath61 . these are also fit with log - normal functions obtaining @xmath64 ms ( @xmath65 ) and @xmath66 ms ( @xmath67 ) . the ratio @xmath68 ( @xmath69 ) is defined as the duty cycle of each burst by @xcite . this value should not be over 1 , because the emission time excludes low fluence intervals in the burst , which sometimes are included in the duration . panels e and f in figure [ duration_dis ] present the distributions of @xmath62 and @xmath63 , which were fit with normal distributions with @xmath70 ( @xmath71 ) and @xmath72 ( @xmath73 ) . both duty cycle distributions have the same mean value of 0.68 . the durations in photon space are estimated using the deconvolved source spectra with the response of the gbm detectors . in principle , since the spectral energies where sgr bursts emit most of their photons are relatively narrow ( @xmath74kev ) there should not be a large difference between photon and count durations . figure [ cr_ph ] shows @xmath75 _ versus _ we notice that in general the count space @xmath0 values tend to be larger by a very small amount , mainly due to the fact that they take into account number of counts irrespective of their energy content . since sgrs have mostly soft spectra , the counts corresponding to the lowest energy photons often do not contribute much flux in the durations ( i.e. , less than 5% ) . the @xmath30 estimates , however , are perfectly aligned along the @xmath77 line , making these spectrally - independent measures for durations ( as also noted for grbs by bissaldi et al . it is also important to note here that these photon durations validate earlier duration measurements , which have been done in count space for all magnetar candidates . and the count space @xmath0 . _ bottom panel _ : the similar plot for @xmath30 . the solid lines are @xmath77 . [ cr_ph],title="fig : " ] and the count space @xmath0 . _ bottom panel _ : the similar plot for @xmath30 . the solid lines are @xmath77 . [ cr_ph],title="fig : " ] we now compare in figure [ duration_com ] the mean values of the count space @xmath0 and @xmath60 of , with those of four other magnetar candidates . sgr j@xmath78 was also observed with gbm and the temporal parameters are estimated using exactly the same procedures as here @xcite . sgrs @xmath9 and @xmath79 were estimated using _ rxte_/pca observations in @xmath80kev by @xcite . axp 1e@xmath81 durations are from @xcite and are also estimated using _ rxte_/pca observations in @xmath80kev . it is obvious from the figure that all durations fall well within the same order of magnitude ( @xmath82 ms ) , indicating a similar origin for the bursts across the magnetar population . ( right hatched bars ) of four sgrs and one axp and of @xmath60 ( left hatched bars ) of four sgrs . the photon space @xmath0 of is indicated with a blank bar . values are marked in ms above each bar . [ duration_com ] ] using the pulse ephemeris reported in @xcite , we have aligned each burst peak to the pulse phase of the spin period of , to search for possible correlations of the burst activity with rotational phase . in an effort to confirm the gbm barycentric correction for this analysis , we compared the barycenter corrected light curves of burst bn080823.020 ( table 1 ) for the gbm and _ rxte_/pca data . note that the pulse ephemeris is based primarily upon _ rxte_/pca data . a cross - correlation of the gbm and _ rxte_/pca burst time histories indicates no significant shift with an upper limit of 4ms . given the 5.76s pulse period of this sgr , even a 4 ms shift is negligible . as a final check , the barycentering software was also tested by epoch folding tte data to obtain a pulse profile of the crab pulsar using the jodrell bank ephemeris ( http://www.jb.man.ac.uk/ pulsar / crab.html ) . we found that the phase of the first peak agreed within 200 @xmath26s of the _ rxte result _ shown by rots et al . ( 2004 ) . burst peak phases were computed using the event times recorded in the tte data in the energy range 8@xmath1560 kev . for each event time history , the burst peak time was defined as the average event time for the six most closely spaced counts in a 4 sec interval surrounding the trigger time . the phase of pulse maximum was defined by fitting an inverted parabola to the folded _ rxte_/pca 2@xmath1510 kev light curve . figure [ phase_offset ] shows the distribution of the phase offsets of all burst peaks relative to the pulse maximum . the average offset ( average of the absolute value ) is 0.285 cycles . for a random phase distribution ( i.e. , null hypothesis ) , one would expect an average offset of 0.25 cycles . a monte carlo simulation of 10@xmath83 realizations for 29 draws from a random distribution shows that the probability of getting an offset value of 0.285 for 29 samples is @xmath84% or roughly @xmath85 . we conclude that there is no correlation of the burst peaks with pulse phase . ] we performed all spectral analysis using rmfit v3.3 and we generated response files for each burst using the gbm response generator _ gbmrsp v1.9 _ ; table [ obs ] lists the detectors used for each event . each event spectrum was restricted within @xmath86kev , as we did not detect any counts above 200kev . however , roughly one third of the bursts have significant emission in the @xmath87kev band : the average count rate in the brightest detector over the @xmath0 interval was larger than 1count / s . most of the bursts were faint and they only had good statistics for time - integrated spectral analysis described below . five events ( including the two saturated bursts mentioned in section [ data ] ) were very bright and we were able to perform time - resolved analysis as described in section [ trs ] . finally , to account for the iodine k - edge effects at 33.2kev , we excluded from our spectral fits the region between @xmath88kev . we fitted the time - integrated spectra of all 29 bursts with several models : a single power law ( spl ) , an optically thin thermal bremsstrahlung ( ottb ) , a single black body ( bb ) , a power law with an exponential cut - off ( compt ) ( e / e_{\rm{piv}})^\lambda,\ ] ] where @xmath89 is the photon number flux in photons s@xmath7 @xmath6 kev@xmath7 , @xmath90 is the amplitude in the same units as @xmath89 , @xmath3 is the peak energy in kev , @xmath91 is the photon index , and @xmath92kev is the pivot energy . ] , a two black body spectrum ( bb+bb ) , a single black body with a power law ( bb+pl ) , a single black body with an ottb ( bb+ottb ) , and finally , a single black body with an exponential cut - off power law ( bb+compt ) . the parameters of the last two models ( bb+ottb , bb+compt ) could not be constrained by most burst data , while a spl was always a bad fit ; all three models were , therefore , rejected from further spectral analysis . to determine the goodness of fit for the remaining models we used the castor modified cash - statistic ( c - stat ) . this is a modified maximum likelihood estimator which asymptotes to @xmath93 , used when there are small numbers of counts / bin ( poisson regime ) , which is the case for most of the sgr events ( especially in the higher energy bins ) . using c - stat we were able to further reject single bb , ottb and bb+pl models . the first two models fit only part of the weaker burst set ; for these the compt model also did not give a significantly smaller c - stat . it was not excluded , however , because this model , contrary to the first two , fit all bursts . moreover , a compt index of @xmath94 or @xmath95 reproduces the bb or ottb spectral shape . the bb+pl model had overall worse c - stat values compared to the remaining two models ( compt and bb+bb ) . the relative goodness of fit among these models is exhibited in figure [ spec_exa ] , which shows spectral fits with compt , bb+bb , ottb , and a single bb of one bright burst from ( bn080826.136 ) . from the figure , we see that the compt and bb+bb models can fit the data equally well ; the residuals , however , of the ottb and the bb model fits are unacceptably large . , title="fig : " ] , title="fig : " ] , title="fig : " ] , title="fig : " ] the compt model fits all 29 bursts well , with the bb+bb model giving equally good fits in only 18 events , where we have enough statistics to constrain the model parameters . to determine whether the compt or the bb+bb model fit the data best , we simulated ( using rmfit ) a large set of bursts with different intensity and spectral shape parameters , using compt and bb + bb as input models , and then fit them with both the compt and bb+bb models ( see van der horst et al . 2011 in preparation , for a detailed description of the simulations ) . we show that the c - stat improvement is not significant to conclude that bb+bb ( with one more parameter ) is better than compt . a set of simulations using xspec gave similar results with rmfit . below we discuss our compt model fits for all 29 bursts . since the bb+bb fits provide significant information on the source parameters @xcite , we also describe the results of these fits for 18 bursts . section [ discussion ] expands on the importance of each model . all bursts were well fit with the compt model . we list the model parameters and statistics in table [ obs ] ( columns 6 - 10 ) . the distributions of index and @xmath3 are displayed in figure [ spec_distribution ] . the top panel shows the spectral index distribution , which is centered around zero and is best fit with a normal function with an average of @xmath96 ( @xmath97 ) . this index distribution clearly excludes a pure ottb or bb fit for all bursts , as such fits would require indices of @xmath98 , @xmath99 , respectively . the bottom panel of figure [ spec_distribution ] exhibits the @xmath3 distribution , which also follows a normal function with mean at [email protected] kev ( @xmath101 kev ) . in both panels , the hatched areas highlight the distribution of 18 bursts also fit with the bb+bb model . this subsample was fit with a normal distribution with @xmath102 kev , @xmath103 kev and @xmath104index@xmath105 , @xmath106 . most bursts here have negative photon index and lower @xmath3 , indicating a softer spectrum . ( _ bottom panel _ ) of the compt model fits for 29 bursts from . the right slashed bars represent the subset of the 18 bursts that can be fit with the bb+bb model as well . the dashed and dotted lines are the best fits with normal distributions of all bursts and of the subset of 18 , respectively . [ spec_distribution],title="fig : " ] ( _ bottom panel _ ) of the compt model fits for 29 bursts from . the right slashed bars represent the subset of the 18 bursts that can be fit with the bb+bb model as well . the dashed and dotted lines are the best fits with normal distributions of all bursts and of the subset of 18 , respectively . [ spec_distribution],title="fig : " ] using the compt spectral fits we estimated the event fluences ( @xmath86 kev , also listed in table [ obs ] ) and plotted in figure [ spec_correlation ] their correlation with spectral indices ( top panel ) and with @xmath3 values ( bottom panel ) . we note that the brightest events have a constant index value of @xmath107 , which progresses to lower ( softer ) values for weaker ones and is widely scattered with larger errors for the faintest events . similarly , a simple trend can not describe the relation between the hardness ( @xmath3 ) and fluence . for bursts with high energy fluence ( e.g. , @xmath108 erg @xmath6 ) , @xmath109 values are constant and cluster around 35 kev , while the values of weaker bursts cluster around 45kev with a large scatter range between 30 and 60 kev . we note here the large span of fluence in our data ( @xmath110 to @xmath111 erg @xmath6 ) , which is one order of magnitude broader than earlier results @xcite . we expand on these trends in section [ discussion ] . ( _ bottom panel _ ) _ versus _ the event energy fluence ( @xmath86 kev ) for 29 bursts from . [ spec_correlation],title="fig : " ] ( _ bottom panel _ ) _ versus _ the event energy fluence ( @xmath86 kev ) for 29 bursts from . [ spec_correlation],title="fig : " ] we now proceed to estimate the cumulative energy fluence during the active period of , shown at the top panel of figure [ energetic ] . the plot levels off at @xmath112 erg @xmath113 , which should be considered a lower limit ( as we have not taken into account event saturation , untriggered events without tte data , and additional bursts seen , e.g. , with _ swift _ ) . assuming that the source is 2 kpc away from the earth , then this fluence would correspond to a total energy of at least 1.8 @xmath114 erg emitted from the magnetar in bursts during this 13-day active period . the bottom panel in figure [ energetic ] presents the differential distribution of energy fluence ( log@xmath115log@xmath116 ) . the best fit with a power law function is overplotted in the figure . the index of the power law is @xmath117 , which corresponds to @xmath118 . this slope is similar to the one estimated with the _ rxte_/pca for sgr@xmath9 @xcite , but differs from all other sgr and axp slope estimates , which are all very close to @xmath119 @xcite . it also differs from the slope estimate of the sgr@xmath9 events detected with the international cometary explorer ( ice ) and the burst and transient source experiment ( batse ) data , which is @xmath119 as well @xcite . however , the gbm fluence range covers the higher end of fluence values , and the analyzed burst sample is most likely incomplete at the lower fluence end ( which contributes most of the events in all other sources ) . . [ energetic],title="fig : " ] . [ energetic],title="fig : " ] eighteen bursts were spectrally fit equally well with the bb+bb model . we chose to also present these results here , although this model is more complicated than the compt model ( with one more parameter ) , because it has been successfully used in the past for the spectral analysis of sgr bursts @xcite ; we will compare our spectral results in section [ discussion ] . assuming that the two bb components arise from two hot spots on the surface or photosphere of the source , each emission area , @xmath120 , can be calculated from the temperature , @xmath121 , of the corresponding black body spectrum as : @xmath122 where _ f _ is the average energy flux per event ( total burst fluence divided by the spectral integration time ) , _ d _ is the distance to the magnetar ( assumed to be 2 kpc ) , and @xmath22 is the stefan boltzmann constant . we plot the emission area of both bb components as a function of _ kt _ in figure [ r2_kt ] . as shown earlier by @xcite for sgr@xmath79 , we also note a clear separation of the two temperatures and emission areas of the cooler and hotter black bodies . the cooler bb has a larger emission area , and the temperature spreads between @xmath123 kev . the emission area of the hotter bb with temperatures in the @xmath124 kev range is much smaller . the emission areas of both bb components have similar evolution through time during the active period of ( figure [ r2_evo ] ) . the relative increase in emission area , however , is higher ( by a factor of ten ) for the hotter bb component . finally , we notice in figure [ lbb_hbb ] that the total fluence of the burst is divided equally between the two black body components ( we find a correlation coefficient of 0.94 corresponding to a chance probability of @xmath125 ) . a single power - law with an index of @xmath126 fits the data well . ] ] ] we performed time - resolved spectral analysis with 8ms temporal resolution for the five brightest bursts of . we binned the data requiring a significance of at least @xmath127 above background for each bin . each bin was then fit with the compt model , which was the best model for the time - integrated spectra . figure [ tr_spec ] displays the light curves of these five events overplotted with their @xmath3 values ( left column ) and the correlation between @xmath3 and energy flux ( right column ) . we see that the @xmath3 follows the light curve for the brightest part of the burst in four out of five cases ( except for the fourth panel from the top ) , and it rises surprisingly at the beginning and the tail end of each event . + + + + to investigate this trend , we plotted in figure [ tr_correlate ] the combined @xmath3 values from all five bursts _ versus _ their fluxes ( top panel ) . we clearly see that @xmath3 rises at both high _ and _ low flux values with a minimum determined with a broken power - law fit at @xmath128erg @xmath129 @xmath130 . the same trend is seen in the lower panel between @xmath3 and fluence with a minimum at @xmath131erg @xmath129 . the indices below and above this critical flux ( fluence ) values are @xmath132 and @xmath133 ( @xmath134 and @xmath135 ) , respectively . an f - test comparison of a broken power law and a single power law fit shows that the former provides a better description of the correlation , with a small probability of chance coincidence @xmath136 for the flux and @xmath137 for the fluence . we also performed a spearman rank test for both branches of each plot and found a _ low _ flux ( fluence ) correlation coefficient of @xmath138 ( @xmath139 ) with a probability of chance coincidence of @xmath140 ( @xmath141 ) . the correlation coefficient and probability for the _ high _ flux ( fluence ) parts are @xmath142 ( @xmath143 ) and @xmath144 ( @xmath145 ) , respectively . the latter probabilities are less significant than the ones for the low flux / fluence parts , because the spearman rank tests involve a smaller number of data points . for the full flux ( fluence ) data sets , the correlations are not very significant , with a coefficient of @xmath146 ( -0.28 ) and probability of @xmath147 ( @xmath148 ) , giving further support to changing trends ( minima ) in the correlations between @xmath3 and flux / fluence . and energy flux in the time - resolved spectra of the five brightest bursts of . ( _ bottom panel _ ) : correlation between the @xmath3 and energy fluence for the same bursts . the solid lines exhibit the best fit to the data with broken power - law functions . [ tr_correlate],title="fig : " ] and energy flux in the time - resolved spectra of the five brightest bursts of . ( _ bottom panel _ ) : correlation between the @xmath3 and energy fluence for the same bursts . the solid lines exhibit the best fit to the data with broken power - law functions . [ tr_correlate],title="fig : " ] ( _ bottom panel _ ) of the compt model for all time - resolved spectra . the dashed lines are the best fits with a normal distribution . [ tr_compt_dis],title="fig : " ] ( _ bottom panel _ ) of the compt model for all time - resolved spectra . the dashed lines are the best fits with a normal distribution . [ tr_compt_dis],title="fig : " ] figure [ tr_compt_dis ] displays the distribution of the compt index and @xmath3 of all time - resolved spectra . both distributions were fit with normal functions with @xmath149 kev ( @xmath150 kev ) and @xmath104index@xmath151 ( @xmath152 ) . these values are very similar to those obtained with the time - integrated fits . finally , we fit the time - resolved spectra with a bb+bb model . retaining only the fits which constrained the bb+bb parameters , we calculated the emission areas for each bb component . similar to the integrated spectra , figure [ tr_r2_kt ] shows that the emission areas follow different behaviors with temperature for the hard and soft bb . interestingly , a comparison with similar results shown in figure 5 of @xcite for sgr@xmath79 reveals that although the @xmath153 values are very close for the two bbs , the emission areas differ by at least one ( and maybe two ) orders of magnitude . these results indicate that either the two sources have different burst energetics or that ( very unlikely ) the distance determination of sgr@xmath79 is off by a large factor . using the 2 kpc distance for the source , we estimate the isotropic luminosity of the two bb components shown in figure [ tr_lbb_hbb ] . the spearman rank correlation coefficient is @xmath154 corresponding to a chance coincidence probability of @xmath155 , indicating that these components are well correlated , as also shown with a power - law fit with index @xmath156 . @xcite have performed a similar analysis for the data of sgr@xmath79 and found a spectral index of @xmath157 . ] erg s@xmath7 ) of the soft and hard bb . [ tr_lbb_hbb ] ] our spectroscopic fitting clearly indicates that the compt and bb+bb models yield superior fits to the other possibilities . the compt model , with its power - law shape curtailed by an exponential turnover , is intended to mock - up the classic unsaturated comptonization spectrum realized in models of accretion disks such as in cyg x-1 or in active galactic nuclei ( see chapter 7 of rybicki & lightman 1979 for a summary of its development as a solution of the kompaneets equation ) . these models use hot , thermal electrons in a corona to repeatedly scatter low - energy photons , heating them gradually up to an energy @xmath158 consistent with the electron temperature @xmath159 , at which point a quasi - exponential spectral turnover emerges as further heating becomes impossible . the power law marks the scale - independence of the compton upscattering , and its slope depends only on the mean energy gain per collision ( @xmath160 for non - relativistic electrons ) and the probability of loss of photons from the scattering zone , being expressed as @xmath161 with the compton @xmath162-parameter in the domain @xmath163 . this parameter is the product of the average fractional energy change per scattering and the mean number of scatterings , and is @xmath164 for a non - relativistic situation for repeated compton upscattering . here @xmath165 is the scattering optical depth . in the context of magnetars , such comptonization can also ensue , but the strong magnetic field now plays an important role . ephemeral coronae of hot electrons could be expected in a dynamic inner magnetosphere . for example , this could be due to intense dissipation of magnetic energy in the closed field line region via field line twisting , i.e. , transient departures from poloidal geometry , as in the considerations of thomson , lyutikov & kulkarni ( 2002 ) , thompson & beloborodov ( 2005 ) , beloborodov & thompson ( 2007 ) , and nobili , turolla & zane ( 2008 ) . such a pumping of energy into electrons in low altitude regions would then be subject to irradiation by the intense bath of surface x - ray emission . the electron coronae would mimic those of their black hole counterparts discussed above , and serve as a comptonization target for the x - rays . temporally , the coronae could be quite variable , resulting in varying or chaotic time profiles such as are observed . if the deposition of energy in hot electrons is persistent over many light - crossing timescales , so also can the upscattered hard x - ray emission be . in this magnetic case , the spectrum would again naturally be a power law truncated by an exponential tail , whose energy is pinned by @xmath159 . the slope does differ somewhat in value from the non - magnetic case because the presence of the intense field anisotropizes the environment . this alters the average fractional energy change per scattering from the isotropic @xmath166 form in rybicki & lightman ( 1979 ) . we note that this picture is similar to that of lyutikov & gavriil ( 2006 ; see also rea et al . 2008 ) , who modeled the steep x - ray tails below 10 kev in quiescent magnetar emission using a resonant cyclotron scattering picture . since the electrons will move along the field lines in the zeroth landau level , the exact kinematics that impact the determination of @xmath167 depend on the colatitude and altitude of the collisions therefore , the effective _ magnetic compton @xmath162-parameter _ , @xmath168 , that could be substituted in eq . ( [ comp_index ] ) , would take a somewhat different value from its non - magnetic cousin , but one anticipates that the range of spectral realizations would be similar to the @xmath169 case . since head - on collisions generally yield greater heating of photons , and these are more likely at lower altitudes , it is expected that interaction zones close to the stellar surface would spawn larger @xmath168 and therefore flatter photon spectra . if there were a coronal radius expansion , this would then translate to an evolutionary steepening of the spectra with time . note that the photon retention probabilities in the coronae will need to be higher than in the lyutikov & gavriil ( 2006 ) scenario to generate the requisite moderate optical depths @xmath165 to match the flat outburst spectra discussed in this paper . the scatterings will take place below the cyclotron resonance unless their altitude is above around 10 stellar radii ( e.g. lyutikov & gavriil 2006 ) . below resonance , the cross section is far inferior to the thomson value ( e.g. see herold 1979 ) . then , the coronal electron density must accordingly be much higher than when @xmath169 in order to establish a sizable optical depth . this provides a possible distinction between the two classes of magnetar emission : the steady @xmath170kev signal may originate at higher altitudes where the resonance is accessed and the local electron density is low , whereas the bursts reported here may be triggered closer to the surface in higher density zones that initially precipitate scattering below the cyclotron resonance . as photons get energized above 20 kev , the altitude where the resonance is accessed is lowered , increasing the cross section for scattering and therefore @xmath168 . note also , that to a considerable extent , polarization mode - switching ( e.g. miller 1995 ) between the two photon polarizations can help increase the opacity , adding a further nuance to consider when modeling the scattering environment . observationally , it is difficult to discern unambiguously the presence of the intense field using a comptonization model scenario in the energy range of data presented in this paper : details of both the field strength and the geometry are subsumed in a single parameter @xmath168 . since the emission in this scenario should be strongly polarized , and the degree of polarization should depend on the interaction geometry , a hard x - ray polarimeter would provide insightful probes into the presence of a strongly anisotropizing super - critical field . the dual blackbody can also be envisaged as a viable alternative from a theoretical standpoint . the moderate thomson depths required to generate flat comptonization spectra could easily be higher , thereby pushing the electron - photon interactions more towards an equilibration . the saturation temperature is then controlled by the largely uncertain total energy dissipated in the inner magnetosphere per hot electron present . since any equilibration will be non - uniform over a coronal volume , there should be a modest temperature gradient throughout , smearing out the continuum . the range of temperatures will not be great because the system is not gravitationally hydrostatic in character . moreover , radiative transfer effects impact the spectral shape and further modify it . accordingly , pure blackbody shapes are not expected . it is quite conceivable that a two - component blackbody fit may well approximate the emergent continuum that is a superposition of distorted blackbodies spanning a small range of temperatures . most probably , due to general thermodynamic considerations , the base of the coronae ( e.g. centered near the source of magnetic dissipation ) should be hotter than the outer layers ( see also lyubarsky 2002 and thompson & duncan 1995 ) . interestingly , the fits here generate a smaller volume associated with the hotter blackbody contribution , consistent with the expectation for coronal structure . yet , somehow , we must obtain a view of the hotter zone , therefore indicating a strongly aspherical coronal geometry . modeling this semi - equilibration is a challenging task for theorists considering the influences of the field , the twisted magnetospheric geometry , and the inherent anisotropy and polarization - dependence of the scattering process . at present , it is not possible to distinguish between this thermal scenario , and a comptonization one . since the compt model fits all events , we used it to derive fluences ( time - integrated and time - resolved ) , and determined using the time - resolved values that the hardness - fluence correlation can be described by a broken power law , with a minimum at a fluence of @xmath171 erg @xmath6 . we have used @xmath172 values to characterize the hardness of the events in our sample . earlier studies of this correlation used hardness ratios to bypass proper spectral fits due to low quality or insufficient data . moreover , these studies @xcite had a very narrow overlap in fluence space . @xcite used 95 sgr@xmath9 events detected with the interplanetary sun earth explorer-3 ( isee-3 ) with fluences ranging between @xmath173 erg @xmath6 , and found a slightly positive correlation between the two quantities ( hardness increasing with fluence ) . on the other hand , @xcite analyzed 159 and 385 events from sgrs @xmath9 and @xmath79 , respectively , observed with _ rxte _ , with fluences ranging between @xmath174 erg @xmath6 , and found the opposite trend ( an anticorrelation between hardness ratios and fluences ) . since the gbm data sample covers both fluence ranges ( @xmath175 erg @xmath6 ) we were able to establish that both trends are indeed correct , and to define the turning point in the @xmath172 - fluence diagram . only two magnetar candidates were observed with gbm to emit a multitude of bursts , thus allowing us to construct their @xmath172 - flux diagrams . we find for the time - resolved data of , that @xmath172 reaches a minimum of @xmath4kev at a flux value of @xmath176 erg @xmath6 s@xmath7 . the second source ( sgrj@xmath8 ; van der horst et al . 2011 in preparation ) has flux values of @xmath177 erg @xmath6 s@xmath7 at a similar @xmath172 minimum . in addition we used the hardness ratio - count rate relationship of sgr@xmath9 found in the _ integral _ data @xcite to derive an approximate flux value of @xmath178 erg @xmath6 s@xmath7 ( using their conversion from count rate to flux ) at the hardness ratio minimum ( figure 3 in * ? ? ? * ) . we then converted these values into isotropic source luminosity , @xmath10 ( see table [ liso ] ) ; we note that ( with the caveat of small number statistics , and uncertainties in the distance measurements ) these values are comparable [ @xmath179 erg s@xmath7 ] , although the @xmath181fields and fluxes at @xmath172 minima vary by approximately a factor of ten between the two gbm sources and sgr@xmath9 . whether these differences reflect intrinsic source properties or instrumental effects is not yet clear . the @xmath172 trend is , however , clearly established in at least two different instrument data sets . the physical interpretation of this trend is beyond the scope of this paper . part of our sample ( 18 bursts ) was fit with the bb+bb spectral function applied to time - integrated intervals ; the remaining 11 events could not be fit due to poor statistics . in addition for five bright events we performed time - resolved spectral analysis . we found that the temperatures , emission areas and fluences ( luminosities ) of the two thermal components exhibit very similar properties in both cases ( time - integrated and -resolved ) . our results are consistent with those presented for intermediate and short bursts from sgr @xmath79 by @xcite and @xcite ; the burst emission areas and temperatures fall into the same region with those of the short bursts of sgr @xmath79 shown in figure 5 of @xcite . we also see a similar trend in the correlation between the luminosities of the two components as described by @xcite , namely that both luminosities increase in tandem . this behavior indicates that the hot and cool bb components may come from two separate emission regions , as also pointed out by @xcite : a smaller but hotter one from the surface of the magnetar , and a larger , cooler one from the star s magnetosphere ( but see also the discussion in section 5.1 ) . we note here that the time - integrated hot bb emission area of 13 bursts ( i.e. , not including the 5 brightest events ; figure [ r2_kt ] ) is similar to the emission area ( @xmath182 km@xmath183 ) of the bb component found in the persistent emission of sgr j1550@xmath155418 during one active bursting episode in january 2009 , believed to originate from a hot spot on the neutron star surface @xcite . finally , we performed a detailed temporal analysis of all 29 bursts and estimated their durations ( @xmath0/@xmath30 ) for the first time in count _ and _ in photon space . both estimates agree within statistics and thus validated all earlier ( count space ) duration estimates of magnetar candidate bursts . the durations ( and for four sgrs also the emission times and duty cycles ) of five magnetar candidates follow a log - normal or normal distribution . we find that events are very similar in average duration with four more magnetar candidates ( three sgrs and one axp ) . however , the @xmath0 distribution of the bursts from axp1e@xmath81 has a factor of two larger dispersion ( @xmath184 , * ? ? ? * ) than those of at least three other sgrs ( @xmath185 , * ? ? ? * , present work ) . bursts have an average duty cycle ( @xmath186 ) larger than other sgrs ( 0.45 , 0.46 for sgrs @xmath79 , @xmath9 , respectively ; * ? ? ? the differences in the intrinsic properties of the sources may be due to differences in the size of the active region responsible for the burst emission . we discuss below the burst energetics and its evolution . similarly to other magnetars , we also do not find a correlation between the pulse phase of the source and the burst peak times . the overall behavior of during its 13-day active period ( 2008 august 22 to september 3 ) is also very interesting . although about half of the bursts ( 16/29 ) occurred on one day ( 2008 august 23 ) , most of the burst energy was emitted from the source during august @xmath187 ( see also figure [ energetic ] ) . the average fluence of the beginning and the later part of the active period was constant and at a lower level . during these three days , gbm detected seven bursts , five of which are also distinguished in that the emission areas from both bb components are the largest . further , one of these five events ( bn080824.054 ) shows a double - peaked structure reminiscent of some bright thermonuclear type i x - ray bursts from accreting neutron stars , which has been interpreted as due to photospheric radius expansion ( pre ) . @xcite studied the effects of pre in high magnetic fields using this event and find that the predicted flux from pre theory is consistent with the one observed , opening the way to determining fundamental parameters of neutron stars , such as their equation of state . this publication is part of the gbm / magnetar key project ( nasa grant nnh07zda001-glast , pi : c. kouveliotou ) . m.g.b . acknowledges support from nasa through grant nnx10ac59a . e.g. and y.k . acknowledge the support from the scientific and technological research council of turkey ( tbitak ) through grant 109t755 . a.v.k . was supported by the bundesministeriums fr wirtschaft und technologie ( bmwi ) through dlr grant 50 og 1101 . alw acknowledges support from a netherlands organisation for scientific research ( nwo ) vidi grant . ramjw acknowledges support from the european research council via advanced investigator grant no . 247295 . 999 aptekar r. l. et al . , 2009 , , 698 , l82 barthelmy s. d. , et al , 2008 , gcn circ . , 8113 , 1 beloborodov , a. m. , & thompson , c. 2007 , , 657 , 967 bibby j. l. et al . , 2008 , , 386 , l23 bissaldi e. et al . , 2011 , arxiv11013325 camilo f. et al . , 2007 , , 666 , l93 enoto t. et al . , 2009 , , 693 , l122 evans p. a. et al . , 2009 , , 397 , 1177 fenimore e. e. , laros j. g. , ulmer a. , 1994 , , 432 , 742 feroci m. et al . , 2004 , , 612 , 408 gavriil f. p. , kaspi v. m. , woods p. m. , 2004 , , 607 , 959 g e. et al . , 1999 , , 526 , l93 g e. et al . , 2000 , , 532 , l121 g e. et al , 2001 , , 558 , 228 g e. , woods p. , & kouveliotou c. 2008 , grb coordinates network # 8118 g e. et al . , 2010 , , 722 , 899 gtz d. et al . , 2004 , a&a , 417 , l45 herold h. 1979 , phys . d , 19 , 2868 holland s. t. , sato g. , 2008 , gcn rep . , 160 , 1 israel g. l. et al . , 2008 , , 685 , 1114 kaneko y. et al . , 2006 , , 166 , 298 kaneko y. et al . , 2010 , , 710 , 1335 koshut t. m. , 1996 , ph . d. thesis , an investigation of temporal characteristics of gamma - ray bursts observed with batse koshut t. m. et al . , 1996 , , 463 , 570 kouveliotou c. et al . , 1993 , , 413 , l101 kumar h. s. 2010 , , 716 , 97 laros j. g. et al . , 1986 , , 322 , 152 lyubarsky , y. e. 2002 , , 332 , 199 lyutikov , m. & gavriil , f. p. 2006 , mnras , 368 , 690 meegan c. et al . , 2009 , , 702 , 791 mereghetti s. et al . , 2005 , , 628 , 938 mereghetti s. , 2008 , , 15 , 225 miller m. c. , 1995 , apj , 448 , l29 mitrofanov i. g. et al . , 1999 , , 522 , 1069 nobili l. , turolla r. , & zane , s. , 2008 , , 386 , 1527 olive j .- f . et al . , 2004 , , 616 , 1148 paciesas w. et al . , 2011 , in preparation rea n. et al . , 2008 , apj , 686 , 1245 rea n. et al . , 2009 , , 396 , 2419 rots , a. h. , jahoda , k. , & lyne , a.g , 2004 , , 605 , l129 rybicki g. b. & lightman a. p. 1979 _ radiative processes in astrophysics ( wiley , new york ) scargle , j. , 1998 , , 504 , 405 scargle , j. et al . , 2011 , in preparation starling r. l. c. et al , 2008 , , 672 , 433 strohmayer t. e. & ibrahim a. , 1998 , aipc , 428 , 947 thompson , c. , & duncan , r. c. 1995 , , 275 , 255 thompson c. & duncan r. c. , 2001 , , 561 , 980 thompson c. , lyutikov m. & kulkarni s. r. , 2002 , apj , 574 , 332 thompson c. & beloborodov a. m. , 2005 , apj , 634 , 565 tiengo a. et al . , 2010 , , 710 , 227 van der horst a. j. et al . , 2011 , in preparation watts a. l. et al . , 2010 , , 719 , 190 woods p. m. , thompson c. , 2006 , in compact stellar x - ray sources , eds . lewin @xmath188 m. van der klis , cambridge astrophysics series , 39 , p.547 woods p. m. et al . , 2008 , atel , 1691 xu y. et al . , 2006 , science , 311 , 54 _ clllccccccc 1 & bn080822.529 & 12:41:56.914 & 8 , 7 , 4 & @xmath189 & @xmath190 & @xmath191 & @xmath192 & @xmath193 & @xmath194 & @xmath195 + 2 & bn080822.647 & 15:36:35.200 & 9 , 10 & @xmath196 & @xmath197 & @xmath198 & @xmath199 & @xmath200 & @xmath201 & @xmath202 + 3 & bn080822.981 & 23:32:57.746 & 2 & @xmath203 & @xmath204 & @xmath205 & @xmath206 & @xmath207 & @xmath208 & @xmath209 + 4 & bn080823.020 & 00:28:09.904 & 3 , 4 & @xmath210 & @xmath211 & @xmath212 & @xmath213 & @xmath214 & @xmath215 & @xmath216 + 5 & bn080823.091 & 02:11:36.630 & 10 , 11 & @xmath217 & @xmath218 & @xmath219 & @xmath220 & @xmath221 & @xmath222 & @xmath223 + 6 & bn080823.174 & 04:10:19.280 & 0 , 1 & @xmath224 & @xmath225 & @xmath226 & @xmath227 & @xmath228 & @xmath229 & @xmath230 + 7 & bn080823.248 & 05:56:31.529 & 2 & @xmath231 & @xmath232 & @xmath233 & @xmath234 & @xmath235 & @xmath236 & @xmath237 + 8 & bn080823.293 & 07:01:09.967 & 3 , 0 , 1 , 5 & @xmath238 & @xmath239 & @xmath240 & @xmath241 & @xmath242 & @xmath243 & @xmath244 + 9 & bn080823.293 & 07:04:22.610 & 3 , 0 , 1 , 5 & @xmath245 & @xmath246 & @xmath247 & @xmath248 & @xmath249 & @xmath250 & @xmath251 + 10 & bn080823.319 & 07:39:32.257 & 9 , 10 & @xmath252 & @xmath253 & @xmath254 & @xmath255 & @xmath256 & @xmath257 & @xmath258 + 11 & bn080823.330 & 07:55:45.690 & 4 , 3 , 8 , 7 & @xmath259 & @xmath260 & @xmath261 & @xmath262 & @xmath263 & @xmath264 & @xmath265 + 12 & bn080823.354 & 08:30:01.633 & 11 , 8 & @xmath266 & @xmath267 & @xmath268 & @xmath269 & @xmath270 & @xmath271 & @xmath272 + 13 & bn080823.429 & 10:18:13.891 & 0 , 1 , 3 , 5 & @xmath273 & @xmath274 & @xmath275 & @xmath276 & @xmath277 & @xmath278 & @xmath279 + 14 & bn080823.478 & 11:27:32.306 & 8 , 4 & @xmath280 & @xmath281 & @xmath282 & @xmath283 & @xmath284 & @xmath285 & @xmath286 + 15 & bn080823.623 & 14:56:23.563 & 10 , 11 & @xmath287 & @xmath288 & @xmath289 & @xmath290 & @xmath291 & @xmath292 & @xmath293 + 16 & bn080823.714 & 17:08:49.038 & 9 , 10 & @xmath294 & @xmath295 & @xmath296 & @xmath297 & @xmath298 & @xmath299 & @xmath300 + 17 & bn080823.847 & 20:19:30.659 & 9 , 10 & @xmath301 & @xmath302 & @xmath303 & @xmath304 & @xmath305 & @xmath306 & @xmath307 + 18 & bn080823.847 & 20:23:42.822 & 9 , 10 & @xmath308 & @xmath309 & @xmath310 & @xmath311 & @xmath312 & @xmath313 & @xmath314 + 19 & bn080823.986 & 23:39:24.472 & 9 , 11 , 7 , 6 & @xmath315 & @xmath316 & @xmath317 & @xmath318 & @xmath319 & @xmath320 & @xmath321 + 20 & bn080824.054 & 01:17:55.394 & 2 , 5 & @xmath322 & @xmath323 & @xmath324 & @xmath325 & @xmath326 & @xmath327 & @xmath328 + 21 & bn080824.346 & 08:18:24.418 & 3 , 4 & @xmath329 & @xmath330 & @xmath331 & @xmath332 & @xmath333 & @xmath334 & @xmath335 + 22 & bn080824.828 & 19:52:51.264 & 2 , 10 & @xmath336 & @xmath337 & @xmath338 & @xmath339 & @xmath340 & @xmath341 & @xmath342 + 23 & bn080825.200 & 04:48:27.405 & 4 & @xmath343 & @xmath344 & @xmath345 & @xmath346 & @xmath347 & @xmath348 & @xmath349 + 24 & bn080825.401 & 09:37:42.158 & 4 , 3 , 8 & @xmath350 & @xmath351 & @xmath352 & @xmath353 & @xmath354 & @xmath355 & @xmath356 + 25 & bn080826.136 & 03:16:14.773 & 8 & @xmath357 & @xmath358 & @xmath359 & @xmath360 & @xmath361 & @xmath362 & @xmath363 + 26 & bn080826.236 & 05:40:19.425 & 9 , 10 & @xmath364 & @xmath365 & @xmath366 & @xmath367 & @xmath368 & @xmath369 & @xmath370 + 27 & bn080828.875 & 20:59:39.966 & 1 , 0 , 5 , 3 & @xmath371 & @xmath372 & @xmath373 & @xmath374 & @xmath375 & @xmath376 & @xmath377 + 28 & bn080903.421 & 10:06:35.329 & 4 , 5 & @xmath378 & @xmath379 & @xmath380 & @xmath381 & @xmath382 & @xmath383 & @xmath384 + 29 & bn080903.787 & 18:53:48.775 & 2 , 10 & @xmath385 & @xmath386 & @xmath387 & @xmath388 & @xmath389 & @xmath390 & @xmath391 + lcccccccc mean & @xmath392 & @xmath393 & @xmath394 & @xmath395 & @xmath396 & @xmath397 & @xmath398 & @xmath399 + @xmath22 & @xmath400 & @xmath401 & @xmath402 & @xmath403 & @xmath404 & @xmath405 & @xmath406 & @xmath407 + weighted mean & @xmath408 & @xmath409 & @xmath410 & @xmath411 & & & & +

we have long speculated , that 6 top + 6 anti - top quarks due to the realtively large size of the top - yukawa coupling would bind exceptionaly strongly by mainly higgs exchange . here we present a surprisingly simple `` calculation '' of the mass of this speculated bound state . even a possible resonance in scattering of two such bound states is speculated . for the `` calculation '' of the masses it is crucial to assume , that our since long speculated principle `` multiple point principle '' is true . this principle says : _ there are several vacua all having almost zero energy density_. further we make an approximation of the higgs yukawa potential essentially replacing the exponential in it by a step - function . the new result means that there are now two independent calls for our bound state having the mass around 750 gev required by our `` new law of nature '' the multiple point principle . it should be remarked that in our picture there is _ no new physics _ in the sense of new fundamental particles , but the `` multiple point principle '' is new in the sense of being not yet accepted . further we get the _ same _ mass within uncertainties as earlier but now from a completly different assumption , except for being from our `` multiple point principle '' . but the two masses are gotten from using _ different _ ( speculative ) vacua occuring in the pure standard model .

1607.07907

recently we have proposed @xcite that the 750gev - resonance@xcite - which now turned out to be only a statistcal fluctuation@xcite - were our long speculated and waited for bound state@xcite of 6 top + 6 anti top quarks . the very suggestive existence of the bound state as well as the in the present article to be presented `` calculations '' of the masses of the bound state and of a compositum of two of these bound states is heavily based on our the even longer ago speculated idea of a new principle , `` multiple point principle '' @xcite , causing restrictions between coupling constants and thus potentially serving as a candidate for solving fine tuning problems@xcite . this `` multiple point principle '' ( = mpp ) says , that there are several vacua with extremely small energy densities . we could also say , that it means that the universe - vacuum is just at some multiple point , where several phases can coexist , much like one at the triple point for water has coexistence of ice , fluid water , and vapor for a common set of intensive variables , pressure and temperature . there may be no real good derivation or argument for our multiple point principle in spite of the fact , that we and others have publiched some attempts to derive this principle @xcite , but all such arguments would have to involve some influence of the future on the passed , at least on the coupling constants , and that would make all such derivations of mpp suspicious . the reader should rather take some previous works - even _ pre_diction(s ) - as well as the results of the present work as _ empirical _ evindence for this new law of nature , the `` multiple point principle '' . the calculation to be presented below is indeed just a slightly renewed version of the calculation delivered in the appendices of our earlier article @xcite , in which it is heavily used , that there should exist a new vacuum degenerate in energy density with the present one and with an approximated structure , as if the s - particles(what we called our bound state , which so much happened to fit the by now digamma fluctuation called in its fashion f , that we shall call it f now ) were sitting as the carbon atoms in the diamond chrystal , as we shall review in section [ modelling ] . colin froggatt and i ended in these appendices@xcite with a mass 260 gev for the bound state , but we certainly did not believe our mass _ pre_diction 260 gev to be very accurate . for the trustworthiness of our whole story the recent work of larisa laperashvili , chitta das and myself @xcite relating the _ mass of _ the speculated _ bound state to _ the degree of instability/_negative energy density _ of the second minimum in the standard model higgs field effective potential calculted without inclusion of our bound state . the point is that in order to achieve just zero energy density ( as our multiple point principle requires ) for the vacuum represented by the second minimum the correction required is just getting right for the mass of the bound state f being in a range very close to 750 gev . we shall return to this work in the last subsection in the conclusion [ larisaetal ] . in the following section [ boundmpp ] we shall review our model of there existing an exceedingly strongly bound system of 6 top + 6 anti top quarks , and of our `` multiple point principle '' fine tuning the coupling constants , so that for instance a condensate of the bound state can fill the vacuum and cause a `` new vacuum '' with the energy density just finetuned to be again remarkably small , of the same order as say the astronomical observation of the energy density(= cosmological constant ) of the vacuum , in which we live . ( this astronomically observed cosmological constant is quite negligible compared to the energy densities of any significance for high energy physics parameters such the bound state mass or the higgs mass ) . a subsection [ mpp ] of this section [ boundmpp ] is assigned to our `` new law of nature'',``multiple point principle '' . in the next section [ modelling ] we then model in a very crude approximation and in a non - relativistic picture the just mentioned new vacuum ] in our model called the `` condensate vacuum '' as containing a bose - condensate of the f bound states by suggesting as a very crude approximation , that this vacuum has a system / a lattice of f(750 ) particles interacting with their neighbors contained in the vacuum . we take the `` atoms''= the f s in this lattice to interact in analogy with the carbon atoms in a diamond chrystal . in order to compare the interactions and the binding energies we ignore the effect , that when a top quark goes around /is bound to a swarm of with the same radius bound quarks and antiquarks , it only `` feels '' the force from about half the number of particles in the swarm . however , we argue in section [ accident ] , that the exchange of what we call `` eaten higgses '' , and which really is exchange of the longitudinal components of weak gauge bosons w and z , _ happens _ with help of gluons also just accidentally to cancel this effect . it is very important for the success of our whole picuture of bound states and a vacuum condensate numerically , that inside the bound state f as well as in the condensate vacuum the effective higgs - mass is appreciably lower than the higgs mass of 125 gev observed experimentally . since , however , the higgs field expectation value inside the bound state and inside the condensate vacuum is significantly lower than in the usual vacuum , the higgs self - interaction indeed cause a smaller higgs mass effectively in these places with many top and anti top around on the average . this deminished effective higgs mass is discussed in section [ effhiggsmass ] . in section [ potentialapproximation ] we then for simplicity make the very crude assumption of approximating the exponential factor @xmath0 in the higgs - yukawa - potential by a step - function , a @xmath1 meaning , that we put the higgs _ mass _ to zero for small distances , while we put the higgs _ yukawa potential _ to zero for large distances . next in section [ mass750 ] we `` calculate '' or rather very crudely estimate the mass of the bound state f _ from the requirement of the mpp assumption of the equality of the energy density of the `` condensate vacuum '' and the vacuum , we live in_. so our new principle mpp is really crucial for our mass prediction ! in section [ mass1c8 ] we also with the same picture discuss the at lhc actually first found possible resonance -of mass 1.8 tev - seen ( ? ) decaying into weak gauge bosons(it is very dobtful ) . we take this resonance to be composite from a couple of f s as very weakly suggested from the mass of the 1.8 tev ( we shall see combining the present work with our earlier work with laperashvili and das @xcite that an f - mass around 800 gev is called for ) possible particle being crudely twice that of f. a little problem for our interpretation of the 1.8 tev state this way may be its relatively small width observed . the problem is dicussed a bit in the subsection [ width ] . in section [ conclusion ] we review and comment our result . in the subsection [ larisaetal ] we summarize , that the value for the bound state estimated in the present article - developping the result of @xcite - and the value for the bound state needed for a quite different multiple point principle requirement coincides remarkably ! the crucial suggestion behind our bound state model of 6 tops + 6 anti tops that since higgs exchange like any other even order tensor particle exchange delivers attracktion between top and top , or top and anti top , or anti top and anti top as well , we get stronger and stronger binding between the top and anti top quarks the more of them we imagine brought together . it is because the top and anti top are the strongest binding quarks , that this type of binding becomes most relevant for the top and anti top . now . however , the quarks are fermions and thus you can not just unlimmited clump arbitrarily many , e.g. top quarks , together . since the top quark has a color degree of freedom taking three values , say : red , blue , and yellow , and a spin degree of freedom , that can be up and down , one can bring _ 3 * 2=6 top quarks into the same orbital state _ , but because of fermi - statistics no more . so there can in a single orbital state be just up to 6 top + 6 anti top . thereby a _ closed shell _ is so to speak formed ( in the nuclear physics sense ) . in the zero higgs mass approximation , which will be effectively valid , when the size of the bound state - the radius - multiplied by the effective higgs mass is small , the attracktion between the top - quarks or between tops and anti tops is quite analogous to that between an atomic nucleus and an electron . so we can for first orientation use the terminology from the quantum mechanical description of atomic physics . approximating the bound state , that we suggest to be possible to form from 6 top + 6 anti top by thinking of each top or anti top going arround a collected object formed from the other 11 quarks , we can talk about different `` orbits '' in the atomic terminology of a main quantum number @xmath2 taking positive integer values and further @xmath3 ( the orbital angular momentum magnitude being @xmath4 ) and @xmath5(the angular momentum around the quantization axis ) . as in atomic physics the particles in the n=1 orbit are bound strongest . analogous to the helium atom having especially high excitation energies , we have because of the color factor 3 and both quark and anti quark an especially stable system being a bound state of 6 top and 6 anti top quarks . whether the binding of such a system of 6 top + 6 anti top now is sufficiently strong to even bind to form a resonance , with a rather small mass ( as we shall need say about 750 gev ) compared to the collective mass of 6 top + 6 anti top , 12 @xmath6 = 12 * 173 gev = 2076 gev , is controversial@xcite . however , we think ourselves@xcite , that making use of a long series of corrections , especially also exchange of the other three components of the higgs than the as simple particle observed component , we can stretch the uncertanties in the calculation so far as to allow such a light bound state to be possibly formed@xcite . these other components of the higgs are really present in the standard model as w s and z longitudinal components . we call them `` eaten higgses '' , but really of course it just means to include weak gauge particle z and w exchanges . it is important for our hope , that the bound state can indeed bind so strongly , that it gets so tightly bound , that the strong higgs fields inside the bound state even can modify the effective mass of the higgs significantly there . _ we _ estimated that a top - yukawa coupling of @xmath7 would be just sufficient to bind an extremely light bound state of the 6 top and 6 anti top , and would match with the experimental top - yukawa @xmath8 . but shuryac et al . @xcite find , that due to the high higgs mass , it can not bind at all for the experimental value of @xmath9 . the whole speculation about our bound state of 6 top + 6 anti top is a priori rather much taken out of the air by itself . however , we have all the time proposed it only connected with another speculation , the `` multiple point principle '' . this is , you could say , a wild guess about simplifying the fine tuning problems of the standard model . in order to formulate just the cosmological constant problem about , why the cosmological constant(= the vacuum energy density ) compared to say planck scale dimensional expectations is so enormously small , one needs an _ assumption _ of the form `` the energy density of vacuum is extremely small ! '' now you could look at the `` multiple point principle '' as an extension of this anyway needed assumption , without really complicating it severely : `` several vacua have extremely small energy densities ! '' . we almost just have put the anyway needed assumption into `` plural '' , or changed the `` quantor '' from `` the physical vacuum ... '' to `` several vacua ... '' . now the real supporting point for this principle is , that although it is not unneccesarily complicated , it is the one , which colin froggatt and myself managed to use to make historically@xcite in 1996 , long before the higgs particle were found , to make a _ pre_diction of the higgs mass of 135 gev @xmath10 10 gev . now our prediction using the same multiple point principle would be 129.4 gev@xcite but with a much smaller uncertainty , comparable to the experimental uncertainty of a few hundred mev . so at first it then looks , that while our original _ pre_diction agreed perfectly within errors , and the multiple point principle were perfectly right , it is today deviating of the order of three standard deviations from matching experiment . this formal disagreement of the theoretical prediction actually occured in spite of , that the better calculations and better top mass moved our prediction _ closer _ to the experimental value 125 gev during the time we had predited it . it is of course only possible that in spite of this development the agreement relative to the uncertainty could become worse , because the uncertainties in calculation and top and higgs masses went down even faster . however , l.v . laperashvili , c. das and myself@xcite found , that the existence of the bound state f of the 6 top + 6 anti top would make a _ little theoretical correction to the mass of the higgs _ being predicted from the multiple point , so that the agreement might indeed be improved to be perfect , if the mass of this bound state is appropriate . a mass of the bound state @xmath11 would be fine for correcting the higgs mass to be observed to agree with a perfect degeneracy of the vacua . it should be stressed , that this successfull higgs - mass _ pre_diction as well as colin d. froggatts and mine controversial argument , that the top - yukawa - coupling @xmath9 in order to allow for a condensate of bound states of 6 top and 6 antitop with energy density close to zero , must be close to the value @xmath12 supports the `` multiple point principle '' as being a principle uphold by nature . the value @xmath13 namely matches with the experimentally determined higgs yukawa coupling @xmath14 . really we just estimated , what the top - yukawa coupling should be in order , that the bound state assumed to exist of 6 top + 6 anti top should have exceptionally low mass . but this should be approximately needed to have the condensating particle have mass close to zero in order for there being two degenerate vacua as required by mpp(=``multiple point principle '' ) . this means , that even if the theoretical arguments for the mpp are not totally convincing , then there is some _ empirical evidence _ pointing in favor of this mpp . and the present article is meant to provide one more such indirect phenomenological support for mpp . since it is very difficult to treat bound states by the true bethe - salpeter@xcite , we tend to use instead non - relativistic appoximations . in spite of the fact that we consider the `` condensate vacuum '' a condensate of these bound states , and that they thus approximately all should be in the same quantum single - particle - state , we tend to think that the exact boson condensate is less important than that the particles have the mutual configuration dimishing the energy density . we thus suppose that for the purpose of estimating the energy density of such a vacuum it is advisable to use rather some chrystal model for this vacuum , so that the way the particles may have arranged themselves relative to each other is hopefully more realistically taken into account . since the constituents - the quarks and anti quarks - are fermions , we imagine , that the different bound states can not penetrate into each other more deeply than this fermi - statistics allows . the fermions in one f neighbor to an other one can not go deeper in than to the level with the main quantum number ( in the atomic physics language ) n=2 . the n=1 level is namely occupied by the constituents of the first f. now there are for n=2 four orbital states , one 2s and 4 2p - states , each of which in the case of say top quarks can contain 2 * 3 = 6 allowing 4 * 6 = 24 top quarks as constituents in the closest neighbor f s to the given one . there are of course also similarly place for 24 anti top quarks in these neighbors . so there is in the closest layer of f s around a given one place for ( 24 + 24)/12 = 4 neighbors . thus the analogy with a material with tetra - valent atoms ( = having configuration number 4 ) such as carbon is called for . therefore diamond is a candidate for a model for this `` condensate vacuum '' . colin froggatt and i already used in an appendix in our article @xcite such a diamond model . via slightly indirect arguments we then very crudely reached a mass- prediction of 285 gev for the bound state . in the present article the estimate is made a bit more directly , and the result comes a bit closer to 750 gev . this analogy - choice means , that we decided to evaluate the energy density of the `` condensate '' , or perhaps rather a chrystal of f particles sitting in a diamond shaped lattice attrackted most importantly to their four nearest neighbors f s . the attracktion potential is suggestively approximately estimable by considering an f(750 ) neighbor to a given one having its 12 top and anti top quarks going around the latter in the main quantum number orbit n=2 . if one could use pure atomic physics and ignore the higgs mass , it would be wellknown that the binding energy in such an n=2 orbit is just 1/4 of the binding energy in the n=1 orbit , the rydberg . the binding energy of one top quark or of one anti top in an f is of course , if it consists of 12 quarks or anti quarks of the same sort @xmath15 for the case of top - quarks with mass @xmath16 and an ansatz of 750 gev for the mass of f as representing the two slightly different estimates given below for what our earlier work suggests@xcite , 710 gev and 850gev . if we ignore the screening effect , which is claimed to be just compensated for in section [ accident ] , the binding of top quark say due to a neighboring f in the diamond - like lattice is thus just 1/4 of this @xmath17 gev , i.e. it is 28 gev ( really 27.65 ) . our calculation below now consists in observing , that such 28 gev per constituent top or anti top means , that a full f is bound to a neighbor by 12 * 28 gev = 331 gev . now there are four neighbors but if we use that , we double count by a factor 2 , and so at the end the binding per f - particle in the chrystal believed to be an alternative vacuum of the type suggested by mpp is 4/2 times this 331 gev . that is to say , that the binding between neighbors in the diamond - like chrystal per f runs up to 4/2 * 331 gev = 662 gev . if this is taking into account our very crude estimates just equal to a mass 750 gev for the bound state . thus in our diamond chrystal model for the `` condensate vacuum '' we find that the einstein(/mass ) energy 750 gev of the f in the supposed vacuum is just cancelled by the binding to their neighbors , accurately as our multiple point principle ( mpp ) requires ! this is a coincience , that only happens for the mass of the resonance having the right value . but it is of course , if taken seriously , an evidence for our hypotesis the multiple point principle . below in section [ mass750 ] we shall calculate , what mass is it , that is required for this cancellation of binding energy against the einstein energy of the f - particles thus allowing the energy density of the `` condensate vacuum '' to be zero . the most important ( attracktional ) interaction causing the bound states such a low mass as @xmath18 750 gev say is in our picture the exchange of higgs particles , meaning that it is the yukawa potential . most simply we would make an approximation , in which the top and anti top constituents are kept together by a higgs field centered around the `` center '' of the f - particle (= the bound state(which _ we _ previously called s ) . then in order that this approximation should be a good one for a top or an anti top far away from the at ( relatively ) long distance from this `` center '' , we must make the strength of the higgs yukawa field around this `` center '' to be 12 or 11 times that of only one top or anti top . however , if at shorter distance we want the true strength of the yukawa potential , we should rather only use as the strength of this the potential from this `` center '' being the half of that , i.e. as the one from rather 11/2 or 12/2 tops or anti tops . in other words , if we decide to work in the crude approximation of the potential being simply mathematically a yukawa potential around a `` center '' , i.e. having the form @xmath19 then the constant @xmath20 should rather be proportional to 11/2 than to 11 , when there are 12 constituents in the bound state f. indeed if all the 11 other constituents than the one considered were concentrated at origo we would have @xmath21 while , if we , as we must realistically take it , have that on the average the constituents contributing to the attracktion of a given quark or anti quark are only the half of them being closer to the center than the given quark itself , then the yukawa coupling coefficient @xmath20 should rather be @xmath22 this complicates the calculation , especially since truly the potential inside the bound state is not of the simple mathematical form ( [ yukawa ] ) , but rather is a sum over several yukawa potentials , one for each of the constituents . now , however , we want in the present article aiming at a surprisingly simple `` calculation '' to consider a somewhat accidental compensation of this reduction in the higgs field strength due to the smear out of the center by the constituents away from the very center by some corrections . the corrections which froggatt and i considered in the article @xcite and which could potentially help are : * exchange of gluons * what we call u - channel exchange . * exchange of `` eaten higgses '' . these are various interactions studied by us in the article @xcite , which for the case of the two first ones at least can in the limit of small higgs mass be included simply by replacing the coefficient @xmath20 of the yukawa potential by a larger value . in our work @xcite we in fact correct the simple ( t - channel ) - which is one to use , if the interacting quark is not exchanging quantum numbers under the higgs exchange like in a u - channel scattering and the color field is screened and other exchanges ignored - yukawa potential @xmath23 by replacing it by a @xmath24 , which is given by @xmath25 with @xmath26 we got @xmath27 . this means that the binding energy ( 4 ) should be corrected to include the gluon exchange force by substituting @xmath28 however , even though this compared to the t - alone @xmath29 corredponding to @xmath30 is already roughly a doubling , the potential does at first to see not change its shape , and thus at first the effect of the effective charge being distributed , rather than concentrated in the center exactly , has not been corrected for . thinking a bit more this is , however , not quite true because : the gluon exchange is actually quite absent , when we ask for the exchange force between two neighboring f - particles , because the f s are from outside seen colorless . so when asking for the potential keeping the tops and anti tops together inside the f - bound state , the gluon force is there , but when asking for the potential between the two f(750 ) s interacting with each other , only the higgs exchange is present . this effect gives in fact some compensation for the mistake one would do by ignoring the smearing out of the central attracktor ( smearing out of the analogue of the atomic nucleus in atomic physics ) . it is more obvious that the third type of correction - `` the eaten higgs exchange '' - is _ not of the same shape _ and thus can make a difference . the point is that , when an `` eaten higgs '' , meaning really exchange of a longitudinal w or @xmath31 , is exchanged , the top will go into a different state , such as a left b - quark state instead . such a changed state does not bind equally strongly as the then missing top or anti top would have done . this means that the modified top or anti top being the left bottom quark or left anti bottom should rather quickly be brought back to become a top or anti top , if a large energy increase shall be avoided , and we of course think of the ground state . but some amplitude for there being a component of the modified top or anti top is calculable in second order perturbation theory . we may think of this interaction expressed just in terms of top and anti top effectively taking into account the second order perturbation , wherein the `` eaten higgs '' has been exchanged _ twise_. such a double eaten higgs exchange effectively represents , rather than the long range yukawa type interaction , a shorter range interaction having rather the form @xmath32 than the yukawa going rather @xmath33 . so the eaten higgs correction will change the effective shape of the potential . it namely contributes mainly only at short distances @xmath34 . at formally very small distances the three eaten higgses couple very similarly to the uneaten one , so one should think it would be like @xmath35 were increased by a factor 4 . but as the distance @xmath34 gets larger the eaten higgs attracktion will completely disappear . but taking it more in conformity with the speculations in our work@xcite it is the combined effect of two higgses ( eaten or not ) which goes by a feynman diagram with four @xmath9 factors , that goes up by the factor 4 being the number of eaten plus uneaten higses compared to the only one uneaten higgs . if we take this correction of the coupling square @xmath35 to mean that it gets increased by a fator 2 in the tight region , while there is approximately no change due to the eaten higgses for large distance r , then we could say : the @xmath35 effectively to be used for the main quantum number n=1 should be increased , at most though up to being doubled , while the force in the main quantum number n=2 orbit should be essentially unchanged by the eaten higgses . this would potentially compensate the effect of the only 11/2 or 12/2 instead of 11 or 12 due to the screening , that half the attrackting quarks or anti quarks are outside the quark , say , to be attrackted and thus do not provide any attracktion . however , as just said the factor 2 increase of the effective @xmath35 due to eaten higgses were rather an upper bound than the most honest estimate . but now it is then very good for obtaining an approximate compensation of making the error of ignoring the smear out of the cloud of the attrackting particles , that also the gluon exchange gives a contribution to the attracktion of the tops and anti tops being bound to form the f , while not contributing to the attracktion between neighboring f s . now what we are really going to use our `` accidental '' cacellation of the smear out of the consisuents for is to obtain the ratio of the binding energy of a constituent in the n=2 orbit to be just 1/4 of that in the n=1 . of course as long as the effetive higgs mass would be positive , there would allways be an error in this use because the n=2 binding energy will be numerically suppresed more the effective higgs mass than the n=1 orbit . the rudiment of the higgs mass non - zero effect will also contribute to counteract to the ignored effect of the smearing out of the constituents . so as a crude , more or less accidental , cancellation we shall simplify our calculation by taking it , that the eaten higgs correction helped by a couple of smaller effects drops out against the effect of half the quark or antiquark sources being outside for the n=1 orbit . this then means that we can formally allow ourselves to calculate _ as if _ there were no sources - outside - effect and no eaten higgs effect , and even as if the gluon attracktion had same coupling in both n=1 and n=2 , being totally absorbable in the higgs exchange . so we could go on very simply both ignoring gluons - absorbed into higgs - and the smearing out of the attrackting cloud , being compensated for by other effects . this would mean , that we could calculate as if : _ each quark were bound by a completely central point yukawa potential and the very same strength could be used then for all the orbits n=1 , n=2 , and so on . _ that is to say we could claim , that after this cancellation - a bit accidentally - the binding energy of say a quark in the zero higgs mass limit would be just bound by a factor 1/4 in the n=2 orbit compared to its binding in an n=1 orbit . the zero higgs mass approximation is , however , not so obvious , and if we can not use it , then of course the binding in the n=2 orbit might not even be 1/4 of that in the the n=1 orbit as we claimed after our assumed cancellation , see section [ accident ] . we already mentioned , that we here would make another rather drastic approximation : we would replace the exponential factor in the yukawa potential by a theta - function like function . that is to say we would in the different situations , meaning respectively * inside the condensate of f s , and * in the ordinary / physical vacuum , in which we live . use _ different effective higgs masses and thus different theta - like functons for the exponential factor in the yukawa - potential _ since the average higgs field in the condensate vacuum is clearly smaller than in the ordinary or physical one , the effective higgs mass will due to self - interaction easily be seen to be also smaller in the condensate than in the ordinary vacuum . this is seen from the following : * since say the top quark gets its mass from the interaction with the vacuum higgs field , it is obvious that a diminishing numerically of the higgs field just around a top quark would lower the energy / mass . this possible adjustment to lower the energy is in fact what brings about the yukawa potential around say a top quark . to minimize the energy by adjusting the higgs field around a top it actually pays energetically to let the higgs field diverge infinitesimally near to the top , but in the surrounding region the higgs field further and further away go back to its usual vacuum expectation value . this is how the yukawa potential comes about . * if the higgs field in some region remains very close to some value @xmath36 say , then small deviations in the higgs field @xmath37 from this @xmath36 will behave as if the higgs mass squared were @xmath38 . now the higgs effective potential actually has such a form , that when the value is numerically lowered the second derivative also becomes lower , end even at some point becomes zero and then negative . it is wellknown that for zero higgs field the effective higgs mass square in our sense here is the tachyonic higgs mass square , i.e. negative . that of course in turn means that the zero higgs mass approximation gets better in the condensate than in the ordinary vacuum . it is therefore possible and we shall assume that - hopefully after a numerical estimate , that it happens to be so approximately - while for the condensate vacuum the zero higgs mass can be used _ including _ the n=2 orbit , i.e. for orbit n=1 _ and _ n=2 , in the ordinary vacuum only the n=1 orbit allows the zero higgs approximation , for the higher ones we let the yukawa potential completely be approximated by zero(this corresponds to infite higgs mass relative to the inverse radius for these higher than or equal to n=3 orbits ) . the here suggested treatment of the yukawa potential is not exactly to take it as a step function , but it is approximately so . to summarise the rule suggested - and to be confirmed by some estimations - : * condensate vacuum in the condensate vacuum we take the higgs mass zero for orbits n=1 , and 2 , and huge , say infinite , for n=3,4 , ... * present vacuum while in the vacuum , we live in , we instead take higgs mass zero ( only ) for n=1 , while infinite or huge for n=2,3,4 , ... the basic requirement used in the present `` calculation '' of the mass of bound state s = f of 6 tops + 6 anti tops is , that in a presumably too naive nonrelativistic thinking the energy density of the condesate vacuum shall be zero . of course this zero should be understood as being compared to the background being identified with the vacuum , in which we live . taking the zero to be relative to energy density in the usual vacuum , means that we only have to include those energy carying ingredients , which are not also present in the usual vacuum , in which we live . since the `` condensate vacuum '' is characterized by its extra f - particles , it is the energy density resulting from these particles and their mutual interactions , that should be added up and required by mpp to be zero . that is to say we shall assume , that the energy in the condensate vacuum per f - particle , i.e. really @xmath39 , is zero . here we used the notation that @xmath40 is the density inside the condensate of the particles f , while the energy density of the condensate , counted relative to the vacuum , we live in , and which have no f s in first thinking ( of course there are some vacuum fluctuations , but that is not counted into neither @xmath40 nor @xmath41 , if the same is present in the vacuum , we live in . ) this mpp - requirement is written @xmath42 we shall indeed follow an appendix of our earlier work@xcite and assume , that the structure of the condensate can be approximated as being a diamond lattice structure , so that there are just @xmath43 , i.e. other f - particles surrounding each one of them in the lattice . when we count all the binding energy per f present in the condensate @xmath44 as being the number of neighbors @xmath45 times the binding of one f to its neighbor @xmath46 , we _ double count _ , because we count the same binding from both the one f and from the other one it binds . therefore we must have the denominator @xmath47 seen in the formula ( [ zerocond ] ) . we made then the approximation , that we can effectively consider it , that the neighboring top quarks and anti topquarks contained in an f neighboring to another one are _ in effect in the n=2 orbit _ of the latter . thus we can take the binding energy of a neighboring f to a given one @xmath46 to be as , if the top and anti tops were in an n=2 orbit or some superposition thereof . thus the binding of the neighbors occur with binding energy @xmath48 . as long as we can take the effective higgs mass for the two lowest orbits n = 1 and 2 to be zero , we can count , that the binding energy , for top say , in the orbit n=2 is just one quarter of that in the n=1 orbit , provided we can use the same potential of the form @xmath33 . but now that were , what our above discussion `` accidental cancellation '' in section [ accident ] should ensure , and so even for an f - particle , which consists of tops and anti tops the ratio of the binding energies should be @xmath49 . from the last step in ( [ zerocond ] ) we easily derive of course @xmath50 in the present picture the idea is , that the resonance doubtfully seen in the decay into z s or w s with mass 1.8 tev should be identified with a resonance consisting essentially of two f - particles , which in turn are the bound states of the 6 tops and 6 anti tops each . already the fact that the mass 1.8 tev is very crudely just that of two f - particles ( with mass as our bound state calculation based on mpp suggests ) , 2 * 750 gev = 1500 gev = 1.5 tev , suggests such a thinking . but now we shall estimate the mass of the 1.8 tev resonance by the following proceedure : _ if _ we could calculate as above that the attracktion in the n=1 and n=2 orbits were as if the higgs mass were zero , we could have one f bound to another one by having all the tops and anti tops of the one going into the n=2 orbit of the other one . then the binding would according to our rule used above be just 1/4 of the binding of one f - particle @xmath51 . now , however , because of the effective higgs mass being smaller in the situation , wherein we have only two f - particles rather than a full condensate , as we thought about above , we should take it that the attracktion in the n=2 disappears , when going from the condensate to the system - the 1.8 particle we hope - of only two f s . this change we then treat as a perturbative correction to the first mass for the two f - bound state of @xmath52 due to the change of the potential energy between the two f - particles disappearing . now in potentials of the @xmath53 form , which we use as our approximation here , the virial theorem allows us to use , that the pototential energy is negative and just twise the binding energy . the energy or mass of the two - f - resonance will thus rather be _ increased _ compared to the mass of two f - particles than decreased by the amount of @xmath54 . that is to say the mass of the resonance - which we want to identify with the 1.8 tev finding - is given as @xmath55 which agrees - accidentally ? - wonderfully with the number 1.8 tev from the experiment ! when you think about our huge higgs mass approximation for the case of binding of just two f - particles , you see that formally there is no interaction at all between the two ! this would not allow a narrow resonance . therefore we need some helping story to excuse that the width after all gets so small , that we experimentally shall conceive of the resonance with mass 1.8 tev as a resonance at all . the excuse suggested is , that during the interaction assumed to be mainly given by higgs exchange in our calculation the two 6tops + 6 anti tops boundstates f are _ partly annihilated _ so as to be in reality not really two true f - particles , but rather some similar structures with a bit smaller numbers of constituents . we could still hope that although such a partial annihilations , that could go back and forth , while the two f s move arround , could help to decrease the decay rate and making the width of the 1.8 tev - resonance smaller , it would not severely modify our crude estimate above . the narower we can speculate the 1.8 tev resonance to be w.r.t . decaying into two f s the bigger we can hope for the partial width into other channels than the channel into two f s . since it is so far only seen in vector boson channels , it is needed to be speculated , that the two f s channel does not take away all the 1.8 tev particles . we have presented an overly simple `` calculation '' leading to there being due to the strong top - yukawa coupling and under the assumtion of a finetuning ensuring a with the normal vacuum degenerate one with a condensate of bound states s = f of six top + six anti top , two `` resonances '' with masses respectively @xmath56 these results were obtained in the philosophy , that the coupling constants - especially say the top - yukawa - coupling @xmath9 - are by the new principle `` multiple point principle '' adjusted / finetuned to make the energy density of a condensate of f - particles - which _ we _ earlier caled s - particles - ( the bound states of 6 top + 6 anti top quarks ) have just the same energy density as the usual vacuum ( in which we live ) . that were to say , that the interaction between the f - particles in the condensate should be just so strong an attracktion between them , that the total energy ( density ) of the condensate just becomes zero ( relative to the normal vacuum ) . that is to say the binding between the neighboring f - states just equals the mass - energies of these f - states . if one therefore considers our good agreement of the masses as an evidence for the truth of the assumptions having been used , then there is a significant evidence for our long speculated `` multiple point principle '' ! it must , however , be admitted that the present very simplified `` calculation '' were based on a very crude treatment of the yukawa potential representing the higgs exchange between the f - particles , which is the ( main ) interaction between these f - particles in our picture . in fact we approximated higgs exchange by letting the higgs be effectively massless , when the top - antitop quarks are in relative orbits with atomic main quantum number n=1 , while we let the higgs potential be totally cut away for n=3 , 4 , ... . for the main quantum number n=2 we made the more complicated assumption of letting it be either as for a zero mass higgs or cut down to zero according to the surroundings , which influence the average higgs field . indeed we took for n=2 the higgs exchange force to be like for massless higgs inside the condensate vacuum , while we put the exchnage potential to be zero , when applied in the two - f - system identified with the 1.8 tev excess seen decaying into weak gauge bosons . thus more severe calculations are to be performed to truly settle , if our calculations are right . it should in this connection be stressed , that since our model is in principle _ only standard model _ extended with our multiple point principle used to restrict the coupling constants , one should in principle be able to calculate whatever one wants . with a relatively strong coupling @xmath9 being the very basis for the whole story there is though of course the complication of not having in principle the basis for perturbation theory . a recent work by larisa laperashvili , chitta das and myself@xcite , in which we have the f - resonance give a little correction to the mass of the higgs , that should be measured relative to the one associated with the running self coupling at the weak scale , improves the agreement with exoeriment of requirement of multiple point principle for yet a vacuum . indeed we have in our picture a third vacuum ( in addition to the usual one and the condensate vacuum with its f(750 ) s in it ) , namely one with a very high higgs field expection value . according to standard model caculations witout our bound state the high higgs field vacuum has a slightly _ negative _ energy density ( compared to the two other vacua ) . however , we find a little correction depending among on other quantities on the mass of the bound state . we found that this bound state mass put to 750 gev would fit wonderfully and consider , that this fact strongly supports the truth of mpp . in fact it turns out that the mass 750 gev for the bound state f is perfect for the correction of ours just to bring the energy density of the vacuum with the high higgs field expectation value from its otherwise slightly negative value to zero . this means that our multiple point pinciple using a quite third vacuum , namely one with a high higgs field expectation value of the order of @xmath57 gev , leads to a need for a particle - our bound state indeed - with a mass about the 750 gev also . this means that now , when the f(750 ) fluctuation digamma once so fashionable turned out being only a statistical fluctuation , then we would nevertheless from multiple point principle get _ two different and essentailly independent reasons for our bound state to have the mass near the value 750 gev_. that is to say we would then claim , well our bound state should be with a mass close to 750 gev . we have already two calculations of this mass in different ways , although both originating from the same principle mpp , but involving quite different data to fit . in fact larisa laperashvili et al . @xcite uses a corection due to the diagram in order to correct the running self coupling of the higgs @xmath58 corresponding to the observed higgs mass 125 gev to go from the value obtained by degrassi et al . @xcite of @xmath59 at the high field scale @xmath60 to the value very accurately zero requied by multiple point principle(=mpp ) . since bound state f is an extended object we must include a formfactor , when using it . defining a quantity @xmath61 denoting the radius of the bound state measured with top quark compton wave length @xmath62 as unit by : @xmath63 we obtain a theoretical estimate @xmath64 crudely confirmed by a slightly different estimate . the dominant diagram / correction - the first and quadratic of the diagrams on the figure just above - is @xmath65 where we have the estimated or measured values @xmath66 using the after all rather small deviation from perfect mpp @xmath67 and requiring it to be cancelled by the correction from the bound state we get the requirement @xmath68 where @xmath69 , @xmath70 , @xmath71 and the factor `` @xmath72 '' were taken in to approximate some neglected diagrams , as the next on the figure . if a nearer study should show that the next diagrams add up to roughly as much as the first one should include the factor @xmath73 to take into account the neglected feynman diagrams correcting the higgs self coupling . the solution w.r.t . the mass of the bound state @xmath74 gives @xmath75 the `` without the @xmath73 '' means what one shall do if the first diagram indeed dominates strongly . in this way we got even two calculations for the bound state mass - using in addition crude estimation - @xmath76 the agreement of the value `` 692 gev '' with the estimate(s ) from the completly different vacuum with the high higgs field `` 850gev '' or `` 710 gev '' is encourraging and a support of our `` multiple point principle '' ! if our numbers are taken as so convincing that our picute should be taken seriously then we would have the consequences : * we must take our `` multiple point principle '' as a true new physical law , even the mechanism behind it may not be clear . * we must accept that the standard model except for `` smaller '' deviations , that are too small to significantly modify the running of the higgs self coupling , is valid all the way to about @xmath77 gev . some see - saw neutrinos may be acceptable , as long as they do not couple too strongly to the higgs to influence the running of its sef coupling . otherwise it would be very accidentally that a pure standard model calculation would give so consisten results . + so there would not be much place for new physics except for the various resonances formed from the boud states , because we now have a new regime of strong interactions that can only be treated by non - perturbative methods . ( the suggestion for the 1.8 tev resonance is an example for how there can be more particles in such a new strong interaction regime ) * there should be seen sooner or laler bound state of the 6 top + 6 anti top with a mass not far from 800 gev . but even if it thus looks a bit sad w.r.t . much new physics , one should not forget that having our multiple point principle established would be a strong element of new physics , perhaps then of an a bit unexpected type . i want to thank the niels bohr institute for allowing me to stay as emeritus . also i thank my near collaborators on the subject of the present article larisa laperashvili , colin d. froggatt and chitta das for the collaboration , without which the present article could never have come to being . further i thank for the comments i have got at mittag lefler insitute and the bled workshop `` beyond the standard models '' where i have had some presentation of essentially the present work . 99 f. dias , s. gadatsch , m. gouzevich , c. leonidopoulos , s. novaes , a. oliveira , m. pierini and t. tomei , `` combination of run-1 exotic searches in diboson final states at the lhc , '' jhep * 1604 * ( 2016 ) 155 doi:10.1007/jhep04(2016)155 [ arxiv:1512.03371 [ hep - ph ] ] . atlas collaboration , phys . b*710 * , 49 ( 2012 ) ; arxiv:1202.1408 . cms collaboration , phys . b*710 * , 26 ( 2012 ) ; arxiv:1202.1488 . cms collaboration , phys . d*92 * , 1 ( 2015 ) . atlas and cms collaborations , phys . * 114 * , 191803 ( 2015 ) . cms - collaboration , cms - pas - exo-16 - 027 , `` search for resonant production of high mass photon pairs using 12.9fb1 of proton - proton collisions at s=13 tev and combined interpretation of searches at 8 and 13 tev '' c. d. froggatt and h. b. nielsen , `` production and decay of 750 gev state of 6 top and 6 anti top quarks , '' arxiv:1605.03909 [ hep - ph ] . l. v. laperashvili , h. b. nielsen and c. r. das , `` new results at lhc confirming the vacuum stability and multiple point principle , '' int . j. mod . a * 31 * ( 2016 ) no.08 , 1650029 doi:10.1142/s0217751x16500299 [ arxiv:1601.03231 [ hep - ph ] ] . y. hamada , h. kawai and k. kawana , `` evidence of the big fix , '' int . phys . a 29 , no . 17 , 1450099 ( 2014 ) [ arxiv:1405.1310 [ hep - ph ] ] . y. hamada , h. kawai and k. kawana , `` weak scale from the maximum entropy principle , '' arxiv:1409.6508 [ hep - ph ] d.l . bennett and h.b . nielsen , int . a*9 * , 5155 ( 1994 ) . bennett and h.b . nielsen , int . a*14 * , 3313 ( 1999 ) . bennett , c.d . froggatt and h.b . nielsen , in proc . conf . on high en- ergy physics , glasgow , scotland , 1994 , eds . p. bussey and i. knowles ( iop publishing ltd . , 1995 ) , bennett , c.d . froggatt and h.b . nielsen , in perspectives in particle physics 94 , eds . d. klabu.car , i. picek and d. tadi.c ( world scientific , singapore , 1995 ) , p. 255 . froggatt and h.b . nielsen , phys . b*368 * , 96 ( 1996 ) . laperashvili , yad . fiz . * 57 * , 501 ( 1994 ) [ phys . atom . nucl . * 57 * , 471 ( 1994 ) ] . froggatt , l.v . laperashvili , r.b . nevzorov and h.b . nielsen , phys . nucl . * 67 * , 582 ( 2004 ) [ yad . fiz . * 67 * , 601 ( 2004 ) ] , arxiv : hep - ph/0310127 . froggatt , l.v . laperashvili and h.b . nielsen , phys . nucl . * 69 * , 67 ( 2006 ) , hep - ph/0407102 . bennett , l.v . laperashvili and h.b . nielsen , _ relation between fine structure constants at the planck scale from multiple point principle _ , in proc . 9th workshop : what comes beyond the standard models , bled , slovenia , eds . m. breskvar et al . ( dmfa , zaloznistvo , ljubljana , 2006 ) , p. 10 , arxiv : hep - ph/0612250 . bennett , l.v . laperashvili and h.b . nielsen , _ finestructure constants at the planck scale from multiple point principle _ , in proc . 10th workshop on what comes beyond the standard model , bled , slovenia , 1727 jul 2007 , arxiv:0711.4681 . froggatt , r.b . nevzorov and h.b . nielsen , _ smallness of the cosmological constant and the multiple point principle _ , j. phys . . ser . * 110 * , 072012 ( 2008 ) , arxiv:0708.2907 . froggatt , r.b . nevzorov , h.b . nielsen , a.w . thomas , phys . b*737 * , 167 ( 2014 ) , arxiv:1403.1001 . froggatt , r.b . nevzorov , h.b . nielsen , a.w . thomas , _ on the smallness of the cosmological constant in sugra models with planck scale susy breaking and degenerate vacua _ , 2015 european physical society conference on high energy physics ( eps - hep 2015 ) , 22 - 29 jul , 2015 , vienna , austria ; arxiv:1510.05379 . h. b. nielsen and m. ninomiya , `` degenerate vacua from unification of second law of thermodynamics with other laws , '' bled workshops phys . * 12 * ( 2011 ) no.2 , 199 [ hep - th/0701018 ] . stillits , cand . thesis at the niels bohr institute c.d . froggatt and h.b . nielsen , _ origin of symmetries _ , world scientific , singapore , 1991 . das and l.v . laperashvili , int . j. mod . a*20 * , 5911 ( 2005 ) . n. cabibbo , l. maiani , g. parisi and r. petronzio , nucl . b*158 * , 295 ( 1979 ) . hung , phys . * 42 * , 873 ( 1979 ) . flores , m. sher , phys . rev . d*27 * , 1679 ( 1983 ) . m. lindner , z. phys . * 31 * , 295 ( 1986 ) . bennett , h.b . nielsen and i. picek , phys . b*208 * , 275 ( 1988 ) . m. sher , phys . rept . * 179 * , 273 ( 1989 ) . m. lindner , m. sher , h.w . zaglauer , phys . b*228 * , 139 ( 1989 ) . arnold , phys . rev . d*40 * , 613 ( 1989 ) . g. anderson , phys . b*243 * , 265 ( 1990 ) . p. arnold and s. vokos , phys . d*44 * , 3620 ( 1991 ) . c. ford , d.r.t . jones , p.w . stephenson , m.b . einhorn , nucl . b*395 * , 17 ( 1993 ) . m. sher , phys . b*317 * , 159 ( 1993 ) . g. altarelli and g. isidori , phys . b*337 * , 141 ( 1994 ) . casas , j.r . espinosa , m. quiros , phys . b*342 * , 171 ( 1995 ) . espinosa , m. quiros , phys . lett . b*353 * , 257 ( 1995 ) . casas , j.r . espinosa , m. quiros , phys . b*382 * , 374 ( 1996 ) . b. schrempp and m. wimmer , prog . * 37 * , 1 ( 1996 ) . froggatt , h.b . nielsen , and y. takanishi , _ standard model higgs boson mass from borderline metastability of the vacuum _ , rev . d*64 * , 113014 ( 2001 ) . v. brancina and e. messina , phys . 111 * , 241801 ( 2013 ) , arxiv:1307.5193 . v. brancina , e. messina and m. sher , phys . rev . d*91 * , 013003 ( 2015 ) , arxiv:1408.5302 . v. brancina , e. messina and a. platania , jhep * 1409 * , 182 ( 2014 ) , arxiv:1407.4112 . g. degrassi , s. di vita , j. elias - miro , j.r . espinosa , g.f . giudice , g. isidori and a. strumia , jhep * 1208 * , 098 ( 2012 ) ; arxiv:1205.6497 . d. buttazzo , g. degrassi , p.p . giardino , g.f . giudice , f. salab , a. salvio , + a. strumia , jhep * 1312 * , 089 ( 2013 ) ; arxiv:1307.3536 . g. isidori , g. ridolfi , a. strumia , nucl . b*609 * , 387 ( 2001 ) . espinosa , g.f . giudice and a. riotto , jcap 0805 ( 2008 ) 002 . j. ellis , j.r . espinosa , g.f . giudice , a. hoecker and a. riotto , phys . b*679 * , 369 ( 2009 ) . j. elias - miro , j.r . espinosa , g.f . giudice , g. isidori , a. riotto , a. strumia , phys . b*709 * , 222 ( 2012 ) . froggatt and h.b . nielsen , _ trying to understand the standard model parameters_. invited talk by h.b . nielsen at the `` xxxi itep winter school of physics '' , moscow , russia , 18 - 26 february , 2003 ; surveys high energy phys . * 18 * , 55 - 75 ( 2003 ) ; hep - ph/0308144 . froggatt , h.b . nielsen and l.v . laperashvili , _ hierarchy - problem and a bound state of 6 t and 6 anti - t , _ in : proceedings of coral gables conference on launching of belle epoque in high - energy physics and cosmology _ ( cg 2003 ) _ , ft . lauderdale , florida , 17 - 21 december , 2003 . froggatt , h.b . nielsen and l.v . laperashvili , int . j. mod a * 20 * , 1268 ( 2005 ) ; hep - ph/0406110 . froggatt and h.b . nielsen , phys . d * 80 * , 034033 ( 2009 ) ; arxiv:0811.2089 . froggatt , h.b . nielsen , _ hierarchy problem and a new bound state _ , in proc . to the euroconference on symmetries beyond the standard model , p.73 , slovenia , portoroz , 2003 ( dmfa , zaloznistvo , 2003 ) ; arxiv : hep - ph/0312218 . froggatt , _ the hierarchy problem and an exotic bound state , _ in : proceedings of 10th international symposium on particles , strings and cosmology , + _ ( pascos 04 ) _ , boston , massachusetts , 16 - 22 aug , 2004 . published in : `` boston 2004 , particles , strings and cosmology '' , pp.325 - 334 ; hep - ph/0412337 . froggatt , l.v . laperashvili and h.b . nielsen , _ a new bound state 6 t + 6 anti - t and the fundamental - weak scale hierarchy in the standard model , _ in : proceedings of 13th international seminar on high - energy physics : _ quarks-2004 _ , pushkinskie gory , russia , 24 - 30 may , 2004 ; hep - ph/0410243 . froggatt , l.v . laperashvili and h.b . nielsen , phys . * 69 * , 67 ( 2006 ) [ yad . fiz . * 69 * , 3 ( 2006 ) ] ; hep - ph/0407102 . froggatt , l.v . laperashvili , r.b . nevzorov and h.b . nielsen , _ the production of @xmath78 bound state at colliders_. a talk given by h.b . nielsen at cern , 2008 , preprint cern - ph - th/2008 - 051 . froggatt , l.v . laperashvili , r.b . nevzorov , h.b . nielsen and c.r . das , _ new bound states of top - anti - top quarks and t - balls production at colliders ( tevatron , lhc , etc . ) _ , report chep - pku-1 - 04 - 2008 , cern ( 2008 ) ; arxiv:0804.4506 . c.d . froggatt and h.b . nielsen , in : proceedings to the 34th international conference on high energy physics ( ichep 2008 ) , 30 jul - 5 aug 2008 , philadelphia , pennsylvania ; arxiv:0810.0475 . das , c.d . froggatt , l.v . laperashvili and h.b . nielsen , int . a * 26 * , 2503 ( 2011 ) ; arxiv:0812.0828 . froggatt , c.r . das , l.v . laperashvili and h.b . nielsen , _ new indications of the existence of the 6 top - anti - top quark bound states in the lhc experiments _ , a talk by l.v . laperashvili at the conference `` quarkonium-2012 '' , moscow , russia , mephi , november , 12 - 16 , 2012 , arxiv:1212.2168 [ hep - ph ] ; yad . * 76 * , 172 ( 2013 ) . froggatt , c.r . das , l.v . laperashvili , h.b . nielsen , int . a*30 * , 1550132 ( 2015 ) ; arxiv:1501.00139 . lhc seminar `` atlas and cms physics results from run 2 '' , talks by jim olsen and marumi kado , cern , 15 dec . http//indico.cern.ch / event/442432/. atlas collaboration , atlas - conf-2015 - 081 , `` search for resonances decaying to photon pairs in 3.2 @xmath79 of pp collisions at @xmath80 = 13tev with the atlas detector '' . cms collaboration , cms pas exo- 15 - 004 , `` search for new physics in high mass diphoton events in proton - proton collisions at 13tev '' ; http//indico.cern.ch / event/442432/. atlas collaboration ( g. aad et al . ) , phys . lett . * 114 * , 081802 ( 2015 ) . particle data group collaboration ( k.a . olive ( minnesota u. ) et al . ) , 2014 , 1676 pp . c*38 * , 090001 ( 2014 ) . atlas , cms , d0 collaborations , arxiv:1403.4427 . bezrukov , m. shaposhnikov , phys . b*659 * , 703 ( 2008 ) . bezrukov , a. magnin , m. shaposhnikov , phys . b*675 * , 88 ( 2009 ) . bezrukov and d. gorbunov , jhep * 1307 * , 140 ( 2013 ) , arxiv:1303.4395 . f.l . bezrukov and m. shaposhnikov , j. exp . . phys . * 120 * , 335 ( 2015 ) [ zh . * 147 * , 389 ( 2015 ) ] , arxiv:1411.1923 . f. bezrukov , j. rubio and m. shaposhnikov , _ living beyond the edge : higgs inflation and vacuum metastability _ , phys . d*92 * 8 , 083512 ( 2015 ) , arxiv:1412.3811 . mihaila , j. salomon and m. steinhauser , phys . * 108 * ( 2012 ) 151602 . k. chetyrkin and m. zoller , jhep * 06 * , 033 ( 2012 ) . f. bezrukov , m.yu . kalmykov , b.a . kniehl , m. shaposhnikov , jhep * 1210 * , 140 ( 2012 ) . froggatt , c.r . das , l.v . laperashvili , h.b . nielsen , int . a*30 * , 21 , 1550132 ( 2015 ) ; arxiv:1501.00139 . macpherson and b.a . campbell , phys.lett . b*306 * , 379 ( 1993 ) ; arxiv : hep - ph/9302278 . a. chodos , r.l . jaffe , k. johnson , c.b . thorn and v.f . weisskopf , phys.rev . d*9 * , 3471 ( 1974 ) . bardeen , m.s . chanowitz , s.d . drell , m. weinstein and t .- m . yan , phys.rev . d*11 * , 1094 ( 1975 ) . kuchiev , v.v . flambaum , e. shuryak , phys . rev . d*78 * , 077502 ( 2008 ) ; arxiv:0808.3632 . kuchiev , v.v . flambaum , e. shuryak , phys . b*693 * , 485 ( 2010 ) ; arxiv:0811.1387 . kuchiev , phys . rev . d*82 * , 127701 ( 2010 ) ; arxiv:1009.2012 . jean - marc richard , _ about the stability of the dodecatoplet _ , few body syst . * 45 * , 65 ( 2009 ) ; arxiv:0811.2711 .

we study soliton solutions to the dkp equation which is defined by the hirota bilinear form , where . the-functions are given by the pfaffians of certain skew - symmetric matrix . we identify one - soliton solution as an element of the weyl group of d - type , and discuss a general structure of the interaction patterns among the solitons . soliton solutions are characterized by skew - symmetric constant matrix which we call the-matrices . we then find that one can have-soliton solutions with being any number from to for some of the-matrices having only nonzero entries in the upper triangular part ( the number of solitons obtained from those-matrices was previously expected to be just ) . partially supported by nsf grant dms0404931 . supported by coe program at faculty of mathematics , kyushu university . ]

nlin0602031

we study soliton solutions of the dkp equation @xcite , @xmath11 where @xmath1 , and @xmath12 and @xmath13 are the usual hirota derivatives . for each @xmath14 , the variables @xmath15 and @xmath16 define the coupled kp equation ( see for example @xcite ) , @xmath17 this set of equations admits a class of particular solutions , called soliton solutions similar to those of the kp equation ( see for example @xcite ) , and in this paper we describe some properties of those solutions . the dkp equation was given as a first member of the dkp hierarchy in @xcite . in @xcite , the coupled kp equation was introduced as an extension of the kp equation whose solutions are given by the pfaffian form . the dkp equation was rediscovered as the pfaff lattice describing the partition function of a skew - symmetric matrix model in @xcite , and was also found as an orbit of some infinite - dimensional clifford group action in @xcite . the @xmath3 functions are given by the pfaffians of @xmath18 skew - symmetric matrices @xmath19 whose entries are denoted by @xmath20 for @xmath21 with @xmath22 , @xmath23 each coefficient @xmath24 gives a sign corresponding to the parity of the permutation , @xmath25 here the elements @xmath20 satisfy @xmath26 where @xmath27 and others are the symmetry parameters . a realization of @xmath20 is given by @xcite , @xmath28 where @xmath29 ( the same for @xmath30 ) and the functions @xmath31 and @xmath32 satisfy the same equations for @xmath33 and @xmath34 evolutions , @xmath35 for an example of finite dimensional solution , we consider @xmath36 with some constants @xmath37 for @xmath38 . the function @xmath39 is the exponential function , @xmath40 where @xmath41 and @xmath42 are arbitrary constants , and throughout this paper we assume the parameters @xmath43 to be ordered as , @xmath44 with those @xmath31 and @xmath32 , the elements @xmath45 become @xmath46 where @xmath47 , and @xmath48 . as a generalization of this form of @xmath20 , we consider an arbitrary @xmath49 skew - symmetric matrix @xmath50 . we then note that the @xmath18 matrix @xmath19 in the @xmath2-function @xmath3 can be expressed as @xmath51 where @xmath52 is a @xmath53 matrix with its transpose @xmath54 , and it is given by @xmath55 in this paper , we discuss a classification problem of several soliton solutions given by ( [ q ] ) in terms of the @xmath5-matrix . note here that @xmath56 for @xmath57 , since the case with @xmath58 gives a trivial solution , i.e. @xmath59 and @xmath60 . also note that one needs @xmath61 for @xmath62 . one should also note that if @xmath63 ( i.e. @xmath64 ) , the corresponding solution satisfies the kp equation . this condition can be of course obtained from the structure of the @xmath5-matrix . for example , if we take @xmath65 , then @xmath66 vanishes identically ( the size of the pfaffian is @xmath67 , but the independent exponentials are three or less ) . this implies that @xmath68 with @xmath65 gives a solution of the kp equation , and it gives either one kp soliton solution or a resonant y - shape kp soliton : with a @xmath69 @xmath5-matrix , we have @xmath70 the function @xmath71 gives one kp soliton solution if one of @xmath72 is zero ( with others being positive ) , and y - shape kp soliton if all @xmath72 are positive . for example , with @xmath73 , we have @xmath74 which leads to @xmath75 the asymptotic values of @xmath76 then take @xmath77 note here that the one soliton exchange the asymptotic values of @xmath78 and @xmath79 , that is , this one soliton permutes the numbers , @xmath80 . we then label one soliton of the kp equation as an element of the permutation group @xmath81 , in this case @xmath82 , the symmetry group of order 3 . we denote the one soliton solution ( [ asoliton ] ) by @xmath83 $ ] , and call this type of soliton a - soliton ( `` a '' stands for type a lie algebra which is the underlying symmetry algebra for kp equation ) . in general , we denote one a - soliton by @xmath84 $ ] , if the function @xmath76 exchanges @xmath85 with @xmath86 . we sometime identify @xmath84 $ ] as an element of the symmetry group . a generic solution of the dkp equation is then obtained for @xmath87 . in particular , we obtain one soliton solution in the case @xmath88 with the @xmath5-matrix having just two nonzero elements in the upper triangular part . for example , we consider the @xmath5-matrix having @xmath89 and all others @xmath90 for @xmath91 . then @xmath92 this gives @xmath93 with @xmath94 . for each asymptotic of @xmath95 , one of the exponential terms in the @xmath2-function becomes dominant . in the case of ( [ onedsol ] ) , we have @xmath96 thus , this one d - soliton exchanges @xmath97 with the values of @xmath76 where @xmath98 . this indicates a weyl - action of @xmath99-type : for example , @xmath100 may be expressed as @xmath101 we denote this d - soliton by @xmath102 $ ] . then identifying @xmath102 $ ] with the permutation of the pairs , @xmath103 , one has the relation @xmath102=[1:3]\cdot[2:4]=[1:4]\cdot[2:3]$ ] where @xmath84 $ ] is the permutation @xmath104 of a - soliton . this relation implies a resonant bifurcation of one d - soliton into two a - solitons ( see section [ twosoliton ] ) . since one d - soliton has four @xmath105-parameters , we need to have @xmath106 parameters to describe @xmath7 d - soliton solution as a solution given by the @xmath2-function @xmath107 . however , we show in this paper that @xmath108 parameter solution can contain up to @xmath8 number of d - solitons ( this is quite different from the solitons of the kp equation ) . in this paper , we study soliton solution of the dkp equation consisting of those a- and d - type soliton solutions . in section [ structure ] , we give a general structure of the @xmath2-functions . then in section [ onesoliton ] , we discuss some details of one d - soliton solutions given by the @xmath5-matrix of size @xmath67 . here we also classify the soliton solutions obtained by @xmath67 @xmath5-matrices . in section [ twosoliton ] , we present the soliton solutions for the case of @xmath109 @xmath5-matrices , and classify the soliton solutions consisting of only d - types . it turns out that the case with @xmath109 @xmath5-matrix can have either two or three d - solitons depending on the values of the parameters @xmath110 . the generic solution given by the @xmath5-matrix having all nonzero entries is then given by four a - solitons . finally we discuss the general case of @xmath4 @xmath5-matrices , i.e. @xmath106 , in section [ multisolitons ] . then we show that the number of d - solitons for those @xmath5-matrices having @xmath9 nonzero entries in the upper triangular part can be any number from @xmath7 to @xmath8 , contrary to the previous study ( see @xcite ) where the number of d - soliton is expected to be just @xmath7 . in @xcite , we showed that the @xmath7-soliton solutions of the kp equation can be classified by the schubert decomposition of the grassmannian gr@xmath111 . we also expect that @xmath7-soliton solutions of the dkp equation can be classified by the similar decomposition of the orthogonal grassmannain ogr@xmath112 for some @xmath6 . however , in this paper , we just present elementary feature of the @xmath7-soliton solutions based on the pfaffian structure of the @xmath2 function , which is already complicated but has several interesting aspects . we plan to discuss a classification problem based on the geometric structure of the orthogonal grassmannain in a future communication . we first note that the @xmath2-function ( [ tau ] ) with the @xmath19 matrix ( [ q ] ) has the following expansion theorem , i.e. the pfaffian version of the binet - cauchy theorem ( see @xcite for the details ) : [ dtaubc ] the @xmath2-functions of ( [ tau ] ) with ( [ q ] ) can be expressed by @xmath113 where @xmath114 is @xmath18 submatrix of the @xmath115 matrix @xmath52 , @xmath116 and @xmath117 is @xmath18 skewsymmetric submatrix of the @xmath4 matrix @xmath118 , @xmath119 the lower triangular part is given by @xmath120 , and is left in blank . also @xmath121 is given by the wronskian determinant , @xmath122 the sign is due to the order ( [ orderp ] ) , i.e. @xmath123 . from this formula of the @xmath2 function , the soliton solutions are completely determined by the nonzero coefficients of the pfaffians pf@xmath124 which are the plcker coordinates of the orthogonal grassmannan gr@xmath125 . for example , if all the coefficients are positive , then we have @xmath126-soliton solution in @xmath127 , that is , we have @xmath128 out - going line solitons in @xmath129 and @xmath130 in - coming line solitons in @xmath131 ( see @xcite ) . as we can see from @xcite that in particular , if @xmath132 we have an @xmath9-soliton solution similar to the kp equation , i.e. all @xmath9 solitons are a - solitons , and all the interactions are of t - types . then the main purpose of this paper is to identify soliton solution of the dkp equation with a certain combination of nonzero coefficients . as the simplest example , the pfaffian pf@xmath133 is given by @xmath134 note here that if the indices in each term have partial overlap , then the product takes the minus sign , e.g. @xmath135 and @xmath136 . this is a key for the classification of four a - solitons ( see section [ twosoliton ] ) . from lemma [ dtaubc ] , one should note that each term in @xmath3 in ( [ tau ] ) is given by the product of @xmath128 exponentials @xmath137 , i.e. @xmath138 where @xmath139 , and they are the indices for the elements of the @xmath5-matrix , i.e. @xmath140 . as we will show , this structure of the @xmath2-functions will be important to identify a soliton solution in the asymptotics @xmath141 . since the coefficients of the @xmath2-function are determined by the @xmath5-matrix , we give some remarks on the structure of the matrix . for a generic @xmath4 skew - symmetric matrix @xmath5 , i.e. @xmath142 , there is a decomposition called _ skew - borel decomposition _ @xcite , @xmath143 where @xmath144 , the group of invertible elements in the set of lower - triangular matrices with nonzero @xmath145 blocks proportional to identity along the diagonal , i.e. @xmath146 the matrix @xmath147 is the @xmath4 skew - symmetric matrix whose @xmath145 diagonal blocks are given by @xmath148 and all the other entries are zero , i.e. @xmath149 a non - generic element of @xmath150 may be obtained by a permutation matrix @xmath151 with @xmath152 ( i.e. @xmath153 ) , that is , we have @xmath154 where @xmath155 is not a generic element in @xmath156 . [ b4 ] consider the case @xmath157 ( @xmath67 @xmath5-matrix ) . then the generic element @xmath5 can be expressed by @xmath158 with @xmath159 then the matrix @xmath5 is given by @xmath160 ( we leave the lower triangular part in blank for skew symmetric matrix . ) if a skew - symmetric matrix has zero at @xmath161-entry ( i.e. @xmath73 ) and others are nonzero , such matrix , say @xmath162 , can not be expressed in this form . in this case , one can consider the form @xmath163 with @xmath164 the matrix @xmath162 is then given by @xmath165 in particular , with @xmath166 , the @xmath67 identity matrix , we have @xmath167 defined by @xmath168 other non - generic element of @xmath5-matrix , say @xmath169 , having @xmath170 can be expressed by @xmath171 , @xmath172 with @xmath173 in particular , for @xmath166 we have @xmath174 defined by @xmath175 in the next section we show that those skew - symmetric matrices @xmath176 and @xmath177 define all the one d - soliton solutions . in general , one can start with @xmath147 to define the @xmath2-functions , that is , @xmath178 with @xmath179 of ( [ q ] ) . then consider a generalization of the @xmath19-matrix by replacing @xmath147 with @xmath180 for some @xmath181 . the role of @xmath182 can be understood as the change of order of the column vectors in @xmath52 , i.e. @xmath183 where the column vector @xmath184 , and @xmath185 indicates the permutation @xmath186 . the signs in @xmath76 should be chosen properly so that the new @xmath2-function @xmath187 is sign - definite , i.e. non - singular ( see section [ twosoliton ] for more details ) . here we discuss the cases with @xmath67 skew - symmetric matrix @xmath5 , and show that the cases include one d - soliton and two a - solitons . in introduction , we show an example of one d - soliton ( [ onedsol ] ) whose @xmath2-function is given by @xmath188 with @xmath147 in ( [ j0 ] ) and @xmath189 in lemma [ dtaubc ] for @xmath157 . this d - soliton is denoted as @xmath102 $ ] , since it exchanges the asymptotic values of @xmath190 as @xmath97 ( recall @xmath98 ) . a general form of the one d - soliton can be expressed as @xmath191 $ ] where @xmath192 and @xmath193 . there are three cases , @xmath194,\quad [ 1,3:2,4],\quad [ 1,4:2,3]\,,\ ] ] which correspond to the d - solitons generated respectively by the skew - symmetric matrices , @xmath176 and @xmath177 in example [ b4 ] as the @xmath5-matrices . recall here that the soliton label @xmath195 $ ] indicates the nonzero elements @xmath72 and @xmath196 in the @xmath5-matrix . namely for the @xmath197-matrix with nonzero entries with 1 at @xmath198 and @xmath199 in the upper - triangular part , the @xmath2-function is given by @xmath200 then the function @xmath190 is @xmath201 with @xmath202 . for each asymptotic of @xmath95 , one of the exponential terms in the @xmath2-function ( [ dtau1 ] ) becomes dominant , and we have @xmath203 thus one d - soliton exchanges @xmath204 with the values of @xmath76 , and is denoted by @xmath191 $ ] , as an element of the weyl group of d - type . one soliton solution @xmath205 is a plane wave having the form , @xmath206 let us denote the wavenumber @xmath207 and the frequency @xmath208 for ( [ 1dsol ] ) by @xmath209 $ ] and @xmath210 $ ] , i.e. @xmath211=(p_{i_2}+p_{j_2}-p_{i_1}-p_{j_1},\ , p_{i_2}^2+p_{j_2}^2-p_{i_1}^2-p_{j_1}^2),\\ { } \\ \omega[i_1,j_1:i_2,j_2]=-(p_{i_2}^3+p_{j_2}^3-p_{i_1}^3-p_{j_1}^3 ) \end{array}\right.\ ] ] the slope ( or velocity ) of the soliton in the @xmath33-@xmath212 plane is given by @xmath213 . the peak of the soliton then determines the line in the @xmath33-@xmath212 plane given by the equation @xmath214 for each @xmath34 . and the velocities of solitons . the intersection point between @xmath215 and @xmath216 gives the velocity of d - soliton @xmath195 $ ] . the solid dots give the velocities of d - solitons , and the open circles give the velocities of a - solitons , e.g. the velocity of @xmath83$]-soliton is @xmath217 which is given by either @xmath218 or @xmath219.,width=264 ] we also note that the velocity @xmath220 of one d - soliton can be obtained as follows : first note that a d - soliton of @xmath195 $ ] gives a line @xmath221 . then setting @xmath222 for this equation , we have @xmath223 with @xmath224 then taking the limit @xmath225 for the equation @xmath221 , the velocity @xmath220 of the d - soliton is determined by @xmath226 , i.e. the intersection point of @xmath215 and @xmath216 . in figure [ uij : fig ] , we show @xmath227 of ( [ uijc ] ) for the parameter @xmath228 . each solid dot gives the velocity of one d - soliton , and each open circle gives the velocity of an a - soliton , e.g. the point at @xmath229 gives the velocity of @xmath230$]-soliton . this implies that if @xmath231 and @xmath232 are only nonzero elements in the @xmath5-matrix , then @xmath68 gives one a - soliton of @xmath230 $ ] , i.e. @xmath233 . figure [ uij : fig ] can be also used to determine the solitons appearing from the @xmath68-function in the asymptotics @xmath225 . for example , if @xmath234 for all @xmath235 , then @xmath68 contains all six terms @xmath236 with @xmath91 . then from figure [ uij : fig ] , we have two a - soliton solutions of t - type , @xmath237 $ ] and @xmath238 $ ] ( see below for the details ) . one d - soliton can be considered as a _ degenerate _ case of two a - solitons ( recall that a - soliton exchanges single letter as @xmath84 $ ] , while d - soliton exchanges two letters as @xmath195 $ ] ) . namely , d - soliton of @xmath195 $ ] can be considered as the `` product '' of two a - solitons of @xmath239 $ ] and @xmath240 $ ] or @xmath241 $ ] and @xmath242 $ ] , i.e. @xmath195=[i : k]\cdot[j : l]=[i : l]\cdot[j : k]$ ] as the product of two permutations . then two a - soliton solutions can be obtained by adding extra exponential terms , i.e. extra nonzero entries to the @xmath5-matrix . if one adds just one term , then we have y - shaped solution satisfying the resonant condition . for example , for the @xmath5-matrix having two nonzero entries @xmath243 and @xmath244 in the upper triangular part , if we add a nonzero entry at either @xmath245 or @xmath199 , we obtain a resonant y - shaped soliton with the resonant condition , @xmath246+{\bf k}[j_1:j_2]&=&{\bf k}[i_1,j_1:i_2,j_2],\\ \omega[i_1:i_2]+\omega[j_1:j_2]&= & \omega[i_1,j_1:i_2,j_2]\ , . \end{array}\right.\ ] ] here the wavenumber @xmath247 $ ] and the frequency @xmath248 $ ] are for the one soliton solution of the kp equation , i.e. @xmath249=(p_{i_1}-p_{i_2 } , p^2_{i_1}-p^2_{i_2}),\quad \omega[i_1:i_2]=p_{i_1}^3-p_{i_2}^3,\ ] ] $ ] . each region , @xmath250 , corresponds to a dominant exponential marked by the element @xmath72 . for example , with @xmath251 , @xmath102$]-d - soliton bifurcates into @xmath237 $ ] and @xmath238 $ ] a - solitons as @xmath252 ( top figure ) . those solitons can be found from figure [ uij : fig].,width=453 ] in figure [ a4:fig ] ( see also example [ 1234dsoliton ] below ) , we show the resonant bifurcation of @xmath102$]-soliton into two a - solitons of @xmath237 $ ] and @xmath238 $ ] . those solitons can be found from figure [ uij : fig ] . for example , in the left figure of figure [ a4:fig ] , we have @xmath253 and @xmath254 from the nonzero entries of the @xmath5-matrix . then the three intersection points of the graphs of those @xmath215 in figure [ uij : fig ] give one d - soliton ( solid dot ) and two a - solitons ( open circles ) . figure [ uij : fig ] also shows that the solid dot indicates the pair of dominant exponentials @xmath255 for @xmath102$]-soliton as @xmath129 , while the open circles show the dominant pairs @xmath256 for @xmath237$]-soliton and @xmath257 for @xmath238$]-soliton as @xmath131 . since there are at least four disconnected regions in the @xmath33-@xmath212 plane divided by two line solitons , one needs to add at least two extra nonzero ( positive ) entries . figure [ a4:fig ] shows the resonant _ bifurcations _ of a d - soliton into two a - solitons . two soliton solution in this figure is a new t - type ( referred to as td - type ) which does not exist in the kp equation ( i.e. @xmath258 ) . the t - type of the kp equation can be obtained by a @xmath5-matrix with det@xmath259 , e.g. @xmath260 note @xmath261 with this matrix due to @xmath262 . one should also note that the generic skew - symmetric @xmath5-matrix having all nonzero entries gives a two a - soliton solution of t - type ( having a resonant hole ) , but it is not a solution of the kp equation . in this case , since those solitons are a - solitons , the function @xmath263 appears only locally in the @xmath33-@xmath212 plane . both @xmath2-functions for t - type two - soliton solutions of kp and dkp equations have six independent exponentials , i.e. @xmath264 for @xmath235 , which is the key to produce t - type resonant interaction of two a - solitons ( see also @xcite ) . [ 1234dsoliton ] consider @xmath102$]-d - soliton which corresponds to the @xmath5-matrix having nonzero entries at @xmath161 and @xmath265 , i.e. @xmath266 in ( [ j0 ] ) : * @xmath237$]- and @xmath238$]-a - solitons ( td - type ) by putting @xmath267 at the entries @xmath268 and @xmath269 . * @xmath270$]- and @xmath230$]-a - solitons ( p - type ) by putting @xmath267 at the entries @xmath135 and @xmath136 . note that the td - type in the case a ) is _ not _ the same as the one of the kp - solitons , i.e. there is no resonance in this case . original t - type of two a - solitons is obtained by the @xmath5-matrix having all nonzero entries and det@xmath259 . again there are three types of two a - solitons for dkp . these types play an important role for the classification problem as a building block . figure [ a5:fig ] shows the diagrams of the three fundamental types of two a - solitons . the middle diagrams of 8-gon connect the types with the corresponding @xmath5-matrices as follows : we have nonzero @xmath72 if the pair @xmath250 is connected in the diagram , and a corresponding a - soliton of @xmath271 $ ] is given by the pair @xmath272 having no connection in the diagram . also note that the corresponding @xmath5-matrix for o - type is not in the generic one given in ( [ genericb ] ) , but in ( [ b1 ] ) ( compare this with the o - type of kp solitons which belong to a schubert cell of codimension one @xcite ) . figure [ tpo : fig ] shows those types of two a - solitons , which are the same as 2-soliton solutions of the kp equation ( note here that we have set @xmath261 for all the types , i.e. @xmath68 gives a solution of the kp equation ) . the parameters are chosen as @xmath273 , and the velocities of solitons can be found from figure [ uij : fig ] . -matrix and the corresponding two - soliton solution , i.e. @xmath274 if and only if the pair @xmath250 is directly connected in the middle diagram , and @xmath271$]-soliton implies that the pair @xmath272 has no direct connection . for example , in o - type , two a - solitons are labeled @xmath83 $ ] and @xmath275 $ ] , then we have the middle diagram to show the connections between the numbers in the pairs , i.e. ( 1,2 ) and ( 3,4 ) are not connected . this diagram implies we have nonzero elements @xmath276 and @xmath277.,width=396 ] ( if @xmath278 , then the dkp solution is also the kp solution ) . however note that the td - type in figure [ a5:fig ] does not have a resonant quadrangle . the figures show the counter plots of the function @xmath279 in the @xmath33-@xmath212 plane for a fixed @xmath34 , and the contrast of the shades shows the different values of @xmath76 which increases in the positive @xmath33-direction.,width=434 ] since each d - soliton has four phases , we need eight phases to describe a two d - soliton solution . the @xmath66-function in the case with @xmath109 @xmath5-matrix is given by @xmath280 where @xmath281 , and the pfaffian is calculated as @xmath282 let us now construct the @xmath5-matrices for two d - soliton solutions which consist of only four nonzero entries in the upper triangular parts . we follow the steps : * take four pairs @xmath283 with @xmath193 from the set @xmath284 , such that @xmath285 there are @xmath286 ways to make those pairs ( compare this with @xmath287 in the case of 4 a - solitons @xcite ) . those four pairs give four nonzero elements in the @xmath5-matrix , i.e. @xmath288 for @xmath289 . * determine the signs of each pairs @xmath290 , so that all the pfaffians in the expansion of @xmath66 associated with the @xmath5-matrix have the same sign . since the pairs @xmath290 are all different , each pfaffian has just one term , denoted by @xmath291 . here the sign depends on the pairs , for examples , @xmath292 in order to determine the sign of the pfaffian @xmath291 , we define : we say that the pairs @xmath293 and @xmath294 with @xmath295 have a _ partial _ overlap , if @xmath296 . otherwise , we say that the pairs have a _ non_-partial overlap ( i.e. total or no overlap ) . let us introduce the sign of overlap between @xmath290 and @xmath297 , @xmath298 then we have @xmath299 . now we have : suppose that the signs @xmath300 satisfiy @xmath301 then one can make the @xmath2-function @xmath66 to be sign - definite , that is , all @xmath302 take the same sign . we denote @xmath303 , so that sgn(pf@xmath304 . the condition that all the terms have the same sign requires @xmath305 for any @xmath306 and @xmath307 . this leads to @xmath308 which gives the assertion . applying the lemma , one can prove : [ 2dsolitons ] let @xmath309 and @xmath310 . then there exist 33 cases of sign - definite @xmath2-functions whose @xmath5-matrices have nonzero entries only at @xmath290 and @xmath311 for @xmath289 . those cases are given by : * for @xmath312 , there are 8 cases with @xmath310 ; @xmath313 * for @xmath314 , there are 4 cases , @xmath315 * for @xmath316 , there are 4 cases , @xmath317 * for @xmath318 , there are 4 cases , @xmath319 * for @xmath320 , there is one case , @xmath321 * for @xmath322 , there are two cases , @xmath323 * for @xmath324 , there are two cases , @xmath325 * for @xmath326 , there is one case , @xmath327 * for @xmath328 , there is one case , @xmath329 * for @xmath330 , there are two cases , @xmath331 * for @xmath332 , there is one case , @xmath333 * for @xmath334 , there is one case , @xmath335 * for @xmath336 , there is one case , @xmath337 * for @xmath338 , there is one case , @xmath339 the case ( n ) corresponds to the @xmath5-matrix @xmath266 defined in ( [ j0 ] ) , and the @xmath5-matrices for other cases can be expressed as @xmath340 for some @xmath341 with appropriate signs in the entries of @xmath182 . those signs are determined so that the @xmath2-function is non - singular . we now classify d - solitons given by the @xmath2-functions associated with the @xmath5-matrices obtained in proposition [ 2dsolitons ] . we obtain : [ 2 - 3dsolitons ] suppose we have the order , @xmath342 with @xmath343 . then as a generic situation , the d - soliton solutions of the dkp equation turns out to be the following two cases ; * two d - solitons of @xmath344 , [ i_2,j_2:i_4,j_4]$ ] or @xmath345 , [ i_2,j_2:i_3,j_3]$ ] , * three d - solitons of @xmath346 , [ i_l , j_l : i_{\alpha},j_{\alpha}]$ ] and @xmath347 $ ] for @xmath348 or @xmath349 ( @xmath350 are all distinct ) . first we note that each @xmath66-function associated with the @xmath5-matrix given in proposition [ 2dsolitons ] has six exponential terms . we denote them as @xmath351 for @xmath352 , where @xmath353 for a fixed @xmath34 . to identify two dominant exponents which determine an asymptotic d - soliton , we set @xmath222 and define @xmath354 ( see section [ onesoliton ] ) . then for large @xmath355 , the velocity @xmath220 of asymptotic soliton can be given by the intersection point of @xmath356 , where as a generic situation only two exponential terms become dominant , say @xmath357 and @xmath358 ( see also @xcite for the details , and figure [ 4768-sol : fig ] below ) . the soliton is then identified as @xmath359 $ ] . then there are two cases where the number of solitons , i.e. the number of balancing pairs of dominant exponentials , is either two or three . to show this , we consider , for example , the case with @xmath360 . we set the parameter @xmath361 so that @xmath362 whith @xmath363 . keeping this order , one can choose @xmath361 , so that we have @xmath364 with @xmath365 . ( this can be done with a large negative value for @xmath366 and a small positive @xmath367 . ) then it is easy to see from the graph of @xmath368 that one can have either two d - soliton of @xmath344 $ ] and @xmath369 $ ] or three d - soliton of @xmath344 , [ i_2,j_2:i_3,j_3]$ ] and @xmath370 $ ] . the parameters @xmath371 can take the same values except @xmath367 for both cases ( figure [ 4768-sol : fig ] gives an example for this , also see below ) . one can also make a different order for @xmath372 , e.g. @xmath373 ( by taking large positive @xmath367 ) . then one can have three d - soliton of @xmath191,[i_2,j_2:i_3,j_3]$ ] and @xmath369 $ ] with the velocities @xmath374 where @xmath375 is the velocity of @xmath359$]-soliton ( recall that those are given by the intersection @xmath376 with @xmath377 ) . thus there is a freedom in the ordering of @xmath372 , and this is a key to generate three d - solitons ( compare with the case for the kp equation @xcite where we have only two a - solitons for six exponential terms in the @xmath2-function ) . in figure [ 4768-sol : fig ] , we illustrate the case of @xmath314 and @xmath378 : here the ordering is @xmath379 . then depending on the parameters @xmath380 s , the number of asymptotic d - solitons is either two or three . the left ( right ) figure shows the case of two ( three ) d - solitons , i.e. there are two ( three ) values of @xmath220 where two exponential terms become dominant . it is obvious that there is no case other than those two in the general case . and @xmath378 . the left figure shows two d - solitons with @xmath381 . the right one shows three d - solitons with the same parameters except @xmath382 . the bottom figures show the graphs of @xmath383 . the dots in the graphs show the dominant two exponential terms , which determine the asymptotic d - solitons . the squares correspond to the intermediate d - solitons forming the resonant quadrangle , and the triangles correspond to the resonant intermediate solitons from the pairs of three asymptotic solitons as shown in the top figures.,width=377 ] we note here that all cases in proposition [ 2dsolitons ] have two - d - soliton solutions , but only some cases can have three - d - soliton solution . in particular , we have : [ case - n ] the case ( n ) in proposition [ 2dsolitons ] can have only two - d - soliton solutions of @xmath384 $ ] and @xmath385 $ ] . in order to prove the proposition , we prepare the following lemma : let us first denote @xmath386 as @xmath387 as before : @xmath388 where @xmath389 for the @xmath5-matrix of the case ( n ) , i.e. @xmath266 . then we have : [ cij ] let @xmath390 be the point of intersection given by @xmath391 . then for a fixed index @xmath392 , we have @xmath393 from @xmath394 , we have @xmath395 let us write @xmath396 and @xmath397 as @xmath398 that is , @xmath399 , and we have @xmath400 suppose @xmath401 ( other cases follow the similar argument ) . then we have @xmath402 it is then easy to show @xmath403 . for the case ( n ) in proposition [ 2dsolitons ] which states , e.g. @xmath404 . the open circles correspond to two d - solitons of @xmath384 $ ] with the velocity @xmath405 and @xmath385 $ ] with @xmath406.,width=226 ] now we give a proof of proposition [ case - n ] : from lemma [ cij ] , we have the order @xmath407 . also note that the slope of @xmath408 is given by @xmath409 , hence @xmath362 . then we obtain qualitative graphs of @xmath387 as shown in figure [ c : fig ] , and from the graphs , we can find the dominant pairs of the exponential terms in @xmath66 for @xmath141 . this shows that the dominant pairs are given by @xmath410 and @xmath411 for @xmath129 , and @xmath412 and @xmath413 for @xmath131 . this implies that we have two d - solitons of @xmath384 $ ] and @xmath385 $ ] ( recall @xmath414 ) . this completes the proof . we also have : [ case - a ] the case with @xmath312 and @xmath415 ( one of the cases in ( a ) of proposition [ 2dsolitons ] ) can have both two and three d - solitons . follow the argument in the proof of theorem [ 2 - 3dsolitons ] . also see the following examples : * for two d - solitons , we take @xmath416 . then the open circles in the left figure in figure [ case - a : fig ] show the dominant exponentials which correspond to the d - solitons of @xmath345=[1,8:4,5]$ ] and @xmath417=[2,7:3,6]$ ] . * for three d - solitons , we take @xmath418 . the three open circles in the right figure in figure [ case - a : fig ] correspond to the two d - solitons of @xmath191=[1,8:2,7]$ ] , @xmath417=[2,7:3,6]$ ] and @xmath369=[2,7:4,5]$ ] . in figure [ 3dsol : fig ] , we show the interaction patterns for those cases shown in figure [ case - a : fig ] ( they are topologically the same as in figure [ 4768-sol : fig ] ) . in section [ multisolitons ] , we will extend propositions [ case - n ] and [ case - a ] to the general cases as propositions [ multi - n ] and [ multi - a ] . recall that a d - soliton can be identified as an element of the weyl group of d - type , @xmath419 . in the case of two d - solitons , we have , for example , @xmath420 and @xmath421 . the group generated by those elements is an abelian subgroup of @xmath419 , and the orbit of this subgroup of @xmath422 represents the asymptotic regions divided by the solitons , i.e. @xmath423 and @xmath424 , which correspond to the dominant exponentials in the @xmath2-function , that is , @xmath425 and @xmath426 , respectively . in the @xmath2-function , we have two more exponential terms labeled by @xmath427 and @xmath428 , i.e. @xmath429 and @xmath430 . those terms become dominant at the points of the resonant quadrangle as in the case of two a - solitons of t - type . namely we have a resonant two d - soliton . the resonant condition is given by , for example , @xmath431+{\bf k}[i_3,j_3:i_4,j_4]={\bf k}[i_1,j_1:i_4,j_4]\\ { } \\ \omega[i_1,j_1:i_3,j_3]+\omega[i_3,j_3:i_4,j_4]=\omega[i_1,j_1:i_4,j_4 ] \end{array}\right.\ ] ] in the case of three d - soliton , for example @xmath191,[i_2,j_2:i_3,j_3]$ ] and @xmath369 $ ] , we have @xmath432 and @xmath433 as the generating elements of the subgroup . then the subgroup contains @xmath434 and @xmath435 , and those are the d - solitons generated by the resonances of pairs , i.e. @xmath436 represents d - soliton of @xmath359 $ ] . each resonant relation can be represented by the product @xmath437 . for @xmath289 . the indices are marked by @xmath438 for @xmath289 . the left figure shows two d - soliton solutions of @xmath439 $ ] and @xmath440 $ ] with @xmath441 , and the right one shows three d - solitons of @xmath442,[2,7:3,6]$ ] and @xmath443 $ ] with @xmath418 . note that only @xmath444 is changed.,width=434 ] for any two d - soliton solution in proposition [ 2dsolitons ] , we have : two d - solitons constructed above are all in resonance , in the sense that there is a resonant quadrangle at the intersection point . namely all the cases are of t - type with one hole . first note that there are six exponential terms in the @xmath2-function . each exponential term becomes dominant in some region in the @xmath33-@xmath212 plane . for example , consider two d - solitons with @xmath384 $ ] and @xmath385 $ ] . then the function @xmath445 takes @xmath446 for @xmath447 , and @xmath448 for @xmath449 . with those d - solitons , we have the regions marked by @xmath450 and @xmath451 . in addition to those exponentials , we also have @xmath452 and @xmath453 . as in the case of two solitons of the kp equation ( see @xcite ) , a resonant interaction of y - shape can occur . for example , @xmath384 , [ 1,2:7,8]$ ] and @xmath454 $ ] solitons are in resonant , and the three regions divided by those solitons are marked by @xmath455 and @xmath453 . then the six exponentials in the @xmath2-function form two d - solitons in resonance ( see the left figure in figure [ 3dsol : fig ] ) . . the function @xmath456 takes the values , @xmath457 as @xmath447 , and @xmath458 as @xmath449 . all six d - solitons in each figure can be identified as the six intersection points of the graphs @xmath459 in figure [ case - a : fig].,width=434 ] in figure 8 of @xcite , isojima et al presented an example of ( two ) d - solitons which seems to have only one intermediate soliton ( no resonant quadrangle ) . however , it turns out that this is _ not _ two d - solitons : first we note that this is the case ( k ) in proposition [ 2dsolitons ] , i.e. @xmath332 and @xmath460 . this can be found by ordering the parameters @xmath461 in figure 8 of @xcite , i.e. relabel @xmath462 in terms of our ordering @xmath463 with @xmath464 and @xmath465 . the parameters chosen in the paper is then given by @xmath466 . with those parameters , we note that @xmath467 , a _ degeneracy _ in the parameters . two d - solitons are given by @xmath468 $ ] and @xmath469 $ ] , and the asymptotic values of @xmath76 are @xmath470 and @xmath471 for @xmath472 . then the resonant quadrangle consists of the d - solitons of @xmath473 , [ 1,4:7,8 ] , [ 2,3:5,6]$ ] and @xmath454 $ ] . however , because of the degeneracy @xmath474 , the @xmath473$]-soliton can not exist , i.e. the function @xmath76 can not change the value across this soliton . also notice that the @xmath475$]- and @xmath468$]-solitons are almost parallel , and so are the @xmath476$]- and @xmath469$]-solitons : for example , the peak of the soliton @xmath475 $ ] is given by @xmath477 , i.e. for a fixed @xmath34 and with the values of @xmath361 , we have the line for @xmath475$]-soliton , @xmath478 which is almost parallel to the line @xmath479 for the @xmath468$]-soliton , @xmath480 thus the soliton solution in figure 8 of @xcite consists of two d - solitons of @xmath468 $ ] and @xmath469 $ ] for @xmath129 and two _ other _ d - solitons of @xmath475 $ ] and @xmath476 $ ] for @xmath252 , and the intermediate soliton is the @xmath454$]-soliton . other intermediate soliton of @xmath473 $ ] may be considered to be located at @xmath481 . for any three d - soliton solutions in proposition [ 2dsolitons ] , we have : any pair of three d - solitons constructed above is in resonance to form an y - shape vertex . this three d - soliton solution has no resonant hole and four resonant y - shape vertices . since there are six exponential terms which mark the six separated regions in the @xmath33-@xmath212 plane bounded by those three solitons , this 3-soliton solution can not have a resonant hole . three solitons are labeled by , for example , @xmath191 , [ i_2,j_2:i_3,j_3]$ ] and @xmath369 $ ] , and any pair of those solitons is in resonance , i.e. @xmath191 $ ] and @xmath417 $ ] have a resonant interaction to generate @xmath482$]-soliton ( see the right figure in fig.[3dsol : fig ] ) . it is easy to see that this resonance appear for any three solitons obtained in proposition [ 2dsolitons ] . since one d - soliton is a degenerate two a - solitons , one can construct 3-soliton solution consisting of two a - solitons and one d - soliton from two d - solitons . one should note that resolution of one d - soliton into two a - solitons can work only for the case of two d - solitons . in the case of three d - solitons , the resolution of one d - soliton affects to other d - solitons , and it can not produce two a - solitons . this can be checked by the asymptotic analysis using the graphs of @xmath215 . consider two d - solitons corresponding to the @xmath5-matrix of the case ( n ) in proposition [ 2dsolitons ] , i.e. @xmath338 and @xmath483 . two d - solitons are @xmath384 $ ] and @xmath385 $ ] . then , for example , two a - soliton of type @xmath484 $ ] and @xmath485 $ ] are obtained by adding two extra nonzero elements at @xmath486 and @xmath487 . this fact can be verified easily from the diagram of t - type ( up - side - down staircase with sides labeled by 1,2,5 and 6 ; see figure [ a5:fig ] ) . the 3-solitons obtained here has only one hole ( see figure [ a7:fig ] ) . one can also get two a - solitons of @xmath488 $ ] and @xmath489 $ ] from @xmath384$]-soliton by adding two nonzero elements at @xmath490 and @xmath491 . notice that those two a - solitons form the diagram of p - type ( cannon shape with sides by 1,2,5 and 6 ; see figure [ a5:fig ] ) . in the same way , one can resolve @xmath385$]-solitons into either the pair of @xmath492 $ ] and @xmath493 $ ] or the pair of @xmath494 $ ] and @xmath495 $ ] , by adding extra nonzero entries to the @xmath5-matrix of the case ( n ) . $ ] -d - soliton resolves into two a - solitons of @xmath484 $ ] and @xmath485 $ ] . d - solitons are expressed as the double lines joined with the double circles.,width=340 ] those solitons can be constructed in the similar way as the case of three solitons ( two a - solitons and one d - soliton ) from two d - solitons . however , the compatibility among those a - solitons separated from d - solitons is somewhat complicated . for example , one can expect that two d - solitons , @xmath384 $ ] and @xmath385 $ ] , generate two a - solitons @xmath484 , [ 2:6]$ ] from @xmath384 $ ] and other two a - solitons @xmath492 $ ] and @xmath493 $ ] from @xmath385 $ ] . the @xmath5-matrices for 3-solitons @xmath484 , [ 2:6 ] , [ 3,4:7,8]$ ] and @xmath384 , [ 3:7 ] , [ 4:8]$ ] may be respectively given by @xmath496 then adding @xmath497 to the @xmath5-matrix in the left ( or @xmath498 to the right one ) , the resulting @xmath5-matrix is expected to give four a - solitons @xmath484,[2:6],[3:7]$ ] and @xmath493 $ ] . however this matrix gives a singular solution , and a correct @xmath5-matrix for this 4-soliton solution may be given by @xmath499 thus in this case we do not have a compatibility between the two sets of 3 solitons obtained from the d - solitons ( in the sense that the @xmath2-function can not stay to be sign - definite ) . a general rule to construct a @xmath5-matrix for four a - solitons from two d - solitons seems to be complicated . here we give a direct way to construct a @xmath5-matrix for a given four a - solitons , based on the @xmath5-matrices obtained in proposition [ 2dsolitons ] : * give four a - solitons labeled by a pair of four numbers , i.e. @xmath500 where @xmath501 and @xmath502 . then each soliton is labeled by @xmath503 $ ] for @xmath289 ( the same as in the kp solitons @xcite ) . this implies that the asymptotic values of @xmath504 are given by @xmath505 for @xmath95 . * take a pair of two solitons , say @xmath506 $ ] and @xmath507 $ ] , and draw the corresponding diagrams shown in figure [ a5:fig ] , and do the same for the other pair . here the choice of the pair should be consistent with the asymptotic values . then adjust the signs for each horizontal edges of the diagrams , so that all the pfaffians have the same sign . if all are satisfied , then the diagrams give the @xmath5-matrix . * example : * consider four solitons with @xmath83 , [ 7:8]$ ] and any other two solitons with the indices from @xmath508 . then one can construct the @xmath5-matrix as follows : the diagram corresponding to @xmath83,[7:8]$ ] is a backward c - shape ( p - type shown in figure [ a5:fig ] ) , having 1,2,7 and 8 on the vertical edge . this implies we have nonzero elements for @xmath509 . now if the other pair is @xmath510,[4:5]$ ] , then the diagram associated to this is a cannon - shape ( o - type ) having 3,4,5,6 on the vertical edges . this gives the nonzero entries @xmath511 . in this example , all those entries in the upper triangular part can take @xmath267 for a sign - definite @xmath2-function . namely we have @xmath512 in the case of two solitons of t - type , @xmath513,[4:6]$ ] , one can use the @xmath5-matrix having the same entries except @xmath514 , and in addition @xmath515 . notice that the nonzero entries has two groups corresponding to two d - solitons @xmath516 $ ] and @xmath517 $ ] , and those groups have no partial overlaps . this structure is true for any cases satisfying the nonsigular condition , i.e. all the pfaffians have the same sign . ( recall that the @xmath5-matrix with @xmath518 , i.e. @xmath519 , also gives three d - solitons of @xmath442,[2,7:3,4]$ ] and @xmath520 $ ] . ) for the same 4-solitons as the previous example , i.e. @xmath83,[3:6],[4:5]$ ] and @xmath521 $ ] , one can also have the following @xmath5-matrix which gives the same @xmath66 function , @xmath522 this matrix can be constructed from the @xmath5-matrix for @xmath316 and @xmath523 , the case ( c ) in proposition [ 2dsolitons ] , ( i.e. this gives two d - solitons with @xmath524 $ ] and @xmath525 $ ] ) . in figure [ a8:fig ] , we show the example of four a - solitons based on two @xmath5-matrices : the case a ) corresponds to the @xmath5-matrix for @xmath526 , and the case b ) for @xmath519 . we take @xmath527 for both cases . -matrix of two d - solitons or three d - solitons . two d - solitons in a ) is given by the @xmath5-martrix with @xmath316 and @xmath523 . three d - solitons in b ) is given by the @xmath5-matrix with @xmath314 and @xmath528 . both cases have the same @xmath527 . the four a - soliton solution in c ) is given by the @xmath5-matrix ( [ 1235b ] ) . note that the @xmath5-matrix of ( [ 1236b ] ) gives the same solution . , width=491 ] as an example of four a - solitons , we consider the case with @xmath529,[2:8],[3:5]$ ] and @xmath530 $ ] , i.e. there are two partial overlaps ( two solitons of t - type ) . we first note that this set of 4-solitons can be split into two groups of 2-solitons having no partial overlaps , @xmath531,[2:8]\}$ ] and @xmath532,[4:6]\}$ ] . since they are both td - type , we have the td - type diagram ( up - side - down staircase ) . this implies we have nonzero entries @xmath533 from the set @xmath531,[2:8]\}$ ] , and @xmath534 from the second set . however this @xmath5-matrix does not give the 4-solitons . in figure [ 2 - 4sol : fig ] , the left figure shows the solution given by this @xmath5-matrix , two d - solitons , @xmath384 $ ] and @xmath385 $ ] with some additional resonances at the intersection region of those solitons . this can be explained as follows : this @xmath5-matrix can be considered as a deformation ( bifurcation ) from the @xmath5-matrix of two d - solitons , @xmath384 $ ] and @xmath385 $ ] , and the @xmath2-function with this @xmath5-matrix gives resonant d - solitons , @xmath102,[1,2:7,8],[3,4:5,6]$ ] and @xmath454 $ ] . then td - type of 2-solitons @xmath529,[2:8]$ ] is obtained from the bifurcation of @xmath535$]-soliton , which is one of the resonant solitons . also note that the exponentials associated to @xmath450 and @xmath451 become dominant for two d - solitons for either @xmath129 or @xmath131 . however those exponentials should not appear for this set of four a - solitons , that is , those four a - solitons can not be obtained from this @xmath5-matrix . this can be fixed by taking following @xmath5-matrix ( i.e. change td - type to t - type ) , @xmath536 in figure [ 2 - 4sol : fig ] , the right figure shows the 4-solitons generated by this @xmath5-matrix . we also mention that the @xmath66-function with the generic @xmath109 @xmath5-matrix having all nonzero entries gives four a - solitons of t - type , that is , they are the same type of the 4-soliton solution of the kp equation . just notice that the number of exponential terms in @xmath66 is @xmath537 , which is the same as the @xmath2-function for the kp equation ( see @xcite ) . then the asymptotic analysis of the @xmath2-functions shows that four a - solitons are given by @xmath484 , [ 2:6 ] , [ 3:7]$ ] and @xmath493 $ ] , which are of t - type . an example of the generic @xmath5-matrix for nonsingular @xmath2-function is given by the skew - symmetric matrix having 1 s in all the entries in the upper triangular part . -matrices given by ( [ 4bb ] ) . the left figure shows two d - solitons obtained by the same matrix except @xmath538 and @xmath539 ( note the extra resonant structure at the intesection region of those d - solitons.,width=340 ] for the general case with arbitrary @xmath106 , we consider the @xmath5-matrix with @xmath9 numbers of nozero entries @xmath140 for @xmath540 and @xmath541 in the upper triangular part . although there is a scheme to find the @xmath5-matrix for sign - definite @xmath107 ( i.e. having all the same sign in the pfaffian coefficients ) , the generalization of proposition [ 2dsolitons ] seems to be difficult . however , we can show the following propositions regarding to the number of d - solitons for two special cases : apply lemma [ cij ] , and find the dominant exponentials by following the argument of `` level of intersection '' introduced in @xcite ( the main idea here is to find the dominant exponents from the graphs of @xmath227 ) . the other case corresponds to the @xmath5-matrix , @xmath542 , where @xmath543 is defined by @xmath544 with @xmath151 , the permutation given by @xmath545\cdot[4:4n-2]\cdots[2n:2n+2]$ ] , i.e. @xmath546 then we have : [ multi - a ] the number of d - solitons generated by the @xmath107-function ( [ tau ] ) with @xmath4 @xmath5-matrix given by @xmath547 of ( [ jw ] ) can be any number from @xmath7 to @xmath8 , depending on the choice of the parameter @xmath361 . here instead of giving a precise proof , we demonstrate a case with @xmath548 : in figure [ m16:fig ] , we show the graphs of @xmath549 for @xmath550 . with a proper choice of the parameter @xmath551 , one can find four different cases of intersection patterns of those lines . then d - solitons can be found at the intersection points marked by open circles in figure [ m16:fig ] , that is , the level of intersection of those points should be four ( i.e. there are four lines above those points ) . for example , d - soliton @xmath552 $ ] is obtained by the intersection of @xmath553 and @xmath554 , and two dominant exponents for this soliton are @xmath555 and @xmath556 which are obtained from four lines above the intersection point @xmath557 ( this is the definition of the level of intersection ; see @xcite for more details ) . we have the following four cases : * four d - solitons ( the top left figure ) of @xmath558 , [ 2,15:7,10 ] , [ 3,14:6,11]$ ] and @xmath552 $ ] with the parameter , @xmath559 * five d - solitons ( the top right figure ) of @xmath558 , [ 2,15:4,13 ] , [ 3,14:6,11 ] , [ 4,13:5,12]$ ] and @xmath560 $ ] with @xmath561 note here that three d - solitons have the common index @xmath562 . the @xmath563$]-soliton in the case ( a ) causes the resonant interaction with two d - solitons @xmath564 $ ] and @xmath560 $ ] . * six d - solitons ( the bottom left figure ) of @xmath565 , [ 2,15:4,13 ] , [ 3,14:6,11 ] , [ 4,13:5,12 ] , [ 4,13:7,10]$ ] and @xmath566 $ ] with @xmath567 * seven d - solitons ( the bottom right figure ) of @xmath568 $ ] for all @xmath550 , except @xmath569 , and @xmath570 one should note that those @xmath361 values do _ not _ give generic cases , since some of the lines @xmath408 are parallel as shown in figure [ m16:fig ] . however , one can make those to be generic by changing slightly the parameters @xmath105 s without breaking the intersection patterns . ( we show those non - generic parameters , since those provide a clear evidence of the existence of those solitons . ) for @xmath571 . each open circle of the intersections of two functions @xmath387 corresponds to a d - soliton given by @xmath572 . for eaxmple , the point for @xmath573 corresponds to the d - soliton of @xmath558 $ ] . the number of open circles depends on the choice of the parameter @xmath551 . notice that only @xmath553 and later @xmath574 are changed to increase the number of solitons ( see the text for the explicit values of @xmath361 ) . , width=453 ] we also remark that all the @xmath7-soliton solutions obtaind by @xmath107 with the @xmath5-matrix given in proposition [ multi - n ] are fully resonant cases like the t - type for the kp equation , that is , the @xmath107 contains @xmath575 exponential terms having the form @xmath576 ( see @xcite ) . finally , we note that in the case of generic @xmath5-matrix having all nonzero entries , the @xmath107 function has the maximal number of exponential terms @xmath577 which has the same structure of the @xmath2-function for fully resonant @xmath9-soliton solutions ( i.e. t - type ) of the kp equation . those @xmath9 solitons are all a - type , and the interaction patterns are topologically the same for both kp and dkp equations . in those cases , the functions @xmath578 appear only locally at the interaction regions among @xmath9-solitons .

we identify protostars in _ spitzer _ surveys of nine star - forming molecular clouds within 1 kpc : serpens , perseus , ophiuchus , chamaeleon , lupus , taurus , orion , cep ob3 , and mon r2 , which combined host over 700 protostar candidates . these clouds encompass a variety of star forming environments , including both low mass and high mass star forming regions , as well as dense clusters and regions of sparsely distributed star formation . our diverse cloud sample allows us to compare protostar luminosity functions in these varied environments . we combine near- and mid - infrared photometry from 2mass j , h , and bands and _ spitzer _ 3.6 , 4.5 , 5.8 , 8.0 , and 24 m bands to create 1 - 24 m spectral energy distributions ( seds ) . using protostars from the _ c2d _ survey with well - determined bolometric luminosities , we derive a relationship between bolometric luminosity , mid - ir luminosity ( integrated from 1 - 24 m ) , and sed slope . estimations of the bolometric luminosities for protostar candidates are combined to create luminosity functions for each cloud . contamination due to edge - on disks , reddened class ii sources , and galaxies is estimated and removed from the luminosity functions . we find that luminosity functions for high mass star forming clouds ( orion , mon r2 , and cep ob3 ) peak near 1 l and show a tail extending toward luminosities above 100 l . the luminosity functions of the low mass star forming clouds ( serpens , perseus , ophiuchus , taurus , lupus , and chamaeleon ) do not exhibit a common peak , however the combined luminosity function of these regions peaks below 1 l . finally , we examine the luminosity functions as a function of the local surface density of ysos . in the orion molecular cloud , we find a significant difference between the luminosity functions of protostars in regions of high and low stellar density , the former of which is biased toward more luminous sources . this may be the result of primordial mass segregation , although this interpretation is not unique . we compare our luminosity functions to those predicted by models and find that our observed luminosity functions are best matched by models which invoke competitive accretion , although we do not find strong agreement of the high mass star forming clouds with any of the models . = 1

1204.1535

stars form in a diverse range of environments within molecular clouds . these cloud environments include crowded , massive clusters heated by o stars ( e.g. in orion ) , smaller clusters without high mass stars ( e.g. in ophiuchus ) , or isolated , cold dark clouds containing low mass stars ( e.g. in taurus ) . the temperatures , densities , and turbulent linewidths of the natal molecular gas can vary systematically between these regions @xcite , and the surface density of young stellar objects ( ysos ) can vary by over two orders of magnitude @xcite . this motivates us to study how these different environments affect the outcomes of the star formation process . despite significant differences in gas temperature , column density , and the turbulent velocities , the star forming regions and clusters in our galaxy exhibit remarkably similar initial mass functions . there may be a few exceptions : @xcite have compared the imfs from taurus with chamaeleon i and ic-348 and find that taurus has a significant excess of stars between 0.6 and 0.8 m@xmath2 . nevertheless , the similarity between the imfs of dark clouds like chamaeleon i , clusters with b - stars such as ic 348 , and massive clusters with o and b stars such as the orion nebula cluster suggests that the imf is remarkably invariant . although the imf averaged over a star forming region or cluster may be universal , there is some evidence that masses are initially segregated within clouds or within clusters . massive stars are found primarily in the center of clusters , although it not clear whether this is the result of primordial mass segregation or dynamical evolution @xcite . the detection of compact groups of massive protostars in the centers of clusters is evidence for primordial mass segregation since such objects would have little time for dynamical evolution @xcite . mass segregation is also observed in small groups of young stars : @xcite studied groups of 20 and 40 stars in the taurus and perseus molecular clouds and found that the most massive members ( in some cases with a mass of only 1 m@xmath2 ) are found preferentially near the centers of the groups . in such small groups , it is unlikely that dynamical evolution has occurred since the relaxation time of the group exceeds 5 myr , longer than the estimated ages of the groups @xcite . these observations are evidence for primordial mass segregation , with the imf in the centers of clusters biased towards higher mass stars . an understanding of the origin of the imf , and the potential primordial mass segregation within star forming regions , requires a better knowledge of both the initial conditions leading to the formation of stars as well as the process by which the gas is subsequently accreted onto the star . work on the initial conditions has focused on the determining the mass function of dense molecular cores that collapse into stars ( hereafter : core mass function or cmf ) . the similarity in the shapes of the imf and cmf has led to the suggestion that the imf reflects the cmf at later stages if a fixed percentage of the core mass is accreted onto the central protostar . although this provides an attractive model for a universal imf , there are several questions regarding this model . first , completeness and signal - to - noise issues may contribute to the observed shape of the cmf @xcite . second , it is unclear whether many of the observed star - less cores are gravitationally bound and capable of forming stars @xcite . third , the most massive cores are observed to form groups of stars , not single stars , and consequently , the high mass end of the cmf may describe a mass function for small stellar groups and not individual stars @xcite . fourth , the collapse of the core may be followed by the infall of gas from the surrounding molecular cloud resulting in a final stellar mass greater than that of the initial core @xcite . finally , it is known that young stars have stellar outflows and winds that mediate the infall of gas onto the central system and the accretion of gas onto the central protostar , thus the final mass must be in part determined by protostellar evolution @xcite . like the imf , a number of investigations have found that the cmf is also remarkably invariant in star forming clouds ( e.g. * ? ? ? . however , these studies may not be able to detect variations in the cmf due to the limited angular resolution of the observations and the relatively small number of detected cores . furthermore , it remains to be seen if the core masses are segregated within molecular clouds . the study presented in this paper is part of a larger effort to build a bridge between the initial conditions of star formation ( the cmf ) and the resulting ensemble properties of the nascent stars ( the imf ) . it is in the protostellar phase that the mass from the collapsing core is accreted onto a star ; thus , protostellar evolution must play a key role in determining the properties of a nascent star . our understanding of protostars is limited by the difficulty in measuring their basic properties . the radiation from the accreting protostar is reprocessed or scattered by a non - spherical infalling envelope and disk . consequently , observations spanning near- to far - ir wavelength coupled with radiative transfer codes are needed to constrain fundamental attributes such as the emitted protostar luminosity , the inner envelope density , and the geometry of the envelope and outflow cavities . for this reason , current work has focused on the most readily measured property of a protostar : its observed luminosity . the determination of the protostellar luminosity function in the nearest clouds by the _ spitzer c2d legacy program has motivated several theoretical studies showing that the luminosity functions provide an important constraint on protostellar evolution @xcite . we present here an observational study of protostellar luminosity functions in nine nearby ( @xmath3 1 kpc ) molecular clouds : orion ( combining the orion a and b clouds ) , mon r2 , cep ob3 , taurus , and the _ c2d _ mapped regions of serpens , perseus , ophiuchus , lupus i , lupus iii , and lupus iv , and chamaeleon ii . this study is motivated by the availability of _ spitzer _ @xmath4 @xmath5 surveys of each of these clouds . with spitzer _ , we are capable of identifying and classifying young stellar objects ( ysos ) and , specifically , protostars using mid - infrared photometry . the protostellar sed peaks longward of the j , h , and k@xmath6 bands , making it imperative to utilize the longer wavelength photometry to identify and characterize protostars @xcite . we utilize photometry from the 3.6 , 4.5 , 5.8 , 8.0 @xmath1 m bands from the infrared array camera ( _ irac _ ) and the 24 @xmath1 m band on the multiband imaging photometer for _ spitzer _ ( _ mips _ ) instruments aboard _ spitzer _ these wavelengths are highly sensitive to the infrared excess exhibited by ysos , with the excess even more pronounced at wavelengths longer than 24 @xmath1 m . after using the mid - ir colors to identify protostars , we use the slope of the sed and the mid - ir luminosity to estimate the bolometric luminosity of the member protostars and construct luminosity functions for each of the clouds . we then estimate the amount of contamination in our protostar sample due to reddened class ii sources , edge - on disks , and galaxies . this study expands on previous work in several ways . the chosen sample of clouds spans a range of environments , from crowded clusters with massive stars to relatively isolated regions of star formation . we present the first protostellar luminosity functions constructed from the _ spitzer _ surveys of the orion , mon r2 and cep ob3 clouds , each of which is forming massive stars . we then compare luminosity functions to test whether the luminosity function is universal , or whether it shows distinct differences between clouds as first suggested in a comparison of the taurus and ophiuchus clouds by @xcite . we also look for spatial variations in the luminosity function within individual clouds by comparing regions of high and low stellar density . finally , we compare our luminosity function with recent models of protostellar accretion . variations in the protostellar luminosity function are of key interest , since they may trace how differences in the environment affect the star formation process . the luminosity of a protostar is the sum of the intrinsic luminosity of the central protostar and the accretion luminosity generated by matter falling onto the protostar . the accretion luminosity is proportional to the mass accretion rate , and the mass accretion rate is dependent on the rate of mass infall from the envelope ( although the accretion rate may not equal the infall rate due to episodic accretion ; * ? ? ? the mass infall rate , in turn , can depend on the properties of the surrounding gas and the physical mechanism driving infall . for example , if the mass infall results from the collapse of thermally supported cores , the infall rate increases with increasing sound speed , and hence , increasing temperature @xcite . on the other hand , if the infall is the result of bondi - hoyle accretion from a larger reservoir of gas , the infall rate increases with gas density and stellar mass @xcite . alternatively , if interactions between protostars are important , the luminosity may depend on the degree of clustering . thus , variations in the luminosity function can be used to understand the physics that mediates infall and accretion . furthermore , variations in the luminosity function may give us some insight into mass segregation . higher protostellar luminosities may imply either higher accretion rates - which can result in higher masses , or higher intrinsic luminosities - which may imply more massive protostars . although the relationship between the current luminosity and the ultimate mass is uncertain for protostars , they have the advantage of being at their birth sites . in contrast , stars used to determine imfs have typically dispersed their birth environment and have moved form their birthsites ( i.e. orion , ic 348 , * ? ? ? * ) . thus , protostars provide the means to more directly connect the properties of a forming star and the environment in which it forms . in section 2 , we overview the data reduction and photometry . the identification of protostars and the determination of their bolometric luminosities is described in section 3 . after estimating the contamination in section 4 , the protostellar luminosity functions are presented and analyzed in section 5 . there we present evidence that the protostellar luminosity functions vary not only between molecular clouds , but also within clouds . section 6 compares our luminosity functions with predicted luminosity functions from a variety of accretion models and speculates on the implication of a spatially varying luminosity function for primordial mass segregation . the photometry of the clouds in our sample was assembled from multiple sources , as described in this section . in addition to using existing source catalogs from the _ c2d _ program , the taurus legacy molecular cloud survey , and guaranteed time observations of taurus , we extracted photometry from _ spitzer _ surveys of the orion , cep ob3 , and mon r2 clouds using psf fitting photometry techniques . for these three clouds , we applied a new approach for identifying saturated stars and measuring their 24 @xmath1 m magnitudes . we also identified saturated stars in the _ c2d _ clouds that are not in the existing catalogs and measured their photometry . finally , we used iras 25 @xmath1 m photometry to estimate 24 @xmath1 m photometry for highly saturated protostars in the taurus cloud . we list the irac and mips photometry for the protostar candidates , identified using the criteria described in section [ sec : criteria ] , in table [ table : table1 ] . aperture photometry of the irac data ( program ids ( pid ) 43 , 50 , 30641 , 50070 ) in orion were taken from megeath et al . ( in prep ) . the irac aperture photometry from cep ob3 and mon r2 ( pid 20403 ) were taken from gutermuth et al . ( in prep ) with no modification . mips data for orion ( pid 43 , 47 , 58 , 30641 , 50070 ) , and cep ob3 and mon r2 ( pid 20403 ) were reduced , calibrated and mosaicked using the mips instrument team s data analysis tool @xcite . an additional mips field in the cep ob3 cloud containing an extension of the cep ob3b cluster ( pid 40147 ) was reduced using cluster grinder @xcite . point spread function ( psf ) fitting photometry was used for the mips 24 @xmath1 m sources in each of these regions due the lower angular resolution in the wavelength band , the resulting confusion with other sources , and also nebulosity present at this wavelength . the psf fitting photometry was performed directly on the 24 @xmath1 m data mosaic . the steps of the psf extraction are described below . first , _ mips _ 24 @xmath1 m sources were identified using the source finding routine in photvis @xcite . the photometry was then extracted by photvis using an aperture size of 5 pixels and sky annulus between 12 and 15 pixels , with each pixel having a size of 1."25 . point - like sources were automatically selected by photvis , but careful visual inspection of the images added a few sources per image , mostly in nebulous regions . a zero - point magnitude of 16.48 was adopted ; this included the aperture correction from 5 pixels to infinity of 1.69975 . the images were in units of dn / s . the sources were detected in photvis above a 6 @xmath7 threshold , then each image was carefully inspected for missing sources . next , we created two psfs using a sample of bright but not saturated stars ; these were selected to be relatively isolated and in regions with little nebulosity . typically 15 sources were used for a given mosaic , but in a subset of mosaics there were fewer than 15 suitable sources available . we used two psfs ; one which would be used to extract photometry from the unsaturated sources and one used for the saturated sources , the latter of which was larger in order to encompass the extended psf wings of the saturated sources . the larger psf used for saturated sources has a radius of 35 pixels and a fitting radius of 10 pixels , while the psf used on unsaturated sources has a radius of 25 pixels and fitting radius of 2 pixels . we then fit the appropriate psf to the sources using the idl implementation of _ nstar _ from the idl astronomy library @xcite . to minimize the effect of crowding on the photometry , the sources were first grouped together in clusters of potentially overlapping source psfs using the routine _ group_. this routine groups together sources which were within 8 pixels of one of the other group members ; the sources in the group were then fit simultaneously using _ nstar_. sources which contained pixels with signals greater than the saturation limit were flagged . to fit the saturated sources , we modified _ nstar _ to ignore saturated pixels , which were flagged by a `` nan '' value in the image . unsaturated sources were fit with the smaller psf while saturated stars were fit with the psf with the larger size and fitting radius to ensure there were enough unsaturated pixels to provide a robust fit . any unsaturated sources which were grouped together with saturated sources were then treated like saturated sources for psf fitting and fit using the larger psf . we then used _ nstar _ to extract fluxes by first extracting fluxes for grouped unsaturated sources and then for grouped saturated sources . afterward , the psfs were scaled by the photometry and subtracted out with the _ substar _ routine ; this process was done twice , once first for the unsaturated and once for the saturated sources . the residual image was then inspected for sources which had not been extracted . these sources , typically one or fewer for each of the mosaics , were added into the photometry . the process of grouping , fitting , and extraction was iterated until we found no more new sources . very saturated sources ( those with many saturated pixels and for which photvis had difficulty estimating fluxes ) may not be easily fit by _ nstar_. these sources were then given estimated magnitudes and run through _ nstar _ using these estimations . after flux extraction , a residual image was created by _ substar _ , and visually inspected for poorly subtracted psfs . saturated sources were typically over - subtracted and were apparent in the residual image . the input magnitudes initially from photvis aperture photometry or from our initial estimation were adjusted until the residual in the image was minimized . photometric uncertainties for these sources are due to the varied environments of these sources , including areas of crowding or nebulosity , and were then found by determining the change in magnitude needed to create a noticeable over- or under - subtracted residual . this constrains the source magnitude to within typically @xmath80.25 @xmath9 . we fit 6 protostar candidates in orion in this way . for some sources , positions needed to be adjusted after _ fitting as well , which was done so that the center of the psf fell directly over the center of the source . since a formal least - square fit and uncertainty could not be obtained , we use the results of this `` fit by eye '' for these sources . three protostar candidates in orion were given positions in this way . the process of grouping , fitting , and extraction was iterated while we found more new sources . since each _ mips _ frame is the subtraction of readouts from the beginning and end of the integration , some very saturated pixels had values below our adopted threshold for saturation . these saturated pixels can be identified by comparing the original image with a fake " image . we built the fake " image by first creating an image of the background nebulosity with the stellar sources subtracted . this was done by first running _ substar _ and smoothing the resulting image to minimize the artifacts from the subtraction , including the reduction of holes " in the image due to the over - subtraction of saturated sources . the subtracted stars were then added back to the smoothed image using one of our two psfs scaled to the best fit photometry . where the resulting simulated data exceeded the saturation limit we masked out pixels . the entire photometric extraction process was then repeated with the newly identified saturated pixels masked . this process was repeated until no new saturated pixels were found . the uncertainties listed in table [ table : table1 ] are the relative uncertainties of the data taken into account the photon noise and variations of the signal in the sky annulus . the 24 @xmath1 m uncertainties are those returned by the nstar program . for the highly saturated stars that are fit by eye " , the uncertainties are given by the change in the magnitude to create a noticeable change in the residual and are noted in table [ table : table1 ] . the reported uncertainties do not include absolute uncertainties due to calibration errors ; these uncertainties are 5% for the irac and mips instruments . photometry for serpens , perseus , ophiuchus , chamaeleon , and lupus were obtained from the cores to disks ( _ c2d _ ) spitzer legacy project 4@xmath10 high - reliability data release . we used only the fluxes that were not band - filled , thus requiring that the source was detected in a specific band before adopting the tabulate photometry . these sources were identified in the _ c2d _ photometry tables as having an entry of ` -2 ' in the `` image type '' column in the high reliability catalog . in addition to the high - reliability catalog , we inspected the 24 @xmath1 m images of these clouds to search for sources that were not included in the _ c2d _ catalog due to saturation . for this purpose , we used the 24 @xmath1 m image of serpens from @xcite and _ c2d _ imagesfor the perseus , ophiuchus , lupus , and chamaeleon ii clouds . by comparing source list positions with the images , we identified sources which do not appear in the _ c2d _ catalog and appeared saturated in the images . we then attempted to fit these sources using the psf technique described in section [ sec : saturation ] . we were successful in obtaining 24 @xmath1 m fluxes for 9 saturated sources across all of the _ c2d _ clouds using the modified version of _ nstar_. we used a sample of unsaturated sources from the ophiuchus cloud to compare our photometry extraction method with that of the _ c2d _ survey . we found in a comparison of 72 sources , the mean difference in @xmath11 which we extracted and @xmath11 from the _ c2d _ data release is 0.0061 @xmath9 with a standard deviation of 0.12 @xmath9 . similarly , we compared our extracted serpens @xmath11 with that from the _ c2d _ data release , and found the median difference among 168 sources to be 0.051 @xmath9 with a standard deviation of 0.13 @xmath9 . these differences are well below the uncertainties in the flux calibration and the uncertainties in the measurements . the positions and 24 @xmath1 m photometry for the saturated sources are listed along with the rest of the photometry in table [ table : table1 ] . also in table [ table : table1 ] are the uncertainties listed for the 2mass , irac , and unsaturated mips sources are adopted from the _ c2d _ catalog . we used taurus photometry from the _ spitzer legacy taurus molecular cloud survey _ data release s14 . we add to our sample 13 sources in the l1551 region from @xcite ; l1551 is a small satellite of the taurus region with several well known protostars which was not covered by the _ spitzer legacy _ map of taurus @xcite . l1551 contains some of the most luminous known protostars in taurus and thus we desired inclusion of these sources in our taurus photometry sample . for 2 protostars from @xcite in l1551 and 6 from the legacy survey which do not have 24 @xmath1 m detections due to saturation , we used iras fluxes at 25 @xmath1 m from the faint source catalog except for l1551 irs5 , which is from the iras point source catalog . we then convert from 25 @xmath1 m flux to 24 @xmath1 m flux using the approximation @xmath12 where k = 0.986 for a cool blackbody ( t = 70 k ) ( _ mips _ handbook , version 2.0 , 2010 ) . the sources receiving their 24 @xmath1 m photometry in this way are l1551 irs5 , ic 2087 ir , 04365@xmath132535 , hl tau , gv tau , 04239@xmath132436 , 04278@xmath132253ab , and 04361@xmath132547ab . in table [ table : table1 ] we list the photometry we have adopted for the taurus protostar candidates . for most sources the photometry is from the taurus legacy survey ; we note in this table the sources for which we use photometry and adopted uncertainties from @xcite , the iras faint source catalog , and the iras point source catalog . taurus is a relatively well - studied cloud and it is important to ensure that known taurus protostars are included or accounted for in our sample . we compare our list with a list of known protostars from @xcite . we find that all but two of the 28 sources listed in their table 1 are also in our photometry sample ; the remaining sources fall off the taurus survey map and l1551 field . we compare the results of our protostar selection criteria with the protostars tabulated in @xcite in section [ sec : criteria ] . [ cols= " < , < , < , < , < , < , < , < , < , < , < , < " , ] [ table : model_ks ] determined for luminosities above 0.1 l@xmath2 . probability given as @xmath14 another two - component model is analyzed by @xcite , particularly a model similar to the two - component turbulent core model of @xcite and the two - component thermal and non - thermal model of @xcite . @xcite find that this model reproduces a realistic imf between 0.1 and 10 m@xmath2 . the model luminosity function shows a peak near 1 l@xmath2 , akin to the luminosity functions from our sample of clouds which form high mass stars . myers ( in prep ) compare the resulting luminosity function with our orion protostellar luminosity function and find good agreement . in conclusion , the shape of the luminosity functions of our high mass and low mass sf clouds are best matched by those produced by the competitive accretion luminosity function of @xcite . we note that there are still significant differences between this luminosity function and the observed luminosity functions . we will further discuss the implications of this in section [ sec : wrapup ] . of particular interest is whether the different luminosity functions imply a difference in the distribution of mass of the emerging stars forming in high and low stellar density regions . this would be a form of primordial mass segregation , with potentially more massive stars forming preferentially in denser regions . we stress again that the bolometric luminosities we estimate are the sum of the intrinsic luminosity of the protostar and the accretion luminosity resulting from gas falling onto the protostar . in low mass protostars , the accretion luminosity probably dominates , but for higher mass objects the intrinsic luminosity becomes increasingly important @xcite . we make no attempt to estimate the contribution of each of these luminosity sources , and thus , it is not clear whether the most luminous protostars are those with the highest intrinsic luminosity or whether they have the highest accretion rates . if they are sources with a higher intrinsic luminosity , then these sources probably are higher in mass , particularly if they follow a stellar birthline @xcite . in the case that the luminosity is dominated by accretion , then a higher luminosity might imply a higher protostellar mass or a higher accretion rate @xcite . this might also imply a higher outcome mass for the protostar ; however , if the accretion rate is higher and the accretion time is shorter , then sources in clusters may exhibit higher luminosities but have the same outcome imf . thus , although one interpretation of the different luminosities is that the high stellar density regions contain protostars that are more massive or that will form stars of higher mass , this is not a unique interpretation . if we are indeed seeing primordial mass segregation , then this would suggest that the presence of massive stars in the dense center of clusters is due to the environment found in these crowded regions , and not the statistical sampling of a constant imf @xcite . the competitive accretion model predicts a correlation between the density of stars in a cluster and the mass of the most massive member @xcite . this correlation is due to the large mass of gas drawn in by the gravity of the entire cluster of stars and accreted onto the most massive members of the cluster . the result is a higher mass accretion rate for the protostars in dense , clustered environments , particularly the most massive stars . although we observe variations in the luminosity function with stellar density , this may in effect be a dependence between the luminosity function and the gas column density . @xcite presented observational evidence for a schmidt - like star formation law in molecular clouds where the star formation rate per unit surface area is proportional to the gas density , squared . in this case , the stellar density may be just tracing the overall column density of the natal gas . consequentially the potential dependence of the imf on stellar density may in fact be a dependence of the imf on gas column density . a dependence between stellar density and gas density is predicted by the competitive accretion model , but could also result from jeans fragmentation in sheet - like clouds @xcite . however , although jeans fragmentation might explain the increasing stellar density with increasing gas density , it can not explain the increasing stellar mass with increasing gas and stellar density since the jeans mass decreases with increasing gas density . an increasing mass accretion rate with increasing gas column density is predicted by turbulent core models , but it is not clear whether such models could also explain the higher stellar densities in the vicinity of higher mass stars @xcite . thus , the competitive model is attractive in that it can explain the observed protostellar luminosity functions and their dependence on stellar density . based on the observational study presented in this paper , we can draw the following conclusions : that the peak of the luminosity function of clouds forming high mass stars is near 1 l@xmath2 while for the clouds not forming high mass stars the luminosity functions show a broad plateau that peaks below 1 l@xmath2 , that the luminosity functions of the high and low mass sf clouds are different , and finally , that the luminosity functions drawn from low and high density regions in orion are different . our comparison of with the current models available in the literature show that none of the available models provide an excellent match to the observed luminosity functions . this is not surprising . the offner & mckee models do not take into account details of protostellar evolution such as the clearing of the envelope and the possibility of some degree of episodic accretion . on the other hand , the @xcite model only considers models staring with the collapse of a singular isothermal sphere . we expect that these results , as well as future luminosity functions derived from herschel data , will drive the development of realistic models of protostellar models . within the available models , the competitive accretion model provides the best match . it best reproduces the shape of the low mass sf cloud and high mass sf cloud luminosity functions . we note that this model works well since it invokes an accretion rate that is dependent on both the instantaneous and final mass of the star ; this is different than models of isothermal sphere collapse where the infall rate is constant and depends only on the sound speed in the gas @xcite . the different luminosity functions in the high and low mass sf clouds can be approximately reproduced with luminosity functions of an ensemble of protostars forming a @xcite imf truncated at upper mass limits of 3 m@xmath2 and 10 m@xmath2 for the high and low mass sf clouds , respectively ( note that we have excluded the regions of high mass star formation from our study of high mass star forming clouds due to saturation ) . furthermore , the variations of the protostellar luminosity function with the spatial density of ysos is also a prediction of the competitive accretion model , where the most massive stars are formed preferentially in the centers of dense clusters of stars @xcite . for these two reasons , we currently favor the competitive accretion model . the protostellar luminosity function , however , does not provide a definitive test for models of protostellar evolution ; these models must be tested on other grounds as well , such as the measured velocities of the protostars and the density of the surrounding gas @xcite and the distribution of sources in a bolometric luminosity and temperature diagram @xcite . only though comprehensive observations examining in detail the many facets of protostars and their environments , can we both test models of protostars and drive the refinement of those models to incorporate all the essential elements of protostellar evolution . we identify 727 protostar candidates in _ spitzer _ surveys of nine star forming clouds within 1 kpc of the sun . the sample includes both nearby dark clouds forming primarily low mass stars , clouds with moderate sized clusters forming intermediate mass stars , and clouds with large clusters forming massive stars . the clouds are ophiuchus , chamaeleon ii , lupus i , iii , and iv , taurus , perseus , serpens , orion , cep ob3 , and mon r2 . with this sample , we have done the following analysis : * we estimate bolometric luminosities based on sed 3 - 24 @xmath1 m slope ( @xmath15 ) and the 1 - 24 @xmath1 m luminosity , @xmath16 . we do this by determining a relationship between @xmath17 and log(@xmath15 ) using _ sources with known bolometric luminosities , then applying the relationship to all protostars in our sample and estimating bolometric luminosities for each of our protostar candidates . * we estimate the amount of contamination from reddened class iis , edge - on disks , and background galaxies . we find that up to 20@xmath18 of our protostar candidate sample are likely edge - on disks , 4@xmath18 are likely reddened class ii objects , and 2@xmath18 are likely galaxies . these contaminating objects are removed from our sample using a statistical approach . * the resulting protostellar luminosity functions for clouds that for high mass stars ( orion , cep ob , and mon r2 ) peak near 1 l@xmath2 . the luminosity functions of each high mass sf cloud has a tail extending toward luminosities upward of 100 l@xmath2 . * the protostellar luminosity function of the low mass sf clouds ( those without high mass stars ) do not show a common peak . the combined taurus / lupus / chamaeleon luminosity function shows a marginal peak near 1 l@xmath2 , the ophiuchus luminosity function shows a marginal peak below 1 l@xmath2 , serpens shows a broad peak near 2.6 l@xmath2 , and perseus do not show a well - defined peak . none of the low mass sf clouds contain protostars at luminosities above 1000 l@xmath2 nor do they show a tail above 100 l@xmath2 . * we find significant differences between the combined low mass sf cloud luminosity function and the combined high mass sf cloud luminosity function . the median probability that they are from the same parent distribution is @xmath14 = -4.66 . thus , we do not expect the luminosity function of protostars in high mass sf clouds are from the same parent distribution as the luminosity function of protostars in low mass sf clouds . * in the orion clouds , there is a very low probability that the protostellar luminosity functions from high and low stellar density regions are drawn from the same parent distribution ; instead the luminosity function becomes increasingly biased to higher luminosities with increasing stellar density . this may be evidence for primordial mass segregation , although there are other possible explanations . a similar tend is seen in mon r2 , but the trend is weak or non - existent in cep ob3 and not present in the low mass star forming clouds . * we compare our luminosity functions with models of protostellar accretion . we do not find a good match between the model luminosity function incorporating episodic accretion of @xcite and our low mass sf cloud luminosity function . the combined luminosity function of both the high mass and low mass sf clouds are best matched by the competitive accretion model as implemented by @xcite , although other models can not be formally ruled out . competitive accretion also predicts a dependence of the accretion rate with stellar density , consistent with the variations of the luminosity function with environment we find in our high mass sf clouds . we conclude that models like competitive accretion which predict mass accretion rates that vary with both the mass of the protostar and the density of stars and gas in the surrounding environment , are best able to describe our observations . the completeness of the protostar catalog is important to assess . our search for protostars in the _ spitzer _ cloud surveys is not limited by sensitivity , but by the @xmath11 cutoff applied to minimize contamination from extragalactic sources . figure 2 shows that we detect faint sources 2 magnitudes or more fainter than our @xmath11 cutoff . furthermore , at the cutoff , the number of 24 @xmath1 m sources is increasing with increasing magnitude . for these reasons , we believe our sample to be complete in most of the surveyed molecular clouds ; however , because of the presence of bright nebulosity , subregions of each cloud may be affected by incompleteness . regions with very bright nebulosity , such as the orion nebula or ngc 2024 , are typically saturated in the _ mips _ 24 @xmath1 m image ; these saturated regions are ignored in this paper s analysis . however , regions with signal levels well below saturation can be incomplete due to the confusion with the spatially varying nebulosity . furthermore , regions with high stellar density can be systematically less complete due to the crowding and brighter nebulosity in clusters ( megeath et al . in prep ) . to the assess the impact on the analysis in this paper , we concentrate on the orion molecular clouds . orion contains the most luminous and nebulous star forming regions of our cloud sample . perhaps more importantly , orion is the one cloud in which the luminosity functions in high and low stellar density regions are statistically different . thus , we need to ascertain whether the difference between regions of high and low stellar density is real or is because of spatially varying completeness . megeath et al . ( in prep . ) measure the completeness using artificial star tests in the orion data . they find that the completeness at a given magnitude varies with position depending on the amount of bright , saturated nebulosity . they parameterize the amount of nebulosity using the root median square deviation , rmedsq = @xmath19 , which is calculated in an annulus around each source . the rmedsq gives a measure of the spatially varying signal from neighboring stars and structured nebulosity surrounding each source ; in the _ spitzer _ bands the variations are dominated by the bright mid - ir nebulosity . using the fraction of synthetic 24 @xmath1 m point sources recovered as a function of rmedsq . ( megeath et al in prep . ) , we have estimated the completeness surrounding each of the protostar candidates . the adopted rmedsq value is the mean value for all the ysos within the clustering length of @xmath20 = 0.19 pc . we then find the fraction of sources detected with magnitudes equal to the cutoff magnitudes . the results of this analysis is shown in figure 18 . for both high and low stellar density regions in orion . for the combined high and low stellar density regions , 55% of orion protostars are in regions where the fraction of stars recovered is @xmath21 0.90 ; nevertheless , both the regions of high and low stellar density have regions where the fraction of recovered sources at @xmath11 = 7 drops to close to 0 . we can reduce the incompleteness by reducing the cutoff magnitude . in figure 18 , we show the same analysis for @xmath11 = 6 . we find that using the @xmath11 = 6 cutoff , 73% are found in regions where @xmath21 0.90 sources are recovered . adams , f. c. , & fatuzzo , m. 1996 , , 464 , 256 allen , l. , et al . 2007 , protostars and planets v , 361 allen , l. e. , et al . 2004 , , 154 , 363 alves , j. , lombardi , m. , & lada , c. j. 2007 , , 462 , l17 andre , p. , ward - thompson , d. , & barsony , m. 1993 , , 406 , 122 andr , p. , menshchikov , a. , bontemps , s. , et al . 2010 , , 518 , l102 ayliffe , b. a. , langdon , j. c. , cohl , h. s. , & bate , m. r. 2007 , , 374 , 1198 bally , j. , lanber , w. d. , stark , a. a. , & wilson , r. w. 1987 , , 312 , l45 bastian , n. , covey , k. r. , & meyer , m. r. 2010 , , 48 , 339 beichman , c. a. , neugebauer , g. , habing , h. j. , clegg , p. e. , & chester , t. j. 1988 , infrared astronomical satellite ( iras ) catalogs and atlases . volume 1 : explanatory supplement , 1 , bonnell , i. a. , & bate , m. r. 2006 , , 370 , 488 bonnell , i. a. , & clarke , c. j. 1999 , , 309 , 461 bonnell , i. a. , & davies , m. b. 1998 , , 295 , 691 bonnell , i. a. , vine , s. g. , & bate , m. r. 2004 , , 349 , 735 bontemps , s. , motte , f. , csengeri , t. , & schneider , n. 2010 , , 524 , a18 brooke , t. y. , huard , t. l. , bourke , t. l. , et al . 2007 , , 655 , 364 calvet , n. , hartmann , l. , kenyon , s. j. , & whitney , b. a. 1994 , , 434 , 330 casertano , s. , & hut , p. 1985 , , 298 , 80 cernis , k. 1990 , , 166 , 315 chabrier , g. 2005 , the initial mass function 50 years later , 327 , 41 cohen , m. , wheaton , w. a. , & megeath , s. t. 2003 , , 126 , 1090 crapsi , a. , van dishoeck , e. f. , hogerheijde , m. r. , pontoppidan , k. m. , & dullemond , c. p. 2008 , , 486 , 245 cunningham , a. j. , klein , r. i. , krumholz , m. r. , & mckee , c. f. 2011 , , 740 , 107 de wit , w. j. , testi , l. , palla , f. , & zinnecker , h. 2005 , , 437 , 247 dunham , m. m. , crapsi , a. , evans , n. j. , ii , bourke , t. l. , huard , t. l. , myers , p. c. , & kauffmann , j. 2008 , , 179 , 249 dunham , m. m. , evans , n. j. , terebey , s. , dullemond , c. p. , & young , c. h. 2010 , , 710 , 470 dunham , m. k. , robitaille , t. p. , evans , n. j. , ii , schlingman , w. m. , cyganowski , c. j. , & urquhart , j. 2011 , , 731 , 90 dzib , s. , loinard , l. , mioduszewski , a. j. , boden , a. f. , rodrguez , l. f. , & torres , r. m. 2010 , , 718 , 610 elmegreen , b. g. 1997 , , 486 , 944 elmegreen , b. g. 1999 , , 515 , 323 enoch , m. l. , evans , n. j. , sargent , a. i. , & glenn , j. 2009 , , 692 , 973 evans , n. j. , ii , rawlings , j. m. c. , shirley , y. l. , & mundy , l. g. 2001 , , 557 , 193 evans , n. j. , et al . 2009 , , 181 , 321 fazio , g. g. , hora , j. l. , allen , l. e. , et al . 2004 , , 154 , 10 fischer , w. j. , et al . 2010 , , 518 , l122 flaherty , k. m. , pipher , j. l. , megeath , s. t. , winston , e. m. , gutermuth , r. a. , muzerolle , j. , allen , l. e. , & fazio , g. g. 2007 , , 663 , 1069 furlan , e. , et al . 2008 , , 176 , 184 gordon , k. d. , rieke , g. h. , engelbracht , c. w. , et al . 2005 , , 117 , 503 greene , t. p. , wilking , b. a. , andre , p. , young , e. t. , & lada , c. j. 1994 , , 434 , 614 gutermuth , r. a. 2005 , ph.d . thesis , gutermuth , r. a. , megeath , s. t. , pipher , j. l. , et al . 2005 , , 632 , 397 gutermuth , r. a. , et al . 2008 , , 674 , 336 gutermuth , r. a. , megeath , s. t. , myers , p. c. , allen , l. e. , pipher , j. l. , & fazio , g. g. 2009 , , 184 , 18 gutermuth , r. a. , pipher , j. l. , megeath , s. t. , et al . 2011 , , 739 , 84 hartmann , l. , dalessio , p. , calvet , n. , & muzerolle , j. 2006 , , 648 , 484 hartmann , l. , cassen , p. , & kenyon , s. j. 1997 , , 475 , 770 hartmann , l. , & kenyon , s. j. 1996 , , 34 , 207 harvey , p. m. , et al . 2006 , , 644 , 307 harvey , p. , mern , b. , huard , t. l. , rebull , l. m. , chapman , n. , evans , n. j. , ii , & myers , p. c. 2007 , , 663 , 1149 hillenbrand , l. a. , & hartmann , l. w. 1998 , , 492 , 540 hunter , t. r. , brogan , c. l. , megeath , s. t. , menten , k. m. , beuther , h. , & thorwirth , s. 2006 , , 649 , 888 jijina , j. , myers , p. c. , & adams , f. c. 1999 , , 125 , 161 kenyon , s. j. , hartmann , l. w. , strom , k. m. , & strom , s. e. 1990 , , 99 , 869 kenyon , s. j. , dobrzycka , d. , & hartmann , l. 1994 , , 108 , 1872 kenyon , s. j. , & hartmann , l. 1995 , , 101 , 117 kharchenko , n. v. , piskunov , a. e. , rser , s. , schilbach , e. , & scholz , r .- d . 2005 , , 438 , 1163 kirk , h. , & myers , p. c. 2011 , , 727 , 64 krumholz , m. r. , cunningham , a. j. , klein , r. i. , & mckee , c. f. 2010 , , 713 , 1120 krumholz , m. r. , mckee , c. f. , & klein , r. i. 2005 , , 438 , 332 lada , c. j. , & lada , e. a. 1991 , the formation and evolution of star clusters , 13 , 3 lada , c. j. , & wilking , b. a. 1984 , , 287 , 610 lada , c. j. , muench , a. a. , rathborne , j. , alves , j. f. , & lombardi , m. 2008 , , 672 , 410 landsman , w. b. 1993 , astronomical data analysis software and systems ii , 52 , 246 larson , r. b. 1985 , , 214 , 379 larson , r. b. 1969 , , 145 , 297 lombardi , m. , lada , c. j. , & alves , j. 2010 , , 512 , a67 luhman , k. l. , mamajek , e. e. , allen , p. r. , & cruz , k. l. 2009 , , 703 , 399 luhman , k. l. , rieke , g. h. , young , e. t. , et al . 2000 , , 540 , 1016 mcclure , m. k. , et al . 2010 , , 188 , 75 mckee , c. f. , & tan , j. c. 2003 , , 585 , 850 megeath , s. t. , allgaier , e. , young , e. , allen , t. , pipher , j. l. , & wilson , t. l. 2009 , , 137 , 4072 megeath , s. t. , wilson , t. l. , & corbin , m. r. 2005 , , 622 , l141 megeath , s. t. , et al . 2004 , , 154 , 367 menten , k. m. , reid , m. j. , forbrich , j. , & brunthaler , a. 2007 , , 474 , 515 mern , b. , et al . 2008 , , 177 , 551 meyer , m. r. , calvet , n. , & hillenbrand , l. a. 1997 , , 114 , 288 miesch , m. s. , & bally , j. 1994 , , 429 , 645 moeckel , n. , & bonnell , i. a. 2009 , , 400 , 657 muzerolle , j. , et al . 2004 , , 154 , 379 myers , p. c. , & fuller , g. a. 1992 , , 396 , 631 myers , p. c. 2009 , , 706 , 1341 myers , p. c. 2011 , , 743 , 98 natta , a. 1993 , , 412 , 761 offner , s. s. r. , & mckee , c. f. 2011 , , 736 , 53 palla , f. , & stahler , s. w. 1991 , , 375 , 288 pineda , j. e. , caselli , p. , & goodman , a. a. 2008 , , 679 , 481 racine , r. 1968 , , 73 , 233 rebull , l. m. , et al . 2010 , , 186 , 259 reid , m. a. , wadsley , j. , petitclerc , n. , & sills , a. 2010 , , 719 , 561 rieke , g. h. , young , e. t. , engelbracht , c. w. , et al . 2004 , , 154 , 25 robitaille , t. p. , whitney , b. a. , indebetouw , r. , wood , k. , & denzmore , p. 2006 , , 167 , 256 romn - ziga , c. g. , lada , c. j. , & alves , j. f. 2009 , , 704 , 183 sadavoy , s. i. , et al . 2010 , , 710 , 1247 sagar , r. 2002 , extragalactic star clusters , 207 , 515 selman , f. j. , & melnick , j. 2008 , , 689 , 816 shu , f. h. 1977 , , 214 , 488 stansberry , j. a. , gordon , k. d. , bhattacharya , b. , et al . 2007 , , 119 , 1038 stetson , p. b. 1987 , , 99 , 191 strom , k. m. , & strom , s. e. 1993 , , 412 , l63 swift , j. j. , & williams , j. p. 2008 , , 679 , 552 terebey , s. , van buren , d. , brundage , m. , & hancock , t. 2006 , , 637 , 811 testi , l. , palla , f. , & natta , a. 1999 , , 342 , 515 ungerechts , h. , & thaddeus , p. 1987 , , 63 , 645 vorobyov , e. i. , & basu , s. 2005 , , 633 , l137 white , r. j. , greene , t. p. , doppmann , g. w. , covey , k. r. , & hillenbrand , l. a. 2007 , protostars and planets v , 117 whitney , b. a. , indebetouw , r. , bjorkman , j. e. , & wood , k. 2004 , , 617 , 1177 whitney , b. a. , wood , k. , bjorkman , j. e. , & cohen , m. 2003 , , 598 , 1079 whitney , b. a. , wood , k. , bjorkman , j. e. , & wolff , m. j. 2003 , , 591 , 1049 whittet , d. c. b. , prusti , t. , franco , g. a. p. , gerakines , p. a. , kilkenny , d. , larson , k. a. , & wesselius , p. r. 1997 , , 327 , 1194 wilking , b. a. , lada , c. j. , & young , e. t. 1989 , , 340 , 823 wilson , t. l. , mauersberger , r. , gensheimer , p. d. , muders , d. , & bieging , j. h. 1999 , , 525 , 343 winston , e. , et al . 2007 , , 669 , 493 young , c. h. , & evans , n. j. , ii 2005 , , 627 , 293 m magnitudes for all sources fitting the protostar criteria and with @xmath22 @xmath21 3 ( black ) . also shown are the histograms of the galaxy contamination estimated from the swire data ( blue ) . the @xmath11 cutoff is shown in red.,scaledwidth=80.0% ] cutoff and the [ 4.5 ] - [ 24 ] color cutoff . the 24 @xmath1 m magnitude is corrected for distance but not reddening . galaxies comprise a distinct clump of fainter sources ( at 24 @xmath1 m ) with colors similar to those of protostars , falling near [ 4.5 ] - [ 24 ] = 6 . symbols and colors are the same as figure [ fig : fig1].,scaledwidth=80.0% ] map . shown are all class iis ( blue ) , rising spectrum protostar candidates ( red ) , and flat spectrum ( green ) protostar candidates with @xmath22 @xmath21 3 ( circles ) and with @xmath22 @xmath3 3 ( diamonds ) , as well as the @xmath22 = 3 contours.,scaledwidth=80.0% ] map . shown are all class iis ( blue ) , rising spectrum protostar candidates ( red ) , and flat spectrum ( green ) protostar candidates with @xmath22 @xmath21 3 ( circles ) and with @xmath22 @xmath3 3 ( diamonds ) , as well as the @xmath22 = 3 contours.,scaledwidth=80.0% ] map . shown are all class iis ( blue ) , rising spectrum protostar candidates ( red ) , and flat spectrum ( green ) protostar candidates with @xmath22 @xmath21 3 ( circles ) and with @xmath22 @xmath3 3 ( diamonds ) , as well as the @xmath22 = 3 contours.,scaledwidth=80.0% ] map . shown are all class iis ( blue ) , rising spectrum protostar candidates ( red ) , and flat spectrum ( green ) protostar candidates with @xmath22 @xmath21 3 ( circles ) and with @xmath22 @xmath3 3 ( diamonds ) , as well as the @xmath22 = 3 contours.,scaledwidth=80.0% ] map . shown are all class iis ( blue ) , rising spectrum protostar candidates ( red ) , and flat spectrum ( green ) protostar candidates with @xmath22 @xmath21 3 ( circles ) and with @xmath22 @xmath3 3 ( diamonds ) , as well as the @xmath22 = 3 contours.,scaledwidth=80.0% ] map . shown are all class iis ( blue ) , rising spectrum protostar candidates ( red ) , and flat spectrum ( green ) protostar candidates with @xmath22 @xmath21 3 ( circles ) and with @xmath22 @xmath3 3 ( diamonds ) , as well as the @xmath22 = 3 contours.,scaledwidth=80.0% ] map . shown are all class iis ( blue ) , rising spectrum protostar candidates ( red ) , and flat spectrum ( green ) protostar candidates with @xmath22 @xmath21 3 ( circles ) and with @xmath22 @xmath3 3 ( diamonds ) , as well as the @xmath22 = 3 contours.,scaledwidth=80.0% ] map . shown are all class iis ( blue ) , rising spectrum protostar candidates ( red ) , and flat spectrum ( green ) protostar candidates with @xmath22 @xmath21 3 ( circles ) and with @xmath22 @xmath3 3 ( diamonds ) , as well as the @xmath22 = 3 contours . most of the region is above @xmath22 = 3.,scaledwidth=60.0% ] map showing both the orion a ( dec @xmath3 -3.5@xmath23 ) and b ( dec @xmath21 -3.5@xmath23 ) clouds . our sample of orion cloud protostars is the combined sample from both of these clouds . shown are all class iis ( blue ) , rising spectrum protostar candidates ( red ) , and flat spectrum ( green ) protostar candidates with @xmath22 @xmath21 3 ( circles ) and with @xmath22 @xmath3 3 ( diamonds ) , as well as the @xmath22 = 3 contours . the lack of ysos near the onc is due to saturation at 24 @xmath1 m in this region.,scaledwidth=60.0% ] map . shown are all class iis ( blue ) , rising spectrum protostar candidates ( red ) , and flat spectrum ( green ) protostar candidates with @xmath22 @xmath21 3 ( circles ) and with @xmath22 @xmath3 3 ( diamonds ) , as well as the @xmath22 = 3 contours . the @xmath22 @xmath3 3 region used to identify likely edge - on disk contamination in cep ob3b is also shown.,scaledwidth=80.0% ] map . shown are all class iis ( blue ) , rising spectrum protostar candidates ( red ) , and flat spectrum ( green ) protostar candidates with @xmath22 @xmath21 3 ( circles ) and with @xmath22 @xmath3 3 ( diamonds ) , as well as the @xmath22 = 3 contours.,scaledwidth=80.0% ] ) relationship . protostars from the _ c2d _ sample with well - constrained bolometric luminosities are plotted as black circles . the taurus protostars with established bolometric luminosities are plotted as red squares , but are not used to determine the fit . left panel : the best fit correlation ( solid line ) to _ c2d _ sources used to derive the relationship using rising spectrum protostars : the square - root of the fraction of mid - ir to the bolometric luminosity vs. log(@xmath15 ) . right panel : the fraction of bolometric to mid - ir luminosity as a function of log(@xmath15 ) . the solid line shows the fit for the rising spectra protostars and adopted ratio for the flat spectrum sources at log(@xmath15 ) @xmath3 -0.5 , which are included in this plot.,scaledwidth=80.0% ] . these distributions look similar , and a k - s test gives the probability that they are from the same parent distribution as 0.83 . right panel : shown in black are estimated luminosities from this work vs. the luminosities from @xcite . shown as squares ( in red ) are taurus sources from @xcite.,scaledwidth=80.0% ] based on the @xmath11 cutoff . the panel showing combined taurus , lupus , and chamaeleon luminosity function also shows the components from each cloud : taurus is the thin red histogram , lupus is the green , and chamaeleon is shown as the blue histogram.,scaledwidth=80.0% ] and [ eqn : flat ] to the de - reddend bolometric luminosity for protostar candidates in all clouds in this survey . right panel : uncorrected ophiuchus protostar candidate luminosity function ( black ) and de - reddened protostar candidate luminosity function ( blue ) using the technique described in section [ sec : reddening].,scaledwidth=80.0% ] nearest neighbor yso ) and calculated bolometric luminosity for all protostar candidates . the final panels show the protostars for orion , cep ob3 , and mon r2 and for tau / lup / cha , perseus , and ophiuchus . orion , cep ob3 , and mon r2 each have protostar candidates above 100 l@xmath2 and show sources at larger nn4 distances only at lower luminosities . the low mass sf clouds do not show this relationship individually or in the combined plot.,scaledwidth=80.0% ] @xmath3 7 @xmath9 ( left ) sources and for @xmath11 @xmath3 6 @xmath9 ( right ) sources . the median k - s probability that the high and low stellar density luminosity functions are from the same parent distribution is 0.0013 for the @xmath11 @xmath3 7 sources and 0.0015 for the luminosity functions of sources with @xmath11 @xmath3 6.,scaledwidth=80.0% ] m detections for orion high stellar density ( green ) and low stellar density ( blue ) ysos . orion disk sources within @xmath20 of a protostar candidate in high or low stellar density ( left panel ) show similar distributions of 24 @xmath1 m detections for sources in high and low stellar density regions and have a k - s probability of 0.61 for @xmath11 @xmath3 7 sources . distributions of @xmath11 for the orion protostar candidates in high and low stellar density ( right panel ) do not show similarities , and have a k - s probability of 0.0089 ( at @xmath11 @xmath3 7 ) that they are from the same parent distribution.,scaledwidth=80.0% ] , and the high mass sf clouds are compared with models corresponding to an upper mass limit of 10 m@xmath2 . the left two panels show the uncorrected luminosity functions with uncorrected @xmath24 ( vertical red line ) . the right two panels show the de - reddened combined luminosity functions and the de - reddened @xmath24 ( vertical red line).,scaledwidth=80.0% ] m sources recovered as a function of the confusion due to nebulosity and crowding for the orion clouds . we use the rmedsq technique of megeath et al . ( in prep . ) to determine the expected fraction of sources detected with magnitudes equal to the cutoff magnitude . we show the distribution of the fraction of detected sources in high stellar densities ( green ) and low stellar densities ( blue ) regions expected . the top row of plots uses a cutoff magnitude @xmath11 = 6 @xmath9 , and the bottom row of plots use @xmath11 = 7 @xmath9 . the left column of plots show the distribution of the expected fractions for all of the protostar candidates . the right column of plots show the expected detection fraction at the cutoff vs. the luminosity of the protostar candidate . we find that the @xmath11 = 6 @xmath9 cutoff has a median expected detection fraction of 0.995 or 0.973 for the high stellar density protostar candidates and 0.999 for the low stellar density protostar candidates , and the @xmath11 = 7 @xmath9 cutoff has a median expected detection fraction of 0.935 or 0.804 for the protostar candidates in high stellar density regions and 0.981 for the protostar candidates in regions of low stellar density.,scaledwidth=80.0% ]

we discuss morphological properties of the large - scale jets in powerful radio sources , which are now observed at radio , optical and x - ray frequencies , in order to determine the origin of their x - ray radiation and the nature of the particle acceleration responsible for their multiwavelength emission . we show that modeling knots of the these objects as stationary regions of energy dissipation within the uniform and continuous jet flow leads to several problems . such problems are especially pronounced if the observed x - ray emission is due to inverse - compton scattering of the cmb radiation . we explore another possibility , namely that the knots represent moving and separate portions of the jet matter , with excess kinetic power . we suggest a possible connection between this scenario and the idea of intermittent / highly modulated jet activity . the proposed model can explain some morphological properties of quasar jets like high knot - to - interknot brightness contrasts , frequency - independent knot profiles and almost universal extents of the knot regions independently of the exact emission mechanism responsible for producing the x - rays . in this context , we consider different possibilities for the production of such x - ray radiation , and discuss the related issue of particle acceleration . we conclude that the appropriate process can not be designated for certain yet . we suggest , however , that x - ray observations of the jet in 3c 120 ( in which intermittent jet activity is quite distinct ) already seem to support the synchrotron origin of the x - ray emission , at least in this object .

astro-ph0401356

growing numbers of large - scale jets in powerful extragalactic radio sources are now being observed at optical and x - ray frequencies ( see , e.g. , * ? ? ? new , and in many ways unexpected , results call for a wide discussion of the physical conditions within these objects , and in particular of the processes responsible for accelerating radiating particles to the required ultrarelativistic energies . until now , few observations had gone beyond the integrated spectral properties of multiwavelength emission from large - scale jets . however , the high angular resolution of hubble space telescope ( hst ) and chandra x - ray observatory ( cxo ) have given us the opportunity to analyze the morphological characteristics of jets at optical and x - ray photon energies , allowing for the comparison of jet morphology at different frequencies . here we discuss the structure of large - scale quasar jets , in order to determine the origin of their x - ray radiation and the nature of the particle acceleration responsible for their multiwavelength emission . cxo has detected about ten quasar jets with observed x - ray luminosities @xmath0 erg s@xmath1 and projected lengths ranging from a few tens up to a few hundred kpc . the x - ray emission is usually much too intense to be explained in terms of synchrotron radiation by the same , single - power - law electron energy distribution that produces the radio - to - optical emission . nor can it be explained in terms of synchrotron - self compton ( ssc ) radiation of a _ homogeneous _ , few - kpc - long knot region , _ unless _ the energy density is far above the equipartition value , which would imply extremely high jet powers ( e.g. , * ? ? ? * ; * ? ? ? * for the case of pks 0637 - 752 ) . this led to the idea that the x - ray emission results from inverse - compton up - scattering of cosmic microwave background ( cmb ) photons ` external inverse - compton ' ( eic ) emission by the low - energy part of the electron energy distribution @xcite . such a model , however , requires highly relativistic jet velocities at distances of a few hundred kpc from the active nuclei ( bulk lorentz factors @xmath2 ) , in apparent disagreement with radio observations ( e.g. , * ? ? ? on the other hand , the eic model was claimed to be additionally supported by energetic considerations @xcite . in this respect , discoveries of large - scale x - ray jets in the gigahertz - peaked ( gps ) objects pks 1127 - 145 and b2 0738 + 313 ( * ? ? ? * ; * ? ? ? * respectively ) , as well as in the high - redshift source gb 1508 + 5714 ( @xmath3 , * ? ? ? * ) are particularly interesting . x - ray emission of large - scale quasar jets , if due to the eic process , has also cosmological implications , as pointed out by @xcite . therefore , it is important to discuss whether comptonization of the cmb within highly relativistic flow is really the most plausible explanation of the cxo quasar jet observations , and to look for possibilities of discriminating between this and other models ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? analysis of the morphology of the emitting regions constitutes an interesting approach to this problem . the most apparent characteristic of quasar jets is their knotty morphology with high knot - to - interknot brightness contrast , but also with distinct ( in some cases ) inter - knot diffuse emission . in addition , knot profiles seem to be frequency - independent , and knot extents are similar when observed at radio , optical and x - ray photon energies . detailed observations of the 3c 273 jet @xcite reveal also that spectral changes along the flow are not correlated with brightness changes this may be a general characteristic of these objects . finally , in some cases spatial offsets between the maxima of the x - ray and radio emission within the knot regions were noted ( e.g. , pks 1127 - 145 , * ? ? ? * ) . it is not clear whether all of these features can be explained in a framework of models involving extended shock waves within continuous jet flow . in fact , we argue that the morphological characteristics can not be explained in this way , and that substantial modifications of the standard picture are required . such modifications are especially needed if the x - ray emission of quasar jets is due to the eic process . we propose that at least some aspects of the hst and cxo observations can be understood in terms of intermittent ( modulated ) jet activity . in this context , we comment on both the external - compton scenario of quasar jet x - ray emission , and on a model of boundary layer acceleration and resulting high - energy radiation @xcite . in particular , in 2 we emphasize the problems with modeling x - ray knots of quasar jets as stationary regions of particle acceleration . in 3 we consider the possibility that the knots are moving sources of non - thermal radiation , propose a possible connection of this scenario to models of intermittent jet activity , and discuss the particle acceleration processes possibly involved in such a scenario . the discussion and final conclusions are presented in 4 and 5 , in the most common version of the eic model , in which the knots are identified with strong stationary shocks through which the jet matter flows continuously , none of the morphological features mentioned in the introduction are straightforwardly expected . the knots appear to be localized along the jet , yet stationary shocks are likely to be highly oblique ( with respect to the jet axis ) and significantly elongated in the flow direction , if the jet is confined by external pressure @xcite . large - scale jets in powerful radio sources display approximately cylindrical geometry , and therefore it is thought that they must be confined . the nature of their confinement is unknown , but it has been shown that the low - density and high - pressure nonthermal medium of the radio lobe can be an efficient confinement agent @xcite . if we are not seeing the shape of the shocks themselves , could the emissivity patterns give rise to the knot morphology ? it seems unlikely , since the cooling time of low - energy electrons responsible for up - scattering cmb photons to the observed x - ray energies is extremely long when compared with the lifetime of electrons emitting synchrotron optical photons in an equipartition magnetic field . in addition , the eic emission is not very sensitive to the magnetic field structure or slow adiabatic losses within the uniform , well - collimated flow . therefore , in the case of the eic model and a continuous jet with knot regions marking the positions of the stationary ( e.g. , reconfinement ) shocks , one should rather expect extended x - ray knot emission , with approximately constant flux along the jet ( if the jet bulk velocity is roughly constant on the large scales ) , and with knot profiles reflecting directly the large differences in radiative and adiabatic cooling times of electrons radiating in radio , optical and x - rays . this is in disagreement with observations , and hence modifications to the eic model are required . one possibility is to assume an inhomogeneous structure of the emission regions , consisting of a number of small , far - from - equipartition and adiabatically expanding clumps of radiating plasma @xcite . the eic model was claimed to be strongly supported by energetic arguments @xcite . let us briefly re - discuss this issue in the context of the clumping scenario . if the jet is composed of a magnetic field of unknown intensity @xmath4 ( measured in the comoving frame or electron lorentz factor @xmath5 , noting instead that they always refer to the emitting plasma rest frame . ] ) , protons , and radiating electrons , one can find a value of the magnetic field that minimizes the total kinetic power of the emission regions for a given observed synchrotron luminosity @xmath6 , observed volume of the emitting plasma @xmath7 , ratio of proton - to - electron energy densities @xmath8 , spectral energy distribution of the radiating ultrarelativistic electrons , bulk lorentz factor @xmath9 , and jet inclination @xmath10 . this is because the total jet power is @xmath11 , where the poynting flux is @xmath12 , the bulk kinetic power carried by the radiating electrons is @xmath13 , and the bulk kinetic power due to the protons is @xmath14 . as the electron energy density for a given total synchrotron luminosity is @xmath15 , one has also @xmath16 and hence @xmath17 . setting @xmath18 one can find that the magnetic field @xmath19 minimizing @xmath20 is the one for which @xmath21 . in other words , one gets condition for the equipartition between ponting flux and the jet kinetic power carried by the particles ( see appendix a ) . it was shown @xcite that this minimum power condition is consistent with x - ray observations of the large - scale jet associated with the quasar pks 0637 - 752 , if its x - ray flux is due to the eic emission of a homogeneous knot region with radius @xmath22 cm and @xmath2 . one should emphasize , however , that this agreement is obtained by invoking certain assumptions about the value of @xmath23 and details of the electron energy distribution . in addition , allowing for inhomogeneous structure of the emission region changes the energetic conclusions . in particular , the magnetic field that minimizes the total power for the uniform knot model , @xmath19 , does not coincide with the minimum energy field evaluated in the framework of the clumping scenario for the radiating plasma , @xmath24 . instead , one has @xmath25 where @xmath26 is the filling factor of the clumps ( see appendix a ) . hence , if the magnetic field estimated from the observed ratio of inverse - compton to synchrotron luminosities agrees with the critical magnetic field computed for the homogeneous knot model , @xmath27 , then any strong clumping of the jet ( i.e. @xmath28 ) will cause large departures from the minimum power condition for the radiating plasma , @xmath29 . in addition , in order to discuss energetics of the whole clumpy jet , one has to make some assumptions about the non - radiating matter between the clumps , which can not be verified by observations . as a result , energetic arguments supporting the eic model can not be applied to the clumping scenario . losing a reliable diagnostic of jet energetics is the price one has to pay for explaining a rapid decrease of the x - ray flux outside the knots . in the expanding , radiating plasma , the magnetic field is expected to decrease in proportion to some power of the emitting region radius , @xmath30 , depending on the assumed magnetic field structure . therefore , one has @xmath31 where @xmath32 and @xmath6 are , respectively , the observed inverse - compton and synchrotron luminosities , @xmath33 is the energy density of the cmb radiation in the comoving frame of the jet ( which is constant along the flow if the bulk velocity of the expanding region is constant ) , and @xmath34 is the energy density of the jet magnetic field . hence , knot profiles in the eic model with adiabatic expansion should always be different when observed at x - ray and radio frequencies ( both produced by the low - energy part of the electron spectrum ) , unless very rapid expansion of the radiating plasma , i.e. , strong clumping of the jet , is involved . in this case , however , the contribution from synchrotron self - compton ( ssc ) emission can not be ignored . note that the main argument in favor of the eic emission model over the ssc model was that the latter requires clumps and departures from equipartition , in order to explain the strong x - ray emission @xcite . the second problem attributed to the eic - clumping model was that rapid expansion of the emitting clumps would inevitably create a large population of very low energy electrons . they should in turn produce optical emission via the eic process , possibly exceeding the synchrotron radiation ( and the observational limits ) at these frequencies . energy losses of the low - energy electrons , radiating via synchrotron radiation in the radio band and via eic emission in the x - ray domain , are dominated by adiabatic cooling , @xmath35 . this process leads to the decrease of the electron energy with linear size of the emitting region @xmath36 according to @xmath37 and also to the decrease of the inverse - compton luminosity at given observed frequency according to @xmath38 where @xmath39 is the spectral index of the electron energy distribution ( appendix b ) . for @xmath40 and two - dimensional expansion ( i.e. , @xmath41 ) , one gets a linear decrease of the observed x - ray luminosity with @xmath36 . in such a case , in order to reduce the x - ray flux by , say , one order of magnitude , one has to allow all the emitting clumps to expand within the single knot region by , at least , one order of magnitude in their linear sizes . thus , electrons up - scattering cmb photons to the observed @xmath42 kev energies , @xmath43 , will end up at energies @xmath44 after such expansion , giving rise to the observed eic emission at @xmath45 ev ( see * ? ? ? this possibility , although not completely excluded , contradicts the synchrotron origin of optical emission from quasar jets , which has been confirmed repeatedly by polarization measurements . hence , the eic model requires a very sharp cut - off in the injected electron energy distribution and/or relatively limited expansion of the clumps both of which call for fine - tuning of the model parameters in order not to overproduce optical inverse - compton radiation . small sizes of the emitting plasma in the clumpy jet can also amplify the energy density of the synchrotron photons to a level leading to domination of the ssc component over the eic one . note that , for the electron energy spectrum usually considered in the eic model , the photon energy range of the ssc emission corresponds roughly with the spectral range of the eic radiation . the ratio of observed ssc to eic luminosity is @xmath46 where @xmath47 is the number of the emitting clumps with radius @xmath48 per knot , @xmath49 is a redshift of the source and @xmath50 is the jet doppler factor for its bulk velocity @xmath51 and inclination @xmath52 ( appendix c ) . it is always possible to adjust the unknown parameters @xmath53 , @xmath48 and @xmath47 in order to obtain @xmath54 . as a result , it is possible to model the observed x - ray emission of large - scale quasar jets by ssc emission , and hence there are no obvious constraints on the large - scale jet parameters from x - ray observations . in particular , the doppler factor of the jet in the ssc model can be smaller than that required by the eic model . modifications to the eic model , which are required in the case of knots treated as stationary regions of energy dissipation , decrease the model s attractiveness . therefore , alternative scenarios postulating a synchrotron origin of the x - ray emission should be considered . however , in this case the x - ray photons would have to be produced by a different electron population than the one responsible for the radio - to - optical emission , as the x - ray flux is usually above the extrapolated radio - to - optical power - law continuum ; or else deviations from a single power - law behavior in the electron energy distribution high energy particle pile - up in particular have to be invoked . furthermore , to reconcile the observed flat x - ray spectral indices with strong cooling of the highly energetic electrons and large observed sizes of the emission regions ( see discussion in * ? ? ? * ) , one must have particle acceleration occurring over extended volumes . this can be illustrated by rewriting the observed propagation length of the electrons radiating synchrotron photons with frequency @xmath55 in a form @xmath56 where @xmath57 is the radiative loss time scale , @xmath58 g , inverse - compton radiative losses on the cmb are included and we put @xmath59 as appropriate for small jet viewing angles @xmath10 ( appendix d ) . the turbulent second - order fermi acceleration process provides a plausible way of obtaining high - energy particle pile - up , if the time scale for particle escape from the acceleration region is much longer than the radiative loss time scale ( other possibilities will be discussed in the later sections ) . this is because such a process favors the particle energy for which the acceleration time scale is equal to the radiative loss time scale . with steady injection of low - energy particles , turbulent acceleration piles up radiating particles at this critical energy . as discussed by @xcite and @xcite , boundary shear layers of large - scale jets are ideal sites for such a process , as the inevitable flow instabilities very likely provide the required amount of turbulence , and a sheared magnetic field configuration does not enable radiating electrons to escape easily from the acceleration region . for conventional assumptions about the jet energetics ( including energy equipartition between the magnetic field and radiating electrons ) , the time scales estimated by @xcite lead to sufficient pile - up of the high - energy electrons to produce synchrotron x - rays with a luminosity comparable to the observed values , without fine - tuning of the jet parameters . in addition , the continuous nature of such an acceleration process can explain the absence of cooling effects in the observed optical and x - ray continuum . boundary layer acceleration , as initially proposed to explain some aspects of the x - ray jet emission observed by cxo , can not solely account for the morphological characteristics of quasar jets , and their knotty nature in particular . in this model the lifetime of the x - ray emitting electrons is extremely short , and hence the emitting region sizes are related to the extent of the particle acceleration sites and not to the spatial scales of the electron cooling , as in the eic and ssc scenarios . also , the interplay between localized shock acceleration and continuous , turbulent energization of the radiating electrons can result in offsets between the radio and x - ray emitting regions , while in the eic and ssc models the maximum of the radio emission should have the same location within the knot as the maximum of the x - ray radiation . as mentioned in the introduction , spatial offsets between radio and x - ray knot maxima are observed in some large - scale quasar jets . however , a simple model of turbulent acceleration does not explain directly why the jets being continuous flows exhibit high knot - to - interknot brightness contrasts , although one can imagine some possible solutions . for example , it was suggested @xcite that the knots can form by the shock compression of the energetic electrons previously accelerated at the boundary layer upstream of the shock , in accordance with the discussion by @xcite . detailed spectral studies at optical frequencies could allow us to distinguish between an inverse - compton and a synchrotron origin for the x - ray emission from large - scale quasar jets . for example , the presence of spectral cut - offs at infrared - to - optical photon energies would argue strongly for the eic model , while spectral flattenings would support synchrotron models involving high - energy particle pile - ups . however , optical emission detected from a few quasar jets , which is most probably synchrotron in nature as indicated by polarization measurements , reveal a variety of spectral shapes , and hence no definitive statements can be made yet . in particular , optical spectral indices of the knots in powerful jets range from @xmath60 in the brightest part of the 3c 273 jet @xcite up to @xmath61 in 3c 279 @xcite . interestingly , spectral hardening at ultraviolet frequencies was noted for the 3c 273 jet @xcite . these observations indicate the presence of @xmath62 electrons producing synchrotron infrared - to - optical photons , for an assumed equipartition magnetic field @xmath63 g , but do not give tight constraints on their spectral distribution . the broad - band energy spectrum of the radiating electrons is usually modeled in the eic scenario as a single power - law , @xmath64 , with spectral index @xmath65 and sharp cut - offs at @xmath66 and @xmath67 ( e.g. , * ? ? ? except for its simplicity , however , such an assumption on a form of @xmath68 has neither theoretical nor observational justification . moreover , it can not be excluded that the electron energy distribution @xmath68 extends up to very high energies , @xmath69 , as seems to be the case in a few kpc - scale fr i jets ( see discussion in * ? ? ? another point is that the broad - band radio - to - x - ray spectra for some of the knots in the large - scale jets associated with quasars 3c 273 and pks 1136 - 135 can not be explained by the eic model , indicating with little doubt the synchrotron origin of the x - rays @xcite . therefore , existing optical spectra of the knots in quasar jets can not yet exclude different scenarios for modeling the radio - to - x - ray continua of these objects . as we have discussed above , modeling knots of large - scale quasar jets as stationary regions of energy dissipation within the uniform and continuous jet flow leads to several problems in explaining the observed x - ray emission , especially in the framework of the eic scenario . below we explore another possibility , namely that the knots represent portions of a relativisticaly moving jet with excess kinetic power . such a situation is expected if the activity of central engines in radio - loud agns is intermittent , or highly modulated . the frequency - independence of the knot sizes , as well as high knot - to - interknot brightness contrasts , can be understood if the radiating particles are present mainly within the separate portions of the jet flow the knots as compared with the interknot regions . this is true almost independently of the emission process involved , and in particular it applies also to the eic model . in addition , in moving knot scenarios the brightness changes along the jet path do not have to correlate with the spectral changes , if they are controlled by distinct physical processes , i.e. , if the former are due to changes in the kinetic energy of the flow resulting from intermittent or modulated jet activity , while the latter are determined by particle acceleration and cooling processes . these facts make the idea of intermittent jet activity promising for explaining the morphologies of cxo quasar jets . the idea of intermittent jet activity was presented in the context of ` partial ' radio jets associated with the powerful radio galaxies 3c 219 @xcite and 3c 288 @xcite . it was also proposed that the restarting jet model can , in a natural way , reconcile the presence of the extended radio emission observed in a few gps sources ( e.g. , 0108 + 388 ) with the inferred young ages of gps compact radio structures , @xmath70 yrs @xcite . we note that extended radio emission around gps objects can sometimes reach mpc scales , as in the cases of b1144 + 352 @xcite and b1245 + 676 @xcite . in addition , as discussed by @xcite , intermittent jet activity can explain the general overabundance of compact radio objects when compared with the number of classical ( extended ) doubles . in particular , @xcite showed that the observed distribution of sizes for extragalactic radio sources can be explained in the framework of a simple evolution scenario if one assumes that the jet - like active period persists only over @xmath71 years , and refreshes on the time scale of @xmath72 years . interestingly , similar time scales can be obtained in the thermal - viscous instability model for the accretion disk @xcite . this suggests that the jet activity and hence the whole evolution of the radio source could be driven by unstable accretion disks surrounding supermassive black holes ( see in this context * ? ? ? non - steady jet activity , albeit on longer time scales than discussed above , was also suggested in order to explain some morphological features observed in the so - called giant radio galaxies ( linear sizes of the order of mpc ; * ? ? ? in a few sources belonging to this class , intermittent jet activity was confirmed directly by observations of double - double radio structure @xcite . in particular , observations of b1834 + 620 @xcite , j0116 - 473 @xcite , and b1545 - 321 @xcite show a pair of ` inner ' radio jets which propagate within a diffuse radio - emitting medium formed by previous jet activity , and which display fr ii radio morphology . such inner radio structures are characterized by lower radio luminosity , larger axis ratios of the lobes , and weaker hot - spots than classical doubles with similar sizes , in agreement with theoretical predictions for restarting bipolar jets @xcite . a restarting jet propagates not within the unperturbed intergalactic or intracluster medium of the active galaxy , but within the interior of the radio lobe created by the previous jet activity ( e.g. , * ? ? ? the number of double - double radio sources is small , probably because this kind of morphology can be formed only if the new pair of jets is created not earlier than @xmath73 yrs after the old ones , in order to allow for an increase in the density of the cold gas in the relict radio cocoon @xcite . otherwise , the rarefied and mostly nonthermal medium formed during the previous period of jet activity prevents formation of new lobes and hot - spots yrs , because after this time the old lobe structure is not expected to remain visible @xcite . ] . for the quasar jets considered here , we require the time scale between the two subsequent jet active epochs to be not longer than @xmath74 yrs . this is because the linear sizes of the knot regions are consistent with a @xmath75-year duration of jet activity , while distances between subsequent knots are on average consistent with quiescent periods an order of magnitude longer . therefore , we characterize the quasar jets in terms of modulation of the jet kinetic power , rather than in terms of restarting flows . this idea is strongly supported by radio observations of such structures ( see , e.g. , * ? ? ? * ) . the respective time scales for activity and quiescence can differ from object to object , but on average they seem to be in agreement with the ones postulated above . in some sources , however , they can be smaller than average , resulting in a more continuous flow ( as in 3c 273 ? ) . we also note that galactic x - ray transients are known to produce radio emission which is widely modeled in terms of multiple jet - like ejection events ( see review by * ? ? ? modulated jet activity of the large - scale jets considered here constitutes an interesting link between quasar and microquasar phenomena , although the physical processes controlling intermittent / highly modulated activity of the central engines in both types of sources can be different . as a result , the time scales for the active and quiescent periods in galactic and extragalactic jet sources do not have to scale in a simple way with the mass of the central compact object . the moving portions of the jet have to be sites of particle acceleration . @xcite rejected this possibility , arguing that either continuous or episodic particle acceleration acting within moving knots can not reproduce the required electron energy distribution . however , as the synchrotron spectrum at frequencies higher than the radio band is unknown in most large - scale quasar jets , such a statement is premature ( see 2.3 ) . note that @xcite base their conclusions on the _ assumed _ electron energy distribution . in reality , many different processes can control the spectral behavior of radiating electrons , and it is not certain from both observational and theoretical points of view how well a single power - law describes the spectrum . in the framework of the intermittent jet activity model considered in this section , turbulent particle acceleration processes due to interaction of the jet with the surrounding medium are expected to lead to the formation of a piled - up distribution of energetic electrons radiating synchrotron x - rays . the formation of internal shocks within the relativistic , highly modulated jet flow is also very probable . in the latter case , one can expect the eic process to dominate production of the x - rays , whereas the deominant radiation mechanism is less clear in the former . realistically , the situation can be quite complex , involving an ensemble of multiple shocks ( formed due to both velocity irregularities and propagation instabilities of the flow ) as well as the interplay between coherent and stochastic particle acceleration processes . the formation of shocks in extragalactic jets is most likely related to intrinsic velocity irregularities of the flow , caused by the non - steady operation of the central engine , as proposed by @xcite . such internal shocks are not stationary , but instead are moving together with the radiating matter , possibly with highly relativistic bulk velocities . internal shocks are typically characterized by a low efficiency of converting the beam kinetic energy to internal energy of radiating particles ( see , e.g. , * ? ? ? * ; * ? ? ? this is a very convenient situation for the objects considered here , because their observed radiative luminosities , if corrected for doppler beaming effects , are several orders of magnitude smaller than the estimated jet kinetic powers ( @xmath76 erg s@xmath1 ) , and hence the fraction of the kinetic energy flux of the jet dissipated to the radiating particles can be small . let us consider the simplest situation , in which two colliding portions of the jet flow differ only in bulk velocity . then , for respective bulk lorentz factors @xmath77 , a double shock structure ( symmetric in the comoving frame of the contact discontinuity for the same densities @xmath78 ) develops . the bulk lorentz factor of the contact discontinuity , which is also the lorentz factor of the shocked radiating plasma , is then @xmath79 knowing solely @xmath80 and @xmath81 ( and assuming an equation of state for the downstream matter ) , one can estimate velocities of the forward and reverse shocks from standard shock - jump conditions ( * ? ? ? * appendix e ) . the extent of the shocked region along the jet , in the contact comoving frame denoted by primes , is @xmath82 where @xmath83 is the velocity of the forward and reverse shock fronts and @xmath84 is the time since formation of the double - shock structure . observations of the knots at the maximum synchrotron frequency @xmath85 can give some constraints on the maximum value for @xmath86 . this is because the time interval @xmath84 can not be larger than the radiative loss time scale for the most energetic electrons radiating at @xmath85 , in order to ensure activity of the whole shocked region at this frequency . therefore , for a small viewing angle of the jet , the observed extent of the shocked region can be limited according to @xmath87 where @xmath88 is the radiative loss time scale for a given synchrotron frequency @xmath85 ( appendix d ) . figures 13 show the parameter @xmath89 as a function of the magnetic field strength in the radiating plasma , for @xmath90 hz and different bulk lorentz factors @xmath80 and @xmath81 . note that , for a wide range of bulk lorentz factors of the colliding portions of the jet , the extent of the double - shock structure in the observer frame can not be larger than @xmath91 kpc , independent of the magnetic field intensity . this value , interestingly , is consistent with the observed lengths of the optical knots in quasar jets . therefore , if dynamical time scales controlling the development of the double - shock structure ensure that @xmath92 for the maximum synchrotron frequency occurring in the optical band , then the eic / internal shock scenario can explain frequency - independent knot profiles . however , there are two important problems with the model discussed above . the first is that radio polarization of the quasar jets indicates a magnetic field which is on average parallel to the jet axis . as an example we mention the jet in pks 0637 - 752 , where the x - ray emission as observed by cxo is strong only in the regions characterized by a longitudinal magnetic field such a situation is inconsistent with the internal shock model , because compression of the magnetic field should result in a transverse orientation ( laing 1980 ) . the second problem is the fine - tuning required between the dynamical time scales controlling development of the double - shock structure , and the radiative loss time scales of the radiating particles , in order to obtain almost universal extents of the optical / x - ray knots , @xmath91 kpc . we now discuss particle acceleration processes related to ensembles of shocks which are not comoving with the radiating matter , i.e. , have a different pattern velocity than the jet . numerical simulations reveal the presence of such a structure at the jet boundary ( e.g. , * ? ? ? * ) , although it is possible for these shocks to encompass the whole jet width . the particle acceleration process related to the ensemble of many shocks can be different from single shock acceleration . this issue was considered in the literature in different contexts starting from @xcite , and discussed further by @xcite and @xcite . physically , what distinguishes between single and multiple shock acceleration is that in the latter case , particles can change their momenta between subsequent shock events due to adiabatic and radiative losses , and can also escape from the regions between of the shocks , thus avoiding further acceleration . as shown by @xcite and @xcite , the interplay between time scales representing radiative cooling , particle escape and advection can result in the formation of different particle energy spectra , with spectral flattenings , curvatures , cut - offs or high - energy bumps ( spectral pile - ups ) . let us consider the appropriate time scales for multiple shock acceleration in the context of the knots in quasar jets . following @xcite , we assume that the time scale for acceleration and subsequent decompression at each shock is much shorter than the time scales connected with radiative losses , @xmath57 , and particle escape from the region between subsequent shocks ( i.e. , the region where the nonthermal radiation is produced ) , @xmath93 , as well as the interval between the subsequent shock acceleration events , @xmath94 . then , the average particle energy distribution resulting from multiple ( roughly identical ) shock acceleration events depends on the ratios of these three characteristic time scales . note first , that if particle escape is controlled by diffusion from the region between the shocks , which is characterized by a thickness @xmath95 , then the condition @xmath96 reads @xmath97 for strong turbulence , where the particle mean free path is comparable to the particle gyroradius , and for inverse - compton losses comparable to synchrotron losses ( appendix f ) . thus , if @xmath95 is of order of the jet radius ( @xmath91 kpc ) , then particle escape is inefficient compared to the radiative losses , independent of particle energy . also , if the mean separation of the shocks is less than @xmath91 kpc , particles can not escape before passing through many shocks , and therefore they experience multiple ( and not single ) shock acceleration . in such a case , as shown by the analysis of @xcite , one should expect the formation of a flat power - law spectrum for particles satisfying @xmath98 , a steep power - law spectrum ( but not necessarily an exponential cut - off ) for particles satisfying @xmath99 , and a spectral pile - up at particle energies for which @xmath100 . now consider what the shock separation has to be in order to obtain a piled - up distribution of particles emitting the observed synchrotron kev photons . if the radiating plasma between the shocks is highly relativistic ( with the respective bulk lorentz factor @xmath101 ) , then the condition @xmath100 reads @xmath102 where @xmath103 is the shock separation as measured in the radiating plasma rest frame ( appendices d and f ) . the parameter @xmath103 as a function of @xmath9 , for the observed critical synchrotron photon energy @xmath104 kev and different magnetic field intensities , is shown on figure 4 . for @xmath105 and @xmath106 g one gets @xmath107 pc . for larger shock separations , pile - ups are expected to occur at lower energies of the synchrotron photons . however , these values should be considered as illustrative only , because the analysis above refers to a very simplified situation , in which we neglect the effects of turbulent second - order fermi acceleration and assume identical properties for all the shocks . for reasonable large - scale jet parameters , however , an ensemble of shock waves present within the extended knot region can result in the formation of a flat power - law electron energy distribution followed by a high - energy pile - up announced by its x - ray synchrotron emission . this situation is similar to the model discussed by @xcite , but here the time scale between subsequent shock events , and not the second - order fermi acceleration time scale , @xmath108 , determines the critical electron energy . the ratio of these two time scales , for @xmath109 and electron lorentz factor @xmath69 , can be rewritten as @xmath110 where @xmath111 is the alfven speed in the radiating plasma ( appendix f ) . therefore , we conclude that an ensemble of oblique shock waves within the knot regions of large - scale quasar jets can act in the analogous way to second - order fermi acceleration , as discussed by @xcite . both processes , which in a realistic situation can be hard to separate , are due to the interaction of the propagating jet with the surrounding medium . both of them can result ( for reasonable jet parameters ) in a pile - up of high - energy electrons radiating x - ray synchrotron photons , and both are consistent with the observed longitudinal magnetic field . inverse - compton scattering of the cmb radiation field within a highly relativistic flow is commonly believed to explain relatively strong x - ray emission detected by cxo from large - scale quasar jets . this model , as initially proposed , is attractive because of its simplicity and energetic efficiency . on the other hand , it is inconsistent with morphological properties of these objects , namely with the frequency - independent knot profiles , _ if _ the jet flow is continuous and the knots are stationary ( e.g. , if they are manifestations of the reconfinement shocks through which the jet matter is flowing at a constant rate ) . it was proposed @xcite that the inconsistency can be resolved by assuming that the radiating jet region consists of small , adiabaticaly expanding clumps . however , as discussed in 2.1 , by invoking an inhomogeneous structure of the emitting regions , one loses the advantages of the eic model over other models proposed in the literature . in particular , when clumps are incorporated into the eic model , emitting regions generally have to be far from the minimum power condition . in this situation , it becomes much more difficult to arrange for the radiating particles to be confined to the required volumes . also , knot profiles will be observed as almost frequency - independent only if the radiating regions expand very rapidly , starting from very compact clumps . such a situation leads to the problem that contributions from the ssc process can not be excluded anymore . furthermore , fine - tuning of the parameters is required in order not to overproduce optical emission by the eic process at the final stage of the clump evolution . one should emphasize that , if the x - rays from large - scale quasar jets modeled as continuous flows with localized and stationary particle acceleration sites are indeed inverse - compton in origin , then clumping of the knot regions is unavoidable . in the case of the eic scenario this is because strong adiabatic losses are required in order to obtain knots instead of uniform and continuous emission along the jet , while in the case of the ssc process it is because the energy density of synchrotron photons within the emitting regions is required to be larger than in the homogeneous case in order to provide the observed level of x - ray emission . therefore , we consider the clumpy jet model ( as proposed by @xcite ) as unlikely . we note that complex substructure of the knots is suggested by the time variability of the kpc - scale jet in the fr i radio galaxy m 87 ( e.g. , * ? ? ? in fr i sources , the x - ray emission observed by cxo is most probably synchrotron in origin . also , detailed studies of the centaurus a jet radio and x - ray radiation seem to indicate a spine - boundary layer morphology @xcite . this can be taken as an indication that dissipation processes acting at the jet boundary can be crucial for understanding the high - energy radiative output of extragalactic large - scale jets , and their x - ray emission in particular @xcite . we stress that the nature of the quasar jet clumps as localized small sites of particle acceleration is unknown . it was proposed @xcite that the clumpy structure can form when the jet crosses a region filled with small clouds of ambient medium , or that the clumps can be related to magnetic reconnection events . regardless of the specific mechanism of x - ray production , the morphological properties of quasar jets ( inferred from multiwavelength observations ) can be understood if the knots are not stationary regions of energy dissipation within the continuous jet flow , but instead are moving , separate portions of the jet matter . in 3.1 we outlined the possible connection between this idea and highly modulated jet activity . in this case , a variety of particle acceleration processes ( including reconnection , acting in a way analogous to second - order fermi acceleration ) can play a role . unfortunately , optical observations , which could be crucial in discriminating among models , can not give definitive answers at the moment . therefore , detailed spectral and morphological studies are still needed . let us mention in this context one especially interesting object , which could be a very promising target for such multiwavelength studies : the large - scale jet associated with the seyfert galaxy 3c 120 . in this object , intermittent jet activity seems to take place , while the x - ray emission of the knot seems to be exceptionally inconsistent with the eic model . 3c 120 ( @xmath112 ) displays large - scale radio structure , with a one - sided bending jet ( observed up to roughly @xmath113 kpc from the core ) and diffuse , amorphous radio emission extending in several directions up to @xmath114 kpc @xcite . vlbi observations reveal complex jet structure with superluminal motions of some jet components at the @xmath115 pc scale ( e.g. , * ? ? ? this , together with the large bending of the flow at kpc - scales , indicate a small jet inclination to the line of sight . at @xmath116 kpc from the host galaxy , an elongated radio knot , oriented perpendicular to the flow axis , is observed . long - term vla monitoring @xcite indicates that this knot is stationary ( or moving only subluminally ) . farther away the jet bends and fades , but is manifested again at @xmath117 from the core as a bright , extended radio knot , oriented obliquely with respect to the jet path . the radio spectral index @xmath118 remains constant along the flow , although brightness variations are large . optical observations reveal polarized emission aligned with some parts of the radio jet in the first few kpc from the core , but with morphology not corresponding directly to the radio one @xcite . surprisingly , the @xmath119 knot turn out to possess an x - ray counterpart with a @xmath120 kev luminosity @xmath121 erg s@xmath1 , as shown by rosat observations @xcite . optical upper limits ( placed below the interpolated radio - to - x - ray continuum ) allow one to estimate the optical - to - x - ray power - law slope for this region as @xmath122 . the eic model applied to the x - ray knot of the 3c 120 jet gives an uncomfortably small viewing angle of the jet , @xmath123 , and a very high bulk lorentz factor @xmath124 @xcite . note that , in such a case , the deprojected distance of the x - ray knot from the host galaxy is on the order of mpc . on the other hand , radio observations together with optical upper limits exclude the possibility that the radio - to - x - ray continuum results from the synchrotron emission of a single power - law electron energy distribution . also , if the x - ray emission is synchrotron radiation , for example due to an additional electron population as suggested by @xcite , the flat spectral index and large extent of the knot call for continuous particle energization . we suggest that the x - ray emission of the knot in the 3c 120 jet is indeed synchrotron in origin , and is a manifestation of the pile - up of high - energy electrons by some particle acceleration process acting continuously within the radiating region . this region can be further identified with a separate portion of the jet matter , possibly moving with relativistic velocities ( but not as high as required by the eic model ) , ejected from the active nucleus some @xmath74 yrs ago . this is consistent with the morphology of the 3c 120 extended radio lobes , which also indicate intermittent jet activity . we propose that the knots observed in large - scale quasar jets in radio , optical and x - rays represent portions of the jet with an excess of kinetic power due to intermittent high activity of the jet engine . the appropriate time scales for jet activity and quiescent epochs ( @xmath71 yr and @xmath125 yr , respectively ) are roughly consistent with the ones estimated in different contexts for powerful radio sources @xcite . the proposed scenario can explain some morphological properties of quasar jets , independently of the exact emission process responsible for production of x - rays . indeed , the dominant emission process can not be designated for certain yet . it is possible that the x - ray emission of quasar jets is due to inverse - compton scattering of the cmb by low - energy electrons accelerated in internal shocks . on the other hand , it is also possible that the observed kev photons are synchrotron in origin , being a manifestation of high - energy electrons piled - up at the critical energy @xmath126 tev due to the second - order fermi process and/or multiple shock acceleration . the fact that an exceptionally high doppler factor is required to explain the 3c 120 x - ray data in the framework of the eic scenario , suggests that synchrotron emission is likely to be the dominant process in this object , at least . whatever emission mechanism prevails in most large - scale quasar jets , intermittent jet activity assures , in a natural way , high knot - to - interknot brightness contrasts , frequency - independent knot profiles , and almost universal extents of the knots ( @xmath91 kpc ) . s , ms and mo were supported by the grant pbz - kbn-054/p03/2001 . s acknowledges also financial support from fundacja astronomii polskiej . mcb acknowledges support from national science foundation grant ast-0307502 . a part of this work was done during the stay of s and ms at jila , university of colorado . the total power of a cylindrical relativistic jet with bulk lorentz factor @xmath127 and radius @xmath128 , consisting of magnetic field with comoving energy density @xmath129 , radiating ultrarelativistic electrons with the comoving energy density @xmath130 and protons with @xmath131 , is @xmath132 where the poynting flux is @xmath133 and the power carried by electrons and protons is , respectively @xmath134 for a given electron energy distribution ( i.e. , the respective moments @xmath135 and @xmath136 ) , the electron energy density can be related to synchrotron luminosity of some radiating volume @xmath137 by @xmath138 hence @xmath139 for a radiating region modeled either as a moving or steady source , one has the transformation to the observer frame @xmath140 where the doppler factor of the jet emission is @xmath141 and @xmath52 is the jet viewing angle ( see , e.g. , * ? ? ? note that , for small jet viewing angles ( @xmath59 ) , the poynting flux and the bulk power of radiating electrons do not behave in opposite ways with respect to @xmath9 and @xmath4 , because @xmath142 while @xmath143 ( cf . * ) . setting @xmath144 one can find that the magnetic field @xmath19 minimizing @xmath20 is the one for which @xmath21 . this value is given by @xmath145 let us next consider a clumpy jet , consisting of @xmath47 identical clumps of radiating matter with comoving volumes @xmath146 , which are present within the region of volume @xmath137 . the clump filling factor can be defined as @xmath147 assuming that the jet velocity is uniform within the jet region under consideration , the poynting flux for the radiating matter is then @xmath148 this follows from the fact that , given the magnetic energy of each clump @xmath149 , the total magnetic energy of the radiating matter flowing through the jet surface @xmath150 is @xmath151 . similarly , the bulk kinetic powers of the emitting electrons and protons ( for an assumed @xmath23 the same as in the homogeneous knot model ) are , respectively , @xmath152 however , one has also @xmath153 where @xmath154 is the intrinsic synchrotron luminosity of each clump . as a result , the bulk kinetic energy of the particles computed for the radiating regions of the clumpy jet is the same as that computed for the homogeneous knot model . hence , the magnetic field minimizing the total power of the radiating clumpy matter is @xmath155 adiabatic losses of ultrarelativistic electrons with lorentz factor @xmath5 , within an expanding region of linear size @xmath36 , obey @xmath156 ( where @xmath157 [ 2/3 ] for three - dimensional [ two - dimensional ] expansion ) , and lead to the evolution of the electron distribution according to @xmath158 here @xmath159 , @xmath64 is the electron energy distribution , @xmath160 is the number of electrons injected impulsively at @xmath161 with initial energies @xmath162 , and @xmath163 ( see , e.g. , * ? ? ? the inverse - compton luminosity due to electrons with given energy @xmath164 is @xmath165_{\rm eic } \propto n'_{\rm e}(\gamma_{\rm eic } ) \ , v ' = n'(\gamma_{\rm eic } , \ , r ) .\ ] ] therefore , the adiabatic decrease of the eic luminosity at a given frequency @xmath166 is @xmath167_{\rm eic } \over [ \nu l_{\nu}(r_0)]_{\rm eic } } = { n'(\gamma_{\rm eic } , \ , r ) \over n'_0(\gamma_{\rm eic } ) } = \left({r \over r_0}\right)^{-a \ , ( s-1 ) } .\ ] ] the ratio of the observed ssc luminosity to the observed eic luminosity is @xmath168 ( e.g. , * ? ? ? the comoving energy density of the cmb radiation at redshift @xmath49 is @xmath169 where @xmath170 erg @xmath171 . the comoving energy density of the synchrotron emission in the clumpy jet , for clumps with radius @xmath48 , is @xmath172 the intrinsic synchrotron luminosity of each clump , @xmath173 , is related to the observed synchrotron luminosity of the whole knot region , @xmath6 , by @xmath174 where @xmath47 is a number of identical clumps . hence , one finally obtains @xmath175 the time scale of radiative losses via synchrotron and inverse - compton processes , where the latter is dominated by comptonization of the cmb , as measured in the comoving frame of the radiating plasma is @xmath176 where @xmath177 ( see appendix c ) . the electron lorentz factor @xmath5 is related to the observed frequency of the synchrotron emission , @xmath178 , by @xmath179 and therefore , for a given observed synchrotron frequency @xmath55 , one obtains @xmath180 due to inverse - compton losses , for a given @xmath55 the propagation length of the synchrotron - emitting electrons as measured in the observer rest frame , @xmath181 , can not be arbitrary large , and passes through a maximum as a function of @xmath182 or @xmath4 @xcite . if the colliding portions of the jet flow differ only in bulk velocities ( @xmath183 and @xmath184 , respectively ) , one can find the bulk velocity of the contact discontinuity from the energy and momentum conservation laws : @xmath185 ( e.g. , * ? ? ? hence , for highly relativistic velocities one has the contact discontinuity bulk lorentz factor @xmath79 after the collision , a double - shock structure develops . this structure is symmetric in the comoving frame of the contact discontinuity ( denoted by primes ) , and therefore the extent of the shocked region along the jet is @xmath186 where @xmath83 is the shock velocity in the contact frame and @xmath84 is the time since the collision . in order to find the shock velocity , one has to solve the shock - jump conditions between the upstream plasma , moving with velocity @xmath183 , and the downstream plasma , moving with velocity @xmath187 . this can be done conveniently in the comoving frame of the upstream plasma ( denoted by @xmath188 ) , to obtain the lorentz factor of the shock @xmath189 ^ 2 \over \hat{\gamma}_{\rm ct } \ , ( 2 - \hat{\gamma}_{\rm ct } ) \ , ( \gamma_{\rm ct}^{(1 ) } - 1 ) + 2}}\ ] ] @xcite . the constant @xmath190 can be identified with the ratio of specific heats of the upstream matter ( we assume @xmath191 ) . next , the shock velocity @xmath192 in the contact frame can be found by transforming @xmath193 to the downstream plasma frame . for large @xmath80 and @xmath81 , @xmath83 is expected to be subrelativistic . the shocked region is a moving radiating portion of the jet flow , and therefore its deprojected observed length is @xmath194 with the requirement @xmath195 , where @xmath57 is the lifetime of the radiatively cooling electrons in the downstream plasma moving with bulk lorentz factor @xmath196 ( appendix d ) , one can write @xmath197 to obtain for small jet inclinations ( i.e. , for @xmath198 ) @xmath199 the time scale for particles to escape from the region of the shock ensemble can be estimated as @xmath200 where @xmath95 is the thickness of the considered region and @xmath201 is the diffusion coefficient of the electrons with mean free path @xmath202 and lorentz factor @xmath5 . assuming that @xmath202 is comparable to the electron gyroradius , @xmath203 one obtains @xmath204 where the radiative loss time scale @xmath57 is given in appendix d. the time scale between subsequent shocks can be estimated as @xmath205 where @xmath103 is the mean separation of the shocks in the comoving frame of the radiating plasma between the shocks ( moving with bulk lorentz factor @xmath101 ) . second - order fermi acceleration can be characterized by the time scale @xmath206 where the velocity of the scattering centers involved in the turbulent acceleration @xmath207 can be identified with the alfven speed , @xmath208 , and the particle gyroradius can be taken for @xmath202 @xcite . hence , one obtains the ratio @xmath209 where @xmath210 . achterberg , a. 1990 , , 231 , 251 aharonian , f. a. 2002 , , 332 , 215 atoyan , a. m. , & aharonian , f. a. 1999 , , 302 , 253 baum , s. a. , odea , c. p. , de bruyn , a. g. , & murphy , d. w. 1990 , , 232 , 19 begelman , m. c. , & cioffi , d. f. 1989 , , 345 , l21 begelman , m. c. , & kirk , j. g. 1990 , , 353 , 66 blandford , r. d. , & mckee , c. f. 1976 , phfl , 19 , 1130 blandford , r. d. , & ostriker , j. p. 1980 , , 237 , 793 bridle , a. h. , perley , r. a. , & henriksen , r. n. 1986 , , 92 , 534 bridle , a. h. , fomalont , e. b. , byrd , g. g. , & valtonen , m. j. 1989 , , 97 , 674 bridle , a. h. , hough , d , h. , lonsdale , c , j. , burns , j , o. , & laing , r , a. 1994 , , 108 , 766 celotti , a. , ghisellini , g. , & chiaberge , m. 2001 , , 321 , l1 chartas , g. , worrall , d. m. , birkinshaw , m. , creistello - dittmar , m. , cui , w. , ghosh , k. k. , harris , d. e. , hooper , d. j. , jauncey , d. l. , kim , d. w. , lovell , j. , marshall . , h. l. , mathur , s. , schwartz , d. a. , tingay , s. j. , virani , s. n. , & wilkes , b. j. 2000 , , 542 , 655 cheung , c. c. , 2002 , , 581 , l15 clarke , d. a. , & burns , j. o. 1991 , , 369 , 308 clarke , d. a. , bridle , a. h. , burns , j. o. , perley , r. a. , & norman , m. l. 1992 , , 285 , 173 dermer , c.d . , & atoyan , a.m. 2002 , , 568 , 81 fender , r. 2003 , in _ stellar x - ray sources _ , eds . w.h.g . lewin and m. van der klis , cambridge university press ( astro - ph/0303339 ) ghisellini , g. , & celotti , a. 2001 , , 327 , 739 gieseler , u. d. j. , & jones , t. w. 2000 , , 357 , 1133 hardcastle , m. j. , worrall , d. m. , kraft , r. p. , forman , w. r. , jones , c. , & murray , s. s. 2003 , , 593 , 169 harris , d. e. , & krawczynski , h. 2002 , , 565 , 244 harris , d. e. , biretta , j. a. , & junor , w. 1997 , , 284 , 21 harris , d. e. , hjorth , j. , sadun , a. c. , silverman , j. d. , & vestergaard , m. 1999 , , 518 , 213 harris , d. e. , biretta , j. a. , junor , w. , perlman , e. s. , sparks , w. b. , & wilson , a. s. 2003 , , 586 , l41 hjorth , j. , vestergaard , m. , sorensen , a. n. , & grundahl , f. 1995 , , 452 , l17 janiuk , a. , czerny , b. , siemiginowska , a. , & szczerba , r. 2003 , , accepted ( astro - ph/0311011 ) jester , s. , rser , h .- j . , meisenheimer , k. , perley , r. , & conway , r. 2001 , , 373 , 447 jester , s. , rser , h .- j . , meisenheimer , k. , & perley , r. 2002 , , 385 , l27 kaiser , c. r. , schoenmakers , a. p. , & rttgering , h. j. a. 2000 , , 315 , 381 komissarov , s. s. , & gubanov , a. g. , 1994 , , 285 , 27 komissarov , s. s. , & falle , s. a. 1997 , , 288 , 833 laing , r. a. 1980 , , 193 , 439 lazzati , d. , ghisellini , g. , & celotti , a. 1999 , , 309 , l13 marcowith , a. , & kirk , j. g. 1999 , , 347 , 391 marecki , a. , barthel , p. d. , polatidis , a. , & owsianik , i. 2003a , pasa , 20 , 16 marecki , a. , spencer , r. e. , & kunert , m. 2003b , pasa , 20 , 46 marshall , h. l. , harris , d. e. , grimes , j. p. , drake , j. j. , fruscione , a. , juda , m. , kraft , r. p. , mathur , s. , murray , s. s. , ogle , p. m. , pease , d. o. , schwartz , d. a. , siemiginowska , a. l. , vrtilek , s. d. , & wargelin , b. j. 2001 , , 549 , l167 melrose , d. b. , & pope , m. h. 1993 , pasa , 10 , 222 melrose , d. b. , & crouch , a. 1997 , pasa , 14 , 251 norman , m. l. , winkler , k .- h . a. , smarr , l. , & smith , m. d. 1982 , , 113 , 285 ostrowski , m. 2000 , , 312 , 579 pope , m. h. , & melrose , d. b. 1994 , pasa , 11 , 175 rees , m. j. 1978 , , 184 , 61 reynolds , c. s. , & begelman , m. c. 1997 , , 487 , 135 sambruna , r. m. , maraschi , l. , tavecchio , f. , urry , c. m. , cheung , c. c. , chartas , g. , scarpa , r. , & gambill , j. k. , 2002 , , 571 , 206 sanders , r. h. 1983 , , 266 , 73 saripalli , l. , subrahmanyan , r. , & udaya shankar , n. 2002 , , 565 , 256 saripalli , l. , subrahmanyan , r. , & udaya shankar , n. 2003 , , 590 , 181 schneider , p. 1993 , , 278 , 315 schoenmakers , a. p. , de bruyn , a. g. , rttgering , h. j. a. , & van der laan , h. 1999 , , 341 , 44 schoenmakers , a. p. , de bruyn , a. g. , rttgering , h. j. a. , van der laan , h. , & kaiser , c. r. 2000a , , 315 , 371 schoenmakers , a. p. , de bruyn , a. g. , rttgering , h. j. a. , & van der laan , h. 2000b , , 315 , 395 schwartz , d. a. 2002 , , 569 , l23 schwartz , d. a. , marshall , h. l. , lovell , j. e. j. , piner , b. g. , tingay , s. j. , birkinshaw , m. , chartas , g. , elvis , m. , feigelson , e. d. , ghosh , k. k. , harris , d. e. , hirabayashi , h. , hooper , e. j. , jauncey , d. l. , lanzetta , k. m. , mathur , s. , preston , r. a. , tucker , w. h. , virani , s. , wilkes , b. , & worrall , d. m. 2000 , , 540 , 69 siemiginowska , a. , & elvis , m. 1997 , apj , 482 , 9 siemiginowska , a. , bechtold , j. , aldcroft , t. l. , elvis , m. , harris , d. e. , & dobrzycki , a. 2002 , , 570 , 543 siemiginowska , a. , stanghellini , c. , brunetti , g. , fiore , f. , aldcroft , t. l. , bechtold , j. , elvis , m. , murray , s. s. , antonelli , l. a. , & colafrancesco , s. 2003a , , 595 , 643 siemiginowska , a. , smith , r. k. , aldcroft , t. l. , schwartz , d. a. , paerels , f. , & petric , a. o. 2003b , , 598 , l15 sikora , m. , madejski , g. , moderski , r. , & poutanen , j. 1997 , , 484 , 108 spada , m. , ghisellini , g. , lazzati , d. , & celotti , a. 2001 , , 325 , 1559 spruit , h. c. 1988 , , 194 , 319 stanghellini , c. , baum , s. a. , odea , c. p. , & morris , g. b. 1990 , , 233 , 379 stawarz , . 2004 , chjaa , submitted ( astro - ph/0310795 ) stawarz , . , & ostrowski , m. 2002 , , 578 , 763 stawarz , . , sikora , m. , & ostrowski , m. 2003 , , 597 , 186 subrahmanyan , r. , saripalli , l. , & hunstead , r.w . 1996 , , 279 , 257 tavecchio , f. , maraschi , l. , sambruna , r. m. , & urry , c. m. 2000 , , 544 , l23 tavecchio , f. , ghisellini , g. , & celotti , a. 2003 , , 403 , 83 walker , r. c. 1997 , , 488 , 675 walker , r. c. , benson , j. m. , & unwin , s. c. 1987 , , 316 , 546 walker , r. c. , benson , j. m. , unwin , s. c. , lystrup , m. b. , hunter , t. r. , pilbratt , g. , & hardee , p. e. 2001 , , 556 , 756 wardle , j. f. c. , & aaron , s. e. 1997 , , 286 , 425

the icecube neutrino observatory pursues a follow - up program selecting interesting neutrino events in real - time and issuing alerts for electromagnetic follow - up observations . in march 2012 , the most significant neutrino alert during the first three years of operation was issued by icecube . in the follow - up observations performed by the palomar transient factory ( ptf ) , a type iin supernova ( sn ) ptf12csy was found away from the neutrino alert direction , with an error radius of . it has a redshift of , corresponding to a luminosity distance of about and the pan - starrs1 survey shows that its explosion time was at least 158 days ( in host galaxy rest frame ) before the neutrino alert , so that a causal connection is unlikely . the _ a posteriori _ significance of the chance detection of both the neutrinos and the sn at any epoch is within icecube s 2011/12 data acquisition season . also , a complementary neutrino analysis reveals no long - term signal over the course of one year . therefore , we consider the sn detection coincidental and the neutrinos uncorrelated to the sn . however , the sn is unusual and interesting by itself : it is luminous and energetic , bearing strong resemblance to the sn iin 2010jl , and shows signs of interaction of the sn ejecta with a dense circumstellar medium . high - energy neutrino emission is expected in models of diffusive shock acceleration , but at a low , non - detectable level for this specific sn . in this paper , we describe the sn ptf12csy and present both the neutrino and electromagnetic data , as well as their analysis .

1506.03115

icecube is a cubic - kilometer - sized neutrino detector installed in the ice at the geographic south pole between depths of 1450 m and 2450 m @xcite . it consists of an array of photon sensors , called digital optical modules ( doms ) , attached to 86 cables , called strings . detector construction started in 2005 and finished in december 2010 . neutrino observation relies on the optical detection of cherenkov radiation emitted by secondary particles produced in neutrino interactions in ice or bedrock near icecube . due to the small neutrino interaction cross - section , the kilometer - scale detector has an effective area of only for muon neutrinos of energy @xcite . as part of the optical follow - up ( ofu ) program , the icecube neutrino observatory records high - energy ( @xmath1 ) neutrino events at a rate of about @xmath2 , about 250 per day . those events are mostly ( @xmath3 ) cosmic - ray induced neutrinos from the atmosphere , referred to as atmospheric neutrinos , with about a 10% contamination of cosmic - ray induced atmospheric muons . routine neutrino analyses in icecube , which are referred to as _ offline _ analyses , are performed after a certain amount of data , e.g. one or several years , has been collected . they benefit from events being put through computationally expensive reconstructions , as well as information on detector performance that become available only days or weeks after data acquisition . in contrast , neutrino analyses running _ online _ will not have access to such information , but have the advantage of being near real - time results are available with a latency of @xmath4 . with such a short latency neutrino analysis , multi - wavelength follow - up observations can be triggered by neutrino events . these follow - up data have the potential to reveal the electromagnetic counterpart of a transient neutrino source , which might otherwise be missed and thus be unavailable for further observations . in addition , the coincident detection of neutrino and electromagnetic emission can be statistically more significant and provide more information about the physics of the source than the neutrino detection alone . another advantage of an online analysis is the prompt availability of the reconstructed neutrino dataset and thus the possibility of fast response analyses . thus , icecube s online neutrino analysis efforts have also enabled fast @xmath5-ray burst ( grb ) searches like the one following grb 130427a , published in a gcn circular @xcite . the online search for short transient neutrino sources ( order of @xmath6 ) is mostly motivated by models of neutrinos from long duration grbs @xcite and from choked jet supernovae ( sne ) @xcite . the two source classes are related : both are thought to host a jet , which is highly relativistic in case of long grbs , but only mildly relativistic in case of choked jet sne . long grb progenitors are conceived to be wolf - rayet stars that have lost their outer hydrogen and helium envelope @xcite , while choked jet sne can still have those outer layers @xcite which are important for efficient high - energy ( he ) neutrino production . the choked jet is more baryon - rich and has a much lower lorentz factor @xmath7 than the grb jet with @xmath8 . it can not penetrate the stellar envelope and remains optically thick , making it invisible in @xmath5-rays . the neutrinos produced at energies can escape nevertheless and may trigger the discovery of the sn in other channels . in particular , for a bright nearby sn , a neutrino detection would enable the acquisition of optical data during the rise of the light curve , strengthening the time correlation between the neutrino burst and the optical supernova via the improved estimate of the explosion time @xcite . mildly relativistic jets may occur in a much larger fraction of core - collapse sne than highly relativistic jets , i.e. grbs @xcite . detections of neutrinos from grbs and choked jet sne would be a remarkable discovery and could provide important insight into the supernova - grb connection and the underlying jet physics @xcite . both sources are expected to emit a short , about long burst of neutrinos @xcite either before or at the time of the @xmath5-ray burst ( if detectable ) @xcite , setting the natural time scale of the neutrino search . after recording the neutrino burst , follow - up observations can be used to identify the counterpart of the transient neutrino source . a @xmath5-ray burst can be identified either via the prompt @xmath5-ray emission lasting up to about @xmath9 @xcite or via the optical and x - ray afterglow lasting up to several hours @xcite . the latter involves instruments of limited field of view and thus requires telescope slew , but has the advantage of much better angular resolution well below an arc minute @xcite . a fast response within minutes to hours is required for a grb afterglow follow - up . a choked jet sn is found by detecting a shock breakout or a sn light curve in the follow - up images , slowly rising and then declining within weeks after the neutrino burst . following this scientific motivation , an online neutrino analysis , targeted at sn and grb afterglow detection , was installed at icecube in 2008 ( see section [ sec : ofu ] ) . in addition to the transient neutrino emission within , as discussed above , sne can be promising sources of high - energy neutrinos over longer time scales . in this paper , the class of type iin sne @xcite is explored further . these are core - collapse sne embedded in a dense circumstellar medium ( csm ) that was ejected in a pre - explosion phase . following the explosion , the sn ejecta plow through the dense csm and collisionless shocks can form and accelerate particles , which may create high - energy neutrinos . this is comparable to a sn remnant , but on a much shorter time scale of @xcite . sne iin ( `` n '' for narrow ) are spectrally characterized by the presence of strong emission lines , most notably h@xmath10 , that have a narrow component , together with blue continuum emission @xcite . the narrow component is interpreted to originate from surrounding hii regions , and the generally slow spectral evolution is due to the presence of a high - density csm @xcite . since interaction of the sn ejecta with the dense csm can lead to the conversion of a large fraction of the ejecta s kinetic energy to radiation , sne iin are on average more luminous than other sne ii @xcite . they generally fade quite slowly and some belong to the most luminous sne @xcite . there is much diversity within the subclass of sne iin @xcite , both spectroscopically and photometrically , which can be explained by a diversity of progenitor stars and mass loss histories prior to explosion @xcite . recently , there have been observations of eruptions prior to sn iin explosions associated with mass loss which explain the existence of the dense csm shells ( e.g. * ? ? ? * ) . in this paper , we report the discovery of a type iin sn in the optical observations triggered by an icecube neutrino alert from march 2012 , and analyze the available neutrino and electromagnetic observations . the sn is already at a late stage at the time of the neutrino detection , which means that the neutrino - sn connection is presumably coincidental . the paper is organized as follows : section [ sec : ofu ] introduces the optical follow - up system of icecube . section [ sec : alert ] gives details about the neutrino alert triggering the follow - up observations that led to the sn discovery . section [ sec : highenergy ] reports limits from a complementary offline neutrino search and x - ray limits . the uv and optical data that were obtained are discussed in depth in section [ sec : lowenergy ] . we finally summarize the results and give a conclusion in section [ sec : conclusion ] . in late 2008 , an online neutrino event selection was set up at icecube , looking for muons produced by charged current interactions of muon neutrinos in or near the icecube detector . the analysis is running in real - time within the limited computing resources at the south pole , capable of reconstructing and filtering the neutrinos and sending alerts to follow - up instruments with a latency of only a few minutes @xcite . the optical ( ofu ) and x - ray ( xfu ) real - time follow - up programs currently encompass three follow - up instruments : the robotic optical transient search experiment ( rotse ) @xcite , the palomar transient factory ( ptf ) @xcite and the swift satellite @xcite . these triggered observations were supplemented with a retrospective search through the pan - starrs1 3@xmath11 survey data @xcite , which is discussed further in section [ sec : lowenergy ] . in addition , there is also a real - time @xmath5-ray follow - up program ( gfu ) targeting slower transients ( time scale of weeks ) , e.g. flaring active galactic nuclei , that is sending alerts to the @xmath5-ray telescopes magic and veritas @xcite . the background of cosmic - ray induced muons from the atmosphere above the detector amounts to @xmath12 muon events per neutrino event . in a first step , it is reduced by limiting the sample to the northern hemisphere , using the earth as a muon shield and selecting only muon tracks that are reconstructed as up - going in the detector . afterwards , cuts on quality parameters similar to those in @xcite are applied to reject mis - reconstructed muon events : after an initial removal of likely noise signals , a chain of reconstructions is performed , which utilizes the spatial and temporal distribution of recorded photons on the doms . starting with a simple linear track algorithm , more advanced reconstructions are employed , seeded with the respective preceding reconstruction . the advanced reconstructions maximize a likelihood that accounts for the optical properties of the ice for photon propagation @xcite . the final reconstruction in this chain is used to select events that are up - going in the detector . at this point , because of the vast amount of atmospheric muons , the data are still dominated by ( down - going ) muons that are mis - reconstructed as up - going . to reject this remaining background , high quality events are selected , where the selection parameters are derived from the value of the maximized track likelihood and from the number and geometry of recorded signals with a detection time compatible with unscattered photon propagation . alternatively , events with a large number of total recorded photon signals are selected . the resulting analysis sample has an event rate of about @xmath2 , of which @xmath3 are atmospheric muon neutrinos that have passed through the earth and @xmath13 are contamination of cosmic - ray induced muons from the atmosphere . in the analysis sample , the median angular resolution of the neutrino direction is about for a multi - tev muon neutrino charged current interaction event , and or less for and higher energies . the angular resolution of the sample is estimated using monte carlo ( mc ) simulation . additionally , an estimator of the directional uncertainty is computed for each event , which is based on the shape of the reconstruction likelihood close to the found maximum @xcite . it is calibrated such that its median matches the mc derived angular resolution . we define the _ bulk _ of a sample as events within the central 90% of the energy distribution . the bulk of the main background contained in the analysis sample , atmospheric neutrinos , has energies between and . in contrast , the bulk of signal events from an unbroken @xmath14 power - law spectrum would have energies @xmath15 . signal neutrinos from grbs are expected to follow a spectrum similar to @xmath14 , which has cut - off energies between @xmath16 to @xmath17 @xcite . choked jet sne are predicted to have lower cut - off energies around @xcite so that the overlap with atmospheric neutrinos is presumably much larger , while iin neutrinos may have higher cut - offs of ( see section [ sec : offline ] ) . in order to suppress background from atmospheric neutrinos , a multiplet of at least two neutrinos within 100 seconds and angular separation of or less is required to trigger an alert . in addition , since 2011 mid - september , a test statistic is used , providing a single parameter for selection of the most significant alerts . it was derived as the analytic maximization of a likelihood ratio following @xcite , for the special case of a neutrino doublet with rich signal content : @xmath18 where the time between the neutrinos in the doublet is denoted as @xmath19 , and their angular separation as @xmath20 . the quantities @xmath21 and @xmath22 depend on the event - by - event directional uncertainties @xmath23 and @xmath24 of the two neutrino events , typically @xmath25 . the angle @xmath26 corresponds to the circularized angular radius of the field of view ( fov ) of the follow - up telescope . it is set to @xmath27 for swift and @xmath28 for rotse and ptf . the test statistic @xmath29 is smaller for more signal - like alerts , which have small separation @xmath20 , small time difference @xmath19 and a high chance to lie in the fov of the telescope . thus , @xmath29 is a useful parameter to separate signal and background alerts . for each follow - up program , a specific cut on @xmath29 is applied in order to send the most significant alerts to the follow - up instruments . in the 2011/12 data acquisition ( daq ) season , which is discussed here , a cut of @xmath30 was used for the rotse follow - up , while cut values of and were used for ptf and swift . multiplets of multiplicity higher than two are passed directly to all follow - up instruments . since the expected background rate is low ( @xmath31 per year ) , each observation of a triplet or higher order multiplet is significant by itself . * rotse * @xcite is a network of four optical telescopes with aperture and @xmath32 fov , located in australia , texas , namibia and turkey . since late 2012 , only the two northern hemisphere telescopes continue operation . rotse is a completely automatic and autonomous system that can receive alerts , perform observations and send resulting data without requiring human interaction . the limiting magnitude of @xmath33 is however insufficient to discover faint or far sne . for instance , for a very bright sn with absolute magnitude , the detection radius is about , while a faint sn with is only visible within a radius of . icecube has been sending @xmath34 alerts per year to rotse , since 2008 december . the first 116 alerts , with a background expectation of @xmath35 alerts , were followed up with a median latency of hours between the neutrino alert and start of the first follow - up observation . * ptf * @xcite is a survey based at the palomar observatory in california , usa . it utilizes the @xmath36 oschin schmidt telescope on mount palomar . the focal plane is equipped with a mosaic of 11 ccds with field of view of . the typical @xmath37-band limiting magnitude of ptf during dark time is about . all the ptf data are reduced using the lbnl real - time pipeline responsible for transient identification and the ipac pipeline described in @xcite . the image photometric calibration is described in @xcite . ptf pursues a number of science goals , most notably the discovery and observation of sne . several other telescopes in palomar and at other locations can be used for photometric and spectroscopic follow - up observation . icecube has been sending @xmath38 alerts per year to ptf , since august 2010 . the first 23 alerts , with a background expectation of @xmath39 alerts , were followed up with a median latency of hours between the neutrino alert and start of the first follow - up observation . * swift * @xcite is a satellite operated by nasa and boards various instruments : a ultraviolet / optical telescope ( uvot ) , a x - ray telescope ( xrt ) and a hard x - ray burst alert telescope ( bat ) . swift s main goal is the discovery and study of grbs , of which it detects about per year @xcite ( one third of all grbs ) . icecube s x - ray follow - up program triggers swift s xrt , which can provide valuable information by observing a grb afterglow in x - rays . the xrt has a fov of only @xmath40 in diameter , hence swift performs seven pointings for each icecube follow - up , resulting in an effective fov of about in diameter . icecube has been sending @xmath41 alerts per year to swift , since february 2011 . the first 18 alerts , with a background expectation of @xmath42 alerts , were followed up with a median latency of hours between the neutrino alert and start of the first follow - up observation ( see * ? ? ? * ) . on 2012 march 30 ( mjd 56016 ) , the most significant alert since initiation of the follow - up program ( significance of @xmath43 , converting the @xmath29 cumulative distribution function value to single - sided gaussian std . deviations ) was recorded and sent to rotse and ptf simultaneously . the significance was also above the threshold for swift ( @xmath44 ) , however it was within swift s moon proximity constraint , which delayed the observations by three weeks . the two neutrino events causing the alert happened on 2012 march 30 at 01:06:58 ut ( mjd 56016.046505 ) and 1.79 seconds later , with an angular separation of . the combined average neutrino direction is at right ascension and declination in j2000 with an error radius of @xmath45 . this average is a weighted arithmetic mean , weighting the individual directions with their inverse squared error , given by the event - by - event directional uncertainty ( s.a . ) . the error @xmath46 is defined after eq . [ eq : alert - llh ] . this assumes that the two neutrino events were emitted by a point source at a single fixed position . a variance of the individual true neutrino directions does not need to be taken into account , since the assumed intrinsic variance is zero in case of a point source . this leads to a relatively small error on the average direction . the main event properties are summarized in table [ tab : properties ] : the occurrence time on 2012 march 30 , the reconstructed muon energy proxy @xmath47 ( see * ? ? ? * ) , and the estimated directional error @xmath48 . the quantity @xmath47 is a fit parameter and serves as a proxy for the muon energy , however it is not an estimator of the true muon energy . the energy @xmath49 of the neutrino that produced the muon is not directly observable , since only the muon crossing the detector is accessible . however , using monte carlo ( mc ) simulated neutrino events , one can use the muon energy proxy @xmath47 to compare with mc events having a similar @xmath47 value . from those mc events , a distribution of the true muon energy @xmath49 can be derived , which depends on the assumed underlying neutrino energy spectrum . @xmath49 probability density functions are filled with true energy values of mc events that have the reconstructed muon energy proxy @xmath47 not more than 10% and the reconstructed zenith angle @xmath50 not more than 0.1 away from the observed values of the two alert events . the median and the central 90% c.l . interval are calculated from the @xmath49 probability density functions . the results are listed in table [ tab : properties ] , where the assumed neutrino spectrum is given in parentheses . cccccc 01:06:58 & 0.96 & & @xmath51 & @xmath52 & @xmath53 + 01:07:00 & 0.66 & & @xmath54 & @xmath55 & @xmath56 follow - up observations at the direction of the neutrino alert were performed with multiple instruments ( see section [ sec : obs ] ) . in the ptf images , a core - collapse supernova , named _ ptf12csy _ , was discovered at right ascension and declination ( j2000 ) , only away from the average neutrino direction , see figs . [ fig : map ] and [ fig : subtraction ] . this was a promising candidate for the source of the neutrinos , but a search of the pan - starrs1 archive ( see section [ sec : obs ] ) revealed that it was already at least 169 observer frame days old , i.e. 158 days in host galaxy rest frame , at the time of the neutrino alert . therefore , it is highly unlikely that the neutrinos were produced by a jet at the sn site , as this is expected to happen immediately after core collapse in the choked jet scenario @xcite . however , steady neutrino emission on a time scale of several months is a possibility and explored in section [ sec : offline ] . the value of the test statistic @xmath29 for the neutrino doublet amounts to @xmath57 . the background distribution of @xmath29 is constructed from experimental data , containing mostly atmospheric neutrinos , by randomly permuting ( shuffling ) the event times and calculating equatorial coordinates , i.e. right ascension and declination , from local coordinates , i.e. zenith and azimuth angle , using the new times . that way , all detector effects are entirely preserved , e.g. the distribution of the azimuth angle , which has more events at angles where detector strings are aligned , and the time distribution , which is affected by seasonal variations . at the same time , all potential correlations between the events in time and space , and thus a potential signal , are destroyed . the false alarm rate ( far ) for an alert with @xmath58 is , calculated via integration of the @xmath29 distribution below . considering the ofu live time of 220.1 days in the data acquisition season of the alert , september 2011 to may 2012 , yields @xmath59 false alerts . hence , the probability , or p - value , for one or more alerts at least as signal - like to happen by chance in this period is @xmath60 . the ofu system had already been sending alerts to ptf for @xmath61 days at the time of the alert . scaling up the number of expected alerts with @xmath62 , one derives a probability of @xmath63 during 460 days . the estimated explosion time of sn ptf12csy does not fall within the _ a priori _ defined time window for a neutrino - sn coincidence of @xmath64 . it is thus not considered an a priori detection of the follow - up program . despite this fact , for illustrative purposes , we calculate the _ a posteriori _ probability that a random core - collapse sn ( ccsn ) of any type , at any stage after explosion , is found coincidentally within the error radius of this neutrino doublet and within the luminosity distance of ptf12csy , i.e. @xmath65 . the number of such random sn detections is @xmath66 where @xmath67 is the solid angle of the doublet error circle ( blue circle in fig . [ fig : map ] ) , which is @xmath68 . for the volumetric ccsn rate @xmath69 , a value of @xmath70 is used , ( see * ? ? ? the control time @xmath71 is the average time window in which a sn is detectable , i.e. brighter than the limiting magnitude . it depends on the distance to the source @xmath72 , the peak absolute magnitude @xmath73 of the sn , the limiting magnitude @xmath74 of the telescope , and the shape of the light curve which is adopted from a sn template web page by p. e. nugent . it is assumed that @xmath73 follows a gaussian distribution with mean @xmath75 and standard deviation @xmath76 , based on @xcite . for ptf , @xmath77 is assumed . the resulting expectation value for coincidental sn detections is @xmath78 , which results in a poisson probability of @xmath79 to detect any ccsn within the neutrino alert s error radius . combining this probability with the probability of @xmath80 for the neutrino alert , fisher s method @xcite delivers a combined p - value of @xmath81 , corresponding to a significance of . for the total live time of 460 days , the combined p - value is @xmath82 which corresponds to a significance . this means , even ignoring the a posteriori nature of the p - value , a chance coincidence of the neutrino doublet and the sn detection can not be ruled out and thus we consider the sn detection to be coincidental . the following section reports about the available high - energy follow - up data . limits on a possible long - term neutrino emission from ptf12csy are set using one year of icecube data . limits on the x - ray flux were obtained using the swift satellite . section [ sec : lowenergy ] deals with the analysis of the low - energy optical and uv data as the sn detection is significant and interesting by itself . type iin supernovae , such as ptf12csy , are a promising class of high - energy transients ( see @xcite ) . the expected duration of neutrino emission from sne iin is months , hence it is extremely unlikely that two neutrinos arrive within less than , so late after the sn explosion . however , to test the possibility of a long - term emission , a search for neutrinos from ptf12csy within a search window of roughly one year is conducted . after the core - collapse of a sn iin , the supernova ejecta are crashing into massive circumstellar medium ( csm ) shells , producing a pair of shocks : a forward and a reverse shock . cosmic rays ( crs ) may be accelerated and multi - tev neutrinos produced , potentially detectable with icecube . the collisionless shocks generating the neutrinos are expected to generate x - rays as well at late times ( see e.g.@xcite ) , but no x - rays were detected for ptf12csy , likely because of the large distance to the sn . following @xcite and @xcite ( see also * ? ? ? * ) , we model high - energy neutrino emission from ptf12csy . as a simplified approach , in order to get an order - of - magnitude estimate of the expected event rate , we perform the following calculation : the csm density profile is calculated using ( * ? ? ? * eq . 4 , eq . the proton spectrum is modeled with a power law index , as in ( * ? ? ? * eq . a7 ) , with a cut - off energy given by the maximum energy of accelerated protons . the latter is determined by comparing the proton acceleration time scale either with the dynamical time scale ( * ? ? ? * eq . 28 ) , or , if @xmath83 energy losses are relevant , with the cooling time scale ( * ? ? ? * eq . 30 ) . the lower of the two maximum energies is the one that needs to be considered . the proton spectrum is normalized to the total cr energy @xmath84 by assuming that a fraction of the kinetic energy of the ejecta @xmath85 is converted into crs , that is @xmath86 ( compare * ? ? ? * eq . 3 , eq . 25 , using csm shell mass @xmath87sn ejecta mass @xmath88 ) . other model parameters are the break - out radius @xmath89 and shock velocity @xmath90 . the expected neutrino spectrum from @xmath83 interaction is derived from the semi - analytical description in @xcite , taking into account the meson production efficiency ( * ? ? ? * eq . 35 , eq . 36 ) . it is distance - scaled and folded with icecube s effective area from @xcite to obtain the expected number of events . it is found that inserting commonly assumed values @xcite of @xmath91 ( with the bolometric energy found in section [ sec : sed ] ) , @xmath92 , @xmath93 , @xmath94 , and @xmath95 , on average only icecube neutrino detections are expected . despite the low expectation for the neutrino fluence , we search for a long - term neutrino signal from ptf12csy in the icecube data . as a more elaborate approach compared to the simplified approximation described above , we test the neutrino emission models a and b given in fig . 1 of @xcite , which are two representative cases of cr accelerating scenarios : model a corresponds to a csm shell with a high density of at a small radius of , while model b is the opposite with a density of at radius . for the ejecta , a kinetic energy of , a velocity of , and a mass of several solar masses , lower than the csm mass , are assumed . model a is close to a scenario explaining superluminous sne iin such as sn 2006gy , while model b is a good description for dimmer , but longer lasting sne like sn 2008iy . both models have a neutrino energy spectrum close to @xmath96 , with a cut - off energy around for model a , around for the forward shock ( fs ) in model b , and around for the reverse shock ( rs ) in model b. in model a , only the reverse shock is of importance for cr acceleration . the suggested emission time scales are @xmath97 for model a and @xmath98 for model b @xcite . figure [ fig : limits ] shows the model fluences scaled down to a luminosity distance of . we analyze about one year of icecube data : the entire icecube 86 strings data acquisition season 2011/12 , from 2011 may 13 to 2012 may 15 . the long search window is motivated by the large uncertainty on the explosion date ( between 2011 march 21 and 2011 october 13 ) as well as the long duration of neutrino emission for some scenarios like @xmath99 days for model b in @xcite . for simplicity , we assume that the entire fluence was emitted during the search window . we use the neutrino sample of the icecube optical follow - up system and perform a statistical point source analysis of neutrino events close to the position of the supernova , based on @xcite . each neutrino candidate event @xmath100 is given both a signal and background probability @xmath101 and @xmath102 which are combined in the likelihood function @xmath103 the variable @xmath104 is the number of signal events contained in the sample , which is fitted to maximize the likelihood , and @xmath105 is the total number of selected neutrino candidate events . the signal probability @xmath101 is the product of the spatial gaussian probability density function ( pdf ) and the energy pdf @xmath106 , i.e. the probability of a signal event having reconstructed energy @xmath107 given the neutrino spectrum @xmath108 of the source ( derived from monte carlo simulation ) : @xmath109 here , @xmath110 is the event s angular error estimate , @xmath111 is the event s reconstructed direction , and @xmath112 the sn or neutrino source position . the background probability @xmath102 contains the energy pdf and a normalization constant for the background from atmospheric neutrinos . we define the test statistic as likelihood ratio @xcite @xmath113 , \label{eq : teststat}\ ] ] serving as a powerful test for separating the null hypothesis from the hypothesis of signal event contribution . here , @xmath114 is the number of signal events that maximizes the likelihood and corresponds to the most likely description of the data . the result of the maximum likelihood fit is @xmath115 both for model a and b , i.e. we see no sign of signal contribution in our neutrino event sample . we set 90% c.l . neyman upper limits ( see @xcite , reprinted in @xcite ) on the tested neutrino fluence models , which amount to @xmath116 and @xmath117 times the fluences given for model a and b above , respectively . the limits are much higher than the fluence prediction because of icecube being insensitive to sne iin at such large distances . figure [ fig : limits ] shows a plot of the tested neutrino fluence and the limits set using of icecube data . this null result and the large distance to the sn further support our conclusion that the sn detection was coincidental . nevertheless , it is interesting to roughly estimate the hypothetical emitted neutrino fluence : we take the median neutrino energies of the two alert neutrinos from table [ tab : properties ] and look up the effective areas for the ofu neutrino sample at the respective energies . from this , we derive a hypothetical neutrino fluence of for a source spectrum @xmath118 ( @xmath119 ) , or for a source spectrum @xmath120 ( @xmath121 ) . assuming that this neutrino fluence was emitted by the sn , this would imply a radiated neutrino energy of @xmath122 or @xmath123 using the luminosity distance of @xmath124 , corresponding to about 15 or about 50 times the radiated electromagnetic energy of @xmath125 ( see section [ sec : sed ] ) . this is higher than what can be expected , since with reasonable assumptions that the explosion energy @xmath126 @xcite and a fraction @xmath127 of it going into crs @xcite , the energy in neutrinos should be on the same order or less than @xmath128 . thus , also with a simple energetic argument , isotropic neutrino emission from ptf12csy causing the neutrino alert is implausible , especially on a time scale of seconds . however , a beamed emission from a jet with a small opening angle of @xmath129 would in principle be possible . the swift satellite observed the supernova four times , on 2012 april 20 ( mjd 56037 ) and around 2012 november 15 ( mjd 56246 ) ( see table [ tab : xray ] ) . we perform source detection using the software developed for the 1sxps catalog @xcite on each of the four observations , and on a summed image made by combining all the datasets . no counterpart to ptf12csy is detected . upper limits are generated for each of these images , following @xcite . a radius circle centered on the optical position of ptf12csy is used to measure the detected x - ray counts @xmath130 at this location , and the expected number of background counts @xmath131 , predicted by the background map created in the source detection process . we then use the bayesian method of @xcite to calculate the upper limit @xmath132 on the x - ray count rate of ptf12csy , using the xrt exposure map to correct for any flux losses due to bad pixels on the xrt detector , and the finite size of the circular region . the upper limit count rate is converted to unabsorbed flux @xmath133 using the heasarc tool webpimms , assuming a black body model with @xmath134 as in @xcite , a galactic hydrogen column density of @xcite and a redshift of @xmath0 . the result is a x - ray flux @xmath135 for the most constraining upper limits , corresponding to a x - ray luminosity of @xmath136 with a luminosity distance of about . using a power - law @xmath118 instead of a black body as an alternative x - ray emission model , the unabsorbed flux upper limit becomes @xmath137 , and hence , @xmath138 . cccccc 56037.15 & 4.9 & 1 & 1.47 & 1.3 & + 56245.29 & 2.0 & 1 & 0.62 & 2.1 & + 56246.04 & 1.2 & 2 & 0.44 & 9.8 & + 56247.62 & 5.0 & 2 & 1.35 & 2.1 & + sum & 13.0 & 6 & 3.71 & 1.3 & comparing with other sne iin , e.g. sn 2008iy @xcite which had a measured x - ray luminosity of @xmath139 or sn 2010jl @xcite with @xmath140 , we can not exclude x - ray emission from ptf12csy with our measured upper limit . however , @xcite suggest that @xmath141 be about @xmath142 of the bolometric luminosity at the time of the shock breakout . with the estimated bolometric luminosity from section [ sec : sed ] around the time of the first swift observations , this implies @xmath143 , well below our x - ray limits . during the follow - up program of the neutrino alert , the first observations were done on 2012 april 03 , 05 , 07 and 09 ( mjd 56020 to 56026 ) by ptf with the palomar samuel oschin 48-inch telescope ( p48 ) @xcite , which is a wide - field schmidt telescope . the images ( see fig . [ fig : subtraction ] ) revealed a so far undiscovered supernova , named _ ptf12csy _ , at a magnitude of @xmath144 in the mould @xmath37-band . more photometric observations were carried out , with the p48 and the palomar 60-inch ( p60 ) telescopes @xcite at the palomar observatory in california , and the faulkes telescope north ( ftn ) @xcite at haleakala on maui , hawaii . spectroscopy was taken as well , with the gemini north multi - object spectrograph ( gmos ) @xcite on the gemini north telescope ( mauna kea , hawaii ) on 2012 april 17 ( mjd 56034 ) and with the low - resolution imaging spectrometer ( lris ) @xcite on the keck i telescope ( kamuela , hawaii ) on 2013 february 09 ( mjd 56332 ) , enabling the identification of the supernova as a type iin sn with narrow emission lines . the spectra are available from wiserep @xcite . p48 data were extracted using an aperture photometry pipeline and are calibrated with 21 close - by sdss stars @xcite . the faint host galaxy was subtracted and the upper limits are at the level . p48 magnitudes are in the ptf natural ab magnitude system , which is similar , but not identical to the sdss system . the difference is given by a color term , which is ignored in this work , except for the conversion of mould @xmath37 to sdss @xmath72 , explained in section [ sec : corr ] . the p60 photometry is tied to the same 21 sdss calibration stars . note that there might be host galaxy contamination in the late - epoch p60 photometry . p60 magnitudes are in the sdss ab magnitude system . the ftn data were processed by an automatic pipeline , without host subtraction , and agree very well with the host - subtracted p60 data taken in the same night . the pan - starrs1 ( ps1 ) telescope was not part of the real - time triggering and response system , but its wide - field coverage provides a useful archive to search retrospectively for detections . ps1 is a @xmath145 telescope located at haleakala on maui in the hawaiian islands , equipped with a fov and a 1.4 gigapixel camera @xcite . in the course of its 3@xmath11 steradian survey , the telescope observes each part of the sky typically times per year @xcite . ps1 first detected ptf12csy on mjd and archived it as object _ pso j104.6365 + 17.2622_. the magnitudes in all ps1 images were obtained with psf fitting within the pan - starrs image processing pipeline @xcite . they are calibrated to typically seven local sdss dr8 field stars . the magnitudes are in the natural ps1 ab system as defined in @xcite , which is similar , but not exactly the same as sdss ab magnitudes . particularly the @xmath146-band can differ . the swift uvot data were analyzed using the publicly available swift analysis tools @xcite , and a source was seen close to the detection threshold . rotse s limiting magnitude of about prevented a detection of the sn in rotse follow - up observations . the photometry is corrected for galactic extinction using @xmath147 and @xmath148 @xcite . the extinction coefficient is converted to the filters effective wavelengths using the algorithm from @xcite , i.e. for @xmath149 , for @xmath150 , for @xmath146 , for @xmath72 , for @xmath151 . the extinction within the host galaxy could not be determined . in fig . [ fig : spectrum_and_filters ] , the gemini north spectrum is overlaid with the applied photometric filters . the strong balmer lines contribute differently to the various filters . for the sed construction ( see section [ sec : sed ] ) , in order to approximate the black body continuum , the contribution of the strongest emission lines , h@xmath10 and h@xmath152 , is removed from the photometry using the gemini north spectrum and the filter curves . for figs . [ fig : lc_abs ] and [ fig : decline ] , the p48 mould @xmath37 magnitudes are converted to sdss @xmath72 by subtracting the h@xmath10 contribution ( as above ) , applying the formulae in @xcite valid for black body spectra , and then re - adding the h@xmath10 contribution to the @xmath72-band . after conversion , the p48 @xmath37 magnitudes are consistent with the p60 sdss @xmath72 magnitudes . the swift uvot data contain host contamination . since no galex data from a pre- or post - sn epoch are available for the host galaxy , no host subtraction can be done in the uv filters of uvot . for the @xmath149 , @xmath153 and @xmath154 filters , the host is subtracted by interpolating the host magnitudes from the sdss dr12 data @xcite to the effective wavelengths of the uvot filters . the earliest detection of ptf12csy was in the pan - starrs1 y - band on 2011 october 13 ( mjd ) , 169 days prior to the neutrino alert in observer frame , corresponding to 158 days in host galaxy rest frame using @xmath0 ( see section [ sec : spec ] ) . the latest non - detection , again in pan - starrs1 , was on 2011 march 21 ( mjd 55641.3 ) in a @xmath155 @xmath151-band frame , 206 days before the first detection ( 193 days in rest frame ) . hence , the explosion time is not well constrained and can be anytime between mjd 55641.3 and mjd . hereafter , we refer to the @xmath156-band detection at mjd as the first detection and use it as day 0 for the light curve . the uncorrected sn light curves with the data available through the icecube optical follow - up program are displayed in fig . [ fig : lc_app ] , including photometry acquired with the swift uvot filters @xmath157 , @xmath158 , @xmath159 , @xmath149 and @xmath153 ; the johnson @xmath150 filter on p60 ; the sdss filters @xmath146 , @xmath72 , @xmath100 with data from p60 , ps1 , and ftn ; the sdss @xmath151 filter on p60 ; mould @xmath37 filter on p48 ; and pan - starrs @xmath156 filter on ps1 . the entire uncorrected photometry in apparent magnitudes , as seen in fig . [ fig : lc_app ] , is also available in table [ tab : lc ] . the light curves are averaged within bins of 10 days width , for each filter and telescope separately . note that , in contrast to most of the other photometry , no host subtraction is performed for the swift uvot magnitudes presented in fig . [ fig : lc_app ] and table [ tab : lc ] . ccccccc & & & & & @xmath146 & p48 + & & & & & @xmath37 & p48 + & & & & & @xmath37 & p48 + & & & & & @xmath37 & p48 + & & & & & @xmath146 & p48 + & & & & & @xmath151 & ps1 + & & & & & @xmath156 & ps1 + & & & & & @xmath100 & ps1 + & & & & & @xmath151 & ps1 + & & & & & @xmath146 & ps1 + & & & & & @xmath72 & ps1 + & & & & & @xmath146 & ps1 + & & & & & @xmath100 & ps1 + & & & & & @xmath156 & ps1 + & & & & & @xmath37 & p48 + & & & & & @xmath151 & ps1 + & & & & & @xmath72 & p60 + & & & & & @xmath151 & p60 + & & & & & @xmath37 & p48 + & & & & & @xmath100 & p60 + & & & & & @xmath150 & p60 + & & & & & @xmath146 & p60 + & & & & & @xmath100 & ftn + & & & & & @xmath72 & ftn + & & & & & @xmath146 & ftn + & & & & & @xmath154 & uvot + & & & & & @xmath149 & uvot + & & & & & @xmath158 & uvot + & & & & & @xmath157 & uvot + & & & & & @xmath159 & uvot + & & & & & @xmath153 & uvot + & & & & & @xmath100 & p60 + & & & & & @xmath72 & p60 + & & & & & @xmath150 & p60 + & & & & & @xmath146 & p60 + & & & & & @xmath100 & ftn + & & & & & @xmath37 & p48 + & & & & & @xmath100 & p60 + & & & & & @xmath72 & p60 + & & & & & @xmath100 & p60 + & & & & & @xmath146 & p60 + & & & & & @xmath146 & p60 + & & & & & @xmath72 & p60 + & & & & & @xmath100 & p60 + & & & & & @xmath72 & p60 + & & & & & @xmath146 & p60 + & & & & & @xmath100 & p60 + & & & & & @xmath72 & p60 + & & & & & @xmath72 & p60 + & & & & & @xmath100 & p60 + & & & & & @xmath146 & p60 + & & & & & @xmath146 & p60 + & & & & & @xmath72 & p60 + & & & & & @xmath100 & p60 + & & & & & @xmath100 & p60 + & & & & & @xmath157 & uvot + & & & & & @xmath159 & uvot + & & & & & @xmath158 & uvot + figure [ fig : lc_abs ] shows the light curve of selected filters in absolute magnitudes , after the photometric corrections . light curves of other exceptional type ii sne are overlaid for comparison : sn iin 2006gy @xcite , one of the most luminous sne ever recorded , and sn 2010jl @xcite , a sn iin that is spectroscopically similar to ptf12csy ( see section [ sec : spec ] ) and shows signs of a collisionless shock in an optically thick circumstellar medium ( csm ) , hinting towards potential high - energy neutrino production @xcite . note that the sn 2010jl light curve is not extinction corrected and the comparison light curves have different reference dates : sn 2010jl is relative to maximum light , 2006gy is relative to the explosion time . a theoretical light curve from pure radioactive decay of @xmath160 ( black dashed line ) is added to the figure as well , scaled to match the observed absolute magnitude of ptf12csy . the brightest observed absolute magnitudes after application of photometric corrections ( see section [ sec : corr ] ) and converting to absolute magnitudes with a distance modulus of @xmath161 ( @xmath0 ) are @xmath162 , @xmath163 , @xmath164 , @xmath165 , and @xmath166 , assuming standard cosmology with hubble parameter @xmath167 , matter density @xmath168 , and dark energy density @xmath169 . while these are lower limits to the peak magnitude due to the sparse sampling , these absolute magnitudes are relatively modest compared to the most luminous sne iin , e.g. sn 2006gy ( @xmath170 ) @xcite or sn 2008fz ( @xmath171 ) @xcite . they are however comparable to the sne iin 2008iy ( @xmath172 ) @xcite , 1988z ( @xmath173 ) @xcite and sn 2010jl ( @xmath174 ) @xcite . the light curves of ptf12csy indicate a plateau within @xmath175 after first detection , and a slow fading afterwards . the corrected absolute magnitude light curves are fitted to obtain the linear decline rates in different photometric filters , during different epochs ( see fig . [ fig : decline ] and table [ tab : decline ] ) . for some epochs and filters , especially @xmath146 and @xmath72 , the decline rates are close to , the decline rate expected for radioactive @xmath176co decay @xcite , while in general decline rates are slower , indicating that at least part of the radiated energy is powered by interaction of the sn ejecta with a dense csm @xcite . cccc @xmath157 & & & + @xmath146 & & & + @xmath72 & & & + @xmath100 & & & + @xmath151 & & & + @xmath156 & & & additionally , radioactive decay of @xmath176co at a still relatively high absolute magnitude of @xmath177 implies a preceding @xmath176ni decay with an extremely bright peak , which was not observed , although the data are quite sparse . assuming that the luminosity is generated by radioactive decay alone and following @xcite , we estimate that @xmath178 of @xmath176ni would be required to provide the bolometric luminosity of at in rest frame ( see section [ sec : sed ] ) . the lower limit on the @xmath176ni mass is set by assuming that the explosion and thus generation of @xmath176ni was at the latest possible time , directly before the first detection . figure [ fig : lc_abs ] shows the corresponding theoretical light curve resulting from the radioactive decay of nickel and cobalt ( black dashed line ) . adopting an earlier explosion time results in an even larger @xmath176ni mass . this is much more than the usual amount of @xmath176ni of , often ( see e.g.@xcite , also @xcite ) . however , extremely superluminous sne might have @xmath176ni masses of that order of magnitude @xcite . we note that in addition to the @xmath176ni mass and luminosity arguments , the spectrum showing intermediate width balmer lines and a continuum appears inconsistent with radioactive decay as well ( see section [ sec : spec ] ) . here we assume that the light curve is powered by conversion of the ejecta s kinetic energy to luminosity through interaction of the ejecta with the csm . following ( * ? ? ? * see also : e.g. @xcite ) , we model the light curve as a power - law of the form @xmath179 . after shock breakout , there is a phase of power - law decline of the luminosity , with an index of typically @xmath180 . this lasts until the shock runs over a csm mass equivalent to the ejecta mass and the shock enters a new phase of either conservation of energy if the density is low enough and the gas can not cool quickly ( the sedov - taylor phase ) , or conservation of momentum if the gas radiates its energy via fast cooling ( the snow - plow phase ) . during the late stage , the light curve will be declining more steeply , in both cases @xcite . since we assume ptf12csy to be powered by interaction , we try to fit the interaction model from @xcite to the light curve data with the least - squares method . we perform the fit within the range of rest frame days , starting at the first r - band detection , and use the r - band light curve scaled with the bolometric luminosity from section [ sec : sed ] . it is found that the power - law index @xmath10 needs to be significantly steeper than in order to reasonably describe the data . it lies in the range of , using the constraint on the explosion time ( see section [ sec : lc ] ) for the temporal zero point of the power - law . the best fit is at @xmath181 , with the explosion time at the lowest allowed value which is the date of the last non - detection . this suggests a very steep csm density profile @xmath182 ( see * ? ? ? * eq . 12 ) , compared to the profile @xmath183 resulting from a wind with steady mass loss . however , the self - similar solutions of the hydrodynamical equations @xcite that are used in ( * ? ? ? * eq . 12 ) are invalid if the csm density profile is steeper than @xmath184 . but nevertheless , as discussed for the late - time light curve in @xcite , probably the profile is steeper than @xmath184 . this leaves us with several possible explanations : 1 . already between rest frame days 93 and 200 , the sn was in the late , e.g. snow - plow , phase . this is consistent with sn 2010jl , where the late - time light curve also shows a power - law index @xmath185 @xcite . assuming that the break in the light curve between power - law phase and late phase occurred just before the first r - band detection at day 93 , and comparing with sn 2010jl @xcite , then this means that the sn was likely already a few hundred days old , and the r - band maximum was about brighter than the observed one ( cf . * fig . 1 ) , consistent with sn 2010jl s r - band maximum . it follows that the power - law phase ended @xmath186 rest frame days after explosion , from which we can derive a swept csm mass of @xmath187 , using ( * ? ? ? * eq . 22 ) and adopting the standard values given for sn 2010jl . the sn is powered by ejecta - csm interaction , but its light curve is declining steeper than a @xmath188 power - law . this is possible , e.g. if spherical symmetry , assumed in @xcite , is broken , if the optical depth is lower than in sn 2010jl or if the csm density profile falls steeper than @xmath189 ( s.a . ) . the sn is not powered by interaction , but by radioactive decay , leading to an exponential light curve decline . however , this appears unlikely , as noted above in section [ sec : decline ] . since the spectra are only roughly calibrated , the spectral energy distribution ( sed ) is approximated from photometric data . for the highest spectral range and number of observations , a window of 10 observer frame days around day 189 , from day 184 to 194 , is used to select data ( day 172.2 to day 181.6 in rest frame ) . photometric corrections are applied , e.g. h@xmath10 and h@xmath152 removed ( see section [ sec : corr ] ) . the data are plotted in fig . [ fig : sed ] as function of the filters effective wavelengths . a black body spectrum is assumed to describe the sed and is fitted to the data . for each filter , a model data point corresponding to the black body spectrum is calculated via integration of the black body spectrum and the filter function following the sdss definition of ab magnitude in ( * ? ? ? * eq . 7 ) . a @xmath190 fit minimizes the difference between the model data and the measured data . the fit results in a reduced @xmath191 and delivers estimates for both the rest frame temperature @xmath192 and the absolute bolometric luminosity @xmath193 of the photosphere emitting the black body radiation : @xmath194 and @xmath195 , where the errors correspond to . applying the stefan - boltzmann law , we can calculate the radius of the black body photosphere from the bolometric luminosity . we estimate it to be @xmath196 . finally , to obtain an estimate on the total radiated energy , the lines contributions to the luminosity have to be added to the continuum luminosity . using the gemini north spectrum , the contribution of the h@xmath10 and h@xmath152 line to the total luminosity is computed and added to the continuum luminosity from the black body fit . this results in an estimated total radiated luminosity of at day 189 in the observer frame , i.e. day 177 in the rest frame . we use the fitted shape of the @xmath100-band light curve ( cf . section [ sec : phot ] , fig . [ fig : decline ] , table [ tab : decline ] ) to extrapolate this value and find a total radiated luminosity of at ( rest frame ) , as used in section [ sec : decline ] , and a total energy of @xmath197 radiated within 400 rest frame days after first detection , comparable to sn 2008iy which had ) @xcite and sn 2010jl with @xcite . this is a lower limit on the total radiated energy , since we lack photometric data between explosion and first detection and do not extrapolate before the first detection . additionally , as discussed below , we are neglecting a possible contribution of x - ray and @xmath5-ray emission to the total radiated energy , which is not considered by the black body spectrum based on the uv and optical data . we recommend to treat these results with caution , since @xcite pointed out that at late times the fraction of energy released from sne iin in x - rays can increase , causing the optical spectrum to deviate from a black body as fewer photons are available in the optical . this can lead to an effective decrease of the estimated photospheric radius . in this context , our estimates of @xmath198 , @xmath199 , and @xmath128 must be treated as lower limits . unfortunately , the x - ray flux from ptf12csy was not detected ( see section [ sec : x - ray ] ) . two spectra were acquired ( table [ tab : speclog ] , fig . [ fig : spectra ] ) . they are dominated by narrow emission lines , characteristic for type iin sne , with a very weak blue continuum , which indicates the old age of the sn . no continuum is visible in the late spectrum . the sn emission lines are primarily hydrogen , the balmer series is visible from h@xmath10 up to h@xmath200 . the oxygen lines oi @xmath201 , oii @xmath202 with fwhm @xmath203 , and oiii @xmath204 with fwhm @xmath205 are very narrow and were most likely produced by circumstellar gas released by the progenitor prior to explosion and then photoionized by uv radiation @xcite . ccccc 56034 & 11 & 175 & 80 & gemini north gmos + 56332 & 290 & 454 & 100 & keck i lris figure [ fig : halpha ] shows a close - up on the h@xmath10 line from both spectra , plotted vs. doppler velocity relative to the rest frame line center . in the early spectrum , the h@xmath10 line peaks at the line center and the line is composed of a narrow , intermediate and broad component with fwhm of @xmath206 , @xmath207 , and @xmath208 respectively , found by fitting a superposition of three gaussian functions to the h@xmath10 profile . this is similar to other sne iin , e.g. sn 1988z and 2008iy @xcite . the h@xmath10 profile with broad and intermediate - width component can be explained as a result of the interaction of the sn ejecta with a two - component wind @xcite . in this model , the broad line component is emitted from shocked sn ejecta expanding in a relatively rarefied wind , while the intermediate component arises from a shocked dense part of the wind , which can either consist of dense clumps or be a dense equatorial wind @xcite . this is an indication of ejecta - csm interaction . the intermediate and broad component of the early spectrum s h@xmath10 line are blueshifted which may indicate formation of dust , as explored by @xcite for sn 2010jl . a more recent study by @xcite , the most comprehensive work on the sn 2010jl emission line blueshifts to date , finds very strong evidence for a wavelength dependence of the blueshift . therefore , the authors conclude that the origin of the blueshifts is most likely the rapid formation of large dust grains , confirming @xcite and having implications on the origin of dust in galaxies . alternatively , @xcite explain the line blueshift in sn 2010jl with a bulk velocity of the emitting gas towards the observer . this is more consistent with observations if the spectral lines are symmetric about a center and if there is no wavelength dependence of the blueshift . the bulk velocity is believed to be the result of radiative acceleration of the gas by flux from the sn . presumably , there are also other possible explanations for the blueshift , e.g. the geometry or density structure of the csm . in case of sn ptf12csy , spectral line blueshift is only clearly visible in the h@xmath10 line , prohibiting the interpretation of the blueshift in favor of any scenario . the late spectrum s h@xmath10 line has a much more complicated structure than the early spectrum s . it does not peak at zero velocity anymore , but the peak is blueshifted and there are many sub - peaks . again , the blueshift might be connected to dust formation or radiative gas acceleration , but other reasons are conceivable as well . at least part of the late h@xmath10 line s complex appearance might be due to superposition of other spectral lines , e.g. nii . apart from that , it is an indication of an inhomogeneous , maybe clumpy , csm structure , and perhaps asymmetric sn explosion . the spectra are compared to template spectra from the padova - asiago supernova archive ( asa ) @xcite using the online tool gelato . the algorithm @xcite divides a spectrum into 11 relevant bins and averages within the bins to classify and compare with the archived spectra . the ptf12csy spectra are de - reddened with @xmath209 and compared to other type ii sne . gelato returns the best 30 matching spectra together with their phases , ordered by quality of fit . for both the gemini north spectrum taken at and the keck i spectrum from , the majority of best matching template spectra come from sn 2010jl . the mean phase of the matching spectra is significantly higher for the keck i spectrum : , versus for the gemini n spectrum . however , the reference dates for the spectra are mostly the discovery date , only rarely the date of maximum light or explosion date . the galaxy hosting ptf12csy is a faint dwarf galaxy designated sdss j065832.82 + 171541.6 , barely visible in the sdss dr12 images . since it has no cataloged redshift , we measure its redshift via oii @xmath202 and oiii @xmath210 lines in the keck i spectrum . the redshift is , corresponding to a luminosity distance of @xmath211 assuming standard cosmology with hubble parameter @xmath212 , matter density @xmath168 , and dark energy density @xmath169 . adopting the luminosity distance from above , the host galaxy has absolute magnitudes of @xmath213 , @xmath214 and @xmath215(corrected for galactic extinction ) . using the luminosity - metallicity relation from ( * ? ? ? * eq . 1 ) , we find a metallicity of @xmath216 , indicating that the host galaxy is quite metal - poor . overluminous sne iin , such as ptf12csy , have been preferentially found to occur in subluminous , low - metallicity galaxies @xcite , such as the host of ptf12csy . this is a trend , which is also observed for long grbs @xcite . statistics are still low and @xcite cautioned that there could be some selection bias due to intrinsically bright sne in faint host galaxies being more easily discovered during surveys doing aperture photometry . however , new surveys performing image subtraction and observing large untargeted fields , e.g. ptf and pan - starrs , provide increasing evidence for this trend , as most of the discovered bright objects would have been luminous enough to be detected in bright galaxies and in searches that are targeted to bright galaxies @xcite . ptf12csy , probably discovered by coincidence in an unbiased way , confirms this emerging trend as well . the sdss dr12has the host galaxy s cataloged center position about away from the found sn position ( see also fig . [ fig : subtraction ] ) . with an apparent diameter of about , corresponding to @xmath217 , this is quite far from the center of the galaxy , i.e.about off - center . the highest significance alert from the icecube optical follow - up program led to a coincidental discovery of the interesting and unusual type iin sn ptf12csy , which was already at least 169 days old . the combined _ a posteriori _ significance of the neutrino doublet alert and the coincident detection of any core - collapse sn within the error radius of the neutrino events ( ) and within the luminosity distance of the sn ( ) is , for the time interval of the icecube data acquisition season 2011/12 . ptf12csy is rare and unusual : with peak absolute magnitudes of @xmath218 , perhaps about -20 , it belongs to the most luminous sne . the sn is most likely powered by interaction of the ejecta with a dense circumstellar medium ( csm ) . the spectrum indicates a complicated structure of the csm . its host galaxy is a faint and metal - poor dwarf galaxy , confirming an observed trend for luminous sne iin . ptf12csy is similar in photometry and spectroscopy to other rare luminous sne iin , e.g. sne 2008iy and 2010jl . the total radiated energy is within the first 400 rest frame days after detection . given the ejecta - csm interaction , high - energy ( he ) cosmic ray production and neutrino emission may be expected on a time scale of , according to @xcite and @xcite . however , the sn is too far away for icecube to detect this emission . a complementary neutrino analysis performed offline , using one year of icecube data which cover most of the optical sn fluence , does not reveal a signal - like accumulation of neutrino events from the sn s position , leading to a very high upper limit of more than 1000 times the tested model fluence , owing to the large distance . due to the long delay of several months between explosion date and neutrinos , the doublet of neutrinos within less than two seconds can not be explained by the formation of a jet shortly after core - collapse according to the choked jet model @xcite . nor can it be explained by the expected he neutrino production from ejecta - csm interaction of sne iin ( s.a . ) . we therefore conclude that the only reasonable explanation is that the sn detection was coincidental and the neutrino doublet was produced by uncorrelated background events of atmospheric neutrinos and/or mis - reconstructed atmospheric muons . however , if there was a delaying mechanism at work , such as the spin - down of a supramassive neutron star that delays its collapse to a black hole , as in the blitzar model @xcite , then the neutrino doublet may have originated from a jet forming in the course of this delayed collapse few hundred days after sn explosion . the coincidental detection of a type iin sn following an icecube neutrino alert demonstrates the capability of the follow - up system to reveal transient he neutrino sources . an advantage of the follow - up paradigm is the prompt availability of multi - messenger information for the identification of the source , as well as the mere statistical significance of a coincidence between a neutrino burst and an electromagnetic transient detection . in this case , the significance is very low due to the delay of several months between explosion and neutrinos . however , this coincidence motivates the continuation of the follow - up program as well as further stacked neutrino analyses of type iin supernovae . fisher , r. a. 1990 , statistical methods , experimental design , and scientific inference : a re - issue of statistical methods for research workers , the design of experiments , and statistical methods and scientific inference ( oxford england new york : oxford university press ) , b. , sapir , n. , & waxman , e. 2012 , in http://dx.doi.org/10.1017/s174392131201304x [ ] , http://adsabs.harvard.edu/abs/2012iaus..279..274k[vol . 279 , death of massive stars : supernovae and gamma - ray bursts , 274281 ]

the dominion radio astrophysical observatory deep field polarization study has been matched with the _ spitzer _ wide - area infrared extragalactic survey of the european large area _ infrared space observatory _ survey north 1 field . we have used vla observations with a total intensity rms ofjy beam to match swire counterparts to the radio sources . infrared color analysis of our radio sample shows that the majority of polarized sources are elliptical galaxies with an embedded active galactic nucleus . using available redshift catalogs , we found 429 radio sources of which 69 are polarized with redshifts in the range of . we find no correlation between redshift and percentage polarization for our sample . however , for polarized radio sources , we find a weak correlation between increasing percentage polarization and decreasing luminosity .

1103.4228

cosmic magnetic fields have become of increasing importance in understanding the universe . magnetic fields play an essential role in star formation and are believed to be part of galaxy formation and evolution . however , little is known about cosmic magnetic field origin , structure , and evolution . do magnetic fields and galaxies evolve together ? how are large scale magnetic fields generated and maintained ? what is the role of magnetic fields in galaxy evolution ? answers to these questions have become one of the five main science goals of the next generation radio telescope , the square kilometer array ( ska , * ? ? ? detecting radio waves provides information on the magnetic field structure of an object . extragalactic radio sources emit synchrotron radiation which is polarized , where the percentage polarization theoretically can be as large as @xmath3% @xcite . synchrotron emission provides the measure of the total field strength and the polarization of the radio wave provides information on the direction of the magnetic field and the degree of ordering . information on the relationship between the environments of the radio sources and their properties can be inferred by the magnetic field structure . understanding of magnetic fields can lead to studies of the evolution of magnetic fields and the environments of radio sources with redshift and luminosity . at @xmath4ghz , active galactic nuclei ( agns ) are detected over a luminosity range between @xmath5w hz@xmath1 and out to redshifts @xmath6 . the majority of extragalactic polarized sources that have been studied are radio - loud agn . agns can be separated into a mixture of extragalactic radio sources which include fanaroff - riley class i and ii radio galaxies @xcite , where the classification is based on morphology and luminosity . @xcite show a model for an agn where an orientation close to the line - of - sight will result in a quasar or blazar while the classic frii is observed when the orientation angle is perpendicular to the line - of - sight . friis dominate the higher luminosity range and are more powerful than fris , however , the boundary divide between friis and fris is unknown as there is considerable overlap between the two classes in the luminosity function @xcite . the classical frii has highly collimated jets originating from the core with the large lobe structure getting brighter as they extend away from the core , with hotspots near the outer edge of the lobe . fris , on the other hand , have brighter emission near the core and decreases further away . the magnetic fields in friis are shown to line up along the jet axes with the majority of the polarized emission originating in the hotspots . on the other hand , in fris the magnetic fields lie perpendicular to the source axes and are shown to wrap around the jet ( see * ? ? ? * and references within ) . recent studies of polarized radio sources have shown that percentage polarization increases with decreasing flux density @xcite . @xcite suggest that a population change of radio sources at fainter flux densities causes the increasing percentage polarization with decreasing flux density . @xcite suggest that the changing fraction of radio - quiet agn is responsible for the increase in polarization at faint flux density levels . @xcite propose that the increase in percentage polarization with decreasing flux is the result of the transition from an frii - dominated population to an fri - dominated population . the cause of this increase in @xmath7 with decreasing flux density remains unknown . polarized observations of resolved and compact radio sources have been investigated by @xcite and @xcite to study the increase in @xmath7 with decreasing flux density . for radio sources with @xmath8mjy , @xcite determined that compact polarized radio sources ( unresolved at @xmath9 ) are more polarized than resolved polarized sources . on the other hand , for radio sources with @xmath10mjy , @xcite found that sources resolved at @xmath11 are more polarized than compact sources . @xcite suggest that the higher degree of polarization may be originating from the unbeamed lobe - dominated structure , not beamed blazars . @xcite study a sample of highly polarized objects to determine the nature of these radio sources . @xcite show that polarized radio sources with percentage polarization @xmath12 are contained in elliptical galaxies and that the sample of highly polarized sources have luminosity densities in the range of @xmath13 w hz@xmath1 . however , @xcite found no difference in the source environments between low polarization sources and ultrahigh polarization sources , indicating that the high polarization must be a result of intrinsic properties of the radio sources and not dependent on optical morphology , redshift , linear size , and radio power . we present an analysis of @xmath4ghz polarization data with the _ spitzer _ wide - area infrared extragalactic survey ( swire , * ? ? ? * ) observations to study the nature of polarized radio sources to investigate possible reasons for the observed increase in @xmath7 with decreasing flux density . we use both spectroscopic and photometric redshifts presented by @xcite and the sloan digital sky survey data release 7 ( sdss dr7 , * ? ? ? we outline the sample selection and data processing in section [ obs ] . section [ cat ] describes the polarized source list containing @xmath4ghz radio parameters and the swire infrared source properties . the nature of our polarized source sample is described in section [ nature ] . the cosmological parameters used throughout this paper are : @xmath14 , @xmath15 , and @xmath16 . spectral index is defined as @xmath17 between @xmath4ghz and @xmath18mhz . the data presented in this paper cover the european large area _ iso _ survey north 1 region ( elais n1 , * ? ? ? * ) centered on @xmath19 and @xmath20 . table [ enobs ] provides a listing of available data in the elais n1 region . the 1.4 ghz radio source sample used in this paper was taken from @xcite and was observed with the dominion radio astrophysical observatory synthesis telescope ( drao st , * ? ? ? the observations achieved a sensitivity of @xmath21jy beam@xmath1 in total intensity and @xmath22jy beam@xmath1 in both stokes @xmath23 and @xmath24 . the survey covers 15.16 deg@xmath25 , centered on @xmath26 and @xmath27 , and contains 958 radio sources down to @xmath28jy beam@xmath1 ( @xmath29 ) , of which 136 are detected in polarization down to bias - corrected @xmath30jy beam@xmath1 ( @xmath31 ) . the angular resolution is @xmath32 at the field center . instrumental polarization levels range from @xmath33 at the field center to @xmath34 at the edges of the mosaic . to obtain cross - identifications with swire counterparts , higher resolution imaging is required to provide accurate positions . high resolution ( @xmath35 ) observations of the drao _ planck _ deep fields elais n1 region were carried out with the vla in 2007 october . the observations were made by the vla in b configuration for seven @xmath36mhz channels centered on @xmath37ghz . the observations lasted a total of @xmath38hours , split into three days of @xmath39hours . the area was covered by 80 pointings , each separated by @xmath40 in a hexagonal pattern and centered at @xmath41 and @xmath42 . fig.@xmath43[obsarea ] shows the vla mosaic geometry with respect to the drao deep field observations and the swire elais n1 area . the image processing was completed within the astronomical image processing system ( aips ) data reduction package . the three days were calibrated and edited separately using the polarized flux calibrator 3c286 and the phase calibrator 1634 + 627 . 3c286 was observed twice each day while the phase calibrator was observed every 20 to 30 minutes . the flux calibration used the baars flux density scale @xcite of @xmath44jy for 3c286 . at the time of the observations , 12 of the antennas had been upgraded to expanded vla ( evla ) specifications . the amplitudes of the evla antennas on the evla - vla baselines were found to occasionally drop on the third day . as a result , all evla - vla baselines were flagged on the third day and otherwise the evla antennas remained in the calibration and imaging observations . all 80 pointings were channel averaged and imaged out to @xmath45 of the primary beam as the mapping efficiency does not improve beyond this point ( see * ? ? ? * ) . to minimize geometric phase distortions , each pointing was split into 61 facets to cover the primary beam area , each with @xmath46 pixels . all facets were cleaned and self - calibrated and then combined into one image using the aips task flatn . flatn was then run a second time to mosaic all 80 pointings together and apply the primary beam correction . the resulting total intensity rms noise level at the center of the mosaic is @xmath0jy beam@xmath1 . as shown in fig.@xmath43[obsarea ] , the vla observations do not completely cover the drao deep field or the swire elais n1 area . for the region outside the the vla mosaic , the faint images of the radio sky at twenty centimeters ( first , * ? ? ? * ) survey was used to provide high resolution ( @xmath47 ) imaging of those drao deep field radio sources down to @xmath48mjy ( @xmath49 ) . classification of each radio galaxy in the study was completed by @xcite who made the distinction between compact and resolved structure using first at an angular resolution of @xmath11 . this classification is used in our analysis . the polarization information for the radio sources comes from the drao deep field source catalog @xcite . the infrared and optical observations of the @xmath50 deg@xmath51 elais n1 region were taken from the swire elais n1 source catalog @xcite . there are 662 drao deep field radio sources covering the swire elais n1 field , of which 553 radio sources have an accurate position to match to swire objects from the vla observations and the first positions , while 109 have only the drao position . the 109 radio sources with only a drao position were covered by either the vla observations or first but were not detected because : ( 1 ) the flux density is below the detection threshold of the vla observations and first ; or ( 2 ) the diffuse emission is resolved out by the higher resolution imaging . cross - identifications between swire and vla / first positions were made using a search radius of @xmath52 around the high resolution vla radio position and taking the closest swire source . the mean positional offset between the radio source position and the swire position is @xmath53 in right ascension and @xmath54 in declination . the probability of a false cross - identification between the radio source and a swire source was determined by shifting the radio positions by @xmath55 in four directions ( up , down , left , and right ) . for each shift in position , the cross - identification procedure was repeated . the mean probability of a false cross - identification was found to be @xmath56 for a search radius of @xmath52 . visual inspection of all 662 radio sources in the swire elais n1 images were completed in order to confirm the matches . an additional complication is that a resolved radio source can have multiple components which can lead to misidentification with a swire source . visual inspection of multiple component radio sources avoided this problem . the 662 drao elais n1 radio sources in the swire elais n1 images were separated into four categories , see table [ swireid ] : 1 . unambiguous - exact matches to a swire elais n1 source , of which we found 444 radio sources . exact matches were made using only those radio sources with high resolution vla or first positioning . ambiguous - there is no high - resolution position ( a result of resolution and sensitivity ) to help determine the position of the swire counterpart or the radio source is extended so that the central galaxy could not be positively identified . there are 166 objects in this category . 3 . infrared faint - radio sources with high resolution imaging but no swire elais n1 counterpart . we found 18 radio sources matching this criteria . these objects could possibly be infrared faint radio sources ( ifrs , * ? ? ? not cataloged - there are 34 radio sources with a _ spitzer _ counterpart that did not meet the swire elais n1 catalog flux density cutoff and were therefore not included in the swire elais n1 catalog . only the radio sources with an unambiguous match in the swire catalog were used for the subsequent analysis . these sources represent @xmath57 of our radio source sample . photometric redshifts for the drao deep field sources were taken from the @xcite photometric redshift catalog ( hereafter rr08 ) of the swire survey . rr08 have cataloged photometric redshifts for 218,117 elais n1 sources . using the 444 radio sources with an exact swire identification , we obtained photometric redshifts for 189 radio sources in rr08 . the best fit photometric redshift was used in order to minimize @xmath58 , resulting in using the @xmath59 solution in all cases . the difference between the photometric redshifts fitted with the @xmath59 template and the freely fitted @xmath60 template was negligible . we also required our photometric redshifts to have : ( 1 ) a fit using four or more photometric bands ; and ( 2 ) a reduced @xmath61 . these limits are similar to those rr08 used to compare photometric redshifts with spectroscopic redshifts and rr08 noted that a @xmath62 is a failure of the photometric method . in order to test the rr08 photometric redshifts , we matched photometric redshifts of sdss dr7 sources for 154 radio sources that also have a photometric redshift in rr08 . fig.@xmath43[sdssplot ] shows the comparison of the photometric redshifts between rr08 and sdss dr7 for radio sources whose position is within @xmath11 in both catalogs . the @xmath63log@xmath64 restriction from rr08 is also shown on the plot and the majority of our radio sources lie within this cutoff . the outliers are predominantly found at @xmath65 , where the sdss dr7 errors are large . the errors in the rr08 photometric redshifts depend on the number of photometric bands available for the fit , the limiting magnitude of the object , and the limiting value of @xmath58 . rr08 determined that the typical rms value is as low as @xmath66 when fitted with 13 photometric bands and as large as @xmath67 when fitted with four photometric bands . for qsos fitted with four or more photometric bands , the uncertainty can be as large as @xmath68 . therefore the photometric redshift uncertainty in the rr08 catalog is very wide and depends heavily on the type of object and the number of photometric bands included in the fit . it should be noted that rr08 noticed a slight systematic overestimation around @xmath69 by 0.1 in elais n1 that was not seen in the other swire fields . rr08 s explanation of this overestimation is the result of a bias in the photometry at fainter magnitudes . for radio sources outside the swire elais n1 region , photometric redshifts were obtained from the sdss dr7 for sources with high resolution vla imaging . if a radio source has a photometric redshift in both rr08 and sdss dr7 , the rr08 photometric redshift was used . we used spectroscopic redshifts instead of photometric redshifts when available . spectroscopic redshifts were obtained for 24 radio sources from rr08 as well as the sdss dr7 . in our drao deep field radio source list we were able to identify one radio source in the @xcite supercluster ; dsx1 - 25 . the corresponding radio source has no detectable polarization and has a @xmath4ghz flux density of @xmath70 mjy , @xmath71 , and @xmath72w hz@xmath1 . in summary , we have obtained redshift information for 429 radio sources , of which 69 are polarized in the drao deep field . to test the completeness of our radio sources with redshifts , we compared the spectral index and percentage polarization distribution of polarized radio sources with and without redshifts , shown in fig.@xmath43 [ alphahist ] and [ pihist ] . in both cases the sample of polarized sources with a redshift is similar to those with no redshift . for the spectral index distribution of polarized sources with and without a redshift , the two - dimensional ks test determined a significance level of 0.19 . the significance level ranges form 0 to 1 and a small value indicates that the cumulative distribution function of the two data sets is significantly different . for the percentage polarization distribution of polarized sources with and without a redshift , the two - dimensional ks test determined a significance level of 0.34 . the luminosity density of the 429 radio sources with redshifts was calculated using the equations from @xcite and including the @xmath73-correction : @xmath74 where @xmath75 is the luminosity distance , @xmath76 is the flux density at @xmath4ghz , @xmath77 is the spectral index ( @xmath78 ) from @xcite and @xmath79 is the redshift . upper limits for the spectral index were only available for 18 of the 69 polarized sources and 270 of the 360 radio sources with no detectable polarization . for sources with only an upper limit to the spectral index , a value of @xmath80 for compact objects and @xmath81 for resolved objects was assumed , based on the mean value for each classification in @xcite . linear polarization @xmath82 is evaluated using stokes @xmath23 and stokes @xmath24 where @xmath83 . the estimate of the bias - corrected polarized flux is @xmath84 for a signal - to - noise ratio greater than 4 @xcite . the percentage polarization is evaluated by : @xmath85 where s is the stokes @xmath86 total flux density of the source . the bias - corrected percentage polarization @xmath87 is determined using @xmath88 instead of @xmath82 . the 69 polarized radio sources with a redshift are listed in table [ catalog ] , which contains the drao deep field name and @xmath4ghz properties from @xcite , as well as the redshift and luminosity . the near - infrared bands of the swire elais n1 catalog provide insight into the type of galaxy that hosts the radio emission . the @xmath89 m infrared wavelength range can distinguish emission between stellar light in the host galaxy and light that is reprocessed by dust and gas @xcite . fig.@xmath43[swirecc ] shows the _ spitzer _ near - infrared color - color diagram of 243 drao deep field host galaxies that were detected in all four near - infrared _ spitzer _ bands : 3.6 , 4.5 , 5.8 , and 8.0@xmath90 m . there are 45 polarized radio sources and 198 sources with no detectable polarization matching these criteria in our sample . the majority of our sources are identified with extended infrared galaxies for which we have selected the swire isophotal integrated fluxes for the four bands in our analysis . the color - color diagram is split into five regions based on the simulations of @xcite . the divisions are determined by the strength of the polycyclic aromatic hydrocarbon ( pah ) bands and the slope of the continuum in various redshift ranges . region 1 preferentially selects sources where the infrared emission is dominated by non - equilibrium emission of very small dust grains , which is interpreted as pah destruction by the hard ultraviolet spectrum of an agn . region 2 is mainly populated by dusty star - forming galaxies at redshift @xmath91 with strong pah bands , as the 3.6 and 8.0@xmath90 m flux contain the strongest pah features at low redshift . region 3 is populated by : ( a ) elliptical galaxies dominated by the starlight of old stellar populations ; and ( b ) galaxies with fainter pah emission . this region can be split into the proposed galaxy types by the lines log@xmath92 for region 3a and log@xmath93 for region 3b as denoted by @xcite and @xcite . region 4 consists of pah - dominated galaxies at redshift @xmath94 . the number of sources found in each swire color - color region is listed in table [ swirenum ] . the polarized radio sources lie mainly in regions 1 and 3 , with @xmath95 of the polarized sample in region 3 . therefore the majority of our _ polarized _ radio source population tend to have host elliptical galaxies , confirming results by @xcite . the polarized sources in region 1 tend to be infrared galaxies with very small dust grains where the radio emission indicates an agn . our color - color diagram distribution differs from @xcite in region 3 . there are polarized sources in region 3 of the earlier diagram that are not included in our diagram . we have adapted a more conservative signal - to - noise polarization cutoff than @xcite which has removed some of these sources in both total intensity and polarization . @xcite noted that there was an excess of polarized radio sources in region 3a compared to radio sources with no detectable polarization . the higher flux density cutoff of our sample does not confirm the earlier result , indicating that the results of @xcite could come from a dependence on the flux density detection level of radio sources . the additional radio sources in region 3a that were detected are not sufficiently bright to enable us to detect the polarized emission . a more sensitive polarization survey should detect the polarization from these fainter radio sources . in our radio source sample , the radio sources that lie in region 2 are radio sources with infrared emission dominated by pah indicating a star - forming galaxy . table [ spiral ] lists the 16 spiral galaxies in our sample with upper limits on @xmath87 for each spiral galaxy using an upper limit of @xmath96 from @xcite . @xcite show that spiral galaxies detected at @xmath97ghz can have @xmath87 as large as @xmath98% indicating that the upper limits on @xmath87 for our spiral galaxies are larger and a deeper polarization survey should detect polarized emission from the majority of these spiral galaxies . however , there are face - on spiral galaxies with clearly visible spiral arms that have small @xmath87 in our list . beam depolarization and not sensitivity could be the reason why these face - on spiral galaxies have no detectable polarization . most radio sources have detectable infrared counterparts . sources without infrared counterparts , the so - called infrared faint radio sources ( ifrs ) were first discovered by @xcite . @xcite provide four possible explanations for what these radio sources are : ( 1 ) agn or star - forming galaxy that have a high redshift or are very heavily obscured ; ( 2 ) agn or starburst in a transitory phase ; ( 3 ) unidentified radio lobe ; or ( 4 ) unknown object , potentially galactic in origin . observations of ifrs have suggested that these objects are high redshift agn of the 3c type at @xmath99 @xcite or compact steep spectrum sources with @xmath65 @xcite . recent observations by @xcite suggest that ifrs are dust - obscured radio - loud agn at @xmath100 . @xcite observed 17 ifrs and found that ifrs are predominantly steep spectrum objects with median spectral index @xmath101 , consistent with high - redshift radio galaxies . they reported polarization emission of three ifrs , one source having @xmath102 and the other two having @xmath103 . we found 18 ifrs in our sample of radio sources , of which five are polarized , see table [ ifrstab ] . there are four polarized ifrs that show structure at arcsecond scales ( fig.@xmath43[ifrsimg ] ) , while there are only two sources with no detectable polarization that show resolved structure . the mean spectral index for our polarized ifrs is @xmath104 . for the 13 sources with no detectable polarization the mean spectral index is @xmath105 . the mean @xmath87 of our polarized ifrs is @xmath106 . the flux densities of our polarized ifrs are @xmath107mjy with one source having @xmath108mjy , while the ifrs with no detectable polarization have @xmath109mjy . our spectral indices and percentage polarization are consistent with previous findings of @xcite . we have examined luminosity density and redshift distributions of our polarized sample to study the effect of increasing @xmath7 with decreasing flux density . fig.@xmath43[lumred ] shows the distribution of luminosity density as a function of redshift for both polarized radio sources and radio sources with no detectable polarization . the range of redshifts in our sample is @xmath110 . as expected , the polarized sources are predominantly the more luminous radio galaxies , as these tend to be the higher flux density sources which have a low @xmath7 detection threshold , a result of the detection limit in polarized flux density . there appears to be an increase in the fraction of polarized sources with @xmath111 . a complete flux limited sample is required to limit the selection effect in polarization . fig.@xmath43[lum_flux ] plots luminosity density and flux density for all the radio sources in the drao deep field . applying a flux density cut of @xmath112mjy to our data , the polarized radio sources that remain in the analysis lie between @xmath113 and @xmath114w hz@xmath1 . fig.@xmath43[pi_flux ] plots the bias - corrected percentage polarization ( @xmath87 ) and @xmath4ghz flux density for all polarized radio sources in the drao deep field . our conservative flux density cutoff decreases the selection effect , i.e. a polarized source can be detected at @xmath115 at any flux density @xmath116mjy . we have applied the flux density cut of @xmath117mjy for subsequent analysis . to study whether redshift contributes to this trend of increasing @xmath87 with decreasing flux density , the polarized radio sources are plotted in fig.@xmath43[redpi ] as a function of percentage polarization and redshift . @xcite split the polarized sources into compact and resolved classifications and suggest that the increase in @xmath87 comes from the resolved sources . using the classification of compact and resolved sources from @xcite , we fit the data with a power - law of the form @xmath118 to the @xmath119 compact sources and @xmath120 resolved sources . for resolved sources , @xmath121 and for compact sources , @xmath122 . the pearson correlation coefficient was found to be @xmath123 and @xmath124 for resolved and compact sources respectively . to measure the statistical dependence of @xmath87 on redshift , the spearman rank correlation coefficient was found to be @xmath125 with a significance level of @xmath126 for resolved sources and @xmath127 with significance level of @xmath128 for compact sources respectively . the significance level ranges from 0 to 1 and smaller values indicates a significant correlation . therefore , in our sample of polarized radio sources there is no correlation between redshift and percentage polarization . in fig.@xmath43[lumpi_res ] we plot luminosity density and percentage polarization for the 12 compact polarized radio sources and for the 34 resolved polarized radio sources with @xmath116mjy . a power - law of the form @xmath129 was fit for both the compact and resolved polarized radio sources separately . the resolved sources were fit with @xmath130 , while the compact sources were fit with @xmath131 . the pearson correlation coefficient was found to be @xmath132 and @xmath133 for resolved and compact sources respectively . to measure the statistical dependence of @xmath87 on luminosity density , the spearman rank correlation coefficient was found to be @xmath134 with a significance level of @xmath135 for resolved sources and @xmath136 with significance level of @xmath137 for compact sources respectively . therefore , @xmath4ghz observations show that resolved polarized radio sources exhibit a weak increase in @xmath87 with decreasing luminosity . the low - luminosity radio sources with low @xmath87 are not detected in our sample , which is a result of our detection limit . summarizing , we found no evidence for a dependence of @xmath87 on z , but we did find a weak dependence of @xmath87 on luminosity density for polarized sources . sources resolved at @xmath11 exhibit the trend of increasing @xmath87 with decreasing luminosity density , while compact sources do not . previous results have shown that @xmath87 increases with decreasing flux density . however , the reason for this effect remains unknown . a systematic change in the properties of radio sources with decreasing flux density has been suggested by many authors as a possible explanation for this increase ( see * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? we have compared both resolved sources and compact sources to study the properties of radio sources and the trend of increasing @xmath87 with decreasing flux density . the distribution of spectral index , percentage polarization , redshift , and luminosity density for the compact and resolved sources with a redshift and @xmath117mjy are shown in table [ radioprop ] . the compact and resolved sources differ in redshift and luminosity , with resolved sources spanning a larger range in both redshift and luminosity density . however , the 12 compact radio sources showed no trend in @xmath87 with redshift or luminosity density , while the 34 resolved radio sources show a weak trend with luminosity density but not with redshift . to understand why the compact sources differ from the resolved sources , we have examined the possible agn classifications of these objects . our compact polarized radio sources could be bl lac objects or compact steep spectrum ( css ) radio sources . @xcite show that bl lacertae objects are beamed flat - spectrum agn of both fri and frii classification , and are known to have high and variable polarization . @xcite studied 47 bl lac objects and found that polarization correlates with spectral index . @xcite showed that there is no such trend . @xcite studied the polarization properties of core - dominated quasars and bl lac objects and found that the median @xmath7 of these sources is similar at @xmath138 . the bl lac objects and core - dominated quasars in the @xcite sample were more highly polarized then the sample fri and frii radio sources . on the other hand , the polarized compact radio source may be css radio sources . css sources are radio sources that are unresolved at arc - second scales and are considered to be young radio sources with ages @xmath139years @xcite . polarization observations of css sources show that these objects are weakly polarized , with the median @xmath87 of @xmath140% at @xmath4ghz , but increasing with frequency @xcite . the young ages of css sources imply that classical radio lobes have not yet formed and hence the radio emission probably originates in the central core of the host galaxy where there is a lot of dense and inhomogeneous gaseous material . therefore the lower degree of percentage polarization may be caused by internal faraday effects like the ` partial coverage ' of the narrow - line region ( nlr ) as mentioned by @xcite . the compact polarized radio sources in our sample can be a mixture of both types of radio sources . the weak trend and poorer fit to compact sources for @xmath87 and luminosity density suggests that these objects are a separate radio population from the resolved polarized sources and do not have dramatically evolving magnetic fields . if , however , these compact objects are css sources and are the younger versions of the powerful radio - loud agn , then this suggests there is magnetic field evolution from the compact radio sources to the resolved radio sources , since resolved radio sources tend to be more polarized ( see * ? ? ? @xcite showed that resolved radio sources are more highly polarized than compact radio sources . the resolved polarized radio sources in our sample were found to have a weak trend of increasing @xmath87 with decreasing luminosity density but no trend with redshift . the result of @xmath87 not correlating with redshift is consistent with the study by @xcite who used @xmath141 nvss polarized radio sources with a redshift obtained from the catalog of quasars and active nuclei by @xcite . visual inspection of the resolved radio sources show a mixture of fris and friis . four resolved polarized radio sources are shown in fig.@xmath43 [ frsample ] to highlight the mixture of fris and friis in our sample . however , the location of the polarized emission is unknown . a study of @xmath87 and luminosity density was carried out by @xcite who found that there is no correlation between @xmath87 and @xmath142 for sources with @xmath143mjy . @xcite find a mean @xmath144% for 143 steep - spectrum polarized radio sources at @xmath4ghz . @xcite studied 26 powerful radio - loud agns and found a difference in rm in the separate lobes of the radio sources , indicating that the difference is internal to the source . the difference in the rm of each lobe and the dispersion in rm both correlate with redshift but not with luminosity or source size , pointing to more turbulent environments for high redshift agn compared to their low redshift counterparts . in the cases of @xcite and @xcite , the radio source samples are high flux density radio sources which are known to be less polarized than lower flux density radio sources . our study consists of polarized radio sources predominantly below @xmath145mjy and our sources lie in the same luminosity range as @xcite . we do not find a trend of percentage polarization with redshift , indicating that our sources may be a different population than those studied by @xcite and @xcite . we are working on high resolution polarimetric observations which will provide the necessary information to determine the location of the polarized emission in both fris and friis . these higher resolution polarimetric observations will help to determine whether fris or friis are more highly polarized . however , we are aware that higher resolution imaging can potentially resolve out any diffuse polarized emission within the sources . we have presented polarization , redshift , luminosity and infrared classification of radio sources from the drao deep field source catalog . there are 662 drao deep field radio sources with @xmath146mjy in the swire elais n1 field of which 553 have high resolution @xmath4ghz positions from follow - up vla observations . the vla observations cover the swire elais n1 region down to a rms noise level of @xmath0jy beam@xmath1 , a factor of 2 deeper than first . of the 553 radio sources with high resolution positions , 444 have an unambiguous match to a swire object . the infrared fluxes from swire show that the majority of polarized radio sources in our sample have hosts that are old elliptical galaxies with an imbedded agn . these galaxies can be either high or low luminosity radio sources . the rest of the polarized radio sources have host galaxies where the infrared emission originates from light reprocessed by dust and gas and where the radio emission comes from the agn . the agn hosted by elliptical galaxies are more highly polarized than the agn with infrared emission from dust and gas . correlating these polarized radio galaxies with redshifts show no trend between @xmath87 and redshift . we were nevertheless able to determine a weak correlation between @xmath87 and luminosity . the trend of increasing @xmath87 with decreasing @xmath142 shows that low - luminosity radio sources are more polarized than high - luminosity radio sources for sources in the flux density range of @xmath147mjy . there is a mixture of polarized fris , friis and compact sources in the two different types of host galaxy , therefore we are unable to determine if fris are more polarized than friis . high resolution polarization imaging will be able to determine the origin of polarized emission in these sources and whether fris or friis are more polarized . we thank r. perley , m. goss , and m. claussen for their support and discussions with vla image processing . j. k. b. acknowledges support from the natural sciences and engineering research council of canada . observations and research on the drao planck deep fields are supported by the natural science and engineering council of canada and the national research council canada . the dominion radio astrophysical observatory is operated as a national facility by the national research council of canada . this research has made use of observations made with the _ spitzer space telescope _ and the nasa / ipac extragalactic database ( ned ) which both are operated by the jet propulsion laboratory , california institute of technology under a contract with nasa . 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/. abazajian k. n. , et al . 2009 , , 182 , 543 beck r. , & gaensler b. m. 2004 , , 48 , 1289 ciliegi p. , et al . 1999 , , 302 , 222 de breuck , c. , van breugel , w. , rttgering , h. j. a. , & miley , g. 2000 , , 143 , 303 fanaroff , b. l. , & riley , j. m. 1974 , , 167 , 31p fan , j .- h . , et al . 2006 , , 58 , 945 fanti , r. , fanti , c. , schilizzi , r. t. , spencer , r. e. , nan rendong , parma , p. , van breugel , w. j. m. , & venturi , t. 1990 , , 231 , 333 garn , t. , green , d. a. , riley , j. m. , & alexander , p. 2008 , , 383 , 75 garn t. & alexander p. 2008 , , 319 , 1000 goodlet , j. a. , & kaiser , c. r. 2005 , , 359 , 1456 grant , j. k. , taylor , a. r. , stil , j. m. , landecker , t. l. , kothes , r. , ransom , r. r. , & scott , d. 2010 , , 714 , 1689 heckman , t. m. , & kauffmann , g. 2006 , , 50 , 677 jackson , c. a. , & wall , j. v. 1999 , , 304 , 160 kharb , p. , shastri , p. , & gabuzda , d. c. 2005 , , 632 , l69 landecker , t. l. , et al . 2000 , , 145 , 509 manners , j. c. , et al . 2003 , , 343 , 293 mantovani , f. , mack , k .- h . , montenegro - montes , f. m. , rossetti , a. , & kraus , a. 2009 , , 502 , 61 mesa , d. , baccigalupi , c. , de zotti , g. , gregorini , l. , mack , k .- h . , vigotti , m. , & klein , u. 2002 , , 396 , 463 middelberg e. , norris r. p. , tingay s. et al . 2008 , , 491 , 435 middelberg e. , norris r. p. , hales c. a. et al . 2011 , , 526 , a8 + norris , r. p. , et al . 2006 , , 132 , 2409 norris , r. p. , afonso , j. , cava , a. , et al . 2011 , , submitted oliver , s. , et al . 2000 , , 316 , 749 pacholczyk , a. g. 1970 , series of books in astronomy and astrophysics , san francisco : freeman , 1970 , p. 106 sadler , e. m. , et al . 2006 , , 371 , 898 saikia , d. j. 1999 , , 302 , l60 saikia , d. j. , & salter , c. j. 1988 , , 26 , 93 schilizzi , r. t. 2004 , , 5489 , 62 shi , h. , liang , h. , han , j. l. , & hunstead , r. w. 2010 , , 1321 simmons j. f. l. & stewart b. g. 1985 , , 142 , 100 sirothia , s. k. , dennefeld , m. , saikia , d. j. , dole , h. , ricquebourg , f. , & roland , j. 2009 , , 395 , 269 stil , j. m. , krause , m. , beck , r. , & taylor , a. r. 2009 , , 693 , 1392 subrahmanyan , r. , ekers , r. d. , saripalli , l. , & sadler , e. m. 2010 , , 402 , 2792 swinbank , a. m. , et al . 2007 , , 379 , 1343 taylor , a. r. , stil , j. m. , grant , j. k. , landecker , t. l. , kothes , r. , reid , r. i. , gray , a. d. , scott , d. , martin , p. g. , boothroyd , a. i. , joncas , g. , lockman , f. j. , english , j. , sajina , a. , & bond , j. r. 2007 , , 666 , 201 tucci , m. , martnez - gonzlez , e. , toffolatti , l. , gonzlez - nuevo , j. , & de zotti , g. 2004 , , 349 , 1267 vaillancourt , j. e. 2006 , , 118 , 1340 veron - cetty , m. p. , & veron , p. 2000 , vizier online data catalog , 7215 , 0 yang , c. , jin - lin , h. , & xiang - tao , h. 2001 , , 25 , 6 cccc 3.6 , 4.5 , 5.8 , @xmath148 m & swire@xmath149irac & 7.3 , 9.7 , 27.5 , @xmath150jy & @xcite + + 24 , 70 , @xmath151 m & swire@xmath149mips & @xmath152jy , 2.75 , @xmath153mjy & @xcite + + 7,15 , 90 , @xmath154 m & _ iso _ & @xmath155mjy & @xcite + + @xmath156kev & _ chandra _ & @xmath157ks & @xcite + + @xmath158,@xmath159,@xmath160,@xmath161,@xmath162 & int@xmath149wfs & 23.4 , 24.9 , 24.0 , 23.2 , @xmath163mag & @xcite + + @xmath164 cm ( @xmath4ghz ) & vla & @xmath165jy & @xcite + & & @xmath166jy & this paper + & vla@xmath149first & @xmath167mjy & @xcite + & vla@xmath149nvss & @xmath168mjy & @xcite + & drao st & @xmath169jy & @xcite + & drao st & @xmath170jy , polarization & @xcite + & drao st & @xmath171jy & @xcite + & drao st & @xmath172jy , polarization & @xcite + + @xmath173 cm ( @xmath174mhz ) & gmrt & @xmath175jy & @xcite + + @xmath176 cm ( @xmath18mhz ) & westerbork st & @xmath98mjy & @xcite + & gmrt & @xmath175jy & @xcite + [ enobs ] lrr unambiguous & 70 ( 65@xmath1778% ) & 374 ( 67@xmath1773% ) + ambiguous & 21 ( 20@xmath1774% ) & 145 ( 26@xmath1772% ) + infrared faint radio sources & 5 ( 5@xmath1772% ) & 13 ( 3@xmath1771% ) + not cataloged & 11 ( 10@xmath1773% ) & 23 ( 4@xmath1771% ) + [ swireid ] lrrrrccc draoj_160016.66 + 533944.82&&@xmath178&@xmath179&@xmath180&@xmath181&@xmath182&@xmath183 + draoj_160505.45 + 550045.50 & 374150&@xmath184&@xmath185&@xmath186&@xmath187&@xmath188&@xmath189 + draoj_160600.00 + 545406.16 & 374873&@xmath190&@xmath191&@xmath192&@xmath193&@xmath194&@xmath195 + draoj_160634.97 + 543455.67 & 365142&@xmath196&@xmath197&@xmath198&@xmath199&@xmath200&@xmath201 + draoj_160721.60 + 534639.58 & 151535&@xmath202&@xmath203&@xmath204&@xmath205&@xmath206&@xmath207 + draoj_160722.80 + 553059.65 & 408285&@xmath208&@xmath209&@xmath210&@xmath211&@xmath212&@xmath213 + draoj_160724.12 + 531408.34&&@xmath214&@xmath215&@xmath216&@xmath217&@xmath218&@xmath219 + draoj_160828.34 + 541031.91 & 175922&@xmath220&@xmath221&@xmath222&@xmath223&@xmath224&@xmath225 + draoj_160830.91 + 534421.98 & 156843&@xmath226&@xmath227&@xmath228&@xmath229&@xmath230&@xmath231 + draoj_160907.44 + 535425.60 & 168323&@xmath232&@xmath233&@xmath234&@xmath235&@xmath236&@xmath237 + draoj_160910.22 + 552632.14 & 415349&@xmath238&@xmath239&@xmath240&@xmath241&@xmath242&@xmath243 + draoj_160922.66 + 561511.20&&@xmath244&@xmath245&@xmath246&@xmath247&@xmath248&@xmath249 + draoj_160944.14 + 543749.37 & 202770&@xmath250&@xmath251&@xmath252&@xmath253&@xmath254&@xmath255 + draoj_160953.30 + 550708.29 & 405410&@xmath256&@xmath257&@xmath258&@xmath259&@xmath260&@xmath261 + draoj_161002.52 + 541637.63 & 190161&@xmath262&@xmath263&@xmath264&@xmath265&@xmath266&@xmath267 + draoj_161012.89 + 544925.90 & 213445&@xmath268&@xmath269&@xmath270&@xmath271&@xmath272&@xmath273 + draoj_161027.53 + 541246.94 & 189954&@xmath274&@xmath275&@xmath276&@xmath277&@xmath278&@xmath279 + draoj_161037.54 + 532430.78&&@xmath280&@xmath281&@xmath282&@xmath283&@xmath284&@xmath285 + draoj_161057.53 + 553528.00 & 431174&@xmath286&@xmath287&@xmath288&@xmath289&@xmath290&@xmath291 + draoj_161119.85 + 552846.31 & 428862&@xmath292&@xmath293&@xmath294&@xmath295&@xmath296&@xmath297 + draoj_161138.18 + 535924.86 & 187983&@xmath298&@xmath299&@xmath300&@xmath301&@xmath302&@xmath303 + draoj_161149.32 + 550049.82 & 231824&@xmath304&@xmath305&@xmath306&@xmath307&@xmath308&@xmath309 + draoj_161159.98 + 555720.23&&@xmath310&@xmath311&@xmath312&@xmath313&@xmath314&@xmath315 + draoj_161223.71 + 552554.91&&@xmath316&@xmath317&@xmath318&@xmath319&@xmath314&@xmath320 + draoj_161229.02 + 550636.72 & 239645&@xmath321&@xmath322&@xmath323&@xmath324&@xmath325&@xmath326 + draoj_161249.32 + 550239.05 & 238910&@xmath327&@xmath328&@xmath329&@xmath330&@xmath331&@xmath332 + draoj_161250.59 + 560358.07&&@xmath333&@xmath334&@xmath335&@xmath336&@xmath337&@xmath338 + draoj_161303.24 + 543224.47 & 218306&@xmath339&@xmath340&@xmath341&@xmath342&@xmath188&@xmath343 + draoj_161320.50 + 541636.80 & 209853&@xmath344&@xmath345&@xmath346&@xmath347&@xmath308&@xmath348 + draoj_161331.25 + 542718.14&&@xmath349&@xmath350&@xmath351&@xmath352&@xmath353&@xmath354 + draoj_161341.76 + 561215.37&&@xmath355&@xmath356&@xmath357&@xmath358&@xmath359&@xmath360 + draoj_161355.42 + 545753.14 & 241939&@xmath361&@xmath362&@xmath363&@xmath364&@xmath365&@xmath366 + draoj_161400.34 + 535715.08 & 46196&@xmath367&@xmath368&@xmath369&@xmath370&@xmath371&@xmath372 + draoj_161412.14 + 554132.75 & 454481&@xmath373&@xmath374&@xmath375&@xmath376&@xmath377&@xmath378 + draoj_161421.62 + 553651.26 & 451943&@xmath379&@xmath380&@xmath381&@xmath382&@xmath383&@xmath384 + draoj_161509.94 + 531015.82&&@xmath385&@xmath386&@xmath387&@xmath388&@xmath389&@xmath390 + draoj_161527.67 + 542712.46 & 230443&@xmath391&@xmath392&@xmath393&@xmath394&@xmath395&@xmath396 + draoj_161537.82 + 534646.96 & 48475&@xmath397&@xmath398&@xmath399&@xmath400&@xmath401&@xmath402 + draoj_161547.14 + 532820.21&&@xmath403&@xmath404&@xmath405&@xmath406&@xmath407&@xmath408 + draoj_161558.82 + 552505.48 & 271343&@xmath409&@xmath410&@xmath411&@xmath412&@xmath413&@xmath414 + draoj_161559.21 + 532416.56&&@xmath415&@xmath416&@xmath417&@xmath418&@xmath419&@xmath420 + draoj_161603.82 + 540434.46 & 60000&@xmath421&@xmath422&@xmath423&@xmath424&@xmath425&@xmath426 + draoj_161616.49 + 541724.65 & 68227&@xmath427&@xmath374&@xmath428&@xmath429&@xmath430&@xmath431 + draoj_161623.93 + 552703.82 & 274949&@xmath432&@xmath433&@xmath434&@xmath435&@xmath308&@xmath436 + draoj_161639.14 + 562032.64&&@xmath437&@xmath438&@xmath439&@xmath440&@xmath441&@xmath442 + draoj_161639.46 + 554525.74&&@xmath443&@xmath444&@xmath445&@xmath446&@xmath447&@xmath448 + draoj_161758.06 + 543518.64 & 85013&@xmath449&@xmath450&@xmath451&@xmath394&@xmath452&@xmath453 + draoj_161807.58 + 544237.87 & 89241&@xmath454&@xmath455&@xmath456&@xmath457&@xmath458&@xmath459 + draoj_161826.28 + 542532.77 & 82324&@xmath460&@xmath461&@xmath462&@xmath463&@xmath272&@xmath464 + draoj_161831.99 + 543834.58 & 88998&@xmath465&@xmath466&@xmath467&@xmath468&@xmath469&@xmath470 + draoj_161833.00 + 543145.77 & 85921&@xmath471&@xmath472&@xmath473&@xmath474&@xmath475&@xmath476 + draoj_161915.31 + 550513.27 & 278015&@xmath477&@xmath478&@xmath479&@xmath480&@xmath481&@xmath482 + draoj_161919.97 + 553600.97 & 298450&@xmath483&@xmath484&@xmath485&@xmath486&@xmath487&@xmath488 + draoj_161923.38 + 540032.98&&@xmath489&@xmath490&@xmath491&@xmath492&@xmath493&@xmath494 + draoj_161924.79 + 555113.57&&@xmath495&@xmath496&@xmath497&@xmath498&@xmath499&@xmath500 + draoj_162006.53 + 543233.11 & 93142&@xmath501&@xmath502&@xmath503&@xmath504&@xmath505&@xmath506 + draoj_162011.28 + 562615.43&&@xmath507&@xmath508&@xmath509&@xmath510&@xmath337&@xmath511 + draoj_162038.02 + 545129.41&&@xmath512&@xmath513&@xmath514&@xmath515&@xmath516&@xmath517 + draoj_162047.54 + 545037.25&&@xmath518&@xmath519&@xmath520&@xmath521&@xmath218&@xmath522 + draoj_162145.50 + 542111.34&&@xmath523&@xmath524&@xmath525&@xmath526&@xmath527&@xmath528 + draoj_162145.98 + 554950.20&&@xmath529&@xmath530&@xmath531&@xmath532&@xmath533&@xmath534 + draoj_162209.53 + 552341.46&&@xmath535&@xmath536&@xmath537&@xmath538&@xmath539&@xmath540 + draoj_162216.46 + 552209.59&&@xmath541&@xmath542&@xmath543&@xmath544&@xmath545&@xmath546 + draoj_162233.58 + 553035.14&&@xmath547&@xmath548&@xmath549&@xmath550&@xmath551&@xmath552 + draoj_162321.72 + 541418.78&&@xmath553&@xmath554&@xmath555&@xmath556&@xmath557&@xmath558 + draoj_162500.26 + 555020.58&&@xmath559&@xmath560&@xmath561&@xmath562&@xmath563&@xmath564 + draoj_162513.99 + 560032.00&&@xmath565&@xmath566&@xmath567&@xmath568&@xmath499&@xmath569 + draoj_162602.54 + 553858.16&&@xmath570&@xmath571&@xmath572&@xmath573&@xmath574&@xmath575 + draoj_162607.27 + 550300.76&&@xmath576&@xmath577&@xmath578&@xmath579&@xmath580&@xmath581 + [ catalog ] lrr 1 & 12 ( 27@xmath5828% ) & 42 ( 21@xmath5823% ) + 2 & 2 ( 4@xmath5823% ) & 28 ( 14@xmath5823% ) + 3a & 20 ( 45@xmath58210% ) & 87 ( 44@xmath5825% ) + 3b & 10 ( 22@xmath5827% ) & 32 ( 16@xmath5823% ) + 4 & 1 ( 2@xmath5822% ) & 9 ( 5@xmath5822% ) + [ swirenum ] lllrrr drao_j160603.24 + 552527.62 & ugc 10214 & sb(s)c pec & @xmath583 & @xmath584 & @xmath58529 + drao_j160736.67 + 535729.92 & cgcg 275@xmath149024 & sbc & @xmath586 & @xmath587 & @xmath58521 + drao_j161035.26 + 561613.40 & cgcg 275@xmath149026 & & 0.06@xmath588 & @xmath589 & @xmath58514 + drao_j161109.62 + 535811.96 & mcg + 09@xmath14926@xmath149063 & sy 2 & 0.10@xmath590 & @xmath591 & @xmath58531 + drao_j161237.03 + 535814.02 & mcg + 09@xmath14926@xmath149066 & & 0.07@xmath588 & @xmath592 & @xmath58521 + drao_j161254.58 + 545524.74 & 2masx j16125415 + 5455261 & sa & 0.12@xmath590 & @xmath593 & @xmath58529 + drao_j161331.51 + 541637.34 & 2masx j16133121 + 5416294 & s0/a & 0.08@xmath590 & @xmath594 & @xmath58517 + drao_j161449.82 + 554509.11 & 2masx j16144902 + 5545120 & & 0.09@xmath588 & @xmath595 & @xmath58536 + drao_j161457.41 + 555224.71 & cgcg 276@xmath149004 & & 0.08@xmath588 & @xmath596 & @xmath5854 + drao_j161459.30 + 535507.43 & sdss j161459.13 + 535507.4 & & 0.15@xmath590 & @xmath597 & @xmath58527 + drao_j161819.44 + 541901.34 & 2masx j16181934 + 5418587 & & 0.13@xmath590 & @xmath598 & @xmath58511 + drao_j161905.64 + 550252.40 & 2masx j16190575 + 5502447 & & 0.16@xmath590 & @xmath599 & @xmath58527 + drao_j162142.36 + 550507.04 & ngc 6143 & sab(rs)bc & 0.06@xmath588 & @xmath600 & @xmath5855 + drao_j162144.90 + 542725.92 & sbs 1620 + 545 & sy 2 & 0.10@xmath588 & @xmath601 & @xmath5857 + drao_j162148.05 + 543927.94 & mcg + 09@xmath14927@xmath149023 & & 0.10@xmath588 & @xmath602 & @xmath5854 + drao_j162211.14 + 550255.90 & mcg + 09@xmath14927@xmath149025 & & 0.08@xmath588 & @xmath603 & @xmath5858 + [ spiral ] lrrrr draoj_160333.60 + 542906.79 & 8.05 @xmath177 0.41 & 1.07 @xmath177 0.20 & 13.30 @xmath177 2.59 & @xmath1491.07 @xmath177 0.21 + draoj_160422.03 + 550546.07 & 22.22 @xmath177 0.69 & & & @xmath1491.26 @xmath177 0.09 + draoj_160530.48 + 540902.09 & 3.56 @xmath177 0.24 & & & @xmath6041.10 + draoj_160552.97 + 551032.52 & 3.12 @xmath177 0.23 & & & @xmath1491.29 @xmath177 0.37 + draoj_160607.30 + 551608.04 & 6.30 @xmath177 0.31 & 1.04 @xmath177 0.15 & 16.43 @xmath177 2.46 & @xmath1490.54 @xmath177 0.53 + draoj_160647.93 + 541510.51 & 2.23 @xmath177 0.18 & & & @xmath6041.42 + draoj_160838.74 + 542751.88 & 1.39 @xmath177 0.13 & & & @xmath6041.73 + draoj_160949.75 + 540833.32 & 1.08 @xmath177 0.13 & & & @xmath6041.91 + draoj_161112.89 + 543317.64 & 2.45 @xmath177 0.14 & & & @xmath6041.35 + draoj_161212.29 + 552302.18 & 360.15 @xmath177 4.20 & 21.90 @xmath177 0.07 & 6.08 @xmath177 0.07 & @xmath1491.01 @xmath177 0.07 + draoj_161225.78 + 545503.04 & 0.84 @xmath177 0.10 & & & @xmath6042.08 + draoj_161324.86 + 553930.64 & 3.50 @xmath177 0.18 & 0.48 @xmath177 0.08 & 13.70 @xmath177 2.46 & @xmath6041.11 + draoj_161647.42 + 535951.00 & 3.35 @xmath177 0.20 & 0.48 @xmath177 0.10 & 14.30 @xmath177 3.08 & @xmath1491.83 @xmath177 0.19 + draoj_161818.38 + 554310.60 & 1.90 @xmath177 0.15 & & & @xmath6041.52 + draoj_161832.09 + 545105.44 & 0.66 @xmath177 0.10 & & & @xmath6042.24 + draoj_162033.94 + 544328.85 & 1.72 @xmath177 0.14 & & & @xmath6041.59 + draoj_162153.35 + 545321.77 & 6.52 @xmath177 0.27 & & & @xmath1490.94 @xmath177 0.30 + draoj_162408.74 + 545218.73 & 6.96 @xmath177 0.35 & & & @xmath6040.64 + [ ifrstab ] lccccc resolves sources & 34 & @xmath605 , @xmath606 & @xmath607 , @xmath608 & @xmath609 , @xmath610 & @xmath611 , @xmath612 + compact sources & 12 & @xmath613 , @xmath614 & @xmath615 , @xmath616 & @xmath617 , @xmath618 & @xmath619 , @xmath620 + [ radioprop ] ghz for all sources in the drao deep field . the horizontal dotted line indicates the flux density detection limit and the horizontal dashed line indicates the flux density cut at @xmath622mjy . the open circles are polarized sources and the dots indicate the radio sources with no detectable polarization . ] ghz flux density and bias - corrected percentage polarization for all polarized sources in the drao deep field . the vertical dashed line is the @xmath622mjy flux density cut and the solid diagonal line is the percentage polarization detection limit . ] _ left _ : the solid line is the best fit power - law with an index of @xmath623 for resolved sources . _ right _ : the solid line is the best fit power - law with an index of @xmath624 for compact sources . ] the solid line is the best fit power - law with an index of @xmath625 . _ right _ : percentage polarization and luminosity relationship for compact polarized sources with @xmath116mjy . the solid line is the best fit power - law with an index of @xmath626 . ]

new forms of empirical formulae that provide an approximate description of the decay of shock pressure with distance in hypervelocity impacts are proposed . these forms , which are intended for use in applications such as large - scale mantle convection models , are continuous and smooth from the point of impact to arbitrarily large distances , thereby avoiding the need to divide the domain into different decay regimes and yielding the maximum pressure in a self - consistent way without resorting to the impedance - match solution . individual fits for different impact velocities as well as a tentative general fitting formula are given , especially for the case of dunite - on - dunite impacts . the temperature effects resulting from the shock are estimated for different decay models , and the differences between them are found to be substantial in some cases , potentially leading to over- or underestimates of impact heating and melt production in modeling contexts like mantle convection , where such parameterizations are commonly used to represent giant impacts . impact processes ; terrestrial planets ; thermal histories ; interiors

1702.05537

hypervelocity impacts on a planet subject the target material to extreme conditions , in particular very high pressures . it has long been known from experimental and numerical studies that the impactor penetrates the target before having deposited all of its energy , so that the center of the region from which the shock wave emerges lies at a certain depth beneath the pre - impact surface . the pressure of the shock varies with some inverse power of distance from this center , but it does not do so in a uniform way . rather , the finite size of the projectile causes pressure in a target region comparable to its own size to vary in a different way than at greater distance ( e.g. , * ? ? ? * ) , and the details of this variation are controlled by the shape of the impactor . in the context of planetary science , the most common configuration considered is that of a spherical impactor hitting a homogeneous half - space . since the pioneering numerical study by @xcite , which investigated this type of configuration , expressions of the form @xmath0 have been used to describe the decay of shock pressure @xmath1 with distance @xmath2 ; @xmath3 is the radius of the impactor , @xmath4 and @xmath5 are fitting constants , and lg is the decadic logarithm . the change in the character of pressure decay implies that the expression eq . [ eq : tradform ] does not apply with the same @xmath4 and @xmath5 everywhere , but that piecewise fits must be carried out in the different domains . the numerical studies by @xcite , @xcite , and @xcite indicate that the pressure drop is much smaller at small @xmath2 than at great distance from the impact site . most studies on the subject distinguish between a near - field and a far - field and perform fits for these two domains ; only @xcite have chosen to introduce an additional narrow mid - field . such expressions for the decay of the impact shock can be used to estimate the heating of the target following the formalism developed by @xcite or a variant thereof @xcite , and they have been used to define the thermal perturbations @xmath6 caused by impacts in planetary bodies that influence the course of evolution of mantle ( and maybe even core ) convection ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? global dynamical evolution models of this type often do not require the highly detailed representation of impacts provided by computationally expensive fully dynamical impact simulations , and their substitution with simple and readily computable first - order representations such as those developed in the following are the principal motivation for this paper . the power - law form of eq . [ eq : tradform ] has been found to work well over limited ranges and stands in a tradition of many other empirical or semi - empirical relations in impact studies that have the form of a power law , and its simplicity makes it easy to implement . however , in this case it also has disadvantages . obviously , the term @xmath7 goes to infinity as @xmath8 , which means that an upper bound of @xmath1 has to be imposed in some form ; the application of the hugoniot equation of state to the idealized situation of two colliding infinite planes offers itself as the most straightforward and familiar analytical model if such a bound is not derived directly from the data . indeed , the near - field region , in which pressure decay with distance is weak , has been treated as isobaric by several workers , and the pressure within this isobaric core has been determined from that so - called impedance - match solution ( e.g. , * ? ? ? while this is probably the best they could do with the available means , it implies that the impedance - match solution is indeed the correct upper bound for @xmath1 and also an acceptable approximation for the supposedly constant value of @xmath1 in the near field or `` isobaric core '' , but this is not always supported , if not directly contradicted , by numerical models . with this notion of a ( quasi-)isobaric central region in mind , @xcite even derived an empirical formula for the radius of the isobaric core from their numerical models . the concept of this bimodality of pressure decay is a helpful qualitative description , but on the other hand the division into distinct domains introduces discontinuities in the radial derivatives of @xmath1 and @xmath6 that are unnatural . these remarks on the shock pressure apply in similar form also to the particle velocity , which is closely related . in this paper , alternative formulations for the pressure decay with distance are considered . these formulations are smooth ( in the sense of having a first derivative ) everywhere , finite , and applicable with a single set of two or three parameters for all @xmath2 . some consideration is also given to the possibility of defining domains such as a `` near field '' and a `` far field '' self - consistently from properties of the fitting model . due to the somewhat tentative character of the results of this paper and because the pressure is probably of more immediate interest for work outside the impact community than the particle velocity , the focus here is exclusively on the pressure . although formulae of the type discussed in this paper have their merits as a fairly easy and computationally inexpensive means of estimating the most important first - order effects of large impacts and thus allow their straightforward inclusion in global convection models , it is important to keep in mind that the assumptions on which they are based , i.e. , the combination of the ( one - dimensional ) impedance - match solution for the pressure with a radially symmetric decay profile , are very idealized . therefore , they are not an accurate method of describing the smaller - scale details of any given impact process or the deviations from radial symmetry ; for that purpose , further assumptions would be needed . in order to avoid the shortcomings of the model function eq . [ eq : tradform ] , an alternative formulation for @xmath9 should be 1 ) positive for all finite @xmath2 and bounded from above ; 2 ) monotonically decreasing to zero for @xmath10 ; 3 ) almost constant at @xmath11 ; and 4 ) asymptotically equal to the form of eq . [ eq : tradform ] for large @xmath2 . requirements 2 and 3 suggest that @xmath9 may have a maximum at @xmath12 , i.e. , @xmath13 ; this may not necessarily be the case , although it would be more consistent with the idealized impedance - match solution for infinite colliding planes , which is briefly summarized in [ app : im ] . the physics involved in an impact is too complicated to be captured in a single equation , and it is likely that different processes dominate at different distances from the impact center , which entail different decay behaviors for @xmath9 ; a change in the exponent of @xmath14 had already been pointed out by @xcite . nonetheless , the four requirements provide a framework in which a heuristic mathematical approximation for easy use in applications that do not demand highest precision can be sought . there are various functions that fulfill these conditions ; here i investigate the three forms , [ eqs : genform ] @xmath15,\end{aligned}\ ] ] which will hereafter be referred to as the `` inverse-@xmath2 '' , the `` inverse polynomial '' , and the `` arc cotangent '' model functions , respectively . other forms , especially generalized gaussian bell functions , have also been considered , but they always seem to decay too strongly with @xmath2 . the coefficients @xmath16 and @xmath17 in the first two forms are coupled by the fact that @xmath18 and @xmath19 , respectively . the impedance - match approximation is expected to be nowhere as accurate as at @xmath12 , and it provides a natural scale for the shock pressure . it is convenient to determine the fit in terms of a normalized pressure @xmath20 , where @xmath21 is the impedance - match solution . indeed , if one decides to impose the impedance - match solution at @xmath12 as a constraint , i.e. , @xmath22 , then the number of free parameters is reduced by one as @xmath23 , @xmath24 , and @xmath25 , respectively , and the model equations become [ eqs : cform ] @xmath26.\end{aligned}\ ] ] both the general forms eqs . [ eqs : genform ] ( normalized with @xmath21 ) and the `` constrained '' forms eqs . [ eqs : cform ] have been applied to various datasets for which both the numerical data and the parameters necessary to determine the impedance - match solution were available . the most important ones are those for dunite projectiles and targets from @xcite for impactor velocities @xmath27 between 10 and 60km / s and from @xcite for @xmath27 between 4 and 10km / s , which match up quite well at 10km / s and can therefore be studied separately as well as in combination ; although @xcite do not give all material properties in the same explicit form as @xcite , their discussion and the good match between both datasets justifies to use the same properties for both . in addition , the data for gabbroic anorthosite or iron projectiles and gabbroic anorthosite targets from @xcite were used in order to test the applicability to different materials and to include a case in which impactor and target consist of different materials , although their datasets are sparser and of lower quality . all data points were extracted from figures in the respective publications and are shown in normalized form in figure [ fig : data ] . the material properties and the resulting @xmath21 are given in table [ tab : matprop ] . lc@xmath27 ( km / s)&@xmath21 ( gpa ) + + 4&55.112 + 5&72.625 + 6&91.632 + 7&112.133 + 8&134.128 + 9&157.617 + 10&182.6 + 20&514.6 + 40&1626.8 + 60&3336.6 + + 5&57.875 + 15&305.071 + 45&2098.224 + + 5&85.593 + 15&471.895 + 45&3349.465 + the fitting procedure was carried out with a levenberg marquardt algorithm and consisted of two steps . first , each dataset for a given @xmath27 was fitted to each of the six model functions as a function of the form @xmath28 to determine the fitting parameters for each @xmath27 ; for the dunite / dunite data and @xmath29km / s , where two complementary datasets were available , both sets were fitted individually as well as in combination . the misfit was calculated as the root - mean - square of the residuals , @xmath30 where @xmath31 is the number of degrees of freedom of the fit ( cf . * ) ; @xmath32 is the number of data points and @xmath33 is the number of fitting parameters , i.e. , 3 for eqs . [ eqs : genform ] and 2 for the constrained forms . the results of selected @xmath1 fits are given in table [ tab : fit1 ] , a complete table is provided as supplementary material . fig . [ fig : fit1 ] shows the data and some fits for dunite and impactor velocities of 4 , 10 , and 60km / s . also shown are the fitting formulae from @xcite for the corresponding velocities ( or the general formula with @xmath27-dependent coefficients for the 60km / s - model ) and the formula from @xcite ; as the latter does not provide a scaling factor for the pressure , i followed what seems to be common practice in the use of this formula , namely to determine the radius of the isobaric core with their formula ( using their values for dunite ) and to set @xmath1 to the impedance - match solution at @xmath34 while applying their decay law scaled to give a continuous graph outside . furthermore , a modified version of their formula was also tried out as a consequence of certain problems with their original formula that will be discussed below . cccccccc@xmath27 ( km / s)&@xmath16&@xmath17&@xmath35&misfit&@xmath36&@xmath37&misfit + + + 4&0.797&1.687&1.213&@xmath38&0.392&0.904&@xmath39 + 5&1.148&2.551&1.285&@xmath40&0.598&1.031&@xmath41 + 6&1.367&2.645&1.307&@xmath42&0.591&0.945&@xmath43 + 7&1.355&1.575&1.299&@xmath44&1.236&1.265&@xmath45 + 8&2.899&4.710&1.590&@xmath46&1.062&1.123&@xmath47 + 9&4.475&7.934&1.747&@xmath48&1.292&1.198&@xmath49 + 10&6.489&13.111&1.883&@xmath50&1.231&1.165&@xmath51 + & 1.311&1.564&0.994&@xmath52&0.976&0.785&@xmath53 + & 1.892&2.566&1.317&@xmath54&1.074&1.023&@xmath55 + 20&5.224&5.888&2.243&@xmath56&3.191&1.861&@xmath57 + 40&18.661&18.352&3.040&@xmath58&20.924&3.124&@xmath59 + 60&34.451&31.635&3.366&@xmath60&70.989&3.884&@xmath61 + + 5&37.691&80.474&2.447&@xmath62&0.616&0.551&@xmath63 + 15&3.109&3.720&1.340&@xmath64&1.850&1.058&@xmath65 + 45&57.067&66.452&4.439&@xmath66&12.287&3.007&@xmath67 + + 5&5.250&9.761&1.440&@xmath68&0.973&0.775&@xmath69 + 15&7.149&9.489&2.335&@xmath70&2.144&1.461&@xmath71 + 45&26.179&31.872&2.996&@xmath72&6.513&2.000&@xmath73 + + + 4&0.280&0.539&1.069&@xmath74&2.155&0.796&@xmath75 + 5&0.253&0.344&1.146&@xmath76&1.433&0.926&@xmath77 + 6&0.317&0.395&1.129&@xmath78&1.528&0.816&@xmath79 + 7&0.388&0.374&1.217&@xmath80&0.745&1.134&@xmath81 + 8&0.385&0.250&1.367&@xmath82&0.942&0.953&@xmath83 + 9&0.351&0.158&1.502&@xmath84&0.795&1.020&@xmath85 + 10&0.311&0.108&1.601&@xmath86&0.824&0.994&@xmath87 + & 0.543&0.727&0.778&@xmath88&1.015&0.631&@xmath89 + & 0.461&0.434&1.111&@xmath90&0.928&0.859&@xmath91 + 20&0.566&0.234&1.834&@xmath92&0.385&1.521&@xmath93 + 40&0.647&0.090&2.511&@xmath94&0.080&2.582&@xmath95 + 60&0.692&0.055&2.804&@xmath96&0.028&3.241&@xmath97 + + 5&0.300&0.028&2.006&@xmath98&1.492&0.447&@xmath99 + 15&0.535&0.349&1.093&@xmath100&0.606&0.867&@xmath101 + 45&0.548&0.032&3.654&@xmath102&0.126&2.483&@xmath103 + + 5&0.336&0.132&1.244&@xmath104&1.025&0.641&@xmath105 + 15&0.480&0.159&1.909&@xmath106&0.522&1.210&@xmath107 + 45&0.525&0.061&2.433&@xmath108&0.209&1.659&@xmath109 + at least some of the fitting parameters @xmath16 , @xmath17 , and @xmath35 are themselves found to be functions of impactor velocity @xmath27 , in agreement with the findings from the earlier studies , although the type of dependence is not always clear and straightforward to cast in a simple form ; it may even depend on the materials involved . two different types of functions are tried , namely linear functions of the form [ eqs : parfit ] @xmath110 for all three parameters and power - law type functions of the form @xmath111 for @xmath16 and @xmath17 ; in the case of @xmath35 , the suggestion by @xcite to use logarithmic model functions of the form @xmath112 seems to be a more appropriate choice for the decay functions eqs . [ eqs : genform ] and [ eqs : cform ] as well , again in agreement with the earlier work . hence , the second step was to try to fit the parameters @xmath16 , @xmath17 , and @xmath35 to each of the model functions eqs . [ eqs : parfit ] for each set of materials ; again , the 10km / s data for dunite were treated individually as well as in combination . moreover , the 20km / s dunite data were excluded from the parameter fit , because especially the values for @xmath16 and @xmath35 are so far off the general trend that they should be regarded as outliers . in some cases , however , the data in this second step have such a markedly non - monotonic trend or show so little variation that no convincing fit can be achieved and that the model function is dismissed on these grounds , or a mean value should be used instead . the results of selected parameter fits of interest carried out are shown in fig . [ fig : parfit ] and given in table [ tab : fit2 ] , the complete data are provided as supplementary material . ( a ) , @xmath17 ( b ) , and @xmath35 ( c ) as a function of impactor velocity @xmath27 , for dunite and using the preferred power law model eq . [ eqs : parfit]b in the case of @xmath16 and @xmath17 and the logarithmic function model eq . [ eqs : parfit]c in the case of @xmath35 . the data points and their error bars represent the parameter values for a given @xmath27 and their standard deviations , based on models by @xcite ( @xmath113 ) and @xcite ( @xmath114 ) ; the data points for @xmath29km / s are from the combined fit to both datasets ( @xmath115 ) . the corresponding fits to the linear model eq . [ eqs : parfit]a are shown on page 3132 of the supplementary material.[fig : parfit],title="fig:",scaledwidth=75.0% ] ( a ) , @xmath17 ( b ) , and @xmath35 ( c ) as a function of impactor velocity @xmath27 , for dunite and using the preferred power law model eq . [ eqs : parfit]b in the case of @xmath16 and @xmath17 and the logarithmic function model eq . [ eqs : parfit]c in the case of @xmath35 . the data points and their error bars represent the parameter values for a given @xmath27 and their standard deviations , based on models by @xcite ( @xmath113 ) and @xcite ( @xmath114 ) ; the data points for @xmath29km / s are from the combined fit to both datasets ( @xmath115 ) . the corresponding fits to the linear model eq . [ eqs : parfit]a are shown on page 3132 of the supplementary material.[fig : parfit],title="fig:",scaledwidth=75.0% ] ( a ) , @xmath17 ( b ) , and @xmath35 ( c ) as a function of impactor velocity @xmath27 , for dunite and using the preferred power law model eq . [ eqs : parfit]b in the case of @xmath16 and @xmath17 and the logarithmic function model eq . [ eqs : parfit]c in the case of @xmath35 . the data points and their error bars represent the parameter values for a given @xmath27 and their standard deviations , based on models by @xcite ( @xmath113 ) and @xcite ( @xmath114 ) ; the data points for @xmath29km / s are from the combined fit to both datasets ( @xmath115 ) . the corresponding fits to the linear model eq . [ eqs : parfit]a are shown on page 3132 of the supplementary material.[fig : parfit],title="fig:",scaledwidth=75.0% ] lcccccc & & + data&@xmath116&@xmath117&misfit&@xmath116&@xmath117&misfit + + + dunite&@xmath118&0.902&@xmath119&0.003&3.336&@xmath120 + & @xmath121&0.673&1.837&0.044&1.631&@xmath122 + & @xmath123&0.580&1.972&0.081&1.478&1.115 + fe / an&0.967&0.549&2.615&0.592&0.993&2.811 + + dunite&0.237&0.013&@xmath124&0.192&0.278&@xmath125 + & 0.511&0.003&@xmath126&0.381&0.144&@xmath127 + & 0.308&0.007&@xmath128&0.187&0.333&@xmath129 + anorth.&0.355&0.005&@xmath130&0.245&0.226&@xmath131 + fe / an&0.361&0.004&@xmath132&0.267&0.186&@xmath133 + + + dunite&@xmath134&1.682&2.425&0.001&4.061&1.316 + & @xmath135&0.609&1.042&0.075&1.477&@xmath136 + & @xmath137&0.509&2.284&0.266&1.161&2.123 + fe / an&4.097&0.597&4.550&1.542&0.789&5.625 + + dunite&@xmath138&0.156&@xmath139&0.097&1.151&@xmath140 + & @xmath141&1.388&13.018&@xmath142&2.985&@xmath143 + & @xmath144&1.122&8.256&@xmath145&2.968&@xmath146 + anorth.&@xmath147&0.305&1.321&0.021&1.676&@xmath148 + fe / an&0.174&0.140&@xmath149&0.170&0.956&@xmath150 + + dunite&0.761&@xmath151&@xmath152&2.826&@xmath153&@xmath154 + & 0.653&@xmath155&@xmath156&26.737&@xmath157&@xmath158 + & 0.398&@xmath159&@xmath160&1.177&@xmath161&@xmath162 + fe / an&0.163&@xmath163&@xmath164&0.216&@xmath165&@xmath166 + + dunite&2.668&@xmath167&@xmath168&11.175&@xmath169&@xmath170 + & 0.956&@xmath171&@xmath172&37.747&@xmath157&@xmath173 + & 1.367&@xmath174&@xmath175&10.428&@xmath176&@xmath177 + anorth.&1.391&@xmath178&@xmath179&6.537&@xmath180&@xmath181 + fe / an&0.978&@xmath171&@xmath182&3.022&@xmath183&@xmath184 + lcccccc & & + data&@xmath116&@xmath117&misfit&@xmath116&@xmath117&misfit + + + dunite&0.670&0.115&@xmath185&0.101&1.664&@xmath186 + & 0.979&0.044&@xmath187&@xmath188&3.045&@xmath189 + & 1.151&0.041&@xmath156&@xmath190&1.954&@xmath191 + fe / an&1.499&0.035&@xmath192&0.339&1.631&@xmath193 + + dunite&0.767&0.046&@xmath194&0.487&0.730&@xmath195 + & 0.445&0.061&@xmath196&@xmath197&3.958&@xmath198 + & 0.702&0.055&@xmath199&@xmath200&2.620&@xmath201 + fe / an&0.812&0.028&@xmath202&@xmath203&1.284&@xmath204 + + dunite&0.654&0.091&@xmath205&0.195&1.326&@xmath206 + & 0.759&0.038&@xmath207&@xmath208&2.597&@xmath209 + & 1.021&0.032&@xmath210&@xmath211&1.539&@xmath212 + fe / an&1.279&0.027&@xmath213&0.396&1.247&@xmath214 + + dunite&0.718&0.033&@xmath215&0.511&0.529&@xmath216 + & 0.335&0.051&@xmath217&@xmath218&3.360&@xmath219 + & 0.624&0.045&@xmath220&@xmath221&2.044&@xmath222 + fe / an&0.671&0.023&@xmath223&@xmath224&1.066&@xmath225 + it is not always obvious whether the linear or the non - linear parameter fits perform better , but by and large , the non - linear formulae eqs . [ eqs : parfit]b and c tend to yield slightly better results and are therefore generally preferred . the more promising fitting models for @xmath9 , i.e. , the inverse-@xmath2 and the arc cotangent models , can be combined with these parameter fits into a tentative general expression in which the corresponding coefficients @xmath16 , @xmath17 , and @xmath35 in eqs . [ eqs : genform]a and c and [ eqs : cform]c are functions of @xmath27 of the form eqs . [ eqs : parfit]b and c , respectively . these fits are shown in fig . [ fig : fit1 g ] , along with the general fitting formulae by @xcite and by @xcite ( in modified form ) . with the general fitting formulae , for the inverse-@xmath2 and arc cotangent decay models ; a version of this plot with error bars is provided on pp . 2930 of the supplementary material . the constrained inverse-@xmath2 curve used a linear @xmath226 , the other a logarithmic one . the dotted vertical lines with labels indicate the positions of the inflexion points of the fitting curves where applicable ( invr : inverse-@xmath2 ; invrc : constrained inverse-@xmath2 ; acot : arc cotangent ; acotc : constrained arc cotangent ) and of the radius of the isobaric core for dunite according to @xcite ( ic : original ; mic : modified).[fig : fit1g],title="fig:",scaledwidth=75.0% ] with the general fitting formulae , for the inverse-@xmath2 and arc cotangent decay models ; a version of this plot with error bars is provided on pp . 2930 of the supplementary material . the constrained inverse-@xmath2 curve used a linear @xmath226 , the other a logarithmic one . the dotted vertical lines with labels indicate the positions of the inflexion points of the fitting curves where applicable ( invr : inverse-@xmath2 ; invrc : constrained inverse-@xmath2 ; acot : arc cotangent ; acotc : constrained arc cotangent ) and of the radius of the isobaric core for dunite according to @xcite ( ic : original ; mic : modified).[fig : fit1g],title="fig:",scaledwidth=75.0% ] with the general fitting formulae , for the inverse-@xmath2 and arc cotangent decay models ; a version of this plot with error bars is provided on pp . 2930 of the supplementary material . the constrained inverse-@xmath2 curve used a linear @xmath226 , the other a logarithmic one . the dotted vertical lines with labels indicate the positions of the inflexion points of the fitting curves where applicable ( invr : inverse-@xmath2 ; invrc : constrained inverse-@xmath2 ; acot : arc cotangent ; acotc : constrained arc cotangent ) and of the radius of the isobaric core for dunite according to @xcite ( ic : original ; mic : modified).[fig : fit1g],title="fig:",scaledwidth=75.0% ] on a general note , all datasets indicate that the numerical models approach the impedance - match solution at @xmath8 the better the higher the velocity of the impactor is ; this was already noticed by @xcite , who suggested that it may be a numerical effect , namely a consequence of the shorter timesteps in models with greater @xmath27 . another conspicuous feature is the fact that the transition between the near and the far field seems to become sharper and the far - field slope becomes steeper as @xmath27 increases ; this fact is reflected by the @xmath27 dependence of the exponent @xmath35 in the model functions . given the fewer fitting parameters of the constrained fitting model functions eqs . [ eqs : cform ] , one would expect them to yield poorer fits to the data than their unconstrained counterparts . indeed , inspection of the curves and the misfits indicates that this tends to be true , in particular at low @xmath27 , but several fits with the constrained inverse-@xmath2 or arc cotangent models perform remarkably well , especially at higher @xmath27 . the best fits are usually achieved with the arc cotangent model , with the inverse-@xmath2 model coming close , whereas the inverse polynomial models give good results only at @xmath227km / s . the following discussion will therefore mostly focus on the arc cotangent and the inverse-@xmath2 models ; due to the marked deterioration at high @xmath27 , the inverse polynomial models are not considered further as general models , although their performance for dunite - on - dunite and iron - on - anorthosite impacts at low @xmath27 is on a par with the other models . the fits to dunite data at three different @xmath27 spanning the entire range of @xmath27 for which data are available ( fig . [ fig : fit1 ] ) show the strengths and weaknesses of the different fitting models . the constrained inverse-@xmath2 and arc cotangent models fit the data from @xcite almost as well as their own three - part fit and have the advantage of approaching a finite value below the impedance - match solution , whereas the near - field fit of @xcite would need to be capped to prevent it from growing to infinity as @xmath8 . the formula by @xcite can obviously not be used at these low @xmath27 at all , because the exponent @xmath35 changes sign at about @xmath228km / s , as has already been observed by @xcite and @xcite ; it had been calibrated with data for @xmath229km / s and should not be used for extrapolation to much lower @xmath27 , just as the formulae by @xcite were calibrated with and optimized for low-@xmath27 data and are not suited for extrapolation to high @xmath27 , as can be seen in the plot for @xmath230km / s . this deterioration has to be kept in mind when using the @xcite formula for planets like mars , where the ( effective ) impact velocities are expected to lie below the calibration range of the formula , because it translates into an overestimate of the shocked and heated region . at 10km / s , where the datasets from @xcite and @xcite can be combined , all formulae perform at least decently . the @xcite fit would be better in the near field if the innermost data points had been used instead of the impedance - match solution , but it is clear that in the far - field , their fit fails to reproduce the steeper decline revealed by the data from @xcite and returns pressures several times too high . by contrast , the innermost data points from @xcite suggest a flatter trend than the data by @xcite in the near field , and their near - field formula , which is apparently based on only two or three points , does not match the older data well . the fitting models of this study reproduce the data quite well for @xmath231 ; beyond that distance , the pressures returned by the inverse-@xmath2 and arc cotangent models begin to lie a bit higher than the data points , whereas the inverse polynomial model continues to give a quite good fit , although its slope may become too steep at distances greater than those sampled by the models . the reason for the deterioration may lie in the fact that the two datasets , although agreeing quite well , introduce conflicting trends in the crucial transition region . i felt unable to decide which data are to be preferred at different @xmath2 but would expect the proposed model functions to work better if one single , consistent dataset for the entire @xmath2 range were available . this highlights that a good sampling of the near field and the far field as well as the transitional region is crucial to achieve a fit that works well at all @xmath2 , and the problem is that especially at low to intermediate @xmath27 , such good sampling is generally not achieved , probably due to the computational cost . nonetheless , the individual fits for each @xmath27 give good approximations at all @xmath27 , especially for the preferred inverse-@xmath2 and arc cotangent functions , as can be seen in fig . [ fig : fit1 ] and table [ tab : fit1 ] ; even for the 10km / s fit , they lie mostly within the misfit range , especially in the critically important intermediate distance range where the shock temperature approaches the solidus ( cf . supplementary material , pp . they should therefore serve their purpose well . the apparently good performance of some of the model functions for @xmath9 at all @xmath27 motivated the construction of @xmath27-dependent parameters to be used with the @xmath1 model . the results as applied to the same dunite datasets are shown in fig . [ fig : fit1 g ] together with the general fits from @xcite and @xcite . at @xmath27 up to 10km / s , the @xcite formula with @xmath27-dependent coefficients works best , whereas the @xcite formula can not be used or has problems in the far field . some of the model functions from this study show promise , but they do not perform quite as well as one might have hoped ; especially the constrained inverse-@xmath2 model disappoints irrespective of the form chosen for @xmath226 and is therefore not considered further for the general fit . a possible reason can be identified by analyzing the parameter fits in fig . [ fig : parfit ] . although a general overall trend can be clearly identified for all three parameters and justifies the application of one function for the entire @xmath27 range , the two datasets also show distinct individual trends . although it has been tried to link them at 10km / s by using the parameter values for the combined fit , it becomes clear that the match between the two datasets is not perfect . their different trends are probably a consequence of their different strengths in terms of spatial sampling . again , a dataset produced from a single model suite with good spatial sampling in all fields would likely make a better fit possible . altogether , the general @xmath27-dependent fit performs clearly less well than the individual fits even for the preferred model functions , and the fitting curves lie outside the misfit range in several instances , albeit often only marginally so ( cf . supplementary material , pp . 2930 ) ; when judging the quality of the fit , one should keep in mind that the distortion due to the logarithmic scale tends to make discrepancies look worse than they are in the far field , where the shock pressure has already dropped to relatively small values . however , the quality of the fit tends to improve towards higher @xmath27 . depending on the accuracy requirements of the problem at hand , end users may decide to use the general @xmath27-dependent fit or interpolate between two of the more accurate individual-@xmath27 fits from table [ tab : fit1 ] if their @xmath27 is not covered by one of the available fits . given the shortcomings of the available data , it is difficult to decide which of the parameter fitting models in eqs . [ eqs : parfit ] are best . in the case of @xmath232 , the power law seems to work a bit better than the linear fit , and it has the benefit that @xmath233 for positive exponents , which is the physically expected behavior for all three pressure decay models , because @xmath16 sets the pressure scale . the properties of @xmath234 are less clear for all three decay models , although for physical reasons a function that does not become zero at @xmath235 is preferable for the inverse-@xmath2 ( and also the inverse polynomial ) model to avoid that @xmath236 for @xmath237 . whereas @xmath234 increases with @xmath27 in those models , it decreases for the arc cotangent models , if a unique trend can be identified at all , i.e. , for a power - law @xmath234 the exponent is negative , and there is a pole at @xmath235 ; however , as the arc cotangent tends to zero for increasing arguments , this is not a problem . the logarithmic form of @xmath226 was proposed by @xcite and also seems to work well in most of the cases investigated here . as @xmath35 is always positive in our fits , it seems that a change of sign is not a desirable feature of a functional form for @xmath226 and would limit the range of @xmath27 where it can be used ; the logarithmic form may thus be problematic as @xmath238 . on the other hand , very low impact velocities are probably of interest only in rather special circumstances in the context where such formulae are applied , because they approach the speed of sound , and the corresponding collision would not be a hypervelocity impact with the concomitant shock phenomena . in the literature it has been common practice to distinguish between a near field and a far field , and possibly some transitional regime , and to draw the boundary between them based on the inspection of the available data . although at least the division into two domains is clearly justified , it is often not possible to define the position of the boundary precisely ; for instance , the fits for the near field and the mid field in the numerical models by @xcite especially at low @xmath27 are based on very few ( @xmath239 ) points , and the regression lines do not cross at the nominal boundaries . appropriate coverage of both the near field and the far field is crucial , because insufficient sampling of the near field and the transitional region can easily cause the fit to yield poor values for @xmath12 and give a wrong impression about the validity of the impedance - match solution . the problem seems to be more salient at low @xmath27 , therefore the fits to the low-@xmath27 data for dunite from @xcite and for the 5km / s data from @xcite must be interpreted with caution in this respect ; on the other hand , the high - velocity data often do not cover the far - field to distances as large as those from the low-@xmath27 @xcite data , and so the latter can be expected to provide more reliable constraints for the far - field decay . ideally , coverage would always be as good as in the combined datasets for dunite at @xmath29km / s . a reliable fit especially in the near field is not only desirable as a `` final product '' for the impact modelers , but it is also important for users of that final product from other areas such as mantle convection modeling . for instance , the shock heating that produces a thermal anomaly and possibly melt in the central area is a function of the shock pressure . the waste heat produced by the unloading of the shocked material has been estimated by @xcite and was reformulated by @xcite as a function of shock pressure as @xmath240 with @xmath241 where @xmath242 is the lithostatic pressure , @xmath243 is the density , @xmath244 is the bulk sound speed , and @xmath245 is a parameter related to the hugoniot curve ( cf . [ app : im ] ) . this relation can be divided by the isobaric heat capacity @xmath246 to yield the temperature increase , which , after correcting for latent heat consumption upon melting and/or vaporization , is commonly used for calculating the post - impact temperature field in large - scale mantle convection models that consider the effects of impacts ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? hence , a poor fit may result in poor estimates of the magnitude of the temperature anomaly and its spatial extent , including the melt - producing region ; even if the temperature is capped at or near the solidus in such models for technical reasons , the size of the shocked region remains a concern . figure [ fig : dtg ] shows the thermal effect for the pressures from fig . [ fig : fit1 g ] . the temperatures in the central part of the target region are of course much higher than can realistically be expected and are to be taken as measures of the waste energy rather than as actual temperatures , because effects like latent heat consumption in various phase transitions are not included here . the parts worth more quantitative consideration are those at lower temperatures , i.e. , where @xmath247k , because they mark the boundaries of the partially molten region ; the actual maximum @xmath6 of interest in this regard depends on the pre - impact temperature and will be higher for shallow depths that affect only the crust than for giant impacts that penetrate into the already hot uppermost mantle . as can be seen from these plots , for low to intermediate @xmath27 the use of different pressure decay models translates into temperature differences of several hundred kelvin in the far field or to differences of several times the impactor radius for the size of the heated and molten region , which corresponds to tens to hundreds of kilometers for giant impacts . the use of a single , continuous , and smooth function does not only make it unnecessary to define explicitly such domains as a near field and a far field and to decide upon boundaries for the purpose of fitting , but in principle offers the opportunity to derive the position of a boundary between a near field and a far field on the basis of well - defined properties of the fitting function ; some basic properties of the inverse-@xmath2 model and the two arc cotangent models , which will be needed below , are given in [ app : derivs ] . keeping in mind the traditional notion that the near field is characterized by a near - constancy of the pressure , the most straightforward possibility is to define a certain fraction @xmath248 of the maximum pressure as the boundary , so that for the inverse-@xmath2 and the arc cotangent models the boundary would simply lie at [ eqs : rphi ] @xmath249{\frac{a}{\varphi p_\mathrm{max}}-b}=r\sqrt[n]{\frac{1-\varphi}{\varphi}b},\\ r_\varphi&=r\sqrt[n]{\frac{\cot\frac{\varphi p_\mathrm{max}}{a}}{b}}=r\sqrt[n]{\frac{\cot\frac{\varphi\pi}{2}}{b}}\end{aligned}\ ] ] respectively . however , inspection of the data and the curves suggests that a useful definition of the boundary should relate it to the curvature , whereby it must be kept in mind that the usual logarithmic representation distorts the curves and may give a partly misleading visual impression of the position of such key points . if the boundary is to be defined as the inflexion point , it may be located at [ eqs : infl ] @xmath250{\frac{n-1}{n+1}b}\\ r_\mathrm{infl}&=r\sqrt[2n]{\frac{n-1}{b^2(n+1)}}\end{aligned}\ ] ] for the inverse-@xmath2 and the arc cotangent models , respectively , under the condition that @xmath251 . as @xmath17 and @xmath35 are functions of @xmath27 , so is the inflexion point . table [ tab : fit1 ] shows that the individual fits with the unconstrained model functions fulfill this requirement in most cases , with @xmath35 usually lying between 1 and 3 , but in constrained models , including the constrained arc cotangent model , @xmath35 is often smaller than 1 . in this case , there is no inflexion point , and the pressure curve has a cusp instead of a flat maximum at @xmath12 . alternatively , the point of strongest negative curvature could be considered for defining the boundary . the curvature of a function @xmath252 is defined by @xmath253 where primes indicate derivatives with respect to @xmath2 . hence , the task is to find the global minimum of @xmath254 , which is a root of its first derivative @xmath255 at which the second derivative is positive . although we know that it lies between the inflexion point and the maximum at @xmath12 , there is unfortunately no analytic solution for the root for any of the functions proposed here , but the derivatives can be determined so that the root can be found numerically . however , its usefulness is limited by the fact that it coincides with the maximum if @xmath256 ( cf . [ app : derivs ] ) , which is often the case for velocities up to 1030km / s . [ fig : n2f ] shows @xmath257 and @xmath258 for @xmath259 as functions of @xmath27 along with the formula by @xcite in original and modified form for the radius of the isobaric core , @xmath260 with @xmath27 in km / s ; the values used here are not their general values , but their values for dunite from their table ii , whereby their erroneous value 0.022 for @xmath261 is replaced by 0.22 . the original coefficients are thus @xmath262 ( @xmath263 ) and @xmath264 , the coefficients for the modified form are @xmath265 ( @xmath266 ) and @xmath267 . given the insufficiencies in the @xmath27-dependence fits of the @xmath16 , @xmath17 , and @xmath35 , the curves should only be seen as a crude guide and are not suited for extrapolation to lower @xmath27 . as already mentioned , the point of maximum curvature detaches itself from the maximum only at higher @xmath27 for dunite , whereas the fit for the anorthosite target models from @xcite suggests that the separation takes place at lower @xmath27 already in that case . this would point to a material dependence , which is plausible , but given the scatter in the @xcite data , it could also be an artifact . it may be tempting to infer a change in dynamic behavior from the existence and position of the point of maximum curvature and/or the inflexion point , but given that the model functions are not derived from physical principles but are ad hoc constructs only designed to fulfill certain physical requirements , such inferences would probably be an overinterpretation . at this point , some remarks on the formula by @xcite with its original coefficients are necessary , because this formula has been used by several studies . the application of their original formula ( but already with corrected coefficient @xmath261 ) to their own data for @xmath29 and 60km / s is shown in fig . [ fig : fit1]b and c and reveals that their equation gives an increasingly bad description of @xmath268 toward higher @xmath27 ; of the four curves available from their paper , only the result for @xmath29km / s was found to be satisfactory . this problem prompted an attempt to determine the coefficients at least for dunite anew following their method , the result of which led to the coefficients of the modified version . @xcite explained , with reference to their figure 4 that they `` estimated the size of the isobaric core by determining the point where the two pressure decay regimes intersect '' . these regimes were defined `` by fitting a line through the two tracer points closest to the impact point for the first regime and by a least - squares fit of a number of tracer points for the second ( avoiding the intermediate region , where the shift from one regime to the other occurs ) . '' the expression `` fitting a line through [ ] two tracer points '' can be understood in two ways : first , as drawing a line going through these two points and beyond , which however should not be called a `` fit '' ; and second , as taking the mean of the two values as a constant , which is not what is usually meant when speaking of `` fitting '' , but would result in an isobaric region sensu stricto . the first method has the disadvantage of being very sensitive to errors in the values of these two points , which may easily translate into substantial shifts in the position of the intersection ; moreover , the fitted line may never intersect the far - field fit at all . i have therefore tried both options and decided to use the second , `` isobaric '' for the modified formula ; it gives the smallest possible radius under the given conditions . for the far - field fit , it is not always clear how many points should be included , as @xcite do not ( and actually can not ) give a specific number . in order to reproduce their @xmath268 value for @xmath29km / s , which can be seen in their figure 8a , one has to include almost all of the remaining points , effectively including the intermediate region they tried to exclude ; indeed , they hardly seem to reach the far field proper in the first place , as the far - field data from @xcite reveal , so that the satisfactory performance of their formula applies only to the intermediate region . using the far - field fit from @xcite at this @xmath27 instead did not give a better overall result , however , and was therefore dismissed . the @xmath1 curve for @xmath269km / s from @xcite features an odd kink at @xmath270 , where the curve flattens again and which is not seen in that form in the other curves and is probably an artifact . if one uses the outermost points for the far - field fit , the result has too shallow a slope and will intersect the near - field line too closely to the origin . their value for @xmath269km / s as plotted in their figure 8a indicates that this is what they did , but a better result would probably have been obtained by using , or at least including , the points at @xmath271 ; i tried all three possibilities and favor the fit only with points between @xmath272 and the kink , which gives the largest @xmath268 . in the fit for @xmath230km / s , two sets of points , @xmath273 and @xmath274 , may be considered , but the difference is minor ; eventually the latter was used . the final modified formula gives results that are worse at @xmath29km / s , but substantially better at all higher @xmath27 . in any case , a general problem with this parameterization that neglects the intermediate region is that the region that is strongly shock - heated to temperatures even above the melting point is artificially augmented in the modified version , whereas the original form results in an underestimate at higher @xmath27 . this can introduce substantial errors in terms of the thermal effect and melt production especially in the case of giant impacts . has dropped to @xmath259 of its peak value ( dashed lines ) as a function of impact velocity @xmath27 for dunite , according to eqs . [ eqs : rphi ] and [ eqs : infl ] , respectively . @xmath17 has the form of eq . [ eqs : parfit]b , and @xmath35 follows eq . [ eqs : parfit]c . for comparison , the formulae for the radius of the isobaric core , @xmath268 , from @xcite and its modified form are also given , with their parameters for dunite.[fig : n2f ] ] another question of interest is whether the pressure decay for an arbitrary material combination for projectile and target can be predicted if the decay law for a given combination is known . the few datasets available ( cf . fig . [ fig : data ] ) suggest that in normalized distance and pressure coordinates , the data for different materials indeed follow fairly similar trends at a given @xmath27 , although the scatter in the @xcite data introduces considerable uncertainty ; nonetheless , similar conclusions had already been reached by the original authors of the respective papers , prompting @xcite to formulate their relation for the decay exponent for all materials investigated ( except ice ) in a uniform way . in the framework of this work , it means that it may be possible to arrive at a decent result for an impact of a projectile made of material a into a target of material a ( or b ) by using an established fit for material c in normalized coordinates and rescaling it with the impedance - match solution for materials a and b , which can be determined independently from the material properties . an example is shown in fig . [ fig : extr ] , where the iron - on - anorthosite impact data from @xcite at @xmath275km / s are shown with their best fits according to tab . [ tab : fit1 ] and with a prediction derived from the dunite - on - dunite impact data . and arc cotangent fits to these data according to tab . [ tab : fit1 ] . the dotted lines are the inverse-@xmath2 and arc cotangent parameterizations from the dunite data , calculated with parameters for @xmath275km / s with eqs . [ eqs : parfit]b and c , and rescaled with the impedance - match solution for an iron - on - anorthosite impact.[fig : extr ] ] the improvement of the phenomenological description of pressure decay notwithstanding , it must be kept in mind that analytical formulae of the type considered here are only idealized , approximate descriptions of some fundamental characteristics of an impact process . as such , they may be suited for use in settings where a first - order description is sufficient and basic characteristics need to be estimated quickly and efficiently , for instance in global mantle convection models that can not resolve much detail of an impact anyway . in such models , which are the incentive for this study , the most important task is to obtain a reasonable estimate of the spatial extent and amplitude of the temperature anomaly for one or several successive impacts ; in such a context , the impact velocity from the formula will usually be replaced by its vertical component , assuming an impact angle of 45 , and effects related to the free surface such as destructive interference of the direct shock wave and its reflection from the surface may be dealt with by `` post - processing '' the calculated shock pressure according to an approach developed by @xcite . the essential end result of the parameterized shock pressure formula is a strong thermal anomaly of a certain size that will usually be located at some depth in the upper mantle of a planet and modify the regional or even the global convection patterns , even though it is common practice for technical reasons to cap the thermal anomaly at a temperature close to the solidus ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? if such models also include the compositional effects of melting , the thermal anomaly generated by the shock induces a compositional anomaly that entails dynamical effects of its own ( e.g. , * ? ? * ; * ? ? ? * ; * ? ? ? * and submitted ) . the multitude of details found in a real impact , much of it on length scales below the resolution of global convection models , is necessarily neglected , and much of it would be destroyed by stirring in the dynamically vigorous aftermath of the impact anyway . accounting for the departure from axisymmetry in oblique impacts or a more accurate description of surface effects in an analytical model would be welcome further improvements , but they are likely to complicate the expressions considerably , if they are possible at all . the parameterized expressions for the shock pressure in an impact commonly used in situations where reasonable estimates must be obtained efficiently have a form that requires an artificial subdivision of the affected volume into different decay regimes and additional assumptions about the maximum pressure . in order to avoid some of these shortcomings , it is suggested to replace the conventional formulations with a type of function that is smooth in the entire domain and fulfills certain physical requirements like boundedness by design . specifically , a form as given in eqs . [ eqs : genform]a or c , i.e. , the inverse-@xmath2 or the arc cotangent model , are found to be useful , and fits have been carried out for dunite - on - dunite impacts at velocities between 4 and 60km / s and for anorthosite or iron - on - anorthosite impacts between 5 and 45km / s on the basis of published numerical models . a tentative parameterization has also been developed in which the fitting parameters that control the shape of the curve are functions of the impact velocity , but is found to give satisfactory results only at large @xmath27 , as a consequence of the limitations in sampling density , the extent of the modeling domain in the numerical models , but also the inhomogeneities in , or slight inconsistencies between , the available datasets . users may therefore still prefer to interpolate linearly between two sets of coefficients from table [ tab : fit1 ] bracketing the target @xmath27 or devise their own fits for eqs . [ eqs : parfit ] to a suitable subset of the coefficients from that table . it seems that a non - linear functional form for the dependence of the fitting parameters on @xmath27 is preferable , but in order to arrive at a definitive decision , a comprehensive survey covering a wide range of impact velocities ( from a few to several tens of km / s ) and a broader choice of materials is needed . such a survey should also establish more firmly the tentative observation from this and earlier studies that the decay law is , at least within the limits of certain material classes , independent from the material in normalized distance pressure space . the application of the decay model functions proposed here to the data of this survey will crucially depend on sufficiently dense sampling at all distances , i.e. , even in the very near field and in the region of strongest curvature . similar model functions can also be applied to the particle velocity . in studies that use a parameterization of pressure decay such as mantle convection models that include large impacts , an accurate representation of the impact - induced temperature anomaly is of great interest with respect to the subsequent dynamics of the mantle and the production of melt . it is for applications of this sort that the model functions proposed here have been developed . it is hoped that they will be improved further , for instance by taking into account the dependence from impact angle , with data from new , dedicated numerical models and replace the hitherto used function developed by @xcite , which was found to be problematic in this respect . i thank julien monteux for some further explanations of his paper . the comments of james roberts , ross potter , and an anonymous reviewer helped to improve the paper , in particular to clarify the purpose and applicability of parameterizations of the type discussed here . this work was funded by grant ru 1839/1 - 1 from the german science foundation ( dfg ) , with additional initial funding from the helmholtz alliance project `` planetary evolution and life '' . the impedance - match solution for the peak pressure generated in the collision of two infinite colliding planes has been given by various authors . this brief summary essentially follows @xcite . the particle velocity @xmath276 in the target is given by @xmath277 with @xmath278 ( e.g. , * ? ? ? 6.5,6.6 ) ; the properties used here are those of the unshocked material . @xmath244 is the bulk sound speed of the material . @xmath245 corresponds to the slope in @xmath279@xmath276 space of the hugoniot curve of a shock front propagating with velocity @xmath279 . it is related to the zero - pressure grneisen parameter @xmath280 by @xmath281 and to the parameters commonly designated as @xmath282 , @xmath283 , @xmath16 , and @xmath17 of the tillotson equation of state by @xmath284 ( * ? ? ? aii.7.4 ) . the shock pressure in the target is then @xmath285 some basic properties of the three model functions of interest are summarized below , with emphasis on @xmath251 , @xmath286 , and @xmath10 , the latter two conditions being imposed by the physical requirement of having a positive pressure and by the radial symmetry of the problem . for the purpose of this discussion , it is more convenient to replace the variable @xmath2 by @xmath287 . obviously , @xmath290 , and as @xmath291 for all @xmath292 for the parameters of interest ( except @xmath293 ) , the point at @xmath294 is effectively a global maximum thanks to the symmetry of the problem , fulfilling one of the requirements made at the beginning . however , for @xmath295 , the second derivative , and hence also the curvature , has a pole at @xmath294 , which means that in this interval the point of maximum curvature coincides with the maximum at @xmath294 ; for @xmath293 , the maximum of @xmath296 turns into a cusp . hence , it is only for @xmath297 that the function will begin to feature a distinct `` flat '' region around the maximum as the point of strongest curvature moves away from the maximum . for @xmath300 , the inverse-@xmath2 model approaches asymptotically the traditional formulation given in eq . [ eq : tradform ] used by previous authors : @xmath301 a similar relation holds also for the inverse polynomial model if @xmath302 . for the same reasons as for the inverse-@xmath2 model , the point at @xmath294 is effectively a global maximum for @xmath305 and a cusp for @xmath293 , and the same distinction has to be made between the domains @xmath295 and @xmath297 . euler found the following expression for the inverse cotangent : @xmath307 ( e.g. , * ? ? ? * ) , where ! ! is the double factorial . substituting @xmath308 and letting @xmath309 yields @xmath310 i.e. , to first order and at large distances , the dependence takes the same form as for the other models and the traditional formulation . ahrens , t. j. , okeefe , j. d. 1977 . equations of state and impact - induced shock - wave attenuation on the moon . in : roddy , d. j. , pepin , r. o. , merrill , r. b. ( eds . ) , impact and explosion cratering , pergamon press , elmsford , n.y . . 639656 . gault , d. e. , heitowit , e. d. 1963 . the partition of energy for hypervelocity impact craters formed in rock . in : proc . 6th hypervelocity impact symp . army , u.s . air force , u.s . navy , cleveland , ohio , vol . 2 , pp .

we derive exact operator average expressions for the first two spectral moments of nonequilibrium green s functions for the falicov - kimball model and the hubbard model in the presence of a spatially uniform , time - dependent electric field . the moments are similar to the well - known moments in equilibrium , but we extend those results to systems in arbitrary time - dependent electric fields . moment sum rules can be employed to estimate the accuracy of numerical calculations ; we compare our theoretical results to numerical calculations for the nonequilibrium dynamical mean - field theory solution of the falicov - kimball model at half - filling .

cond-mat0511063

the problem of strong electron correlation is one of the most challenging problems in condensed matter physics . it is interesting because many materials with important properties for applications derive those properties from the delicate balance between minimizing the kinetic and the potential energies in strongly correlated materials . most theoretical work on this problem has focused on equilibrium properties and linear response , with only limited work available on the nonequilibrium system ( which is most easily attained when driven by an external electric field ) . two commonly studied models of strong electron correlation are the hubbard model@xcite and the falicov - kimball model@xcite ( equivalent to the hubbard model with zero hopping parameter for the down spin electrons ) . despite tremendous efforts , the exact equilibrium solution of these models is known only in some limiting cases like one dimension , where the bethe anzatz technique@xcite can be successfully applied to the hubbard model , and in the infinite - dimensional case where both models can be solved@xcite with dynamical mean - field theory ( dmft ) . since the exact solution of these problems is challenging to attain , exact results in the form of sum rules can be quite valuable in determining the fidelity of different approximation techniques ( be they analytic , variational , perturbative , or numerical approximations ) . this problem was analyzed in equilibrium for the hubbard model by steve white@xcite and used to check the accuracy of a quantum monte carlo solution to the two - dimensional hubbard model . can similar results be found for nonequilibrium situations , like the case of a strongly correlated material in an uniform external electric field ( with arbitrary time dependence ) ? the answer to this question has , surprisingly , not been discussed much in the literature . it is well known that the canonical anticommutation relation for fermion creation and annihilation operators leads to the integral of the spectral function being 1 in both equilibrium and nonequilibrium situations . it is also well known that the conventional proof that the spectral function is nonnegative in equilibrium ( arising from a lehmann representation ) , does not apply to the nonequilibrium case , so the spectral function can be negative . it is also known that in the limit of a steady state , the spectral function recovers its nonnegativity . in this work , we examine the theoretical problem of the first few moments of the spectral functions in the presence of an external electric field . it turns out that the equilibrium results for the first few moments are quite similar to the nonequilibrium results , implying they can be used effectively to determine the accuracy of different approximation techniques in solving nonequilibrium problems . dynamical mean - field theory has been employed to solve many of the models of strongly correlated electrons . to date , most of this work has focused on equilibrium properties . schmidt and monien@xcite , made a first attempt to solve nonequilibrium dmft via second - order perturbation theory for the hubbard model . their theoretical formulation evaluated the case of a spatially uniform , but time dependent scalar potential , which unfortunately does not correspond to any electric field . more recently , a generalization of the brandt - urbanek solution@xcite for the localized electron spectral function allows for an exact numerical solution of the nonequilibrium problem for the falicov - kimball model@xcite within the dmft framework . the procedure works directly in time by discretizing the continuous matrix operators and solving the nonlinear equations by iterating matrix operations on the discretized operators . the approach has been tested against the equilibrium solution@xcite , and nonequilibrium results for the quenching of bloch oscillations will be presented elsewhere . in this contribution , we derive operator identities for the first two spectral moments of the nonequilibrium green s functions when the strongly correlated material is in the presence of a spatially uniform electric field with arbitrary time dependence . these identities are found for both the falicov - kimball and the hubbard model . our results are valid for any spatial dimensionality . in the general case , the moments depend on time , which is expected because the field can be turned on at any time ; surprisingly , the first two moments of the local retarded green s function are time independent , and have the same form as in equilibrium . this last result is particularly surprising because the local retarded green s function is a nontrivial oscillating function of time in the noninteracting case@xcite . the rest of the paper is organized as follows . we derive analytical expressions for the spectral moments for the falicov - kimball model and the hubbard model in section ii . the application of nonequilibrium dmft to the falicov - kimball model follows in section iii . section iv contains results for the green s functions and spectral moments in this case . a summary of our results and our conclusions appear in section v. we begin our formal developments with a derivation of exact expressions for the zeroth , first and second spectral moments of the retarded and lesser green s functions ; the analysis is performed for the spinless falicov - kimball model and the spin one - half hubbard model in an external arbitrary time - dependent spatially uniform electric field . the hamiltonian for both models has the following form ( in the absense of an external field ) : @xmath0 in the case of the falicov - kimball model,@xcite the hamiltonian in eq . ( [ h ] ) describes a system which consists of two kinds of spinless electrons : itinerant @xmath1-electrons with a nearest neighbor hopping matrix @xmath2 and localized @xmath3-electrons with a hopping matrix equal to zero ( @xmath4 ) . we normally take the hopping matrix to be between nearest neighbors only , but this is not a requirement . we do assume the matrix elements are all real and that the hopping matrices are hermitian . the on - site interaction between the two electrons is equal to @xmath5 . double occupation by a @xmath1 or @xmath3 electron is forbidden by the pauli exclusion principle . the chemical potentials are @xmath6 and @xmath7 for the @xmath1- and @xmath3-electrons , respectively . we will set @xmath8 in our calculations ; it plays no role in the spectral moments of the @xmath1 particles , which we will be evaluating in this contribution . the spectral moments of the @xmath3-electrons in equilibrium were worked out in ref . . in the case of the hubbard model,@xcite the hamiltonian in eq . ( [ h ] ) describes a system of spin - up @xmath9-electrons , and spin - down @xmath10-electrons with equal hopping matrix elements @xmath11 and chemical potentials @xmath12 . the local coulomb repulsion between electrons with different spins is @xmath5 . the electric field @xmath13 can be described by a vector potential @xmath14 in the hamiltonian or temporal gauge ( where the scalar potential vanishes ) : @xmath15 we assume that vector potential @xmath14 is smooth enough , that the magnetic field produced by @xmath14 can be neglected . for simplicity we shall assume that the vector potential is spatially uniform ( independent of * r * ) . note that if we have any time dependence to the electric field , then neglecting the magnetic field will violate maxwell s equations . in most situations these magnetic field effects are small enough that they can be neglected in a first analysis , and added back later via either perturbative or gradient - based approaches . the electric field is coupled to the electrons via the peierls substitution,@xcite which involves modifying the hopping matrix elements by a phase that depends on the line integral of the vector potential : @xmath16 \nonumber\\ & = & t_{ij}\exp\left [ -\frac{ie { \bf a}(t)}{\hbar c } \cdot ( { \bf r}_{i}-{\bf r}_{j } ) \right ] , \label{peierls1}\end{aligned}\ ] ] @xmath17 \nonumber\\ & = & t_{ij}^{f}\exp\left [ -\frac{ie { \bf a}(t)}{\hbar c } \cdot ( { \bf r}_{i}-{\bf r}_{j } ) \right ] . \label{peierls2}\end{aligned}\ ] ] the second line in each equation follows for spatially uniform vector potentials . note that the hamiltonian in a field , @xmath18 , is identical in form to that shown in eq . ( [ h ] ) , but it uses the hopping matrices in eqs . ( [ peierls1 ] ) and ( [ peierls2 ] ) . note also that @xmath19 for the falicov - kimball model . the `` peierls substituted '' hamiltonian in eq . ( [ h ] ) , with the hopping matrix elements in eqs . ( [ peierls1 ] ) and ( [ peierls2 ] ) , has a simple form in the momentum representation : @xmath20 c_{{\bf k}}^{\dagger}c_{{\bf k}}\nonumber\\ & + & \sum_{{\bf k } } \left[\epsilon^{f}\left({\bf k}-\frac{e{\bf a}(t)}{\hbar c}\right ) -\mu_{f}\right ] f_{{\bf k}}^{\dagger}f_{{\bf k } } \nonumber \\ & + & u\sum_{{\bf p},{\bf k},{\bf q } } f_{{\bf p}+{\bf q}}^{\dagger}c_{{\bf k}-{\bf q}}^{\dagger } c_{{\bf k}}f_{{\bf p } } , \label{hk}\end{aligned}\ ] ] where the fermionic creation and annihilation operators now create or annihilate electrons with well - defined momentum . the free electron energy spectra in eq . ( [ hk ] ) satisfy @xmath21 , \nonumber\end{aligned}\ ] ] for the hubbard model . in the case of the falicov - kimball model , the @xmath22 term vanishes . we shall consider the spectral moments for the retarded @xmath23 and the lesser @xmath24 green s functions ; the symbol @xmath25 is the anticommutator and the operators @xmath26 and @xmath27 are in the heisenberg representation , where all time dependence is carried by the operators and the states are time - independent . any heisenberg representation operator @xmath28 is connected with a corresponding schr " odinger representation operator @xmath29 via @xmath30 e^{(i/\hbar ) \mathcal{h}(0)(t - t_{0})}o_{s } \nonumber \\ \times e^{-(i/\hbar ) \mathcal{h}(0)(t - t_{0 } ) } \left [ \mathcal{t } \exp\left\{-(i/\hbar)\int_{t_{0}}^{t}d{\bar t}\mathcal{h}_{i}({\bar t } ) \right\ } \right ] , \label{heisenbergoperator}\end{aligned}\ ] ] where @xmath31 is the time - independent part of the hamiltonian [ in eq . ( [ h ] ) with hopping matrix elements given by their field - free constant values ] , and @xmath32 is the time - dependent part of the hamiltonian , which includes the interaction with an external field [ but expressed in the interaction representation as detailed in eq . ( [ interactionoperator ] ) below ] : that is , we define the time - dependent piece in the schr " odinger representation via @xmath33 and then re - express in the interaction representation . note that the interaction representation operator is defined to be the schr " odinger representation operator evolved under the time - independent hamiltonian [ _ i.e. _ , the middle three terms in eq . ( [ heisenbergoperator ] ) ] . hence , the time - dependent piece of the hamiltonian in the interaction representation is expressed by @xmath34 in terms of the schr " odinger operator ; in this form , there is the bare time dependence arising from the time dependence of the fields , plus the time dependence inherited by the operators , as we go from the schr " odinger representation to the interaction representation . the symbol @xmath35 ( @xmath36 ) in eq . ( [ heisenbergoperator ] ) is the time - ordering ( anti - time - ordering ) operator . we prepare our system to be in equilibrium prior to the field being turned on , hence the quantum statistical averages in eqs . ( [ gr ] ) and ( [ gl ] ) are defined with respect to the zero - field ( equilibrium ) hamiltonian @xmath31 : @xmath37 / \mathcal{z } , \label{average}\ ] ] where the partition function satisfies @xmath38 , \label{z}\ ] ] and @xmath39 is the inverse temperature of the original equilibrium distribution . as was already mentioned in section i , the retarded and the lesser green s functions form an independent green s function basis . any other green s function can be expressed in terms of these two functions . this is in contrast to the equilibrium case , where only one green s function is independent ( because the fermi - dirac distribution function is determined by the equilibrium condition ) . calculations in the nonequilibrium case are complicated by the fact that the green s functions in eqs . ( [ gr ] ) and ( [ gl ] ) are functions of two time variables , contrary to the equilibrium case , where they only depend on the relative time difference ( because the equilibrium system is time - translation invariant ) . it is convenient to transform the two - time dependence of the green s functions from @xmath40 and @xmath41 to wigner coordinates , which use the average time @xmath42 and the relative time @xmath43 ( do not confuse the average time @xmath44 with the temperature @xmath45 ) . next , the relative time dependence is fourier transformed to a frequency , and the additional ( average ) time evolution of different quantities is described by the average time coordinate @xmath44 ; in equilibrium , there is no @xmath44 dependence . for example , the spectral function for the retarded and the lesser green s functions can be defined as @xmath46 where @xmath47 is the respective green s function from eq . ( [ gr ] ) or ( [ gl ] ) with @xmath48 and @xmath49 ; @xmath50 is equal to @xmath51 for the retarded green s function and @xmath52 for the lesser green s function so that the spectral functions are nonnegative in equilibrium . in general , the spectral function depends on the average time , because the system no longer has time - translation invariance when a field is turned on at a specific time . we define the @xmath53th moment of the retarded and lesser spectral function [ in eq . ( [ ar < ] ) ] @xmath54 to be @xmath55 it is easy to show that this expression is equivalent to @xmath56 begin by noting that @xmath57 so that @xmath58 substituting eq . ( [ munr<4 ] ) into the first line of eq . ( [ munr < ] ) then yields eq . ( [ munr<2 ] ) . before proceeding with the evaluation of analytical expressions for the spectral moments from eq . ( [ munr<2 ] ) , we note that the integration over frequency in eq . ( [ munr<2 ] ) can be evaluated , yielding the following expression , which connects the spectral moments to the derivative of the green s function with respect to relative time @xmath59 at zero relative time : @xmath60_{t=0^{+ } } . \label{munr<5}\ ] ] this formula assumes that the green s function is a differentiable function , which is true in most cases of interest . despite the fact that the expressions in eqs . ( [ munr<2 ] ) and ( [ munr<5 ] ) are formally equivalent , it is preferable to use one or the other in specific cases . in the case of the retarded green s function , the well - known expression for the zeroth spectral moment can be found from eqs . ( [ munr<5 ] ) and ( [ gr ] ) : @xmath61_{t=0^{+}}\nonumber\\ & = & -2{\rm im}\left [ -i\theta ( t ) \left\langle\left\ { c_{{\bf k}}\left ( t+\frac{t}{2}\right ) , c_{{\bf k}}^{\dagger}\left ( t-\frac{t}{2}\right)\right\ } \right\rangle \right]_{t=0^{+ } } \nonumber \\ & = & \left\langle\left\ { c_{{\bf k}}(t ) , c_{{\bf k}}^{\dagger}(t)\right\}\right\rangle = 1 . \label{mu0r}\end{aligned}\ ] ] in the derivation of eq . ( [ mu0r ] ) , we used the anticommutation relation for heisenberg operators and the fact that the theta function is equal to @xmath62 when its argument is equal to zero . it is more convenient to use eqs . ( [ munr<2 ] ) and ( [ gr ] ) to evaluate the expression for the first moment of @xmath63 : @xmath64 . \label{mu1rderiv}\end{aligned}\ ] ] taking the time derivative in eq . ( [ mu1rderiv ] ) gives @xmath65 . \label{mu1rderiv2}\end{aligned}\ ] ] the first term in eq . ( [ mu1rderiv2 ] ) is equal to zero , because the integral over time is equal to @xmath66 , therefore its imaginary part vanishes . the second term in eq . ( [ mu1rderiv2 ] ) can be simplified by performing an integration over @xmath67 and replacing the time derivatives of the operators by their commutators with the hamiltonian , according to the heisenberg equation of motion @xmath68 $ ] , where @xmath69 is the total hamiltonian including the effects of the time - dependent field . this yields @xmath70,c_{{\bf k}}^{\dagger}(t ) \right\ } \right\rangle\right .\nonumber\\ & -&\left . \left\langle\left\ { c_{{\bf k}}(t),[c_{{\bf k}}^{\dagger}(t),\mathcal{h}(t ) ] \right\}\right\rangle\right ) . \label{mu1rder3}\end{aligned}\ ] ] evaluation of the commutators of the fermi - operators with the hamiltonian and the subsequent anticommutators in eq . ( [ hk ] ) gives the following expression for the first spectral moment of the retarded green s function @xmath71 where @xmath72 is the average number of @xmath3 ( @xmath73)-electrons in the system ; this number of electrons does not depend on the average or the relative time , because the total electron number for each species of electron is conserved . similarly , the expression for the second moment of the retarded green s function can be found from eqs . ( [ munr<2 ] ) and ( [ gr ] ) : @xmath74,\mathcal{h}(t ) ] , c_{{\bf k}}^{\dagger}(t ) \right\}\right\rangle \right . \nonumber \\ & -&2\left\langle\left\ { [ c_{{\bf k}}(t),\mathcal{h}(t)],[c_{{\bf k}}^{\dagger}(t ) , \mathcal{h}(t ) ] \right\}\right\rangle\nonumber\\ & + & \left . \left\langle\left\ { c_{{\bf k}}(t),[[c_{{\bf k}}^{\dagger}(t ) , \mathcal{h}(t)],\mathcal{h}(t ) ] \right\}\right\rangle \right ) . \label{mu2rderiv2}\end{aligned}\ ] ] details of the derivation are presented in the appendix . evaluating the commutators and anticommutators in eq . ( [ mu2rderiv2 ] ) gives @xmath75^{2}\label{mu2r } \\ & + & 2u\left[\epsilon \left({\bf k}-\frac{e{\bf a}(t)}{\hbar c}\right ) -\mu\right]n_{f}+u^{2}n_{f}. \nonumber \end{aligned}\ ] ] the moments of the local retarded green s function @xmath76 are obtained by summing the corresponding spectral moment functions @xmath77 over @xmath78 @xmath79 performing the summations for eqs . ( [ mu1r1 ] ) and ( [ mu2r ] ) yields the following local moments : @xmath80 these results coincide with those derived previously for the hubbard model in the equilibrium case@xcite . since the hopping matrix is always chosen to be traceless in our models , the sum of the energy @xmath81 in eqs . ( [ mu1r1 ] ) and ( [ mu2r ] ) is equal to zero . the expression for the zeroth local moment @xmath82 has the same form as the expression for the zeroth spectral moment in eq . ( [ mu0r ] ) , since the zeroth spectral moment is momentum - independent . hence , the zeroth and the first two local moments of the retarded green s function in an arbitrary external time - dependent homogeneous electric field are all time - independent ! this is a nontrivial result because the retarded green s function strongly depends on the average time . in particular , the retarded green s function is an oscillating function of time@xcite when @xmath83 . furthermore , the moments do not depend on the electric field at half - filling because the chemical potential is not changed by the field ! it is nt obvious whether the chemical potential would be changed by the field off of half - filling . in the case of half - filling , where @xmath84 and @xmath85 , the expressions in eqs . ( [ mu0r ] ) , ( [ mu1rdensity ] ) and ( [ mu2rdensity ] ) acquire an even simpler form : @xmath86 if one examines the moments for gauge - invariant green s functions,@xcite then the local moments are unchanged , and the spectral function moments are modified by a time - dependent shift of the momentum wavevector . we do nt include those formulas here , because they just involve such a simple shift . the corresponding moments of the lesser green s functions are found by a similar analysis . using eqs . ( [ munr<5 ] ) and ( [ gl ] ) we find @xmath87 c_{{\bf k}}(t)\right\rangle\right . \nonumber\\ & -&\left . \left\langle c_{{\bf k}}^{\dagger}(t)[c_{{\bf k}}(t),\mathcal{h}(t ) ] \right\rangle \right ) , \label{mu1<deriv}\\ \mu_{2}^{<}({\bf k},t)&=&\frac{1}{2}{\rm re } \left ( \left\langle [ [ c_{{\bf k}}^{\dagger}(t),\mathcal{h}(t)],\mathcal{h}(t ) ] c_{{\bf k}}(t)\right\rangle\right . \nonumber\\ & -&\left.2\left\langle [ c_{{\bf k}}^{\dagger}(t),\mathcal{h}(t ) ] [ c_{{\bf k}}(t),\mathcal{h}(t)]\right\rangle \right . \nonumber \\ & + & \left . k}}^{\dagger}(t)[[c_{{\bf k}}^{\dagger}(t ) , \mathcal{h}(t)],\mathcal{h}(t ) ] \right\rangle \right ) , \label{mu2<deriv}\end{aligned}\ ] ] where @xmath88 is the momentum distribution function for the @xmath1 @xmath89-electrons . note that a commutator term depending on the derivative of the hamiltonian with respect to time can be shown to cancel , so it is not included in the second moment expression above . evaluation of the commutators of the operators with the hamiltonian [ in eq . ( [ hk ] ) ] in eqs . ( [ mu1<deriv ] ) and ( [ mu2<deriv ] ) gives @xmath90n_{c}({\bf k},t ) \nonumber \\ + u\sum_{{\bf p},{\bf q}}\left(\left\langle f_{{\bf p}+{\bf q}}^{\dagger } c_{{\bf k}-{\bf q}}^{\dagger } c_{{\bf k}}f_{{\bf p}}\right\rangle + \left\langle f_{{\bf p}+{\bf q}}^{\dagger } c_{{\bf k}}^{\dagger } c_{{\bf k}+{\bf q}}f_{{\bf p}}\right\rangle\right ) . \label{mu1<}\end{aligned}\ ] ] and @xmath91^{2}n_{c}({\bf k},t ) + \frac{3}{2}u\left[\epsilon \left({\bf k}-\frac{e{\bf a}(t)}{\hbar c}\right ) -\mu\right ] \sum_{{\bf p},{\bf q}}\left(\left\langle f_{{\bf p}+{\bf q}}^{\dagger } c_{{\bf k}-{\bf q}}^{\dagger } c_{{\bf k}}f_{{\bf p}}\right\rangle + \left\langle f_{{\bf p}+{\bf q}}^{\dagger } c_{{\bf k}}^{\dagger } c_{{\bf k}+{\bf q}}f_{{\bf p}}\right\rangle\right ) \nonumber \\ & + & \frac{1}{2}u\sum_{{\bf p},{\bf q } } \left[\epsilon \left({\bf k}-{\bf q}-\frac{e{\bf a}(t)}{\hbar c}\right ) -\mu\right ] \left\langle f_{{\bf p}+{\bf q}}^{\dagger}c_{{\bf k}-{\bf q}}^{\dagger } c_{{\bf k}}f_{{\bf p}}\right\rangle + \frac{1}{2}u\sum_{{\bf p},{\bf q } } \left[\epsilon \left({\bf k}+{\bf q}-\frac{e{\bf a}(t)}{\hbar c}\right ) -\mu\right ] \left\langle f_{{\bf p}+{\bf q}}^{\dagger}c_{{\bf k}}^{\dagger } c_{{\bf k}+{\bf q}}f_{{\bf p}}\right\rangle \nonumber \\ & + & \frac{1}{2}u\sum_{{\bf p},{\bf q } } \epsilon^{f}\left({\bf p}+{\bf q}-\frac{e{\bf a}(t)}{\hbar c}\right ) \left [ \left\langle f_{{\bf p}+{\bf q}}^{\dagger}c_{{\bf k}}^{\dagger } c_{{\bf k}+{\bf q}}f_{{\bf p}}\right\rangle - \left\langle f_{{\bf p}+{\bf q}}^{\dagger}c_{{\bf k}-{\bf q}}^{\dagger } c_{{\bf k}}f_{{\bf p}}\right\rangle \right ] \nonumber \\ & -&\frac{1}{2}u\sum_{{\bf p},{\bf q } } \epsilon^{f}\left({\bf p}-\frac{e{\bf a}(t)}{\hbar c}\right ) \left [ \left\langle f_{{\bf p}+{\bf q}}^{\dagger}c_{{\bf k}}^{\dagger } c_{{\bf k}+{\bf q}}f_{{\bf p}}\right\rangle - \left\langle f_{{\bf p}+{\bf q}}^{\dagger}c_{{\bf k}-{\bf q}}^{\dagger } c_{{\bf k}}f_{{\bf p}}\right\rangle \right ] \nonumber \\ & + & \frac{1}{2}u^{2}\sum_{{\bf p},{\bf q},{\bf p},{\bf q } } \left [ \left\langle f_{{\bf p}+{\bf q}}^{\dagger}f_{{\bf p } } f_{{\bf p}+{\bf q}}^{\dagger}f_{{\bf p } } c_{{\bf k}-{\bf q}-{\bf q}}^{\dagger}c_{{\bf k}}\right\rangle + 2\left\langle f_{{\bf p}+{\bf q}}^{\dagger}f_{{\bf p } } f_{{\bf p}+{\bf q}}^{\dagger}f_{{\bf p } } c_{{\bf k}-{\bf q}}^{\dagger}c_{{\bf k}+{\bf q}}\right\rangle \right . \nonumber \\ & + & \left . \left\langle f_{{\bf p}+{\bf q}}^{\dagger}f_{{\bf p } } f_{{\bf p}+{\bf q}}^{\dagger}f_{{\bf p } } c_{{\bf k}}^{\dagger}c_{{\bf k}+{\bf q}+{\bf q}}\right\rangle \right ] . \label{mu2<}\end{aligned}\ ] ] in eq . ( [ mu2 < ] ) , we have suppressed the time label @xmath44 corresponding to the time at which all operators are evaluated . we continue to suppress this time label in some equations below ; this should not cause any confusion . the expressions in eqs . ( [ mu0<])([mu2 < ] ) for the lesser spectral moments are more complicated than the corresponding retarded moments . however , they simplify in the case of the local green s function , where we find @xmath92n_{c}({\bf k},t)\nonumber\\ & + & 2u\sum_{i}\left\langle f_{i}^{\dagger}f_{i}c_{i}^{\dagger}c_{i}\right\rangle , \label{mu1<density}\\ { \tilde \mu}_{2}^{<}(t ) & = & 2\sum_{{\bf k } } \left [ \epsilon \left({\bf k}-\frac{e{\bf a}(t)}{\hbar c}\right)-\mu \right]^{2}n_{c}({\bf k},t ) \nonumber \\ & + & 2u\sum_{{\bf k},{\bf p},{\bf q } } \left [ \epsilon\left({\bf k}-\frac{e{\bf a}(t)}{\hbar c}\right)\right.\nonumber\\ & ~&\quad + \left . \epsilon\left({\bf k}-{\bf q}-\frac{e{\bf a}(t)}{\hbar c}\right ) \right ] \left\langle f_{{\bf p}+{\bf q}}^{\dagger}f_{{\bf p } } c_{{\bf k}-{\bf q}}^{\dagger}c_{{\bf k}}\right\rangle \nonumber \\ & -&2u(2\mu - u)\sum_{i}\left\langle f_{i}^{\dagger}f_{i}c_{i}^{\dagger}c_{i } \right\rangle , \label{mu2<density}\end{aligned}\ ] ] with @xmath93 being the time - independent particle density of the @xmath1 ( @xmath94)-electrons . in order to save space , and make the equations more transparent , we use a mixed real - space / momentum - space representation for the operators in eqs . ( [ mu1<density ] ) and ( [ mu2<density ] ) . note that the first moment involves one correlation function and the second moment involves two correlation functions . note further that the value of the first moment in eq . ( [ mu1<density ] ) is equal to twice the average value of the hamiltonian . the last term in eq . ( [ mu2<density ] ) is equal to zero in the case of half - filling ( @xmath95 ) . the second two moments of the lesser green s functions do appear to depend both on the average time and the electric field ( although our empirical evidence at half filling suggests the second moment may be independent of average time see the numerical results below ) . there is an interesting observation that can be made about the first moment , and its relation to the current driven by the electric field and the phenomenon of bloch oscillations . in the limit where @xmath5 is small , one can evaluate the correlation function in eq . ( [ mu1<density ] ) via a mean - field theory decoupling ( @xmath96 ) . for example , at half filling , the first moment will be equal to twice the average kinetic energy ( including the shift by the vector potential needed to construct the actual kinetic energy from the bandstructure ) plus a correction of order @xmath97 . if the current oscillates , we expect the average kinetic energy to oscillate as well . hence , for small @xmath5 there is a correlation between oscillations in the first moment of the local green s function and oscillations of the current . the correlation function @xmath98 that appears in eqs . ( [ mu1<density ] ) and ( [ mu2<density ] ) can be determined for the falicov - kimball or hubbard model via the equation of motion , because it is related to the total energy of the hamiltonian . to show how this works , we first provide the derivation for the equilibrium case using an imaginary - time formalism . begin with the definition of the green s function in real space @xmath99 with a similar result for the spin - down electrons in the hubbard model . here we have @xmath100 . taking the imaginary - time derivative of the local green s function gives @xmath101c^\dagger_i(0)\right\rangle\nonumber\\ & = & -\delta(\tau)+t\sum_\delta g_{i+\delta i}(\tau)+\mu g_{ii}(\tau)\nonumber\\ & + & u\left\langle\mathcal{t}f_i^\dagger f_ic_i(\tau)c_i^\dagger(0)\right\rangle , \label{eq : eom}\end{aligned}\ ] ] where the symbol @xmath102 denotes the translation vector to a nearest - neighbor site and @xmath103 as a subscript refers to the lattice site that is the nearest neighbor of site @xmath104 translated by the nearest - neighbor translation vector @xmath102 . hence we determine the correlation function via @xmath105 . \nonumber \end{aligned}\ ] ] using the matsubara frequency representation @xmath106 with @xmath107 the fermionic matsubara frequency , allows us to determine a simple expression for the correlation function . note that @xmath108 and that the green s function in momentum space satisfies @xmath109 $ ] , to find that the correlation function simplifies to @xmath110 in the limit of infinite dimensions , the self - energy is a local function , and hence has no momentum dependence . then the sum over momentum can be performed by changing from a sum over momentum to an integral over the noninteracting density of states . this produces the local green s function , and we are left with the final form for dmft : @xmath111 in numerical calculations , it is more convenient to evaluate the summation in eq . ( [ eq : corr_fun3 ] ) via the formally equivalent expression with the hartree - fock contribution to the self - energy removed @xmath112g_n , \label{eq : corr_fun4}\ ] ] because the matsubara summation converges faster . in the nonequilibrium case , one can perform a similar analysis , but now one has to work with the nonequilibrium green s functions as functions of real time variables . using the standard equations of motion , and definitions for nonequilibrium green s functions , one can show , after some significant algebra , that the correlation function can be expressed as @xmath113 \nonumber\\ & \times&g^<_{\bf k}(t_1,t_2)\biggr |_{t_2=t_1 } \nonumber\\ & = & -i\sum_{\bf k}\int dt\left [ \sigma^r_{\bf k}(t_1,t)g^<_{\bf k}(t , t_1)\right . \nonumber\\ & + & \left . \sigma^<_{\bf k}(t_1,t)g^a_{\bf k}(t , t_1)\right ] , \label{eq : corr_noneq}\end{aligned}\ ] ] where the final result is written in terms of retarded , lesser and advanced green s functions and self - energies in the presence of the field . in the dmft limit , the self - energies have no momentum dependence . this expression appears like it can have average time dependence , but we can not say that it definitely does , because there could be a cancellation of the time dependence . as a check of eq . ( [ eq : corr_noneq ] ) , we evaluate it in equilibrium , to show it yields the same result as eq . ( [ eq : corr_fun3 ] ) . when we are in equilibrium , the correlation function is independent of time and the green s functions and self - energies depend only on the time difference of their two time variables . hence , we can perform a fourier transform by using the convolution theorem to transform eq . ( [ eq : corr_noneq ] ) into @xmath114 . \label{eq : corr_fun_eq}\end{aligned}\ ] ] using the fact that the lesser functions satisfy @xmath115 and @xmath116 in equilibrium , allows us to transform eq . ( [ eq : corr_fun_eq ] ) into @xmath117 , \label{eq : corr_fun_eq2}\ ] ] which is equal to the analytic continuation of eq . ( [ eq : corr_fun3 ] ) from the imaginary axis to the real axis . the correlation function @xmath118 which enters eq . ( [ mu2<density ] ) is more complicated to evaluate , because it can not be expressed in terms of a simple equation of motion for the single - particle green s functions . because of it s complex nature , we will evaluate it only for the falicov - kimball model in equilibrium . while it is certainly true that it can be evaluated for the falicov - kimball model in the presence of a field , some of the formal details become quite complicated , and take us away from the main theme of this work , so we do not perform such an analysis here . our starting point , then , is the operator average @xmath119 \left\langle f_{{\bf p}+{\bf q}}^{\dagger}f_{{\bf p } } c_{{\bf k}-{\bf q}}^{\dagger}c_{{\bf k}}\right\rangle , \label{eq : 2corr_fun}\ ] ] where we have set the vector potential @xmath120 equal to zero . we use a fourier transform to express the localized electrons in terms of their real - space operators . then elitzur s theorem@xcite ensures that the operator expectation value vanishes if the two localized electrons are not at the same lattice site ( _ i.e. _ , there is no spontaneous hybridization in the falicov - kimball model for nonzero temperature ) . this allows us to perform the summation over the momentum variable @xmath121 and gives us @xmath122 e^{i{\bf q}\cdot{\bf r_i } } \left\langle f_i^{\dagger}f_i c_{{\bf k}-{\bf q}}^{\dagger}c_{{\bf k}}\right\rangle . \label{eq : 2corr_fun2}\ ] ] next , we express the bandstructure in terms of the summation over nearest neighbor translation vectors @xmath102 : @xmath123/\sqrt{d}$ ] , and introduce fourier transforms for the itinerant electrons to real space . this allows us to sum over the remaining momenta , yielding @xmath124 . \label{eq : 2corr_fun3}\ ] ] the statistical averages in eq . ( [ eq : 2corr_fun3 ] ) have already been evaluated.@xcite the procedure is to imagine adding a small field @xmath125 to the hamiltonian , and evaluate the expectation value with the localized particle number via a derivative with respect to the field strength @xmath126 ( then set the field to zero to evaluate the average ) . this gives @xmath127 \left [ \left\langle c_i^{\dagger}c_{i+\delta}\right\rangle+ \left\langle c_{i+\delta}^{\dagger}c_i\right\rangle\right ] \nonumber\\ & = & -\frac{t^*}{\sqrt{d}}\sum_{i\delta}\left [ \frac{1}{\beta } \frac{\partial}{\partial h_i}+\langle w_i\rangle\right ] \nonumber\\ & ~&\times \left [ g_{i+\delta i}(\tau\rightarrow 0 ^ -)+g_{ii+\delta}(\tau\rightarrow 0 ^ - ) \right ] . \label{eq : 2corr_fun4}\end{aligned}\ ] ] now we follow the derivation in ref . , which evaluates the derivatives from the following : @xmath128 g_{i+\delta i}(\tau\rightarrow 0 ^ -)\label{eq : 2corr_fun5}\\ & ~&=\frac{1}{\beta}\sum_n\sum_{jk } g_{i+\delta j}\left [ -\frac{1}{\beta } \frac{\partial}{\partial h_i}+\langle w_i\rangle\right ] g^{-1}_{jk}(i\omega_n)g_{ki}(i\omega_n)\nonumber\\ & ~&=\frac{1}{\beta}\sum_n\left [ g_{i+\delta i } \frac{1}{\beta } \frac{\partial}{\partial h}\sigma(i\omega_n ) g_{ii}(i\omega_n)+\langle f^\dagger f\rangle g_{ii}(i\omega_n)\right ] \nonumber\end{aligned}\ ] ] where the derivative with respect to the field @xmath129 acts on the local self - energy . after some long and complicated algebra , that derivative can be determined , which yields our final result for the correlation function @xmath130\sigma_n . \label{eq : 2corr_fun6}\end{aligned}\ ] ] this result is quite similar to the previous result for the other correlation function , except now we have an extra weighting factor of @xmath131 in the summation over momentum . in this section , we examine the time - dependence of the local green s function for the falicov - kimball model on an infinite - dimensional hypercubic lattice . we consider the case of half - filling with the system being coupled to an external homogeneous time - dependent electric field . the time dependence is taken to be particularly simple , at @xmath132 a constant field is instantly turned on . the formalism involves generalizing the dmft to the nonequilibrium case . the way to do this is based on a kadanoff - baym approach in real time , where continuous matrix operators are discretized along the kadanoff - baym contour , and operator manipulations are carried out on the discretized matrices using standard linear - algebra approaches . a short description of this technique , including a benchmark against the well - known equilibrium solutions has already appeared@xcite ; further details of this approach will appear elsewhere . to test our formulas for the first few moments , we compare numerically calculated local moments to the exact moments in eqs . ( [ mu0rdensityhalffilling][mu2rdensityhalffilling ] ) and ( [ mu0<density][mu2<density ] ) . the action for the falicov - kimball model is quadratic in the conduction electrons . hence the feynman path integral over the kadanoff - baym contour can be expressed by a determinant of a continuous matrix operator whose arguments are defined on the contour . since the concentration of static particles on each site is conserved , the trace over the fermionic variables can be straightforwardly taken . this is what allows the nonequilibrium dmft problem to be solved , but the technical details are complicated . the first thing that needs to be noted is that the self - energy remains local even in the presence of a field . this follows by applying langreth s rules@xcite to the perturbation theory , which state that every nonequilibrium diagram can be related to an equilibrium diagram , but now one must perform the analysis over the kadanoff - baym contour rather than over the finite imaginary time interval . since the perturbative analysis of the equilibrium problem shows the self - energy to be local@xcite in equilibrium , it remains local in the nonequilibrium case as well . we couple the system in eq . ( [ hk ] ) to an external electric field along the unit - cell diagonal direction in real space ; this yields the following vector potential for the electric field@xcite : @xmath133 the bandstructure for noninteracting electrons coupled to the electric field in eq . ( [ a ] ) has a simple form : @xmath134 with the energy functions defined to be @xmath135 and @xmath136 and @xmath137 is the renormalized hopping parameter@xcite in the limit of @xmath138 ; we take @xmath139 . because the green s functions now depend on two energies , the summation over the infinite - dimensional brillouin zone can be replaced by a double integral over a joint density of states for the two energies . this joint density of states ( dos ) is@xcite @xmath140 . \label{rho2}\end{aligned}\ ] ] the numerical integration over the joint dos is performed by an averaged gaussian integration with 54 and 55 points for each energy axis : @xmath141 where @xmath142 are gaussian weights which correspond to the @xmath143 energy points @xmath144 . since the green s functions often depend on the energy as @xmath145 , the gaussian quadrature rule in eq . ( [ gaussian ] ) fails to give correct results when @xmath1 is on the order of ( or larger than ) the inverse of the grid spacing of the energy points near @xmath146 . in this case , the sum over discrete points contains a systematic contribution of terms which do not cancel each other , and leads to an overestimated value to the integral . one way to efficiently correct this is to average two gaussian summations with numbers of gaussian points equal to @xmath143 and @xmath147 because the gaussian points interleave each other , and act like a step size about half as big as either sum alone , and they give somewhat better accuracy than performing the integral with @xmath148 points , because a subset of those points are at such large absolute value that the gaussian weight is small enough that it can be safely neglected . in order to calculate the nonequilibrium local green s functions , one needs to self - consistently solve a system of equations@xcite which connects these functions with the corresponding local self - energy @xmath149 and an effective dynamical mean - field @xmath150 ; these equations are similar in form to the equilibrium case,@xcite but now all the functions are discrete matrices of two time variables defined on the kadanoff - baym contour ( see fig . [ fig : contour ] ) . details of the algorithm and the nonequilibrium dmft equations will be given elsewhere@xcite . first , we present results for the local moments in equilibrium , when the system of the dmft equations is solved in the frequency representation using the brandt - mielsch approach@xcite ( for details , see ref . ) . plots of the local retarded and lesser dos for different values of @xmath5 are shown in figs . [ fig : ret_eq ] and [ fig : less_eq ] . the metal - insulator transition occurs at @xmath151 ; the insulator has anomalous properties because there is no real gap instead the dos is exponentially small in a gap region around the chemical potential , and vanishes only at the chemical potential . the moments for the retarded and lesser green s functions are calculated by directly integrating the green s functions multiplied by the corresponding power of frequency ; we use a step size of @xmath152 and a rectangular quadrature rule . the results for half - filling are presented in table [ tab1 ] . one can immediately see that the numerical results for the first moment of the lesser green s function are in an excellent agreement with the exact expressions for the moment ( the zeroth and second moments agreed exactly with their exact expressions ) . we also calculated the retarded moments and they agreed exactly with the exact values , so we do nt summarize them in a table . note that there appears to be a relation between the second moments of the retarded and lesser green s functions at half filling . this relation ceases to hold off of half filling . for example , in the case with @xmath153 , @xmath154 , and @xmath155 we find the following moments ( exact results in parentheses ) : @xmath156 @xmath157 ; @xmath158 @xmath159 ; @xmath160 @xmath161 ; and @xmath162 @xmath163 . these results are more indicative of the general case ( where @xmath164 ) . note that one needs to use many matsubara frequencies in the summations to get good convergence for the average kinetic energy and for the second correlation function ( we used 50,000 in this calculation with @xmath165 ) . the majority of our numerical error comes from the difficulty in exactly calculating those results ; indeed , the exact result , calculated from the operator averages on the matsubara frequency axis is probably less accurate than the direct integration of the moment on the real axis . one can improve this situation somewhat by working on the real axis to calculate the different operator averages , but we wanted to indicate the accuracy under the most challenging circumstances . we feel our final results are quite satisfactory , and indicate these sum rules do hold . [ cols="^,^,^,^,^,^,^ " , ] to conclude our numerical analysis , we consider the spectral moments of the lesser green s function in the case where it is approximated by the generalized kadanoff - baym ( gkb ) ansatz@xcite . the idea of the gkb is to represent the lesser green s function in terms of the distribution function ( determined by the @xmath166 limit of the lesser green s function ) and the full two - time retarded and advanced green s functions : @xmath167 . \label{kadanoffbaym}\end{aligned}\ ] ] the expression for the spectral moments of the lesser green s function , approximated by eq . ( [ kadanoffbaym ] ) , can be found by taking relative time derivatives of @xmath168 as shown in eq . ( [ munr<5 ] ) . similar to the derivation performed in section ii , one finds ( the hat denotes the gkb approximation ) @xmath169,c_{{\bf k}}^{\dagger}(t)\right\ } \right\rangle\right . \nonumber\\ & -&\left . \left\langle\left\ { c_{{\bf k}}(t),[c_{{\bf k}}^{\dagger}(t),h ] \right\}\right\rangle\right ) \nonumber \\ & + & { \rm im}\frac{\partial n_{c}({\bf k},t)}{\partial t } , \label{mu1<gkb}\\ \hat \mu_{2}^{<}({\bf k},t ) & = & \frac{1}{2}n_{c}({\bf k},t ) \left ( \left\langle\left\ { [ [ c_{{\bf k}}(t),h],h],c_{{\bf k}}^{\dagger}(t)\right\ } \right\rangle \right . \nonumber \\ & -&2\left . \left\langle\left\ { [ c_{{\bf k}}(t),h ] , [ c_{{\bf k}}^{\dagger}(t),h]\right\ } \right\rangle\right . \nonumber\\ & + & \left . \left\langle\left\ { c_{{\bf k}}(t),[[c_{{\bf k}}^{\dagger}(t),h],h ] \right\}\right\rangle\right ) \nonumber\\ & -&\frac{1}{2}{\rm re } \frac{\partial^{2}n_{c}({\bf k},t)}{\partial t^{2}}. \label{mu2<gkb}\end{aligned}\ ] ] comparison of the expressions in eqs . ( [ mu0<gkb])([mu2<gkb ] ) for the gkb moments with the corresponding exact expressions in eqs . ( [ mu0r ] ) , ( [ mu1rder3 ] ) and ( [ mu2rderiv2 ] ) for the moments of @xmath170 allows us to connect the lesser gkb spectral moments with the exact retarded spectral moments : @xmath171 for @xmath172 , 1 , and 2 . since the time derivatives of the momentum distribution function are real valued , the last term in eq . ( [ gkbretarded ] ) is nonzero only for @xmath173 . thus , the lesser gkb spectral moments ( with @xmath172 , 1 , and 2 ) are equal to the corresponding retarded spectral moments multiplied by @xmath174 plus a term involving the second time derivative of the momentum distribution function for the case of @xmath173 . comparison of eq . ( [ gkbretarded ] ) and the exact expressions for the retarded spectral moments [ in eqs . ( [ mu0r ] ) , ( [ mu1r1 ] ) and ( [ mu2r ] ) ] with the exact expressions for the lesser moments [ in eqs . ( [ mu0 < ] ) , ( [ mu1 < ] ) and ( [ mu2 < ] ) ] allows us to conclude that the lesser spectral moments for the gkb approximated green s function correspond to an approximation of the exact spectral moments by evaluating the operator averages with a mean - field approximation ( the second moment contains an additional term with the second time derivative of the momentum distribution function ) . this indicates that the gkb approximation will fail as the correlations increase , because it produces the wrong moments to the spectral functions . note that this does not say the gkb approach is a mean field theory approach , it is not . in this paper , we derived a sequence of spectral moment sum rules for the retarded and lesser green s functions of the falicov - kimball and hubbard models . our analysis holds in equilibrium and in the nonequilibrium case of a spatially uniform , but time - dependent external electric field being applied to the system . our results are interesting , because they show there is no average time dependence nor electric field dependence to the first three moments of the local retarded green s functions . this implies that the value , slope and curvature of the local retarded green s functions , as functions of the relative time , do not depend on the average time or the field . such a result extends the well - known result that the total spectral weight ( zeroth moment ) of the green s function is independent of the field . it also implies that one will only see deviations of green s functions from the equilibrium results when the relative time becomes large . such an observation is quite useful for quantifying the accuracy of nonequilibrium calculations . we showed some numerical results illustrating this effect for nonequilibrium dmft calculations in the falicov - kimball model . the case for the lesser green s function is more complicated , and there is both average time dependence and field dependence apparent in those moments , although it appears that the second moment at half filling may not depend on average time . we also examined a common approximation employed in nonequilibrium calculations , the so - called generalized kadanoff - baym ansatz . we find the moments in that case are similar in form to the exact moments , except they involve a mean - field - theory decoupling of correlation function expectation values when one evaluates the operator averages that yield the sum - rule values . this implies that the gkb approximation must fail as the correlations increase , and such a mean - field decoupling becomes inaccurate , because it will have the wrong spectral moments . we hope that use of these spectral moments will become common in nonequilibrium calculations in order to quantify the errors of the calculations for small times . we believe they can be quite valuable in checking the fidelity of numerical calculations and of different kinds of approximate solutions . we would like to acknowledge support by the national science foundation under grant number dmr-0210717 and by the office of naval research under grant number n00014 - 05 - 1 - 0078 . we show the explicit steps needed to derive eq . ( [ mu2rderiv2 ] ) by using eqs . ( [ munr < ] ) and ( [ gr ] ) for the second spectral moment of the retarded green s function . in this case , the second spectral moment is equal to this expression is equivalent to @xmath176 \nonumber \\ & + & \frac{1}{\pi } \int_{-\infty}^{\infty}d\omega { \rm im}\int_{-\infty}^{\infty } dt e^{i\omega t } i\theta ( t ) \left [ \left\langle\left\ { i^{2}\frac{\partial^{2}}{\partial t^{2 } } c_{{\bf k}}\left(t+\frac{t}{2}\right ) , c_{{\bf k}}^{\dagger}\left ( t-\frac{t}{2}\right ) \right\}\right\rangle \right . \nonumber \\ & + & \left . 2\left\langle\left\{i\frac{\partial}{\partial t } c_{{\bf k}}\left(t+\frac{t}{2}\right ) , i\frac{\partial}{\partial t } c_{{\bf k}}^{\dagger}\left ( t-\frac{t}{2}\right ) \right\}\right\rangle + \left\langle\left\ { c_{{\bf k}}\left(t+\frac{t}{2}\right ) , i^{2}\frac{\partial^{2}}{\partial t^{2 } } c_{{\bf k}}^{\dagger}\left ( t-\frac{t}{2}\right ) \right\}\right\rangle \label{a1}\end{aligned}\ ] ] the first two integrals in eq . ( [ a1 ] ) are equal to zero , that is , @xmath177 and @xmath178 = 0 . \label{a3}\end{aligned}\ ] ] to prove eq . ( [ a2 ] ) , we note that the delta - function derivative satisfies @xmath179 hence , we can transfer the time derivative of the delta function into a time derivative of the other factors in the integrand . this time derivative has two terms : ( i ) the first term is the derivative of the exponential , which introduces an additional factor of @xmath180 ( the operator average then becomes trivial when we set @xmath166 , since the anticommutator is equal to 1 ) and ( ii ) the second term which involves derivatives of the creation and annihilation operators . performing the integration over @xmath59 by using the delta function then yields @xmath181 , c_{{\bf k}}^{\dagger}(t ) \right\}\right\rangle\right . \nonumber \\ & ~&-\left . \left\langle\left\ { c_{{\bf k}}(t ) , [ c_{{\bf k}}^{\dagger}(t),\mathcal{h(t ) } ] \right\}\right\rangle \right]\biggr ) , \end{aligned}\ ] ] where we replaced derivatives with respect to time by commutators with the hamiltonian . the first term has no imaginary part , so it vanishes , as do the second two terms , since one can easily show the difference of the two operators is hermitian , and hence has a real expectation value . this completes the proof of eq . ( [ a2 ] ) . to prove ( [ a3 ] ) , we first perform the integration over @xmath59 . the result is equal to @xmath182 \nonumber \\ & = & -\frac{1}{\pi } \int_{-\infty}^{\infty}d\omega { \rm i m } \frac{1}{2 } \left\langle\left [ \left\ { [ c_{{\bf k}}(t),\mathcal{h}(t ) ] , c_{{\bf k}}^{\dagger } ( t)\right\ } \right.\right.\nonumber\\ & -&\left.\left . \left\{c_{{\bf k}}(t),[c_{{\bf k}}^{\dagger}(t),\mathcal{h}(t)]\right\ } \right]\right\rangle . \label{a7}\end{aligned}\ ] ] the operator in the second line of eq . ( [ a7 ] ) is the same operator we saw above ; it is hermitian so the imaginary part of the statistical average vanishes . this proves ( [ a3 ] ) . to complete the derivation of the second spectral moment , we note that only the last integral term in eq . ( [ a1 ] ) can be nonzero . substituting the operator time derivatives by their commutators with the hamiltonian finally gives the expression in eq . ( [ mu2rderiv2 ] ) of the second spectral moment for the retarded green s function . j. k. freericks , v. m. turkowski and v. zlati ' c , `` real - time formalism for studying the non - linear response of `` smart '' materials to an electric field . '' to be published in _ proceedings of the 2005 users group conference , ieee computer society _ , nashville , tennessee ( 2005 ) .

we propose distributed algorithms to automatically deploy a team of mobile robots to partition and provide coverage of a non - convex environment . to handle arbitrary non - convex environments , we represent them as graphs . our partitioning and coverage algorithm requires only short - range , unreliable pairwise `` gossip '' communication . the algorithm has two components : ( 1 ) a motion protocol to ensure that neighboring robots communicate at least sporadically , and ( 2 ) a pairwise partitioning rule to update territory ownership when two robots communicate . by studying an appropriate dynamical system on the space of partitions of the graph vertices , we prove that territory ownership converges to a pairwise - optimal partition in finite time . this new equilibrium set represents improved performance over common lloyd - type algorithms . additionally , we detail how our algorithm scales well for large teams in large environments and how the computation can run in anytime with limited resources . finally , we report on large - scale simulations in complex environments and hardware experiments using the player / stage robot control system .

1011.1939

coordinated networks of mobile robots are already in use for environmental monitoring and warehouse logistics . in the near future , autonomous robotic teams will revolutionize transportation of passengers and goods , search and rescue operations , and other applications . these tasks share a common feature : the robots are asked to provide service over a space . one question which arises is : when a group of robots is waiting for a task request to come in , how can they best position themselves to be ready to respond ? the distributed _ environment partitioning problem _ for robotic networks consists of designing individual control and communication laws such that the team divides a large space into regions . typically , partitioning is done so as to optimize a cost function which measures the quality of service provided over all of the regions . _ coverage control _ additionally optimizes the positioning of robots inside a region as shown in fig . [ fig : cover_example ] . this paper describes a distributed partitioning and coverage control algorithm for a network of robots to minimize the expected distance between the closest robot and spatially distributed events which will appear at discrete points in a non - convex environment . optimality is defined with reference to a relevant `` multicenter '' cost function . as with all multirobot coordination applications , the challenge comes from reducing the communication requirements : the proposed algorithm requires only short - range gossip " communication , i.e. , asynchronous and unreliable communication between nearby robots . territory partitioning and coverage control have applications in many fields . in cyber - physical systems , applications include automated environmental monitoring @xcite , fetching and delivery @xcite , construction @xcite , and other vehicle routing scenarios @xcite . more generally , coverage of discrete sets is also closely related to the literature on data clustering and @xmath0-means @xcite , as well as the facility location or @xmath0-center problem @xcite . partitioning of graphs is its own field of research , see @xcite for a survey . territory partitioning through local interactions is also studied for animal groups , see for example @xcite . a broad discussion of algorithms for partitioning and coverage control in robotic networks is presented in @xcite which builds on the classic work of lloyd @xcite on optimal quantizer selection through centering and partitioning . " the lloyd approach was first adapted for distributed coverage control in @xcite . since this beginning , similar algorithms have been applied to non - convex environments @xcite , unknown density functions @xcite , equitable partitioning @xcite , and construction of truss - like objects @xcite . there are also multi - agent partitioning algorithms built on market principles or auctions , see @xcite for a survey . while lloyd iterative optimization algorithms are popular and work well in simulation , they require synchronous and reliable communication among neighboring robots . as robots with adjacent regions may be arbitrarily far apart , these communication requirements are burdensome and unrealistic for deployed robotic networks . in response to this issue , in @xcite the authors have shown how a group of robotic agents can optimize the partition of a convex bounded set using a lloyd algorithm with gossip communication . a lloyd algorithm with gossip communication has also been applied to optimizing partitions of non - convex environments in @xcite , the key idea being to transform the coverage problem in euclidean space into a coverage problem on a graph with geodesic distances . distributed lloyd methods are built around separate partitioning and centering steps , and they are attractive because there are known ways to characterize their equilibrium sets ( the so - called centroidal voronoi partitions ) and prove convergence . unfortunately , even for very simple environments ( both continuous and discrete ) the set of centroidal voronoi partitions may contain several sub - optimal configurations . we are thus interested in studying ( discrete ) gossip coverage algorithms for two reasons : ( 1 ) they apply to more realistic robot network models featuring very limited communication in large non - convex environments , and ( 2 ) they are more flexible than typical lloyd algorithms meaning they can avoid poor suboptimal configurations and improve performance . there are three main contributions in this paper . first , we present a discrete partitioning and coverage optimization algorithm for mobile robots with unreliable , asynchronous , and short - range communication . our algorithm has two components : a _ motion protocol _ which drives the robots to meet their neighbors , and a _ pairwise partitioning rule _ to update territories when two robots meet . the partitioning rule optimizes coverage of a set of points connected by edges to form a graph . the flexibility of graphs allows the algorithm to operate in non - convex , non - polygonal environments with holes . our graph partition optimization approach can also be applied to non - planar problems , existing transportation or logistics networks , or more general data sets . second , we provide an analysis of both the convergence properties and computational requirements of the algorithm . by studying a dynamical system of partitions of the graph s vertices , we prove that almost surely the algorithm converges to a pairwise - optimal partition in finite time . the set of pairwise - optimal partitions is shown to be a proper subset of the well - studied set of centroidal voronoi partitions . we further describe how our pairwise partitioning rule can be implemented to run in anytime and how the computational requirements of the algorithm can scale up for large domains and large teams . third , we detail experimental results from our implementation of the algorithm in the player / stage robot control system . we present a simulation of 30 robots providing coverage of a portion of a college campus to demonstrate that our algorithm can handle large robot teams , and a hardware - in - the - loop experiment conducted in our lab which incorporates sensor noise and uncertainty in robot position . through numerical analysis we also show how our new approach to partitioning represents a significant performance improvement over both common lloyd - type methods and the recent results in @xcite . the present work differs from the gossip lloyd method @xcite in three respects . first , while @xcite focuses on territory partitioning in a convex continuous domain , here we operate on a graph which allows our approach to consider geodesic distances , work in non - convex environments , and maintain connected territories . second , instead of a pairwise lloyd - like update , we use an iterative optimal two - partitioning approach which yields better final solutions . third , we also present a motion protocol to produce the sporadic pairwise communications required for our gossip algorithm and characterize the computational complexity of our proposal . preliminary versions of this paper appeared in @xcite and @xcite . compared to these , the new content here includes : ( 1 ) a motion protocol ; ( 2 ) a simplified and improved pairwise partitioning rule ; ( 3 ) proofs of the convergence results ; and ( 4 ) a description of our implementation and a hardware - in - the - loop experiment . in section [ sec : prelim ] we review and adapt coverage and geometric concepts ( e.g. , centroids , voronoi partitions ) to a discrete environment like a graph . we formally describe our robot network model and the discrete partitioning problem in section [ sec : algorithm ] , and then state our coverage algorithm and its properties . section [ sec : convergence ] contains proofs of the main convergence results . in section [ sec : results ] we detail our implementation of the algorithm and present experiments and comparative analysis . some conclusions are given in section [ sec : conclusion ] . in our notation , @xmath1 denotes the set of non - negative real numbers and @xmath2 the set of non - negative integers . given a set @xmath3 , @xmath4 denotes the number of elements in @xmath3 . given sets @xmath5 , their difference is @xmath6 . a set - valued map , denoted by @xmath7 , associates to an element of @xmath3 a subset of @xmath8 . we are given a team of @xmath9 robots tasked with providing coverage of a finite set of points in a non - convex and non - polygonal environment . in this section we translate concepts used in coverage of continuous environments to graphs . let @xmath10 be a finite set of points in a continuous environment . these points represent locations of interest , and are assumed to be connected by weighted edges . let @xmath11 be an ( undirected ) weighted graph with edge set @xmath12 and weight map @xmath13 ; we let @xmath14 be the weight of edge @xmath15 . we assume that @xmath16 is connected and think of the edge weights as distances between locations . [ rem : discretization ] for the examples in this paper we will use a coarse _ occupancy grid map _ as a representation of a continuous environment . in an occupancy grid @xcite , each grid cell is either free space or an obstacle ( occupied ) . to form a weighted graph , each free cell becomes a vertex and free cells are connected with edges if they border each other in the grid . edge weights are the distances between the centers of the cells , i.e. , the grid resolution . there are many other methods to discretize a space , including triangularization and other approaches from computational geometry @xcite , which could also be used . in any weighted graph @xmath16 there is a standard notion of distance between vertices defined as follows . a _ path _ in @xmath17 is an ordered sequence of vertices such that any consecutive pair of vertices is an edge of @xmath17 . the _ weight of a path _ is the sum of the weights of the edges in the path . given vertices @xmath18 and @xmath0 in @xmath17 , the _ distance _ between @xmath18 and @xmath0 , denoted @xmath19 , is the weight of the lowest weight path between them , or @xmath20 if there is no path . if @xmath17 is connected , then the distance between any two vertices in @xmath17 is finite . by convention , @xmath21 if @xmath22 . note that @xmath23 , for any @xmath24 . we will be partitioning @xmath10 into @xmath9 connected subsets or regions which will each be covered by an individual robot . to do so we need to define distances on induced subgraphs of @xmath16 . given @xmath25 , the _ subgraph induced by the restriction of @xmath17 to @xmath26 _ , denoted by @xmath27 , is the graph with vertex set equal to @xmath26 and edge set containing all weighted edges of @xmath17 where both vertices belong to @xmath26 . in other words , we set @xmath28 . the induced subgraph is a weighted graph with a notion of distance between vertices : given @xmath29 , we write @xmath30 note that @xmath31 we define a _ connected subset of @xmath10 _ as a subset @xmath32 such that @xmath33 and @xmath34 is connected . we can then partition @xmath10 into connected subsets as follows . [ def : conpartitions ] given the graph @xmath35 we define a _ connected @xmath36partition of @xmath10 _ as a collection @xmath37 of @xmath9 subsets of @xmath10 such that 1 . @xmath38 ; 2 . @xmath39 if @xmath40 ; 3 . @xmath41 for all @xmath42 ; and 4 . @xmath43 is connected for all @xmath42 . let @xmath44 to be the set of connected @xmath36partitions of @xmath10 . property ( ii ) implies that each element of @xmath10 belongs to just one @xmath43 , i.e. , each location in the environment is covered by just one robot . notice that each @xmath45 induces a connected subgraph in @xmath16 . in subsequent references to @xmath43 we will often mean @xmath46 , and in fact we refer to @xmath47 as the _ dominance subgraph _ or _ region _ of the @xmath48-th robot at time @xmath49 . among the ways of partitioning @xmath10 , there are some which are worth special attention . given a vector of distinct points @xmath50 , the partition @xmath51 is said to be a _ voronoi partition of q generated by c _ if , for each @xmath43 and all @xmath52 , we have @xmath53 and @xmath54 , @xmath55 . note that the voronoi partition generated by @xmath56 is not unique since how to apportion tied vertices is unspecified . for our gossip algorithms we need to introduce the notion of adjacent subgraphs . two distinct connected subgraphs @xmath43 , @xmath57 are said to be _ adjacent _ if there are two vertices @xmath58 , @xmath59 belonging , respectively , to @xmath43 and @xmath57 such that @xmath60 . observe that if @xmath43 and @xmath57 are adjacent then @xmath61 is connected . similarly , we say that robots @xmath48 and @xmath62 are adjacent or are neighbors if their subgraphs @xmath43 and @xmath57 are adjacent . accordingly , we introduce the following useful notion . for @xmath63 , we define the _ adjacency graph _ between regions of partition @xmath64 as @xmath65 , where @xmath66 if @xmath43 and @xmath57 are adjacent . note that @xmath67 is always connected since @xmath16 is . we define three coverage cost functions for graphs : @xmath68 , @xmath69 , and @xmath70 . let the _ weight function _ @xmath71 assign a relative weight to each element of @xmath10 . the _ one - center function _ @xmath68 gives the cost for a robot to cover a connected subset @xmath32 from a vertex @xmath72 with relative prioritization set by @xmath73 : @xmath74 a technical assumption is needed to solve the problem of minimizing @xmath75 : we assume from now on that a _ total order _ relation , @xmath76 , is defined on @xmath10 , i.e. , that @xmath77 . with this assumption we can deterministically pick a vertex in @xmath3 which minimizes @xmath68 as follows . [ def : centroid ] let @xmath10 be a totally ordered set , and let @xmath32 . we define the set of generalized centroids of @xmath3 as the set of vertices in @xmath3 which minimize @xmath68 , i.e. , @xmath78 further , we define the map @xmath79 as @xmath80 . we call @xmath81 the _ generalized centroid _ of @xmath3 . in subsequent use we drop the word generalized " for brevity . note that with this definition the centroid is well - defined , and also that the centroid of a region always belongs to the region . with a slight notational abuse , we define @xmath82 as the map which associates to a partition the vector of the centroids of its elements . we define the _ multicenter function _ @xmath69 to measure the cost for @xmath9 robots to cover a connected @xmath9-partition @xmath64 from the vertex set @xmath83 : @xmath84 we aim to minimize the performance function @xmath69 with respect to both the vertices @xmath56 and the partition @xmath64 . we can now state the coverage cost function we will be concerned with for the rest of this paper . let @xmath85 be defined by @xmath86 in the motivational scenario we are considering , each robot will periodically be asked to perform a task somewhere in its region with tasks appearing according to distribution @xmath73 . when idle , the robots would position themselves at the centroid of their region . by partitioning @xmath17 so as to minimize @xmath70 , the robot team would minimize the expected distance between a task and the robot which will service it . we introduce two notions of optimal partitions : centroidal voronoi and pairwise - optimal . our discussion starts with the following simple result about the multicenter cost function . [ prop : optimal - for - hgeneric ] let @xmath63 and @xmath50 . if @xmath87 is a voronoi partition generated by @xmath56 and @xmath88 is such that @xmath89 , then @xmath90 the second inequality is strict if any @xmath91 . proposition [ prop : optimal - for - hgeneric ] implies the following necessary condition : if @xmath92 minimizes @xmath69 , then @xmath93 and @xmath64 must be a voronoi partition generated by @xmath56 . thus , @xmath70 has the following property as an immediate consequence of proposition [ prop : optimal - for - hgeneric ] : given @xmath63 , if @xmath94 is a voronoi partition generated by @xmath95 then @xmath96 this fact motivates the following definition . @xmath63 is a _ centroidal voronoi partition _ of @xmath10 if there exists a @xmath97 such that @xmath64 is a voronoi partition generated by @xmath56 and @xmath98 . the set of _ pairwise - optimal partitions _ provides an alternative definition for the optimality of a partition : a partition is pairwise - optimal if , for every pair of adjacent regions , one can not find a better two - partition of the union of the two regions . this condition is formally stated as follows . @xmath63 is a _ pairwise - optimal partition _ if for every @xmath66 , @xmath99 the following proposition states that the set pairwise - optimal partitions is in fact a subset of the set of centroidal voronoi partitions . the proof is involved and is deferred to appendix [ sec : appendix_c ] . see fig . [ fig : voronoi ] for an example which demonstrates that the inclusion is strict . [ prop : optpair ] let @xmath51 be a _ pairwise - optimal partition_. then @xmath64 is also a _ centroidal voronoi partition_. for a given environment @xmath10 , a pair made of a centroidal voronoi partition @xmath64 and the corresponding vector of centroids @xmath56 is locally optimal in the following sense : @xmath70 can not be reduced by changing either @xmath64 or @xmath56 independently . a pairwise - optimal partition achieves this property and adds that for every pair of neighboring robots @xmath100 , there does not exist a two - partition of @xmath101 with a lower coverage cost . in other words , positioning the robots at the centroids of a centroidal voronoi partition ( locally ) minimizes the expected distance between a task appearing randomly in @xmath10 according to relative weights @xmath73 and the robot who owns the vertex where the task appears . positioning at the centroids of a pairwise - optimal partition improves performance by reducing the number of sub - optimal solutions which the team might converge to . we aim to partition @xmath10 among @xmath9 robotic agents using only asynchronous , unreliable , short - range communication . in section [ sec : model ] we describe the computation , motion , and communication capabilities required of the team of robots , and in section [ sec : problemformulation ] we formally state the problem we are addressing . in section [ sec : algorithm ] we propose our solution , the _ discrete gossip coverage algorithm _ , and in [ sec : illustrative ] we provide an illustration . in sections [ sec : convprop ] and [ sec : computation ] we state the algorithm s convergence and complexity properties . our discrete gossip coverage algorithm requires a team of @xmath9 robotic agents where each agent @xmath42 has the following basic computation and motion capabilities : 1 . agent @xmath48 knows its unique identifier @xmath48 ; 2 . agent @xmath48 has a processor with the ability to store @xmath16 and perform operations on subgraphs of @xmath102 ; and 3 . agent @xmath48 can determine which vertex in @xmath10 it occupies and can move at speed @xmath103 along the edges of @xmath16 to any other vertex in @xmath10 . the localization requirement in ( c3 ) is actually quite loose . localization is only used for navigation and not for updating partitions , thus limited duration localization errors are not a problem . the robotic agents are assumed to be able to communicate with each other according to the _ range - limited gossip communication model _ which is described as follows : 1 . given a communication range @xmath104 , when any two agents reside for some positive duration at a distance @xmath105 , they communicate at the sample times of a poisson process with intensity @xmath106 . recall that an homogeneous poisson process is a widely - used stochastic model for events which occur randomly and independently in time , where the expected number of events in a period @xmath107 is @xmath108 . [ rem : comm ] ( 1 ) this communication capability is the minimum necessary for our algorithm , any additional capability can only reduce the time required for convergence . for example , it would be acceptable to have intensity @xmath109 depend upon the pairwise robot distance in such a way that @xmath110 for @xmath105 . ( 2)we use distances in the graph to model limited range communication . these graph distances are assumed to approximate geodesic distances in the underlying continuous environment and thus path distances for a diffracting wave or moving robot . assume that , for all @xmath111 , each agent @xmath42 maintains in memory a connected subset @xmath47 of environment @xmath10 . our goal is to design a distributed algorithm that iteratively updates the partition @xmath112 while solving the following optimization problem : @xmath113 subject to the constraints imposed by the robot network model with range - limited gossip communication from section [ sec : model ] . in the design of an algorithm for the minimization problem there are two main questions which must be addressed . first , given the limited communication capabilities in ( c4 ) , how should the robots move inside @xmath10 to guarantee frequent enough meetings between pairs of robots ? second , when two robots are communicating , what information should they exchange and how should they update their regions ? in this section we introduce the _ discrete gossip coverage algorithm _ which , following these two questions , consists of two components : 1 . the _ random destination & wait motion protocol _ ; and 2 . the _ pairwise partitioning rule_. the concurrent implementation of the random destination & wait motion protocol and the pairwise partitioning rule determines the evolution of the positions and dominance subgraphs of the agents as we now formally describe . we start with the random destination & wait motion protocol . ' '' '' width height .4pt * random destination & wait motion protocol * ' '' '' width height .4pt each agent @xmath114 determines its motion by repeatedly performing the following actions : agent @xmath48 samples a _ destination vertex _ @xmath58 from a uniform distribution over its dominance subgraph @xmath43 ; agent @xmath48 moves to vertex @xmath58 through the shortest path in @xmath43 connecting the vertex it currently occupies and @xmath58 ; and agent @xmath48 waits at @xmath58 for a duration @xmath115 . ' '' '' width height .4pt if agent @xmath48 is moving from one vertex to another we say that agent @xmath48 is in the _ moving _ state while if agent @xmath48 is waiting at some vertex we say that it is in the _ waiting _ state . the motion protocol is designed to ensure frequent enough communication between pairs of robots . in general , any motion protocol can be used which meets this requirement , so @xmath48 could select @xmath58 from the boundary of @xmath43 or use some heuristic non - uniform distribution over @xmath43 . if any two agents @xmath48 and @xmath62 reside in two vertices at a graphical distance smaller that @xmath116 for some positive duration , then at the sample times of the corresponding communication poisson process the two agents exchange sufficient information to update their respective dominance subgraphs @xmath43 and @xmath57 via the pairwise partitioning rule . ' '' '' width height .4pt * pairwise partitioning rule * ' '' '' width height .4pt assume that at time @xmath111 , agent @xmath48 and agent @xmath62 communicate . without loss of generality assume that @xmath117 . let @xmath47 and @xmath118 denote the current dominance subgraphs of @xmath48 and @xmath62 , respectively . moreover , let @xmath119 denote the time instant just after @xmath49 . then , agents @xmath48 and @xmath62 perform the following tasks : agent @xmath48 transmits @xmath47 to agent @xmath62 and vice - versa initialize @xmath120 , @xmath121 , @xmath122 , @xmath123 compute @xmath124 and an ordered list @xmath125 of all pairs of vertices in @xmath126 compute the sets + @xmath127 + @xmath128 * if * @xmath129 + @xmath130 * then * @xmath131 @xmath132 ' '' '' width height .4pt some remarks are now in order . ( 1 ) the pairwise partitioning rule is designed to find a minimum cost two - partition of @xmath126 . more formally , if list @xmath125 and sets @xmath133 and @xmath134 for @xmath135 are defined as in the pairwise partitioning rule , then @xmath133 and @xmath134 are an optimal two - partition of @xmath126 . \(2 ) while the loop in steps 4 - 7 must run to completion to guarantee that @xmath133 and @xmath134 are an optimal two - partition of @xmath126 , the loop is designed to return an intermediate sub - optimal result if need be . if @xmath43 and @xmath57 change , then @xmath70 will decrease and this is enough to ensure eventual convergence . ( 3)we make a simplifying assumption in the pairwise partitioning rule that , once two agents communicate , the application of the partitioning rule is instantaneous . we discuss the actual computation time required in section [ sec : computation ] and some implementation details in section [ sec : results ] . ( 4)notice that simply assigning @xmath133 to @xmath48 and @xmath134 to @xmath62 can cause the robots to `` switch sides '' in @xmath126 . while convergence is guaranteed regardless , switching may be undesirable in some applications . in that case , any smart matching of @xmath133 and @xmath134 to @xmath48 and @xmath62 may be inserted . ( 5)agents who are not adjacent may communicate but the partitioning rule will not change their regions . indeed , in this case @xmath133 and @xmath134 will not change from @xmath47 and @xmath118 . some possible modifications and extensions to the algorithm are worth mentioning . in case the robots have heterogeneous dynamics , line 5 can be modified to consider per - robot travel times between vertices . for example , @xmath136 could be replaced by the expected time for robot @xmath48 to travel from @xmath137 to @xmath138 while @xmath139 would consider robot @xmath62 . here we focus on partitioning territory , but this algorithm can easily be combined with methods to provide a service in @xmath10 as in @xcite . the agents could split their time between moving to meet their neighbors and update territory , and performing requested tasks in their region . the simulation in fig . [ fig : sim_four ] shows four robots partitioning a square environment with obstacles where the free space is represented by a @xmath140 grid . in the initial partition shown in the left panel , the robot in the top right controls most of the environment while the robot in the bottom left controls very little . the robots then move according to the random destination & wait motion protocol , and communicate according to range - limited gossip communication model with @xmath141 ( four edges in the graph ) . the first pairwise territory exchange is shown in the second panel , where the bottom left robot claims some territory from the robot on the top left . a later exchange between the two robots on the top is shown in the next two panels . notice that the cyan robot in the top right gives away the vertex it currently occupies . in such a scenario , we direct the robot to follow the shortest path in @xmath16 to its updated territory before continuing on to a random destination . after 9 pairwise territory exchanges , the robots reach the pairwise - optimal partition shown at right in fig . [ fig : sim_four ] . the expected distance between a random vertex and the closest robot decreases from @xmath142 down to @xmath143 . the strength of the discrete gossip coverage algorithm is the possibility of enforcing that a partition will converge to a pairwise - optimal partition through pairwise territory exchange . in theorem [ th : main ] we summarize this convergence property , with proofs given in section [ sec : convergence ] . [ th : main ] consider a network of @xmath9 robotic agents endowed with computation and motion capacities ( c1 ) , ( c2 ) , ( c3 ) , and communication capacities ( c4 ) . assume the agents implement the _ discrete gossip coverage algorithm _ consisting of the concurrent implementation of the _ random destination & wait motion protocol _ and the _ pairwise partitioning rule_. then , a. [ item : well - posedness ] the partition @xmath144 remains connected and is described by @xmath145 and b. [ item : convergence ] @xmath144 converges almost surely in finite time to a pairwise - optimal partition . by definition , a pairwise - optimal partition is optimal in that @xmath70 can not be improved by changing only two regions in the partition . for simplicity we assume uniform robot speeds , communication processes , and waiting times . an extension to non - uniform processes would be straightforward . in this subsection we explore the computational requirements of the discrete gossip coverage algorithm , and make some comments on implementation . cost function @xmath146 is the sum of the distances between @xmath18 and all other vertices in @xmath43 . this computation of one - to - all distances is the core computation of the algorithm . for most graphs of interest the total number of edges @xmath147 is proportional to @xmath148 , so we will state bounds on this computation in terms of @xmath149 . computing one - to - all distances requires one of the following : * if all edge weights in @xmath16 are the same ( e.g. , for a graph from an occupancy grid ) , a breadth - first search approach can be used which requires @xmath150 in time and memory ; * otherwise , dijkstra s algorithm must be used which requires @xmath151 in time and @xmath150 in memory . let @xmath152 be the time to compute one - to - all distances in @xmath43 , then computing @xmath146 requires @xmath153 in time . [ prop : computation ] the motion protocol requires @xmath150 in memory , and @xmath153 in computation time . the partitioning rule requires @xmath154 in communication bandwidth between robots @xmath48 and @xmath62 , @xmath154 in memory , and can run in any time . we first prove the claims for the motion protocol . step 2 is the only non - trivial step and requires finding a shortest path in @xmath43 , which is equivalent to computing one - to - all distances from the robot s current vertex . hence , it requires @xmath153 in time and @xmath155 in memory . we now prove the claims for the partitioning rule . in step 1 , robots @xmath48 and @xmath62 transmit their subgraphs to each other , which requires @xmath154 in communication bandwidth . for step 3 , the robots determine @xmath156 , which requires @xmath154 in memory to store . step 4 is the start of a loop which executes @xmath157 times , affecting the time complexity of steps 5 , 6 and 7 . step 5 requires two computations of one - to - all distances in @xmath126 which each take @xmath158 . step 6 involves four computations of @xmath68 over different subsets of @xmath126 , however those for @xmath133 and @xmath134 can be stored from previous computation . since @xmath159 and @xmath160 are strict subsets of @xmath126 , step 5 takes longer than step 6 . step 7 is trivial , as is step 8 . the total time complexity of the loop is thus @xmath161 . however , the loop in steps 4 - 7 can be truncated after any number of iterations . while it must run to completion to guarantee that @xmath133 and @xmath134 are an optimal two - partition of @xmath126 , the loop is designed to return an intermediate sub - optimal result if need be . if @xmath43 and @xmath57 change , then @xmath70 will decrease . our convergence result will hold provided that all elements of @xmath125 are eventually checked if @xmath43 and @xmath57 do not change . thus , the partitioning rule can run in any time with each iteration requiring @xmath158 . all of the computation and communication requirements in proposition [ prop : computation ] are independent of the number of robots and scale with the size of a robot s partition , meaning the discrete gossip coverage algorithm can easily scale up for large teams of robots in large environments . this section is devoted to proving the two statements in theorem [ th : main ] . the proof that the pairwise partitioning rule maps a connected @xmath9-partition into a connected @xmath9-partition is straightforward . the proof of convergence is more involved and is based on the application of lemma [ lem : finite - lasalle ] in appendix [ sec : appendix_a ] to the discrete gossip coverage algorithm . lemma [ lem : finite - lasalle ] establishes strong convergence properties for a particular class of set valued maps ( set - valued maps are briefly reviewed in appendix [ sec : appendix_a ] ) . we start by proving that the pairwise partitioning rule is well - posed in the sense that it maintains a connected partition . to prove the statement we need to show that @xmath162 satisfies points ( i ) through ( iv ) of definition [ def : conpartitions ] . from the definition of the pairwise partitioning rule , we have that @xmath163 and @xmath164 . moreover , since @xmath165 and @xmath166 , it follows that @xmath167 and @xmath168 . these observations imply the validity of points ( i ) , ( ii ) , and ( iii ) for @xmath162 . finally , we must show that @xmath169 and @xmath170 are connected , i.e. , @xmath162 also satisfies point ( iv ) . to do so we show that , given @xmath171 , any shortest path in @xmath172 connecting @xmath138 to @xmath173 completely belongs to @xmath133 . we proceed by contradiction . let @xmath174 denote a shortest path in @xmath175 connecting @xmath138 to @xmath173 and let us assume that there exists @xmath176 such that @xmath177 . for @xmath178 to be in @xmath134 means that @xmath179 . this implies that @xmath180 this is a contradiction for @xmath171 . similar considerations hold for @xmath134 . the rest of this section is dedicated to proving convergence . our first step is to show that the evolution determined by the discrete gossip coverage algorithm can be seen as a set - valued map . to this end , for any pair of robots @xmath181 , @xmath182 , we define the map @xmath183 by @xmath184 where @xmath185 and @xmath186 . if at time @xmath187 the pair @xmath100 and no other pair of robots perform an iteration of the pairwise partitioning rule , then the dynamical system on the space of partitions is described by @xmath188 we define the set - valued map @xmath189 as @xmath190 observe that can then be rewritten as @xmath191 . the next two propositions state facts whose validity is ensured by lemma [ lemma : onmotionprotocol ] of appendix [ sec : appendix_b ] which states a key property of the random destination & wait motion protocol . [ prop : tk ] consider @xmath9 robots implementing the discrete gossip coverage algorithm . then , there almost surely exists an increasing sequence of time instants @xmath192 such that @xmath193 for some @xmath194 . the proof follows directly from lemma [ lemma : onmotionprotocol ] which implies that the time between two consecutive pairwise communications is almost surely finite . the existence of time sequence @xmath192 allows us to to express the evolution generate by the discrete gossip coverage algorithm as a discrete time process . let @xmath195 and @xmath196 , then @xmath197 where @xmath189 is defined as in . given @xmath198 , let @xmath199 denote the information which completely characterizes the state of discrete gossip coverage algorithm just after the @xmath0-th iteration of the partitioning rule , i.e. , at time @xmath200 . specifically , @xmath199 contains the information related to the partition @xmath201 , the positions of the robots at @xmath200 , and whether each robot is in the _ waiting _ or _ moving _ state at @xmath200 . the following result characterizes the probability that , given @xmath199 , the @xmath202-th iteration of the partitioning rule is governed by any of the maps @xmath203 , @xmath204 . [ prop : pi ] consider a team of @xmath9 robots with capacities ( c1 ) , ( c2 ) , ( c3 ) , and ( c4 ) implementing the discrete gossip coverage algorithm . then , there exists a real number @xmath205 , such that , for any @xmath206 and @xmath204 @xmath207\geq \bar{\pi}.\ ] ] assume that at time @xmath208 one pair of robots communicates . given a pair @xmath209 , we must find a lower bound for the probability that @xmath210 is the communicating pair . since all the poisson communication processes have the same intensity , the distribution of the chance of communication is uniform over the pairs which are `` able to communicate , '' i.e. , closer than @xmath116 to each other . thus , we must only show that @xmath210 has a positive probability of being able to communicate at time @xmath208 , which is equivalent to showing that @xmath210 is able to communicate for a positive fraction of time with positive probability . the proof of lemma [ lemma : onmotionprotocol ] implies that with probability at least @xmath211 any pair in @xmath212 is able to communicate for a fraction of time not smaller than @xmath213 where @xmath214 and @xmath107 are defined in the proof of lemma [ lemma : onmotionprotocol ] . hence the result follows . the property in proposition [ prop : pi ] can also be formulated as follows . let @xmath215 be the stochastic process such that @xmath216 is the communicating pair at time @xmath0 . then , the sequence of pairs of robots performing the partitioning rule at time instants @xmath192 can be seen as a realization of the process @xmath217 , which satisfies @xmath218 \geq \bar{\pi}\ ] ] for all @xmath204 . next we show that the cost function decreases whenever the application of @xmath219 from changes the territory partition . this fact is a key ingredient to apply lemma [ lem : finite - lasalle ] . [ lemma : tdecr ] let @xmath220 and let @xmath221 . if @xmath222 , then @xmath223 . without loss of generality assume that @xmath100 is the pair executing the pairwise partitioning rule . then @xmath224 according to the definition of the pairwise partitioning rule we have that if @xmath225 , @xmath226 , then @xmath227 from which the statement follows . we now complete the proof of the main result , theorem [ th : main ] . note that the algorithm evolves in a finite space of partitions , and by theorem [ th : main ] statement ( [ item : well - posedness ] ) , the set @xmath44 is strongly positively invariant . this fact implies that assumption ( i ) of lemma [ lem : finite - lasalle ] is satisfied . from lemma [ lemma : tdecr ] it follows that assumption ( ii ) is also satisfied , with @xmath70 playing the role of the function @xmath126 . finally , the property in is equivalent to the property of _ persistent random switches _ stated in assumption ( iii ) of lemma [ lem : finite - lasalle ] , for the special case @xmath228 . hence , we are in the position to apply lemma [ lem : finite - lasalle ] and conclude convergence in finite - time to an element of the intersection of the equilibria of the maps @xmath203 , which by definition is the set of the pairwise - optimal partitions . to demonstrate the utility and study practical issues of the discrete gossip coverage algorithm , we implemented it using the open - source player / stage robot control system @xcite and the boost graph library ( bgl ) @xcite . all results presented here were generated using player 2.1.1 , stage 2.1.1 , and bgl 1.34.1 . to compute distances in uniform edge weight graphs we extended the bgl breadth - first search routine with a distance recorder event visitor . to evaluate the performance of our gossip coverage algorithm with larger teams , we tested 30 simulated robots partitioning a map representing a @xmath229 portion of campus at the university of california at santa barbara . as shown in fig . [ fig : large_sim ] , the robots are tasked with providing coverage of the open space around some of the buildings on campus , a space which includes a couple open quads , some narrower passages between buildings , and a few dead - end spurs . for this large environment the simulated robots are @xmath230 on a side and can move at @xmath231 . each territory cell is @xmath232 . in this simulation we handle communication and partitioning as follows . the communication range is set to @xmath233 ( 10 edges in the graph ) with @xmath234 . the robots wait at their destination vertices for @xmath235 . this value for @xmath236 was chosen so that on average one quarter of the robots are waiting at any moment . lower values of @xmath236 mean the robots are moving more of the time and as a result more frequently miss connections , while for higher @xmath236 the robots spend more time stationary which also reduces the rate of convergence . with the goal of improving communication , we implemented a minor modification to the motion protocol : each robot picks its random destination from the cells forming the open boundary is the set of vertices in @xmath43 which are adjacent to at least one vertex owned by another agent . ] of its territory . in our implementation , the full partitioning loop may take @xmath237 seconds for the largest initial territories in fig . [ fig : large_sim ] . we chose to stop the loop after a quarter second for this simulation to verify the anytime computation claim the 30 robots start clustered in the center of the map between engineering ii and broida hall , and an initial voronoi partition is generated from these starting positions . this initial partition is shown on the left in fig . [ fig : large_sim ] with the robots positioned at the centroids of their starting regions . the initial partition has a cost of @xmath238 . the team spends about 27 minutes moving and communicating according to the discrete gossip coverage algorithm before settling on the final partition on the right of fig . [ fig : large_sim ] . the coverage cost of the final equilibrium improved by @xmath239 to @xmath240 . visually , the final partition is also dramatically more uniform than the initial condition . this result demonstrates that the algorithm is effective for large teams in large non - convex environments . over time for the simulation in fig . [ fig : large_sim].,height=125 ] fig . [ fig : large_sim_cost ] shows the evolution of @xmath70 during the simulation . the largest cost improvements happen early when the robots that own the large territories on the left and right of the map communicate with neighbors with much smaller territories . these big territory changes then propagate through the network as the robots meet and are pushed and pulled towards a lower cost partition . we conducted an experiment to test the algorithm using three physical robots in our lab , augmented by six simulated robots in a synthetic environment extending beyond the lab . our lab space is @xmath241 on a side and is represented by the upper left portion of the territory maps in fig . [ fig : experiment ] . the territory graph loops around a center island of desks . we extended the lab space through three connections into a simulated environment around the lab , producing a @xmath242 environment . the map of the environment was specified with a @xmath243 bitmap which we overlayed with a @xmath244 resolution occupancy grid representing the free territory for the robots to cover . the result is a lattice - like graph with all edge weights equal to @xmath244 . the @xmath244 resolution was chosen so that our physical robots would fit easily inside a cell . additional details of our implementation are as follows . we use erratic mobile robots from videre design , as shown in fig . [ fig : robot ] . the vehicle platform has a roughly square footprint @xmath245 , with two differential drive wheels and a single rear caster . each robot carries an onboard computer with a 1.8ghz core 2 duo processor , 1 gb of memory , and 802.11 g wireless communication . for navigation and localization , each robot is equipped with a hokuyo urg-04lx laser rangefinder . the rangefinder scans @xmath246 points over @xmath247 at @xmath248 with a range of @xmath249 meters . our mixed physical and virtual robot experiments are run from a central computer which is attached to a wireless router so it can communicate with the physical robots . the central computer creates a simulated world using stage which mirrors and extends the real space in which the physical robots operate . the central computer also simulates the virtual members of the robot team . these virtual robots are modeled off of our hardware : they are differential drive with the same geometry as the erratic platform and use simulated hokuyo urg-04lx rangefinders . we use the ` amcl ` driver in player which implements adaptive monte - carlo localization @xcite . the physical robots are provided with a map of our lab with a @xmath250 resolution and told their starting pose within the map . we set an initial pose standard deviation of @xmath251 in position and @xmath252 in orientation , and request localization updates using @xmath253 of the sensor s range measurements for each change of @xmath254 in position or @xmath255 in orientation reported by the robot s odometry system . we then use the most likely pose estimate output by ` amcl ` as the location of the robot . for simplicity and reduced computational demand , we allow the virtual robots access to perfect localization information . each robot continuously executes the random destination & wait motion protocol , with navigation handled by the ` snd ` driver in player which implements smooth nearness diagram navigation @xcite . for ` snd ` we set the robot radius parameter to @xmath256 , obstacle avoidance distance to @xmath257 , and maximum speeds to @xmath258 and @xmath259 . the ` snd ` driver is a local obstacle avoidance planner , so we feed it a series of waypoints every couple meters along paths found in @xmath16 . we consider a robot to have achieved its target location when it is within @xmath260 and it will then wait for @xmath235 . for the physical robots the motion protocol and navigation processes run on board , while there are separate threads for each virtual robot on the central computer . as the robots move , a central process monitors their positions and simulates the range - limited gossip communication model between both real and virtual robots . we set @xmath141 and @xmath261 . these parameters were chosen so that the robots would be likely to communicate when separated by at most four edges , but would also sometimes not connect despite being close . when this process determines two robots should communicate , it informs the robots who then perform the pairwise partitioning rule . our pairwise communication implementation is blocking : if robot @xmath48 is exchanging territory with @xmath62 , then it informs the match making process that it is unavailable until the exchange is complete . the results of our experiment with three physical robots and six simulated robots are shown in figs . [ fig : experiment ] and [ fig : exp_cost ] . the left column in fig . [ fig : experiment ] shows the starting positions of the team of robots , with the physical robots , labeled 1 , 2 , and 3 , lined up in a corner of the lab and the simulated robots arrayed around them . the starting positions are used to generate the initial voronoi partition of the environment . the physical robots own the orange , blue , and lime green territories in the upper left quadrant . we chose this initial configuration to have a high coverage cost , while ensuring that the physical robots will remain in the lab as the partition evolves . in the middle column , robots 1 and 2 have met along their shared border and are exchanging territory . in the territory map , the solid red line indicates 1 and 2 are communicating and their updated territories are drawn with solid orange and blue , respectively . the camera view confirms that the two robots have met on the near side of the center island of desks . the final partition at right in fig . [ fig : experiment ] is reached after @xmath262 minutes . all of the robots are positioned at the centroids of their final territories . the three physical robots have gone from a cluster in one corner of the lab to a more even spread around the space . . the total cost @xmath70 is shown above in black , while @xmath68 for each robot is shown below in the robot s color . ] [ fig : exp_cost ] shows the evolution of the cost function @xmath70 as the experiment progresses , including the costs for each robot . as expected , the total cost never increases and the disparity of costs for the individual robots shrinks over time until settling at a pairwise - optimal partition . in this experiment the hardware challenges of sensor noise , navigation , and uncertainty in position were efficiently handled by the ` amcl ` and ` snd ` drivers . the coverage algorithm assumed the role of a higher - level planner , taking in position data from ` amcl ` and directing ` snd ` . by far the most computationally demanding component was ` amcl ` , but the position hypotheses from ` amcl ` are actually unnecessary : our coverage algorithm only requires knowledge of the vertex a robot occupies . if a less intensive localization method is available , the algorithm could run on robots with significantly lower compute power . in this subsection we present a numerical comparison of the performance of the discrete gossip coverage algorithm and the following two lloyd - type algorithms . this method is from @xcite and @xcite , we describe it here for convenience . at each discrete time instant @xmath263 , each robot @xmath48 performs the following tasks : ( 1 ) @xmath48 transmits its position and receives the positions of all adjacent robots ; ( 2 ) @xmath48 computes its voronoi region @xmath43 based on the information received ; and ( 3 ) @xmath48 moves to @xmath264 . this method is from @xcite . it is a gossip algorithm , and so we have used the same communication model and the random destination & wait motion protocol to create meetings between robots . say robots @xmath48 and @xmath62 meet at time @xmath49 , then the pairwise lloyd partitioning rule works as follows : ( 1 ) robot @xmath48 transmits @xmath47 to @xmath62 and vice versa ; ( 2 ) both robots determine @xmath265 ; ( 3 ) robot @xmath48 sets @xmath169 to be its voronoi region of @xmath126 based on @xmath266 and @xmath267 , and @xmath62 does the equivalent . for both lloyd algorithms we use the same tie breaking rule when creating voronoi regions as is present in the pairwise partitioning rule : ties go to the robot with the lowest index . our first numerical result uses a monte carlo probability estimation method from @xcite to place probabilistic bounds on the performance of the two gossip algorithms . recall that the chernoff bound describes the minimum number of random samples @xmath268 required to reach a certain level of accuracy in a probability estimate from independent bernoulli tests . for an accuracy @xmath269 and confidence @xmath270 , the number of samples is given by @xmath271 for @xmath272 and @xmath273 , at least 116 samples are required . figure [ fig : bad_start ] shows both the initial territory partition of the extended laboratory environment used and also a histogram of the final results for the following monte carlo test . the environment and robot motion models used are described in section [ sec : implementation ] . starting from the indicated initial condition , we ran 116 simulations of both gossip algorithms . the randomness in the test comes from the sequence of pairwise communications . these sequences were generated using : ( 1 ) the random destination & wait motion protocol with @xmath58 sampled uniformly from the open boundary of @xmath43 and @xmath235 ; and ( 2 ) the range - limited gossip communication model with @xmath141 and @xmath261 . the cost of the initial partition in fig . [ fig : bad_start ] is @xmath274 , while the best known partition for this environment has a cost of just under @xmath275 . the histogram in fig . [ fig : bad_start ] shows the final equilibrium costs for 116 simulations of the discrete gossip coverage algorithm ( black ) and the gossip lloyd algorithm ( gray ) . it also shows the final cost using the decentralized lloyd algorithm ( red dashed line ) , which is deterministic from a given initial condition . the histogram bins have a width of @xmath276 and start from @xmath277 . for the discrete gossip coverage algorithm , @xmath278 out of @xmath279 trials reach the bin containing the best known partition and the mean final cost is @xmath280 . the gossip lloyd algorithm reaches the lowest bin in only @xmath237 of @xmath279 trials and has a mean final cost of @xmath281 . the decentralized lloyd algorithm settles at @xmath282 . our new gossip algorithm requires an average of @xmath283 pairwise communications to reach an equilibrium , whereas gossip lloyd requires @xmath284 . based on these results , we can conclude with @xmath285 confidence that there is at least an @xmath286 probability that 9 robots executing the discrete gossip coverage algorithm starting from the initial partition shown in fig . [ fig : bad_start ] will reach a pairwise - optimal partition which has a cost within @xmath287 of the best known cost . we can further conclude with @xmath285 confidence that the gossip lloyd algorithm will settle more than @xmath287 above the best known cost at least @xmath288 of the time starting from this initial condition . comparing discrete gossip coverage algorithm ( black bars ) , gossip lloyd algorithm ( gray bars ) , and decentralized lloyd algorithm ( red dashed line ) . for the gossip algorithms , 116 simulations were performed with different sequences of pairwise communications . the decentralized lloyd algorithm is deterministic given an initial condition so only one final cost is shown . the initial cost for each test is drawn with the green dashed line . ] figure [ fig : multi_compare ] compares final cost histograms for @xmath289 different initial conditions for the same environment and parameters as described above . each initial condition was created by selecting unique starting locations for the robots uniformly at random and using these locations to generate an initial voronoi partition . the initial cost for each test is shown with the green dashed line . in 9 out of 10 tests the discrete gossip coverage algorithm reaches the histogram bin with the best known partition in at least @xmath290 of @xmath279 trials . the two lloyd methods get stuck in sub - optimal centroidal voronoi partitions more than @xmath287 away from the best known partition in more than half the trials in 7 of 10 tests . we have presented a novel distributed partitioning and coverage control algorithm which requires only unreliable short - range communication between pairs of robots and works in non - convex environments . the classic lloyd approach to coverage optimization involves iteration of separate centering and voronoi partitioning steps . for gossip algorithms , however , this separation is unnecessary computationally and we have shown that improved performance can be achieved without it . our new discrete gossip coverage algorithm provably converges to a subset of the set of centroidal voronoi partitions which we labeled pairwise - optimal partitions . through numerical comparisons we demonstrated that this new subset of solutions avoids many of the local minima in which lloyd - type algorithms can get stuck . our vision is that this partitioning and coverage algorithm will form the foundation of a distributed task servicing setup for teams of mobile robots . the robots would split their time between servicing tasks in their territory and moving to contact their neighbors and improve the coverage of the space . our convergence results only require sporadic improvements to the cost function , affording flexibility in robot behaviors and capacities , and offering the ability to handle heterogeneous robotic networks . in the bigger picture , this paper demonstrates the potential of gossip communication in distributed coordination algorithms . there appear to be many other problems where this realistic and minimal communication model could be fruitfully applied . given a set @xmath291 , a set - valued map @xmath292 is a map which associates to an element @xmath293 a subset @xmath294 a set - valued map is non - empty if @xmath295 for all @xmath296 . given a non - empty set - valued map @xmath219 , an evolution of the dynamical system associated to @xmath219 is a sequence @xmath297 where @xmath298 for all @xmath299 a set @xmath300 is _ strongly positively invariant _ for @xmath219 if @xmath301 for all @xmath302 . [ lem : finite - lasalle ] let @xmath303 be a finite metric space . given a collection of maps @xmath304 , define the set - valued map @xmath292 by @xmath305 . given a stochastic process @xmath306 , consider an evolution @xmath307 of @xmath219 satisfying @xmath308 assume that : 1 . there exists a set @xmath309 that is strongly positively invariant for @xmath219 ; 2 . there exists a function @xmath310 such that @xmath311 , for all @xmath312 and @xmath313 ; and 3 . there exist @xmath314 and @xmath315 such that , for all @xmath316 and @xmath317 , there exists @xmath318 such that @xmath319 \geq p. $ ] for @xmath320 , let @xmath321 be the set of fixed points of @xmath322 in @xmath323 , i.e. , @xmath324 . if @xmath325 , then the evolution @xmath307 converges almost surely in finite time to an element of the set @xmath326 , i.e. , there exists almost surely @xmath327 such that , for some @xmath328 , @xmath329 for @xmath330 [ lemma : onmotionprotocol ] consider @xmath9 robots implementing the discrete gossip coverage algorithm starting from an arbitrary @xmath51 . consider @xmath187 and let @xmath144 denote the partition at time @xmath49 . assume that at time @xmath49 no two robots are communicating . then , there exist @xmath331 and @xmath332 , independent of @xmath144 and the positions and states of the robots at time @xmath49 , such that , for every @xmath333 , @xmath334\ge \alpha.$ ] our goal is to lower bound the probability that @xmath48 and @xmath62 will communicate within the interval @xmath340 . to do so we construct _ one _ sequence of events of positive probability which enables such communication . consider the following situation : @xmath48 is in the _ moving _ state and needs time @xmath341 to reach its destination @xmath58 , whereas robot @xmath62 is in the _ waiting _ state at vertex @xmath59 and must wait there for time @xmath342 . we denote by @xmath343 ( resp . @xmath344 ) the time needed for @xmath48 ( resp . @xmath62 ) to travel from @xmath58 ( resp . @xmath59 ) to @xmath137 ( resp . @xmath345 ) . let @xmath346 be the event such that @xmath48 performs the following actions in @xmath347 without communicating with any robot @xmath348 : next , we lower bound the probability that event @xmath355 occurs . recall the definition of @xmath356 from sec . [ sec : model ] . since a robot can have at most @xmath357 neighbors , the probability that ( i ) of @xmath355 happens is lower bounded by @xmath358 for ( ii ) , the probability that @xmath48 chooses @xmath137 is @xmath359 , which is lower bounded by @xmath360 . then , in order to spend at least @xmath361 at @xmath137 , @xmath48 must choose @xmath137 for @xmath362 consecutive times . finally , the probability that during this interval @xmath48 will not communicate with any robot other than @xmath62 is lower bounded by @xmath363 the probability that ( ii ) occurs is thus lower bounded by @xmath364 combining the bounds for ( i ) and ( ii ) , it follows that @xmath365\geq \bigl(\tfrac{1}{{\left|q\right|}}\bigr)^{\lceil \frac{\delta}{\tau } \rceil } e^{-{{\lambda_{\textup{comm}}}}(\delta+\tau ) n}.\ ] ] the same lower bound holds for @xmath366 $ ] , meaning that @xmath367&={\mathbb{p}}\left[e_{i}\right]\ , { \mathbb{p}}\left[e_{j}\right ] \geq \bigl(\tfrac{1}{{\left|q\right|}}\bigr)^{2 \lceil \frac{\delta}{\tau } \rceil } e^{-2 { { \lambda_{\textup{comm}}}}(\delta+\tau ) n}.\end{aligned}\ ] ] if event @xmath368 occurs , then robots @xmath48 and @xmath62 will be at adjacent vertices for an amount of time during the interval @xmath347 equal to @xmath369 since @xmath343 and @xmath344 are no more than @xmath370 , we can conclude that @xmath48 and @xmath62 will be within @xmath116 for at least @xmath236 . conditioned on @xmath368 occurring , the probability that @xmath48 and @xmath62 communicate in @xmath347 is lower bounded by @xmath371 . a suitable choice for @xmath214 from the statement of the lemma is thus @xmath372 it can be shown that this also constitutes a lower bound for the other possible combinations of initial states : robot @xmath48 is _ waiting _ and robot @xmath62 is _ moving _ ; robots @xmath48 and @xmath62 are both _ moving _ ; and robots @xmath48 and @xmath62 are both _ waiting_. to create a contradiction , assume that @xmath51 is a pairwise - optimal partition but not a centroidal voronoi partition . in other words , there exist components @xmath43 and @xmath57 in @xmath64 and an element @xmath138 of one component , say @xmath373 , such that @xmath374 choose @xmath57 such that for all @xmath375 @xmath376 let @xmath377 be a shortest path in @xmath17 connecting @xmath137 to @xmath345 and let @xmath378 be the first element of the path starting from @xmath379 which is not in @xmath57 . let @xmath380 be such that @xmath381 . in the first case , we again have a contradiction using the same logic above with @xmath178 in place of @xmath138 . in the second case , we must further consider whether there exists a @xmath389 such that every vertex in @xmath389 is also in @xmath390 . if there is not such a path , then @xmath391 and we again have a contradiction as above . if there is such a path , then we can instead repeat this analysis using using @xmath380 in place of @xmath62 and considering the path formed by this @xmath389 and the vertices in @xmath383 after @xmath178 . since the next vertex playing the role of @xmath178 must be closer to @xmath138 , we will eventually find a vertex which creates a contradiction . r. smith , j. das , h. heidarsson , a. pereira , f. arrichiello , i. cetnic , l. darjany , m .- e . garneau , m. howard , c. oberg , m. ragan , e. seubert , e. smith , b. stauffer , a. schnetzer , g. toro - farmer , d. caron , b. jones , and g. sukhatme , `` usc cinaps builds bridges , '' _ ieee robotics & automation magazine _ , vol . 17 , no . 1 , pp . 2030 , 2010 . s. yun , m. schwager , and d. rus , `` coordinating construction of truss structures using distributed equal - mass partitioning , '' in _ international symposium on robotics research _ , ( lucerne , switzerland ) , aug . 2009 . l. c. a. pimenta , v. kumar , r. c. mesquita , and g. a. s. pereira , `` sensing and coverage for a network of heterogeneous robots , '' in _ ieee conf . on decision and control _ , ( cancn , mxico ) , pp . 39473952 , dec . 2008 . r. cortez , h. tanner , and r. lumia , `` distributed robotic radiation mapping , '' in _ experimental robotics _ ( o. khatib , v. kumar , and g. pappas , eds . ) , vol . 54 of _ springer tracts in advanced robotics _ , pp . 147156 , springer , 2009 . f. bullo , r. carli , and p. frasca , `` gossip coverage control for robotic networks : dynamical systems on the the space of partitions , '' _ siam journal on control and optimization _ , aug available at http://motion.me.ucsb.edu/pdf/2008u-bcf.pdf . j. w. durham , r. carli , p. frasca , and f. bullo , `` discrete partitioning and coverage control with gossip communication , '' in _ asme dynamic systems and control conference _ , ( hollywood , ca , usa ) , pp . 225232 , j. w. durham , r. carli , and f. bullo , `` pairwise optimal coverage control for robotic networks in discretized environments , '' in _ ieee conf . on decision and control _ , ( atlanta , ga , usa ) , pp . 72867291 , dec . 2010 . b. gerkey , r. t. vaughan , and a. howard , `` the player / stage project : tools for multi - robot and distributed sensor systems , '' in _ int . conference on advanced robotics _ , ( coimbra , portugal ) , pp . 317323 , june 2003 .

the possibility of translating logic programs into functional ones has long been a subject of investigation . common to the many approaches is that the original logic program , in order to be translated , needs to be _ well - moded _ and this has led to the common understanding that these programs can be considered to be the `` functional part '' of logic programs . as a consequence of this it has become widely accepted that `` complex '' logical variables , the possibility of a dynamic selection rule , and general properties of non - well - moded programs are exclusive features of logic programs . this is not quite true , as some of these features are naturally found in _ lazy _ functional languages . we readdress the old question of what features are exclusive to the logic programming paradigm by defining a simple translation applicable to a wider range of logic programs , and demonstrate that the current circumscription is unreasonably restrictive . _ keywords : _ logic programming , functional programming , lazy evaluation . _ acm computing classification system : _ d.1.1 ( applicative functional programming ) ; d.1.6 ( logic programming ) ; d.3.2 ( language classifications ) f.3.3 ( studies of program constructs ) programs )

cs0003070

the possibility of translating logic programs into functional ones has long been a subject of investigation . among the different proposals @xcite . such systems are usually devised for one of the following purposes : for proving program properties , for providing better insight on the relation between functional and logic languages , or to a minor extent for improving program performance . common to all the approaches mentioned is that the original logic program , in order to be translated , needs to be _ well - moded _ and this has led to the common understanding that these programs can be considered to be the `` functional part '' of logic programs . this is confirmed by the following statement in @xcite : `` @xmath0 the class of functionally moded ( well - moded and simply moded ) programs can be rightly considered _ the _ functional core of logic programs '' . well - moded programs have , among other features , a straightforward left - to - right dataflow model ( see @xcite ) and prohibit the use of _ logical variables _ to their full potential such as in complex logical data structures like difference - lists . as a consequence of this it is now widely accepted that `` complex '' logical variables , the possibility of a dynamic selection rule , and general properties of non - well - moded programs are exclusive features of logic programs . this is not quite right . at least , not to the extent that one is brought to think . in this paper we show , among other things , that logical structures such as difference lists have a natural counterpart in _ lazy _ functional programs ; i.e. that most programs using difference - lists are functional in nature . this shows immediately that many common non - well - moded programs are functional in nature and that well - modedness is thus not a necessary attribute of those logic programs behaving functionally . we do this by employing a straightforward literal translation of moded logic programs into haskell , a lazy functional language . furthermore , we use the same translation system to show that some programs requiring a dynamic scheduling mechanism are also intrinsically functional . summarizing , in this paper we readdress the old question of what features are exclusive to the logic programming paradigm and demonstrate that the current circumscription is unreasonably restrictive . due to space constraints we omit preliminaries and assume that the reader is acquainted with the terminology and the main results of logic programming theory ( see @xcite ) . in this paper we use over - lined characters to indicate ( a possibly empty ) sequence of objects , so @xmath1 can denote a sequence @xmath2 of terms , @xmath3 a sequence of variables and @xmath4 a sequence of atoms ( i.e. a query ) . to avoid confusion with built - in symbols , we use @xmath5 to indicate syntactic equivalence . in what follows we study logic programs executed by means of the _ ld - resolution _ , which consists of the sld - resolution combined with the leftmost selection rule . an sld - derivation in which the leftmost selection rule is used is called an _ ld - derivation_. this section is partially borrowed from @xcite , we refer to the appendix and to @xcite for further information over well - moded logic programs . [ def : mode ] consider an @xmath6-ary relation symbol @xmath7 . by a _ mode _ for @xmath7 we mean a function @xmath8 from @xmath9 to the set @xmath10 . if @xmath11 , we call @xmath12 an _ input position _ of @xmath13 and if @xmath14 , we call @xmath12 an _ output position _ of @xmath7 ( both w.r.t . @xmath8 ) . an @xmath6-ary relation @xmath15 with a mode @xmath8 will be denoted by @xmath16 for example , for programs ` member ` and ` append ` we typically have the following listings and modes : mode member(in , in ) . + member(el , [ el| _ ] ) . + member(el , [ _ |rest ] ) @xmath17member(el , rest ) . + mode append(in , in , out ) . + append ( [ ] , _ , [ ] ) . + append([h|tail ] , list , [ h|tail ] ) @xmath17append(tail , list , tail ) . modes indicate how the arguments of a relation should be used . we assume that to each relation symbol is associated a unique mode . multiple modes can be obtained by simply renaming the relations . in presence of modes , we require the programs and the queries to be somehow naturally _ consistent _ wrt them . before we introduce the notion of consistency we have to provide some further notation . when writing an atom as @xmath18 we now assume that @xmath19 is a sequence of terms filling in the input positions of @xmath7 and @xmath20 is a sequence of terms filling in the output positions . thus , for notational simplicity , we assume that the input positions come first . let us call _ producing _ the input position of the head and the output positions of the body atoms , and _ consuming _ the other positions of a clause , we have the following definition a clause ( query ) is _ consistent _ iff every variable occurs in at least one producing position . the last lp notion we need is the one of _ plain _ program . here and in the sequel , a set of terms is called _ linear _ if every variable occurs at most once in it . in other words , a sat of terms is linear iff no variable has two distinct occurrences in any of the terms and no two terms have a variable in common . [ def : plain ] a clause @xmath21 is called _ plain _ if 1 . @xmath22 is a linear family of variables ; 2 . @xmath23 is linear . a query @xmath24 is called _ plain _ iff the clause @xmath25 is , where ` q ` is any ( dummy ) atom of zero arity . a program is called _ plain _ if every clause of it is . thus a plain program is a program in which producing positions are filled in by variables and in which a variable occurs in at most one producing position . condition ( i ) is similar to , though less restrictive , than the one of simply moded programs as defined in @xcite ( as we do not impose an ordering constraint ) . our translation requires programs to be consistent and plain . this is far less restrictive than well - modedness plus simple - modedness , and as we shall see allows us to capture a broader segment of functional behaviour found in logic programs . indeed , it is now perhaps _ too _ lenient , but suffices for the goals of this paper to broaden the characterization of logic programs . regarding the ( non ) restrictiveness of the concepts of plain and consistent programs , we have the following : [ rem : sm ] it is important to realize that most programs are plain , and that non plain programs can naturally be transformed into equivalent plain ones , virtually all consistent programs are either plain or safely translatable into a plain form . this is also practically demonstrated by the fact that the language mercury employs a pre - processing phase in which all programs are translated into a _ superheterogeneous _ form ( which is very similar to the form of plain programs ) . concerning ( ii ) , we can always transform a consistent program @xmath26 into an equivalent consistent program @xmath27 which satisfies it . for instance , for the member program defined above , we can transform its first clause into member(el , [ head|rest ] ) @xmath17el = = head . it is also worth noticing that append is already input - linear . our translation system maps logic programs into lazy functional programs , which are written in ( a subset of ) haskell @xcite . the subset we use includes the proposed extension of _ pattern guards _ @xcite , which we describe below . the programs we are going to generate are built as sets of _ equations _ , each of the following form : f s@xmath28 @xmath0 s@xmath29 @xmath30 guard@xmath31 , @xmath0 , guard@xmath32 = result@xmath28 + @xmath33 + @xmath30 guard@xmath34 , @xmath0 , guard@xmath35 = result@xmath34 + @xmath30 otherwise = result@xmath36 where is a function symbol , are _ parameters _ and @xmath37 are _ guard qualifiers_. the ` @xmath30 ' introduces a guard , and the ` @xmath38 ' acts as a logical conjunctive . pattern matching may take place on the parameters . without pattern guards , the guard qualifiers would have to be boolean expressions ; that is , we would only return @xmath39 if the associated guards @xmath40 all evaluate to true . the semantics of haskell dictates that definitions and guards are tried in sequential order . the situation with pattern guards is somewhat different . in fact , _ patterns guards _ can also contain _ let - expressions _ ( which are defined as usual ) and _ pattern - matching expressions _ which are expressions of the form , and whose semantics is the following : if matches with then the variables in are appropriately instantiated , and the pattern - matching guard returns true , otherwise it return false . consider : f z @xmath30 let x = g z + @xmath38 [ y ] @xmath17x + @xmath38 y @xmath41 10 = ( true,1 ) + @xmath30 otherwise = ( false,0 ) here , the tuple will only be returned if the two qualifiers in the first guard succeed . that is , 1 ) the argument can be pattern matched to a list of one element ( denoted by , which also binds to this element ) , and 2 ) the boolean condition is true . in all this , the value of @xmath42 is determined by @xmath43 . if any of these fail , then the second guard is tried . in this case , the special guard will be tried , which always succeeds . a qualifier can also be introduce recursive bindings ; this will become clear in the sequel . a more detailed example is presented in appendix [ app : more ] . as explained below , we need to capture the fact that a predicate might succeed ( possibly returning a computed answer substitution ) , or fail . to do so , we introduce a new datatype @xmath44 in our haskell programs by : data result @xmath45 = suc @xmath45 @xmath30 fail that is , the datatype @xmath44 has two constructors , @xmath46 and @xmath47 , the latter of which can be applied to some term . note that the haskell programs which we generate are not the obvious programs that a functional programmer would write this is not the intention . they do , however , do what the logic programmer intended . all of the programs given in this paper can be compiled by any haskell compiler supporting pattern guards . we also want to mention that despite the fact that pattern guards are an important feature of our translation , we could do without them at the price of less elegant translation . a haskell compiler will usually regard these as syntactic sugar anyway , and compile them into more basic primitives already found in haskell . this implies that all the statements we are going to give in the sequel are true regardless of the availability of a pattern guard construct in the target language . in logic programming , queries can succeed , loop or fail . this third possibility is of crucial importance , since it is often used as a control mechanism . as an example , one can consider the following programming scheme : p(x ) @xmath17generate(x ) , test(x ) . where ` test ` verifies that the value produced by ` generate ` is appropriate , and failure and backtracking take care of the ill - formed terms . another common scheme is the following one : p(x ) @xmath17test_a , x = 1 . + p(x ) @xmath17test_b , x = 2 . where @xmath48 and @xmath49 model a typical ` case ` statement , and the selection of the right branch is done via the failure and backtracking mechanism . nevertheless , relations which are `` not supposed to fail '' are quite common in logic programming . we say that a relation is `` not supposed to fail '' if when called in a `` correct '' way produces at least one answer . examples of such relations are ` sort ` , ` flatten ` and ` append ` , ( this latter , for calls of the form @xmath50 , where @xmath51 and @xmath52 are lists , and @xmath53 is a new variable will always produce one answer ) . the ubiquity of predicates which are not supposed to fail is confirmed by the fact that mercury requires the programmer to specify for each relation symbol , whether it might fail or not . this information is then used to generate optimized code . we do the same thing for our translation , and from now on we assume that the set of predicate symbols is partitioned into two disjoint sets , namely non - test : : predicates , which , when called , are expected to produce at least one answer ( such as append ) , and test : : predicates , which when called are allowed to report no answer , i.e. , to fail immediately or succeed ( such as member and @xmath54 ) . thus , we have the following definition . [ def : partitioning ] a _ partitioning _ is a map from the set of predicate symbols into the set @xmath55 . let @xmath26 be a program and @xmath56 be a set of queries , we say that @xmath26 is _ correct _ wrt . @xmath56 iff for every @xmath57 every , time that a non - test atom @xmath58 is selected in a ld - derivation of @xmath4 in @xmath26 then @xmath58 has at least one successful ld - derivation . thus every program is correct wrt . the trivial partitioning in which all predicates are _ test_. checking correctness is orthogonal to the purposes of this paper , but we should mention that it can be done either using abstract interpretation @xcite or on modes and types @xcite ; also mercury employs a system based on modes and types in order to check that the programs are consistent ( modulo non - termination ) wrt . the partitioning provided by the programmer . the partition into test and non - test predicates exposes the implicit failure mechanism present in logic programs . our translation will transform non - test predicates as ordinary functions , but transform test predicates by returning something of the type , allowing us to indicate failure . essentially if a function fails in the logic programming sense , then a value of @xmath59 will be returned . each `` value '' returned from a test predicate is only every used in a function if it is successfully matched against @xmath60 indicating that no failure occurred(it was not @xmath59 ) . the combination of plain programs with a partition makes it easy to identify the logic programs that can be mapped to functions , and which functions to enhance by mimicking the implicit failure mechanism . now , let @xmath7 be a predicate symbol with mode @xmath61 then @xmath7 can naturally be translated into a function of type @xmath62 where @xmath63 and @xmath64 are appropriate haskell types . here we will not bother further with the type that the translated predicate has : the haskell compiler will be able to infer it autonomously ; what it is important to see is that the haskell counterpart of ` p ` is a function which maps a tuple with @xmath12 elements into a tuple containing @xmath65 elements , possibly embedded in the @xmath66 datatype depending on whether ` p ` is a non - test predicate or not . we shall employ the value @xmath46 to denote the functional counterpart of failure , where @xmath67 looks like an empty tuple but which is in actual fact the only element in the unit type . ] . our translation method requires the program to be translated to be _ consistent _ and _ plain_. of these conditions , _ consistency _ is the only crucial one , in fact as stated in remark [ rem : sm ] it is ( virtually ) always possible to transform a consistent program into an equivalent program which is plain ; moreover , most programs are plain already . now , we can transform the logic program into a haskell one via a simple syntactic transformation . first , we have to translate variables , terms and predicate symbols ; this is done in a straightforward way : one just has to respect the syntactic conventions of the two languages ( uppercase and lowercases , and built - in predicates ) . of course , predicate symbols are transformed into non - constructor function symbols . in the sequel we use sans - serif characters for haskell constructs and typewriter font for logic programming ones , for instance , @xmath68 denote the haskell counterpart of the lp terms @xmath69 . [ def : translation ] let @xmath70 be a logic program , and + @xmath71 + be the set of clauses of @xmath70 defining predicate ` p ` , where the predicates @xmath72 are test predicates and the predicates @xmath73 are the non - test ones . here we assume that the clauses had been renamed apart , i.e. , that they share no variables . * if @xmath7 is a _ predicate , then the _ translation _ of the above section into haskell is the following script ( for the moment the underlined parts have to be treated as if the underline was nt there ) : + pro1pro1pro1pro1pro1= p ( @xmath74 ) @xmath30 ( t@xmath28 ) @xmath17(x ) , + suc ( o@xmath31 ) @xmath17p@xmath31(i@xmath31 ) , + @xmath33 + suc(o@xmath75 ) @xmath17p@xmath75(i@xmath75 ) , + let ( v@xmath31 ) = q@xmath31(u@xmath31 ) , + @xmath33 + let ( v@xmath76 ) = q@xmath76(u@xmath76 ) = + = ( s@xmath28 ) + @xmath30 ( t@xmath77 ) @xmath17(x ) , + @xmath33 + = ( s@xmath77 ) + @xmath33 + + if one of the clauses of the above section is a unit clause , ( i.e. @xmath78 ) or if its body contains no test predicates then the corresponding line has the trivial guard . + note that sequences may be empty , so if the predicate had no output positions , then @xmath79 would be the term @xmath80 . * if @xmath7 is a _ non - test _ predicate , then translation of the above section corresponds to the above script _ after removal _ of the underlined parts ; namely we have to eliminate from it the otherwise statement and the s from the return values . clearly , list constructions and built - in predicates need to be handled separately , in particular , a test predicate of the form @xmath81 will be transformed to the term which returns either @xmath82 on success and @xmath46 on failure . we will abuse the notation and also call this function @xmath83 . [ exa : append ] let us now consider the program append . it is already plain , so , assuming ` append ` to be non - test predicate , its translation is : append ( x1,x2 ) @xmath30 ( [ ] , list ) @xmath17(x1 , x2 ) + = list + @xmath30 ( ( x : xs ) , list ) @xmath17(x1 , x2 ) + @xmath38 let tail = append ( xs , list ) + = x : tail notice that nothing prohibits us from declaring ` append ` as a _ test _ predicate , if we do so we , the result of the translation is : append ( x1,x2 ) @xmath30 ( [ ] , list ) @xmath17(x1 , x2 ) + = suc list + @xmath30 ( ( x : xs ) , list ) @xmath17(x1 , x2 ) + @xmath38 suc tail @xmath17append ( xs , list ) + = suc ( x : tail ) + @xmath30 otherwise = fail in practice , the first program is more efficient than the second one ( though , this can differ per compiler ) , and it is more lazy than the second one . this is further explained in the following _ [ rem : strictness ] the adopted partitioning has a natural influence on the strictness of the resulting haskell code . consider the differences in the above translations of : if is declared as non - test then its translation will contain a let - expression in the guard while if it is declared as test then , in its place , we will find the guard now , while ( 1 ) is a let expression whose bound expression will only be invoked if ( and to the extent that ) the value of is demanded , the second is a guard , which has to be satisfied in order for the function it appears in to return a value . indeed , in ( 2 ) the term will always be reduced until it is completely computed , i.e. until it either reaches either `` success '' or `` failure '' . in this sense ( 2 ) is strict , while ( 1 ) is lazy . this behaviour is quite natural if one considers the following : since test atoms might fail , we can not trust their partial answers until we have computed whether they ll succeed or not . this implies that they always have to be fully `` computed '' , therefore forcing a strict computation . on the other hand non - test predicates are guaranteed to eventually succeed , so their computation might be stopped at the moment that we have reached a partial result which is `` sufficient for our purposes '' . therefore the non - test predicates naturally fit the lazy model of computation . we are at last in a position to demonstrate our thesis that the set of logic programs considered as functional needs to be expanded . as issues , we consider logic variables , dynamic scheduling and backtracking in turn . the dynamics of some of the programs we are going to present in this section is unavoidably rather complex , we apologize for the inconvenience and ask the reader to resort to patience and understanding . _ logical variables _ are one of the peculiarities of logic programming . most of the time , they are used in a standard way , that is just as variables in an imperative language this is the case for instance when the program is well - moded . nevertheless there are many important situations in which logical variables are exploited in all their power . a typical such case is in the presence of _ difference structures _ such as _ difference lists_. here we show that even when used in a truly `` logical '' way , logical variables are in many cases not an exclusive feature of logic programs . the following polish flag problem example ( incidentally , a simplified version of dijkstra s dutch flag problem ) , reads as follows : given a list of objects which are either red or white , rearrange it in such a way that the red elements appear first and the white ones appear after them . the following program is inspired by @xcite , we have replaced `` @xmath84 '' by `` , '' , thus splitting a position filled in by a difference - list into two positions . because of this change in some relations , additional arguments are introduced . polish(inlist , redwhites ) @xmath17 + distribute(inlist , redwhites , whites , whites , [ ] ) . + distribute ( [ ] , reds , reds , whites , whites ) . + distribute([x|xs ] , [ x|reds0 ] , reds , whites0 , whites ) @xmath17red(x ) , + distribute(xs , reds0 , reds , whites0 , whites ) . + distribute([x|xs ] , reds0 , reds , [ x|whites0 ] , whites ) @xmath17white(x ) , + distribute(xs , reds0 , reds , whites0 , whites ) . + mode polish(in , out ) : non - test . + mode distribute(in , out , in , out , in ) : non - test . where we assume that predicates red and white are appropriately defined elsewhere in the program and have mode ` red(in ) : test ` . this program is plain and consistent , and by translating it we obtain polish inlist @xmath30 t @xmath17inlist , + let ( redwhites , whites ) = distribute(t , whites , [ ] ) + = redwhites + distribute(is , rstail , wstail ) @xmath30 ( [ ] , rs , ws ) @xmath17(is , rstail , wstail ) + = ( rs , ws ) + @xmath30 ( x : xs , rs , ws ) @xmath17(is , rstail , wstail ) + , let ( rshead , wshead ) = distribute(xs , rs , ws ) + , suc ( ) @xmath17red x + = ( x : rshead , wshead ) + @xmath30 ( x : xs , rs , ws ) @xmath17(is , rstail , wstail ) + , let ( rshead , wshead ) = distribute(xs , rs , ws ) + , suc ( ) @xmath17white x + = ( rshead , x : wshead ) + where and are again defined elsewhere in the program . this program runs perfectly well . notice that the definition of employs a _ circular data structure _ : in fact the variable appears both on the left hand side and on the right hand side of the expression @xmath85)$ ] , in the guard . circular data structures were first advocated by bird @xcite in order to avoid multiple traversal of data structures , and since then have become a standard tool of _ lazy _ functional languages . it is worth remarking that the above program ( and most other programs employing circular structures ) would not function properly if we had used a _ strict _ functional language . it is important to notice that the original logic program employs logical variables in a highly non - trivial way . this is confirmed by the fact that the program is not well - moded . the fact that a program using difference - lists actually presents a functional behaviour is not incidental . consider an atom containing a difference - list @xmath86 ( for the sake of simplicity , we assume that it does not have any other argument ) , as above , we split this position in two , and obtain @xmath87 . now , the whole idea of having difference lists , is that when a computation starting in ( an instance of ) @xmath87 will succeed , it will report a computed answer substitution ( c.a.s . ) @xmath88 such that @xmath89 is ( `` points to '' ) the tail of @xmath90 . this implies that for all @xmath91 if @xmath92 is ground , then @xmath93 is ground as well . typically , after @xmath86 has succeeded with c.a.s . @xmath88 , @xmath89 will eventually be unified with a ground ( classical ) list ( or with the head of another difference - list , in which case the reasoning continues by considering the tail of this second difference - structure ) . after this unification has taken place , @xmath90 is going to be a classical list , which can be employed as normal . in this sense we have that @xmath94 depends on @xmath95 , therefore @xmath94 has to be considered output and @xmath95 input , and the above atom should be translated into ; the only problem is that @xmath95 is an input in disguise , in the sense that when @xmath96 is called , @xmath95 is typically not ( yet ) ground ( in other words , it is not yet known ) . however , this is hardly a problem when we consider _ lazy _ functional languages ( at the same time , it is the reason why programs with difference lists can not be easily translated into a strict functional language ) . one could argue that the above reasoning could be completely reversed , starting by saying that `` @xmath0 for all @xmath91 if @xmath97 is ground , then @xmath92 is ground as well , @xmath0 , thus if @xmath90 is unified to a ground list , then @xmath98 will become a ground list , and this shows that @xmath95 depends on @xmath94 , and that therefore the above atom should be translated into '' ( which would fail to function ) . this is in principle true ( dependencies in lp are always bidirectional ) , however , this property is never used , indeed it can not be used in practice for the following simple reason : after succeeding with c.a.s . @xmath88 , we typically have that @xmath99 $ ] and that @xmath100 . now , while it is always possible to unify @xmath89 with any ground list @xmath101 , trying to do this with @xmath90 will almost certainly lead to failure ( unless @xmath102 $ ] is a prefix of @xmath103 ) . therefore difference - lists are virtually always employed in a _ directional _ fashion . another example of a program using logical variables which can be safely translated into haskell is given in the following section . of course , one can find an example of a program using difference - lists which would not work in haskell ; actually , counterexamples are extremely easy to contrive : variables in lp are always adirectional , and if one fully exploits this will always obtain a program which has no functioning functional counterpart . we do nt want to deny this , on the contrary : here we are interested in how programs are _ usually _ used , and in pointing out that some standard methodologies which are normally considered as applicable only to lp , are actually not so . of course not all programs using logical variables are translated correctly : typical such examples are the programs which incrementally fill in a data structures such as in the _ eight queen _ example and in the ` sequence ` example in appendix [ sec : sequence ] ( these programs use unification in a crucial way , and this is confirmed by the fact that they are not _ consistent _ ) . other examples are circular programs such as the following one p(x ) @xmath17eq(x , x ) . + eq(x , x ) . moded as follows : ` p(in : ground):non - test ` and ` eq(out : ground , out : ground):non - test ` . this program is circular in a non - well - founded way , and this , when translated , yields a program which is not _ productive_. we can safely conclude that difference lists have a natural counterpart in the circular structures of lazy functional programming . another prominent property of logic programming is the possibility of having a dynamic selection rule , possibly guided by appropriate _ delay declarations_. let us consider the following example , which , given the list xs of integer values , del_max(xs , zs ) produces the list zs by deleting all the occurrences of its maximum element . del_max(xs , zs ) @xmath17find_max_and_del(xs , max , zs , max ) . + % find_max_and_del(inlist , el , outlist , max ) + % max is the maximum element of the list inlist , and + % outlist is obtained from inlist by deleting all the occurrences of el from it + find_max_and_del ( [ ] , _ , [ ] , 0 ) . + find_max_and_del([x | xs ] , el , ys , max ) @xmath17 + find_max_and_del(xs , el , zs , max ) , + sup(x , max , max ) , + del_if_first([x | zs ] , el , ys ) . + del_if_first([el | zs ] , el , zs ) . + del_if_first([x | zs ] , el , [ x|zs ] ) @xmath17x @xmath104 el . + mode del_max(in : list[int ] , out : list[int ] ) : non - test . + mode find_max_and_del(in : list[int],in : int , out : list[int],out : int):non - test . + mode sup(in : int , in : int , out : int ) : non - test . % defined in the obvious way + mode del_if_first(in : list[int ] , in : int , out : list[int ] ) : non - test . it is worth noticing that the program uses logical variables in a nontrivial way . this is confirmed by the fact that it is not well - moded . specifically , the variable ` max ` in the first clause is used as an asynchronous communication channel between processes , as the atom find_max_and_del(xs , max , max , zs ) uses ` max ` as input value that it has to produce itself . furthermore , the program requires an appropriate _ dynamic scheduling_. in fact , when run with a standard left - to - right selection rule , the query del_max(ts , zs ) ( @xmath105 being a list of natural numbers ) leads to a run - time error ( or to an incorrect answer ) , and , provided that we fix this problem , to a very inefficient computation . the first problem ( concerning the runtime error ) is due to the fact that the computation will soon a goal of the form ` del_if_first(ts , el , zs ) ` , where ` ns ` @xmath106)$ ] is a non - empty list of integers , and ` el ` and ` zs ` are distinct variables . at that point the interpreter will proceed and might reach the call @xmath107 , which being ` el ` a variable will flounder`== ` in order to implement @xmath104 . for instance in eclipse and swi - prolog the call ` n = = el ` fails , and a subsequent call ` n \backslash== el ` succeeds ( ! ) in eclipse by returning the empty c.a.s . and in swi - prolog by instantiating ` el ` to an apparently random numeric value . of course , this behaviours lead to solutions which are almost always incorrect . ] . in other words , this program can not be run with the normal leftmost selection rule . the second problem ( concerning program s inefficiency ) is due to the fact that the query del_max(xs , zs ) could return the list @xmath108 in linear time ( scanning @xmath109 only once ) , however , it is easy to see that if we employ any fixed selection rule , the program has to go through a remarkable amount of backtracking , which makes it run in quadratic time on the length of the input list , where @xmath110 is still a variable . for this reason @xmath111 will take a guess and delete an element which will usually turn out not to be the correct one , and this will eventually produce backtracking . since the number of needed backtracking steps is linear in the size of the input list , this decreases the performance of the program from linear to quadratic time . in order to avoid this problem , we have to make sure that an atom of the form @xmath112 will be selected in the derivation _ only when _ @xmath110 will be instantiated to a ground term . we can do this by employing the above delay declarations . ] . both problems can be solved by employing a _ dynamic selection rule _ and by prohibiting the selection of certain atoms until their arguments are sufficiently instantiated using for instance the following _ delay declarations _ @xcite : delay sup(x , y , _ ) until ground(x ) @xmath113 ground(y ) . + delay @xmath104(x , y ) until ground(x ) @xmath113 ground(y ) . + delay del_if_first([x|xs ] , el , _ ) until ground(x ) @xmath113 ground(el ) for instance , the first declaration will suspend any call to ` sup(t , s , v ) . ` until ` t ` and ` s ` are ground terms . delay declarations have become an important standard control tool and are implemented in various versions of prolog ( for instance in sixtus prolog and in eclipse @xcite ) and in the language gdel @xcite . now , let us for a moment not bother about the delay declarations and translate this program into haskell . we obtain the following script . del_max as @xmath30 xs @xmath17as + , let ( zs , maxel ) = find_max_and_del ( xs , maxel ) + = zs + find_max_and_del ( x1,x2 ) @xmath30 ( [ ] , el ) @xmath17(x1 , x2 ) + = ( [ ] , 0 ) + @xmath30 ( x : xs , el ) @xmath17(x1,x2 ) + , let ( zs , maxel ) = find_max_and_del ( xs , el ) + , let maxel = max x maxel + , let ys = del_if_first ( x : zs , el ) + = ( ys , maxel ) + del_if_first ( x1 , x2 ) @xmath30 ( [ ] , el ) @xmath17(x1 , x2 ) + = [ ] + @xmath30 ( x : xs , el ) @xmath17(x1,x2 ) + , x = = el + = zs + @xmath30 ( x : xs , el ) @xmath17(x1,x2 ) + , x /= el + = x : zs this program works fine , and his runtime complexity is linear in the size of the input . we can therefore state that the lazy computational mechanism compensates for the lack of control over dynamic scheduling , without which the above logic program could not be run or would have a quadratic complexity . thus although the mechanism of lazy evaluation and delay declarations are quite different ( actually , they are the opposite : the call - by - need mechanism determines which term _ has _ to be reduced , while delay declarations determine which atoms _ should not _ be resolved ) , they often accomplish the same thing . the fact that lazy evaluation here plays a crucial role is confirmed by the fact that , if we had declared all predicates to be test predicates ( thus forcing strictness , as explained in the remark [ rem : strictness ] ) the translated program would not function properly . thus again we are in presence of a program exploiting logical variables in a complex way which nevertheless has a natural translation into haskell . another outstanding feature of logic programs is their backtracking mechanism , which virtually implements a _ do nt know _ nondeterministic system . in the light of the above examples , we believe that nondeterminism is by far the most important and the mostly used peculiar feature of the logic programming paradigm . we do nt want to challenge this , on the contrary . at the same time , it is important for us to show to which extent a ( lazy ) functional program can mimic a logic program which uses backtracking . consider the following program . backtracker(x ) : - producer_a(y ) , picky_modifier(y , x ) . + backtracker(x ) : - producer_b(y ) , picky_modifier(y , x ) . + producer_a(``a '' ) . + producer_b(``b '' ) . + picky_modifier(``b '' , `` c '' ) . the adopted mode and partitioning is backtracker(out ) : non - test + producer_a(out ) : non - test ( and the same for producer_b + picky_modifier(in , out ) : test its translation is the following : backtracker @xmath30 let y = producer_a + , suc x @xmath17picky_modifier y + = x + @xmath30 let y = producer_b + , suc x @xmath17picky_modifier y + = x + producer_a = `` a '' + producer_b = `` b '' + picky_modifier x @xmath30 x = = `` b '' + = suc `` c '' + @xmath30 otherwise = fail + the haskell translation is able to report all the correct answers , even though in lp for the query @xmath114 in order to return the answer x = `` a '' , the interpreter has to go through some backtracking . notice in fact that the above logic program is not deterministic . consider now the following program scheme : p(x ) @xmath17generate(x ) , test(x ) . it is immediate to translate it and to check that if ` generate ` has more than one solution then the translation does not behave as the logic program does : while the query : - p(x ) succeeds provided that one of the solutions of ` generate(x ) ` satisfies ` test(x ) ` , the haskell translation manages to report one answer only so in the unlikely case that the first solution founded by ` generate(x ) ` satisfies ` test(x ) ` ; in all other cases @xmath115 reduces to @xmath46 . the key factor for the translation to work correctly we need to avoid logic programs in which consistent queries might originate sld trees with more than one successful ( sub- ) branch . there exists techniques based on list - comprehension in order to translate logic programs into functional programs in such a way that the resulting program will ( eventually , lazily ) report the list of all the answers that the initial logic program would . in those cases , however , one can clearly not talk of a _ literal _ translation , which is the starting point of our research ( programs able to return more than one answer are in our opinion intrinsically logic programs , and therefore do not belong to our target ) . to be precise , a non - deterministic logic program can be safely translated onto haskell provided it is _ input discriminative _ , as defined as follows : [ def : id2 ] let @xmath70 be a program , @xmath116 be its least herbrand model , and p@xmath28(i@xmath28 , o@xmath28 ) = @xmath17@xmath117@xmath28 , @xmath118@xmath28 . + @xmath33 + p@xmath119(i@xmath119 , o@xmath119 ) @xmath17@xmath117@xmath119 , @xmath118@xmath119 . be the complete set of the rules of @xmath70 ( where the conjunction @xmath117@xmath120 contains only test predicates and @xmath118@xmath120 contains only non - test predicates ) . we say that @xmath70 is _ input discriminative _ if for each @xmath121 $ ] , such that @xmath122 we have that * for each ground @xmath88 such that @xmath123 we have that @xmath124 . it is worth noticing that this concept of input discriminative program is rather less restrictive than the concept of _ deterministic _ program , and that input - discrimitative programs might still require non - trivial ( non - shallow , see ( * ? ? ? * ch . 6 ) ) backtracking . we could say that these programs admit some _ shallow nondeterminism_. summarizing , there is a point to be remarked , that strictly speaking the haskell translation of a program can _ always _ mimic the backtracking taking place in the original logic program . what the haskell translation can__not _ _ do is report multiple answers . the feature of logic programs of being able of reporting more than answer , and how this is handled in the different translation systems is a topic which deserves a separate discussion . regarding this issue , the literature on papers presenting a translation from logic to functional programs can be divided in two main groups . on one side we find papers which are not concerned with the nondeterminism ( or the backtracking ) mechanism of logic languages @xcite , these papers are usually mainly concerned in providing a transformation system which allows one to prove program properties such as termination of the original logic program . for this reason they focus on obtaining a translation which maps only the non - failing computations correctly . in these papers the failure and backtracking mechanism are disregarded during the translation . on the other side , we find @xcite , in which the authors propose a translation in which the full ( prolog - like ) computational mechanism is preserved , including the possibility of having multiple answer for the same query and the possibility of failure . this is achieved by letting a query return the _ list _ of computed answer substitutions , where the empty list corresponds to the failing case , in the same way advocated by wadler @xcite . the lazy computational mechanism then takes care of computing only those answers which are necessary , and backtracking is faithfully rendered by a standard list - comprehension schema . the translation system we have employed lies somewhere in the middle between those two methods . our goal was to take also failure into account , yet retaining a literal translation system , in which the computational mechanism of the resulting functional program is as similar as possible to the one of the original logic program . of course we can only correctly translate programs which do not return more than one answer for the same query ( at the same time , it is important to notice that these programs do nt have to be deterministic ; for instance ` member ` is nondeterministic ) . in our opinion , the possibility of returning more than one answer is to be considered a _ peculiar _ one of the lp paradigm , and the fact that it can be emulated by functional programs does not obliterate our position . the goal of our research was to investigate to which extent some features considered peculiar of the logic programming paradigm are really so . for this purpose we have devised a simple literal system which enabled us to translate logic programs into the lazy functional language haskell . it is known ( see also @xcite ) that _ if we restrict our attention to non - failing , non - backtracking computation _ then well - moded simply moded programs have a natural counterpart in a functional language . the properties of being well - moded and simply moded indicates a manner of use of variables in logic programming which is undoubtedly `` functional '' . to this statement we want to add that well- and simply moded programs can be considered as _ strictly _ functional , as they can be safely translated into a _ strict _ functional language . in this paper we have shown that in a _ lazy _ functional language , this picture broadens significantly , and some of the features that were in the light of the results above commonly considered as exclusive of the logic programming paradigm , can be naturally found in a lazy functional language such as haskell . in particular , we have shown that the use of complex logical variables in data structures such as difference lists ( or such as in program in ` del_max ` ) find a natural counterpart in the circular structures @xcite of lazy functional programs . these structures that were commonly considered as `` structurally logical '' are thus not so . we can then attempt a rough classification of logic programs according to the level of complexity at which they employs their variables ( backtracking and nondeterminism is not considered here ) . we then have the following division . 1 . _ strictly functional _ programs which use variables in a standard ( imperative - like ) way . these are characterized by being _ well - moded _ ( or by being so after permutation of the clause s body atoms ) . 2 . _ lazy functional _ programs which admit a safe translation into haskell : i.e. programs which can be translated into haskell ( via the syntactic translation ) and whose operational behaviour is isomorphic to the one of their functional counterpart . intrinsically logical _ programs which do not admit a safe translation into haskell with our translation scheme . this raises the interesting question of how large is the class of intrinsically logical programs . without pretending to be able to characterize extensively this limit , it is interesting to notice that programs which are plain and consistent and which either admit a layered mode @xcite , or are s - well - typed programs @xcite are safely translatable into haskell ( modulo the possibility of backtracking , which is discussed in the sequel ) . as argued in @xcite , we believe that these programs actually encompass the majority of actual programs which use logical variables in a non - elementary way . we think that a classification and understanding of these levels might be useful both to enhance the performance of logic languages ( as already done to some extent in the language mercury ) and to prove more precise program properties . furthermore , we have also addressed another logical feature : the possibility of dynamic scheduling . in theory in lp _ any _ atom is selectable as all selection rules yield the same successful derivations . in practice this does not work , and adopting a random selection rule would in the best case yield to an explosion of the search space ; for this reason prolog uses a fixed left - to - right selection rule , a feature which is either explicitly or implicitly always exploited by the programmers . however , some programs ( like ` delmax ` above , or concurrent - like programs ) are not correct under a fixed search rule . in these cases the `` right '' selection strategy is enforced by the use of appropriate delay declarations ( d.d . ) , which serve to indicate which atoms in a query should not be resolved . the implementation of d.d . is rather costly , as atoms are continuously being suspended and forced . here we have seen one example in which the lazy evaluation mechanism of haskell achieves the same effect of the use of d.d .. as we have pointed out , call - by - need can be regarded as a dynamic selection strategy , which is however based on an principle opposite to the one of d.d . in the sense that call - by need determines which term has to be reduced , while delay declarations determine which atoms should _ not _ be resolved . a naturally arising question here is whether it is possible to implement in logic programming languages a selection rule which is `` driven '' by a call - by - need mechanism , instead of `` restricted '' by the use of delay declarations . this could possibly lead to reduction of the suspension overhead and thus to performance improvements . the difficulty in implementing such a search rule lies in the fact that in lp it is not clear which output values depend on which input values ( actually , it is not clear what is input and what it is output to start with ) , so in order to implement such an intelligent selection strategy , one would need some sophisticated analysis tools which might either be based on abstract interpretation ( with tools similar to the ones of @xcite ) , or on refined versions of modes such as the ones described in @xcite . other works related to this subjects are @xcite . we have also discussed the fact that logic programs allow backtracking . we have seen that strictly speaking backtracking computations can be easily mimicked by the functional language by an appropriate use of the guards ; what can not be ( easily ) mimicked in haskell is the possibility of returning multiple answers , at least not unless one uses additional constructs such as the list - of - successes method @xcite . an interesting research direction might be to define appropriate monadic structures ( such as those in @xcite ) to capture the failure or success and returning of multiple arguments . this would broaden the set of logical programs which we can capture with our simple translation scheme , without adding signifant complexity to it . in conclusion , we have demonstrated with a simple , literal translation scheme that several features considered as belonging specifically to logic programming are found naturally in lazy functional programming , dismissing the folklore that the functional core of logic programming is contained in the set of well- and simple moded programs . dlgh97 k. r. apt and s. etalle . on the unification free prolog programs . in a. borzyszkowski and s. sokolowski , editors , _ proceedings of the conference on mathematical foundations of computer science ( mfcs 93 ) _ , lecture notes in computer science , pages 119 , berlin , 1993 . springer - verlag . apt and e. marchiori . reasoning about prolog programs : from modes through types to assertions . , 6(6a):743765 , 1994 . k. r. apt . introduction to logic programming . in j. van leeuwen , editor , _ handbook of theoretical computer science _ , volume b : formal models and semantics , pages 495574 . elsevier , amsterdam and the mit press , cambridge , 1990 . r. s. bird . . , 21:239250 , 1984 . j. boye and j. mauszyski . directional types and the annotation method . , 33:179220 , 1997 . h. coelho and j. c. cotta . . springer - verlag , berlin , 1988 . m. codish , s. k. debray , and r. giacobazzi . compositional analysis of modular logic programs . in acm , editor , _ twentieth annual acm sigplan - sigact symposium on principles of programming languages _ , pages 451464 . acm press , 1993 . s. debray , p. lpez - garca , and m. hermenegildo . . in lee naish , editor , _ proceedings of the 14th international conference on logic programming _ , pages 4862 , cambridge , 1997 . mit press . p. dembinski and j. maluszynski . -parallelism with intelligent backtracking for annotated logic programs . in _ proceedings of the international symposium on logic programming _ , pages 2938 , boston , 1985 . s. etalle and m. gabbrielli . layered modes . in f. de boer and m. gabbrielli , editors , _ proc . jicslp96 post - conference workshop on verification and analysis of logic programs _ , 1996 . tehcnical report tr-96 - 31 , dipartimento di informatica di pisa . s. etalle and f. van raamsdonk . beyond success and failure . in j. jaffar , editor , _ proceedings of the 1998 joint international conference and symposium on logic programming ( jicslp-98 ) _ , pages 190204 . mit press , june1519 1998 . h. ganzinger and u. waldmann . . in michal rusinowitch and jean - luc rmy , editors , _ conditional term rewriting systems , third international workshop _ , lncs 656 , pages 430437 . springer - verlag , 1992 . p. m. hill and j. w. lloyd . . the mit press , 1994 . p. hudak ( ed ) , s. l. peyton jones ( ed ) , and p. l. wadler ( ed ) . report on the programming language haskell a non - strict purely functional language , version 1.2 . , 27(5):1162 , may 1992 . s. lttringhaus - kappel . . phd thesis , friedrich - wilhelms - universitt bonn , 1992 . j. w. lloyd . . symbolic computation artificial intelligence . springer - verlag , berlin , 1987 . second edition . m. marchiori . . in giorgio levi and mario rodrguez - artalejo , editors , _ algebraic and logic programming , 4th international conference , alp94 _ , volume 850 of _ lncs _ , pages 223241 . springer - verlag , 1994 . m. marchiori . . in _ proceedings of the seventh international conference on functional programming languages and computer architecture ( fpca95 ) _ , pages 5565 , la jolla , california , 1995 . acm press . l. naish . an introduction to mu - prolog . technical report 82/2 , the university of melbourne , 1982 . . mit press , 1990 . s. l. peyton jones . a new view of guards . http://www.dcs.gla.ac.uk/@xmath125simonpj/ , 1997 . d. pedreschi and s. ruggieri . . technical report tr-97 - 21 , dipartimento di informatica , 1997 . u. s. reddy . transformation of logic programs into functional programs . in _ international symposium on logic programming _ , pages 187198 , silver spring , md , february 1984 . atlantic city , ieee computer society . m. r. k. krishna rao , d. kapur , and r. k. shyamasundar . . , 34(1):141 , 1998 . using program transformation to obtain methods for eliminating backtracking in fixed - mode logic programs . technical report 7 , universidad nacional autonoma de mexico , instituto de investigaciones en matematicas aplicadas y en sistemas , 1991 . l. sterling and e. shapiro . . mit press , 1986 . f. van raamsdonk . . in lee naish , editor , _ proceedings of the 14th international conference on logic programming _ , pages 168182 . mit press , 1997 . p. wadler . . in _ proceedings of the second conference on functional programming languages and computer architecture _ , lecture notes in computer science , pages 113128 . springer - verlag , 1985 . p. wadler . the essence of functional programming . in _ 19th annual symposium on principles of programming languages ( popl ) _ , santa fe , new mexico , 1992 . mark wallace and andr veron . two problems two solutions : one system eclipse . in _ proceedings iee colloquium on advanced software technologies for scheduling _ , london , april 1993 . the following concept is essentially due to dembinski and maluszynski @xcite ; we use here an elegant formulation due to rosenblueth @xcite . [ def : wm ] a clause @xmath126 is called _ well - moded _ if for @xmath127 $ ] @xmath128 a query @xmath129 is called _ well - moded _ iff the clause @xmath130 is , where @xmath131 is any ( dummy ) atom of zero arity . a program is called _ well - moded _ if every clause of it is . it is important to notice that the first atom of a well - moded goal is ground in its input positions and a variant of a well - moded clause is well - moded . furthermore , the notion of of well - modedness , is `` persistent '' , as shown by the following lemma . recall that a ld - resolvent is a resolvent in which the leftmost atom per the query is the selected one , and that an ld - derivation is a derivation obtained employing the leftmost selection rule , [ lem : wm - persist ] an ld - resolvent of a well - moded goal and a well - moded clause that is variable - disjoint with it , is well - moded . the next result is originally due to dembinski and maluszynski and follows directly from the definition of well - moded program . [ cor : wm ] let @xmath26 and @xmath129 be well - moded , and let @xmath132 be an ld - derivation of @xmath129 in @xmath26 . all atoms selected in @xmath132 contain ground terms in their input positions . that is , in presence of well - moded programs and queries , if we use a left - to - right computation schema we are sure that every time that we select an atom , the `` value '' of his input arguments has already been fully computed . this shows that well - moded programs have a straightforward left - to - right data - flow . under certain conditions well - moded programs are also _ unification - free_. to show this , we need a definition first . the following notion was first defined in @xcite . [ def : sm ] a clause @xmath133 is called _ simply moded _ if if @xmath134 is a linear family of variables and for @xmath135 $ ] @xmath136 a query @xmath129 is called _ simply moded _ iff the clause @xmath137 is , where @xmath131 is any ( dummy ) atom of zero arity . a program is called _ simply moded _ if every clause of it is . thus , assuming that in every atom the input positions occur first , a clause is simply moded if all output positions of every body atom are filled in by distinct variables , which do not occur earlier in the body nor in an input position of the head . it is worth noticing that as shown by the little survey in @xcite most programs are already simply - moded and that often non simply - moded programs can naturally be transformed into simply - moded ones , for instance the non - simply - moded clause @xmath138).$ ] can be transformed into @xmath139 = list'$ ] . the property of being simply moded is also `` persistent '' in the sense that the resolvent of a simply moded query with a simply moded clause is simply moded . in @xcite it is proven that if the program and the query are simply moded , then they generate an ld - derivation which is _ unification - free _ , i.e. that each time an atom @xmath140 is selected and resolved in it via a clause @xmath141 , then the unification of @xmath140 and @xmath142 does not really require a full unification algorithm , but can always be reduced to a _ double matching : _ one ( `` from '' @xmath140 `` to '' @xmath142 ) for the input positions and a second one ( `` from '' @xmath142 `` to '' @xmath140 ) for the output ones . this result clearly shows that simply and well - moded logic programs are functional in nature ( besides for the possibility of reporting multiple answers , of course ) . the following ( nonsense ) program embodies most of the concepts we use : append ( [ ] , x ) = suc x + append ( x1,x2 ) @xmath30 ( x : xs ) @xmath17x1 + @xmath38 suc tail @xmath17append ( xs , x2 ) + @xmath38 let newtail = ( x : tail ) + = suc newtail + @xmath30 otherwise = fail in a call of , the first equation will be tried first . here , will be pattern matched to the empty list . if this succeeds , then is matched to ( since both are variables , this will always succeed ) , and finally is returned . if the pattern match above failed , then the first guard will be tried , which tries to pattern match the first element of the tuple to a list with at least one element . should this succeed , the result of a recursive call to is matched against , and if successful is bound to the variable @xmath143 , followed by the returning of @xmath144 . if either of the pattern matches failed , then the second guard will be tried . this example is provided by the prolog formalization of a problem from coelho and cotta ( * ? ? ? * pag . 193 ) : arrange three 1 s , three 2 s , ... , three 9 s in sequence so that for all @xmath145 $ ] there are exactly @xmath12 numbers between successive occurrences of @xmath12 . sublist(xs , ys ) @xmath17 xs is a sublist of the list ys . + sublist(xs , ys ) @xmath17 app ( _ , zs , ys ) , app(xs , _ , zs ) . + sequence(xs ) @xmath17 xs is a list of 27 elements . + sequence ( [ _ , _ , _ , _ , _ , _ , _ , _ , _ , _ , _ , _ , _ , _ , _ , _ , _ , _ , _ , _ , _ , _ , _ , _ , _ , _ , _ ] ) . + question(ss ) @xmath17 ss is a list of 27 elements forming the desired sequence . + question(ss ) @xmath17 + sequence(ss ) , + sublist([1,_,1,_,1 ] , ss ) , + sublist([2,_,_,2,_,_,2 ] , ss ) , + sublist([3,_,_,_,3,_,_,_,3 ] , ss ) , + sublist([4,_,_,_,_,4,_,_,_,_,4 ] , ss ) , + sublist([5,_,_,_,_,_,5,_,_,_,_,_,5 ] , ss ) , + sublist([6,_,_,_,_,_,_,6,_,_,_,_,_,_,6 ] , ss ) , + sublist([7,_,_,_,_,_,_,_,7,_,_,_,_,_,_,_,7 ] , ss ) , + sublist([8,_,_,_,_,_,_,_,_,8,_,_,_,_,_,_,_,_,8 ] , ss ) , + sublist([9,_,_,_,_,_,_,_,_,_,9,_,_,_,_,_,_,_,_,_,9 ] , ss ) .

we present results from the most recent ( 20022004 ) observing campaigns of the eclipsing system kh 15d , in addition to re - reduced data obtained at van vleck observatory ( vvo ) between 1995 and 2000 . phasing nine years of photometric data shows substantial evolution in the width and depth of the eclipses . the most recent data indicate that the eclipses are now approximately 24 days in length , or half the orbital period . these results are interpreted and discussed in the context of the recent models for this system put forward by winn et al . ( 2004 ) and chiang & murray - clay ( 2004 ) . a periodogram of the entire data set yields a highly significant peak at 48.37 0.01 days , which is in accord with the spectroscopic period of 48.38 0.01 days determined by johnson et al . ( 2004 ) . another significant peak , at 9.6 days , was found in the periodogram of the out - of - eclipse data at two different epochs . we interpret this as the rotation period of the visible star and argue that it may be tidally locked in pseudosynchronism with its orbital motion . if so , application of hut s ( 1981 ) theory implies that the eccentricity of the orbit is = 0.65 0.01 . analysis of the uves / vlt spectra obtained by hamilton et al . ( 2003 ) shows that thesin of the visible star in this system is 6.9 0.3 km s . using this value ofsin and the measured rotation period of the star , we calculate the lower limit on the radius to ber , which concurs with the value obtained by hamilton et al . ( 2001 ) from its luminosity and effective temperature . here we assume that = 90 since it is likely that the spin and orbital angular momenta vectors are nearly aligned . one unusually bright data point obtained in the 1995/1996 observing season at vvo is interpreted as the point in time when the currently hidden star ( b ) made its last appearance . based on this datum , we show that star b is 0.46 0.03 mag brighter than the currently visible star a , which is entirely consistent with the historical light curve ( johnson et al . 2005 ) . finally , well - sampled and data obtained at the ctio / yale 1-m telescope during 2001/2002 show an entirely new feature : the system becomes bluer by a small but significant amount in very steady fashion as it enters eclipse and shows an analogous reddening as it emerges from eclipse . this suggests an extended zone of hot gas located close to , but above , the photosphere of the currently visible star . the persistance of the bluing of the light curve shows that its length scale is comparable to a stellar radius .

astro-ph0507578

kh 15d , an extraordinary pre - main - sequence eclipsing binary system , is located in the young ( 24 myr ) open cluster ngc 2264 ( @xmath11 @xmath12 760 pc ) . in 1970 , a russian circular announced the discovery of eight new variable stars in ngc 2264 ( badalian & erastova 1970 ) . kh 15d was among these , and was given the name svs 1723 . it was found to vary by 1.1 mag and was classified as an irregular variable . in 1972 , the new variables discovered by badalian & erastova ( 1970 ) were included in the 58th name - list of variable stars ( kukarkin et al . 1972 ) , and kh 15d was given the name v582 mon . prior to october 2000 , kh 15d had been observed in various surveys of the open cluster ngc 2264 . flaccomio et al . ( 1999 ) and park et al . ( 2000 ) both observed kh 15d at least once while the system was bright and included it in their catalogs , each identifying it as # 391 and # 150 , respectively . earlier surveys of ngc 2264 ( herbig 1954 ; walker 1956 ; vasilevskis , sanders , & balz 1965 ) did not identify kh 15d as a cluster member most likely because of its proximity to the b2 iii star , hd 47887 ( see fig . its light was simply lost in the bright glow of the b2 iii star . this object gained its notoriety as kh 15d when it was reported to have unusual properties in 1995 based on a project at the van vleck observatory ( vvo ) to photometrically monitor star - forming regions ( kearns & herbst 1998 ) . this object was observed to undergo eclipses that were remarkably deep ( @xmath133.5 mag in @xmath14 ) and unusually long ( @xmath1316 days in 1999/2000 ) . of central interest was the evolution of the shape and duration of the eclipses that had been observed over a 5-year period ( hamilton et al . 2001 ) . out of eclipse , the observable star in this system was determined to have a spectral type of k6/k7 ( hamilton et al . 2001 ; agol et al . 2004 ) , while during eclipse , the observed spectrum was simply an attenuated version of that out of eclipse . an international campaign to monitor this intriguing system was begun during july 2001 . the primary goal of this project was to obtain much information about the structure of the intervening material . continued photometric observations from around the world provided nearly complete coverage of three consecutive eclipses during the 2001/2002 observing season ( herbst et al . these consecutive eclipses demonstrated that , in addition to the yearly changes observed in the phased data , differences in the shape of each individual eclipse could be observed over the course of a season . the shapes of ingress and egress were successfully modeled by the steady advance or retreat of a knife edge " across a limb - darkened star . in order to match the duration of ingress and egress , however , herbst et al . ( 2002 ) determined that the occulting edge must be inclined by @xmath1315@xmath8 to the motion of the star . these authors also reported that the object s color becomes slightly bluer during eclipse , which indicates that most ( if not all ) of the light received during eclipse is due to scattering . polarization measurements made by agol et al . ( 2004 ) and gary schmidt ( private communication ) show that out of eclipse , the system exhibits low polarization consistent with zero . during eclipse , however , the polarization is observed to increase to roughly 2% across the optical spectrum . these results support the conclusion that the star is likely completely eclipsed so that the flux during eclipse is due entirely to scattered radiation , and that the obscuring material is most likely made up of relatively large grains ( @xmath1310 @xmath15 m ; agol et al . 2004 ) . high - resolution spectra obtained during the december 2001 eclipse ( hamilton et al . 2003 ) revealed that the system is still undergoing accretion and driving a bipolar outflow . recently , extended molecular hydrogen emission near 2 @xmath15 m was observed spectroscopically by deming , charbonneau , & harrington ( 2004 ) . given the spatial extent of the emission , and the observed h@xmath16 line profile , deming et al . ( 2004 ) suggest that the ambient h@xmath16 gas is being shocked by a bipolar outflow from the star and/or disk . further evidence for an extended and well - collimated outflow was provided by tokunaga et al . ( 2004 ) , who obtained broad - band and narrow - band infrared images of kh 15d , showing a jet - like h@xmath16 emission filament that extends out to @xmath17 from the object . the position angle ( @xmath1320 ) of the in - eclipse polarization is nearly parallel to this h@xmath16 filament ( agol et al . 2004 ) . many hypotheses have been proposed since 2001 to explain the evolving eclipses of kh 15d . most of these involve circumstellar material in some way . hamilton et al . ( 2001 ) , herbst et al . ( 2002 ) , winn et al . ( 2003 ) , and agol et al . ( 2004 ) all proposed the idea of an edge - on circumstellar disk with a warp that periodically passes in front of the star . barge & viton ( 2003 ) proposed that the eclipses were caused by an orbiting vortex of solid particles , and grinin & tambotseva ( 2002 ) suggested that an asymmetric common envelope in a binary system could accurately reproduce the eclipses . historical studies by winn et al . ( 2003 ) found that observations made between 1913 and 1950 were consistent with no eclipses . johnson & winn ( 2004 , hereafter jw04 ) discovered that between 1967 and 1982 , the system alternated between bright and faint states with the same period as observed today , but 180@xmath8 out of phase with the current eclipses . additionally , the eclipse depth was much smaller ( only 0.67 @xmath0 0.07 mag in @xmath14 instead of 3.5 mag ) , and when out of eclipse , the system was brighter by 0.9 @xmath0 0.15 mag than it is today ( jw04 ) . while herbst et al . ( 2002 ) first suggested the possibility of a binary companion , the light curve between 1967 and 1982 led jw04 to believe that the light from a second brighter star once contributed to the flux coming from this system , and that the second star is now completely obscured . these results have been confirmed and extended by data obtained at various observatories between 1954 and 1997 ( johnson et al . 2005 ) . models of the kh 15d system that explained the historical and modern - day light curves were put forward almost simultaneously by winn et al . ( 2004 ) and chiang & murray - clay ( 2004 ) . each model involves an eccentric pre - main - sequence binary whose motion periodically carries it behind a precessing circumbinary disk . these models are supported by the results of a recent high - resolution spectroscopic monitoring program of kh 15d : johnson et al . ( 2004 ) have observed radial - velocity variations that are consistent with a binary companion with an orbital period in agreement with the 48-day photometric period . in summary , kh 15d is now known to be a pre - main - sequence binary system with a strongly eccentric orbit of 48.38 days . the brighter component b is currently totally obscured by the circumbinary ring or disk , while only the fainter member a emerges from behind the obscuring material for just less than half the period . to explain the secular variations observed in the light curve over the past 9 years , it has been suggested that the eclipses occur whenever the motion of a star carries it behind the ring , with a period equal to that of the binary orbital period . if this ring is also precessing , it would be possible to explain the changing length of the eclipse . in this paper we present the analysis of photometric data on kh 15d obtained at vvo between 1995 and 2000 , color data obtained at the cerro tololo inter - american observatory ( ctio ) during the 2001/2002 season , and the results of the 20022004 observing campaigns as compared to those of 2001/2002 . section 2 describes the principal sources of optical photometric observations , the data reduction , and the resulting photometry for each participating observatory in the 20022004 campaigns . a comparison of individual data sets is shown in section 3 , while the results and analysis of these data are presented in section 4 . a brief discussion of our findings in light of the recent models is also presented , while a more complete interpretation is deferred to a future paper ( winn et al . 2005 , in preparation ) . the principal sources of optical photometric data analyzed here are telescopes of 0.62.2 m aperture . these include the us naval observatory s ( usno ) flagstaff observing station in arizona , mount maidanak observatory ( mmo ) in uzbekistan , vvo in connecticut , the european southern observatory ( eso ) and ctio in chile , kitt peak national observatory ( kpno ) in arizona , wise observatory in israel , teide observatory in the canary islands , and konkoly observatory in hungary . we also obtained data with automated telescopes at the tenagra observatory in nogales , arizona , and the katzman automatic imaging telescope ( kait ) , located at lick observatory atop mt . hamilton in california . while attempts were made to encourage uniformity in observing and reduction procedures within the initial group of participating observatories during the 2001/2002 season , the excitement that this object ignited inspired observers at telescopes around the world to contribute data to this project during the subsequent seasons . variations in seeing , image quality , sky brightness , flat - fielding procedures , and ccd characteristics made it impossible to enforce substantial uniformity . although the photometry parameters varied widely between data sets for the 20022004 observing seasons , which are discussed in detail here , we compared data obtained at different observatories on the same nights to search for consistency , as described below . it should also be noted that not all observatories used the same reference stars in their data reduction . we begin by listing the contributing observatories , telescopes , ccd parameters , filters , and dates in table 1 . ( ut dates are used throughout this paper . ) most observations were made with a cousins @xmath14 filter to reduce the nebular contamination of ngc 2264 , as well as the contribution from the nearby b2 iii star , hd 47887 , which is @xmath18 away . here we will discuss the 20022004 data in detail . exposure times varied per observatory according to telescope and ccd pixel size . observatories with similar observing procedures and reduction / photometry techniques are grouped together . photometry parameters for each dataset from 20022004 are included in table 2 . attempts were made to observe kh 15d nearly every clear night during the 2002/2003 observing season at the usno , flagstaff station . @xmath19 , @xmath20 , @xmath21 , and @xmath14 filters were used while kh 15d was bright , and only @xmath20 , @xmath21 , and @xmath14 during eclipse . in 2003/2004 , multiple observations were obtained during three consecutive egresses using only the @xmath20 , @xmath21 , and @xmath14 filters . exposure times generally ranged from 2 to 3 minutes , depending on whether the object was bright or faint . all frames were bias - subtracted and flat - fielded in real time by local code . photometry was performed on individual frames with an aperture radius that depended on the full width at half - maximum ( fwhm ) of unblended stars in the image . to maximize the signal - to - noise ratio ( s / n ) , a radius of @xmath22 was employed . the background parameters were set such that the scattered light from hd 47887 was minimized ( see table 2 ) . differential photometry of kh 15d was performed using seven reference stars selected from all - sky photometry obtained with the 1.0 m telescope at the usno , flagstaff station . a finding chart for the 7 comparison stars is shown in figure 1 , and the adopted standard johnson @xmath23 and kron - cousins @xmath24 magnitudes for each comparison star is given in table 3 . to determine the magnitude for each frame ( @xmath19 , @xmath20 , @xmath21 , and @xmath14 ) , the @xmath25 magnitude difference was calculated for each non - saturated standard star . this difference was then applied to the @xmath26 magnitude of kh 15d . no color terms were used since the usno filters are well matched to the standard system and the local standards span the colors of kh 15d . the resultant uncertainty includes the spread in the @xmath27 due to not including a color term . kh 15d was generally observed once every clear night at vvo , tenagra , and kait . typically , five 1-minute exposures were obtained per night ; however , during ingress and egress , vvo increased the number of exposures . all images obtained at vvo and tenagra were bias - subtracted , corrected for dark current , and flat - fielded using standard iraf tasks ( ccdproc ) . images obtained at kait were bias - subtracted , corrected for dark current , and flat - fielded automatically using a local code ( li et al . 2003 ) . after each image was processed , sets of five were averaged to increase the s / n , and the result was aligned to a single reference image using an idl ( interactive data language ) code . the fwhm for stars in each image was measured using the iraf task imexamine , and was used to assess the seeing on that night . any image that exhibited a very large fwhm ( @xmath13 4.5@xmath28 or larger ) was not used . for the analysis performed here , the average fwhm for the observing season was calculated and used to determine the size of the aperture radius utilized in the photometry process . this method was chosen so that all the data from one season could be processed and analyzed most simply in batch mode . we find that the standard deviation from this mean is typically less than 0.5 pix . to maximize the s / n , a radius of @xmath29 was chosen . instrumental magnitudes and uncertainties were computed for all the local comparison stars ( see fig . 1 ) and kh 15d using the iraf task phot . calibration of the instrumental magnitudes from one observation to the next was accomplished by means of a set of comparison stars , which were averaged together to produce a synthetic stable comparison . however , since even these stars are variable at some level , the final set of stars used were identified by an iterative process , and represent the most stable ones , with a typical standard deviation of 0.005 mag based on their night - to - night scatter . stars a , c , e , and g ( see fig . 1 ) were found to be most stable and were used as the comparison stars for both the 2002/2003 and 2003/2004 seasons at vvo . due to saturation issues , stars a , c , e , and f were used for the tenagra images , while only stars c , e , and f could be used for the kait data due to the small size of the detector . a time series of differential magnitudes relative to the synthetic comparison star was then produced . to place kh 15d on the standard magnitude scale , the adopted @xmath14-band magnitudes ( converted to flux ) of each comparison star were averaged to produce an @xmath30 comparison magnitude . this was then added to the differential magnitudes calculated for kh 15d during the 2002/2003 and 2003/2004 observing seasons . kh 15d was observed multiple times on nearly every clear night at mmo ( uzbekistan ) during the 20022004 observing seasons . exposure times typically ranged from 2 to 5 minutes , and all observations were made with a bessell @xmath14 filter . raw frames were bias - subtracted and flat - fielded using the imred package within the iraf environment . sky flat - field images were usually taken for each night , but in some cases only dome flats were available . differential magnitudes for kh 15d were computed in reference to star 16d from kearns et al . ( 1997 ) . in order to place these differential magnitudes onto our standard scale , we needed to know the true magnitude of 16d . thus , 16d was photometered during the 20012004 seasons at vvo in addition to the 7 local standards and kh 15d ( see section 2.2.2 ) , and the weighted average flux of 16d for the season was converted to a magnitude . this magnitude was then added to the differential magnitudes computed for 15d at mmo . the weighted average @xmath14 magnitudes computed for 16d between 2001 and 2004 are given in table 4 . since error bars were not produced when computing the differential magnitudes , we could not assign formal photometric uncertainties to these data . kh 15d was observed multiple times on nearly every clear night at konkoly observatory during the 2002/2003 observing seasons . exposure times ranged from 2 to 20 minutes , but were typically 5 minutes . all observations were made with a cousins @xmath14 filter . raw frames were bias - subtracted and flat - fielded using the imred package within the iraf environment . sky flat - field images were usually taken for each night , but in some cases only dome flats were available . the differential magnitudes of kh 15d were measured in reference to local secondary standard stars c , d , and f ( see fig . 1 ) . to determine the raw magnitudes of stars in the frames , aperture photometry was performed using the iraf / daophot package ( stetson 1990 ) . the fwhm of the stellar profile was determined for each night and an aperture of @xmath22 was chosen . the ring from which we estimated the background contribution to the measured fluxes started 2 pixels from our aperture and had a width of 5 pixels in every case . the differential magnitudes were calculated by defining the variable s brightness as @xmath31 . these differential magnitudes ( @xmath32 ) were transformed onto the standard photometric system by taking the adopted @xmath14 magnitudes ( converted to flux ) of each comparison star and averaging them together to produce an @xmath30 comparison magnitude . this was then added to @xmath32 to get the true @xmath14 magnitude of kh 15d . no color term was applied since the konkoly @xmath14-band filter is well matched to the standard system . finally , three cousins @xmath14 magnitudes of kh 15d were calculated on each frame by adding the magnitudes of the comparison stars to the differential magnitudes . we calculated the brightness by averaging the three different fluxes determined for kh 15d . the uncertainty of the brightness of kh 15d was estimated from the standard deviation of the three individual magnitudes . kh 15d was observed multiple times every clear night during the 2002/2003 observing season at the teide observatory . exposure times varied from 5 to 20 minutes , and all observations were made with a cousins @xmath14 filter . raw frames were bias - subtracted and flat - fielded using the imred package within the iraf environment . sky flat - field images were usually taken for each night , but in some cases only dome flats were available . differential photometry of kh 15d was performed using an aperture of @xmath33 . the sky background was computed using a circular annulus with an inner radius of @xmath34 and a width of 8 pixels . differential photometry and calibration from instrumental magnitudes to real ones was accomplished with respect to the reference stars 16d , 17d , 27d , 31d , and 39d from kearns et al . ( 1997 ) . data obtained at vvo between the years of 1995 and 2000 were re - examined with parameters that were consistent with those of the 20012004 observing seasons . this was done so that an absolute comparison between the earliest data and the most recent data could be made . one difficulty in completing this task was the fact that from the fall of 1995 through the spring of 1998 , vvo used a chip that was only 512 @xmath35 512 pixels in size . therefore , only one of the current `` standard '' stars ( d ) was available , and it also happened to be the most variable star . therefore , we consulted the original photometric studies of ngc 2264 conducted at vvo ( kearns et al . 1997 ; kearns & herbst 1998 ) and examined 5 of their quoted comparison stars ( 16d , 17d , 21d , 27d , and 31d ) . since these stars did not have known magnitudes , each was photometered on the images taken at vvo during the 20012004 observing seasons , allowing us to determine their magnitude on our standard scale . table 4 lists the weighted average @xmath14 magnitude and uncertainties for each star during 20012004 . once a standard magnitude was known for each comparison star , all 5 stars and kh 15d were photometered on the images obtained at vvo between 1995 and 1998 . table 5 lists the comparison stars used for each season . the eso observations of kh 15d contributed here were the byproduct of a larger photometric monitoring program of ngc 2264 ( lamm et al . full details describing the observing procedure , reduction methods , and photometry parameters can be found in lamm et al . . in summary , monitoring of ngc 2264 , and hence of kh 15d , took place during a period of two months between 30 december 2000 and 01 march 2001 with the wide field imager ( wfi ) on the mpg / eso 2.2 m telescope on la silla ( chile ) . all observations were carried out with a cousins @xmath14-band filter . the wfi consists of a mosaic of 4 @xmath35 2 ccds with a total array size of 8k @xmath35 8k . exposures of 5 s , 50 s , and 500 s were obtained , but the data used in this contribution come only from the 500 s images . image processing was done separately for each of the individual wfi chips using standard iraf tasks . each image was bias - subtracted using the overscan region in each frame , and flattened using an illumination - corrected dome flat . the data presented here are based on differential photometry relative to a set of non - variable reference stars . the daophot / apphot task was used to measure the brightness of kh 15d . the aperture was chosen to be 8 pixels ( 1.9@xmath28 ) in diameter for all measurements in order to maximize the s / n . the sky was calculated as the median of an annulus with an inner diameter of 30 pixels ( 7.1@xmath28 ) and a width of 8 pixels centered on kh 15d . because only differential magnitudes were calculated for the eso data , we computed the average out - of - eclipse magnitude for the 2000/2001 season based only on the data obtained at vvo . an appropriate out - of - eclipse magnitude was then added to the eso differential magnitudes such that these data could be placed on our standard scale . kh 15d was observed extensively during the 2001/2002 observing season with the ctio / yale 1 m telescope . all observations were made with a single instrument ( andicam ; see depoy et al . 2003 for a description ) using johnson @xmath20 and @xmath14 filters . kh 15d was observed on nearly every usable night between 30 august 2001 and 18 april 2002 . typically , several images were obtained each night using 300 s exposures . the data were reduced in the usual manner using standard iraf routines : an overscan ( bias ) was subtracted and a flat - field was applied to all the images . kh 15d was measured relative to two nearby stars shown in figure 2 . labeled as 1 " and 2 , " they are also known as 5d and 25d from kearns et al . ( 1997 ) , with @xmath14 = 13.250 @xmath0 0.001 mag and @xmath14 = 16.464 @xmath0 0.005 mag , respectively . we find that neither comparison star varied by more than 0.005 mag over the duration of the observations of kh 15d . a 3.6@xmath28 diameter aperture was synthesized on each of the ccd images for both of the reference stars and kh 15d using standard iraf aperture photometry routines . the sky was determined in a 4@xmath28 to 6@xmath28 annulus immediately adjacent to each of the sources . we calibrated the measurements of kh 15d relative to the two stars using images obtained at the 1.3-m mcgraw - hill telescope of the michigan - dartmouth - mit ( mdm ) observatory . the data were obtained on the photometric night of 19 october 2003 . the scale for the 1024 @xmath35 1024 pixel ccd that we used is 0.5@xmath28 pixel@xmath4 . throughout the night we observed photometric standard stars taken from landolt ( 1992 ) at a range of airmasses . near the end of this night we also obtained 12 @xmath20-band and 7 @xmath14-band images of the kh 15d field . we reduced the data using standard iraf techniques , including bias - subtraction and flat - fielding using twilight - sky flats obtained during the evening twilight of the same night . we performed aperture photometry on the mcgraw - hill data , using a 20 pixel ( 10@xmath28 ) diameter aperture for both the standard stars and the target stars in the kh 15d field . the seeing throughout the night was roughly 1.4@xmath28 , so aperture effects should be small . a photometric solution was found using all 36 standard - star measurements in @xmath20 and @xmath14 . we solved only for the zero point and extinction coefficient terms and held the color term constant at zero . this solution was then applied to the kh 15d field stars , and the calibration was used to convert the relative photometry to the standard system . we estimate that ignoring the color term in our calibration observations and between the mdm and the ctio / yale observations gives a systematic error in the kh 15d light curve of @xmath360.02 mag . this is based on intercomparison of other standard - star measurements . this paper combines a number of data sets obtained at various observatories , analyzed by different means , as described in the preceding sections . before beginning the process of analyzing the results of these data , random and systematic errors between data sets were assessed . to accomplish this , we selected all the photometry obtained at different places on the same julian date . if there were multiple observations made at a given observatory , a nightly mean was calculated . this nightly mean was then used in the analysis described below . out - of - eclipse and in - eclipse observations were analyzed separately since the photometric errors are typically much larger during eclipse owing to the faintness of the object . additionally , times of ingress and egress were avoided since the magnitude of the system changes too rapidly for comparison of data , even as close as one day apart . each season is addressed separately . one of the main objectives of the observing campaign of 2001/2002 was to study , in detail , the eclipse of december 2001 . the original contributing observatories included wise , vvo , mmo , usno , and kpno , and the results were reported by herbst et al . we now have data from the ctio / yale 1 m telescope that has provided us with perhaps the highest sampling rate yet of the eclipses that occurred during the 2001/2002 season . these are shown in figure 3 . the central reversal is evident in these data , as well as some variation in the shape of ingress / egress from eclipse to eclipse . additionally , substantial variability is seen outside of eclipse , as explored further in section 4.1.2 . since these data were originally obtained through johnson @xmath20 and @xmath14 filters , we could not immediately combine them with the rest of our cousins @xmath14-band data . the coverage of nearly 5 eclipses , including dense sampling throughout ingress and egress , provided excellent color information about this system . we therefore examined the johnson colors separately from the other data sets and present the results in section 4 . however , in order to add these johnson data to the full @xmath14-band light curve , we computed a transformation from johnson @xmath14 to cousins @xmath14 by comparing nearly simultaneous johnson @xmath20-band and cousins @xmath14-band data that were obtained at the usno ( see next section ) . in 2002/2003 , seven different observatories contributed to the light curve of kh 15d ( see table 1 ) . to evaluate the systematic differences between each contribution , we began by selecting all the data that were obtained on the same julian dates outside of eclipse , for the entire observing season . we define out - of - eclipse " to correspond to phases between 0.3 and 0.7 for both the 2002/2003 and 2003/2004 seasons . the out - of - eclipse 2002/2003 data are shown in figure 4 . for each julian date , the mean brightness of the star was formed based on all measurements . these contributions might include a single observation from one observatory and a mean observation ( due to multiple exposures throughout the night ) from another . the difference between each observatory s measurement and the total nightly mean was then calculated and examined as a function of the total mean brightness . a histogram indicating the number of times an observatory s measurement differed from the mean by 0.01 mag , 0.02 mag , etc . is shown in figure 5 . in general , there appears to be good agreement between the data sets . most observations differ from the mean by @xmath130.01 mag , which is typically the size of the estimated random errors associated with the photometry . thus , we can say that the data sets are consistent with each other to within @xmath130.02 mag . variations in the light curve on the order of @xmath130.08 mag are observed out - of - eclipse ( see fig . 4 ) , which we take to be real , and not an artifact of combining the data sets . this will be explored further in section 4 . figure 6 shows multiple data obtained on the same julian dates while in deep eclipse during 2002/2003 . we defined deep eclipse " to correspond to phases between 0.83 and 0.17 for both 2002/2003 and 2003/2004 , with phase = 0.0 ( and 1.0 ) representing the approximate phase for mid - eclipse . the same analysis described above was performed on these data . figure 7 shows that the differences of each observatory s observation from the mean is on average @xmath130.15 mag , which is consistent with the photometric observational errors . in 2003/2004 , only four observatories contributed to the light curve of kh 15d ( see table 1 ) . there were significantly fewer multiple observations out - of - eclipse during 2003/2004 as compared to 2002/2003 . we compared these individual measurements in the same manner as discussed above . the agreement between these data sets is not as good as was seen in the previous season . on average , the data from tenagra were @xmath130.04 mag brighter than those from vvo , and the data obtained at vvo were @xmath130.02 mag brighter than those obtained at mmo ( although that is not always the case ) . there are generally only 2 observations contributing to the calculation of the mean on any given night . therefore , the difference between each observatory s measurement and the mean gives us a sense of the systematic differences between each data set . the average difference is @xmath130.02 mag , which is slightly larger than the typical photometric errors ( @xmath130.01 mag or smaller ) . the in - eclipse data from 2003/2004 showed a great deal of scatter , but for the most part , appear to be correlated with each other . on average , the data depart from the calculated mean by @xmath130.15 mag , which is on the order of the quoted photometric errors associated with the in - eclipse measurements . overall , the data sets from both 2002/2003 and 2003/2004 mesh fairly well , and do not appear to be grossly different from each other . therefore , we feel confident that the shape of the light curve and the depth of the eclipses are accurate to within @xmath130.02 mag out - of - eclipse and @xmath130.15 mag in deep eclipse . the full data set now contains 6694 measurements of cousins @xmath14 magnitude covering nine seasons of observation between 1995 october and 2004 march , as well as additional color measurements . magnitudes in @xmath14 are given in table 6 and colors are given in table 7 . both tables , in full , are available electronically . in this section we describe the light and color variations and propose a rotation period for the star . we also provide some qualitative guidance to the interpretation of the variations based on the models of winn et al . ( 2004 ) and chiang & murray - clay ( 2004 ) . detailed modeling of the photometric ( and spectroscopic ) data will be done by winn et al . we turn first to a discussion of the light variations in @xmath14 , where most of the data are available . figure 8 shows the data from the last two seasons , which is the primary addition to the data set in this paper . clearly , we have obtained good coverage of a total of nine eclipses over two seasons . results indicate that the trends noticed by hamilton et al . ( 2001 ) ( reduction in amplitude of the central reversal and widening of the eclipse ) are continuing , and it would appear that the duration of the eclipse is now about one - half of the period ( compare with fig . the shapes of ingress and egress are not identical and show some level of variation , not only from eclipse to eclipse , but also from year to year . since 2001 , the duration of the eclipse has been growing by roughly 2 days year@xmath4 . however , as will be seen in section 4.1.2 , this has not always been the case . in earlier years , such as from 1998/1999 to 1999/2000 , the change in eclipse length was close to 1 day . in figure 9 , the last two seasons are compared with the seven previous ones for which modern , ccd photometry is available . some of the features and a distinct secular variation of the light curve can be seen . here , it appears that the out - of - eclipse brightness of the star has not changed by more than @xmath130.1 mag since 1995 , while the depth of the eclipse has grown by @xmath131 mag over that same interval . in fact , it appears as though the eclipse depth has been increasing at a rate of @xmath130.2 mag year@xmath4 . to determine the best estimate of the mean @xmath14 magnitude out - of - eclipse , we looked solely at the data obtained at vvo since it was the only observatory from which we had data for every year . we began by calculating a mean @xmath14 magnitude per year to see if there was any variation in the out - of - eclipse brightness with time . the phases that corresponded to the out - of - eclipse state changed as the width of the eclipse evolved with time ( discussed below ) , and are listed in table 8 . figure 10 shows the mean @xmath14 magnitudes versus observing season at vvo . the straight line represents a simple linear fit to the data . it appears as if the currently visible star might also be fading a bit , but we can not be sure whether this is a significant trend . as shown below , the increasing depth of the eclipse is accompanied by an increased time duration and a diminishing height for the central brightness reversal . the current data show that these trends , noticed previously ( see hamilton et al . 2001 ; herbst et al . 2002 ) , have continued in the last two seasons . during the first season of monitoring at vvo there was one unusually bright data point , taken on only the sixth night that the system was observed . repeated checks have shown that this is a valid datum and can not be discarded as an error . as will be shown below , its phase also corresponds to a point close to mid - eclipse . we now interpret that point as the only observation in our data set during which the relatively unobscured light from the currently invisible star ( b ) was seen directly . this point occurred close to periastron when star b was peaking above the disk for just about the last time . on that night , the system s light was only due to star b , since star a was hidden behind the disk . if this interpretation is correct , it means that the magnitude of star b is equal to or brighter than @xmath14 = 14.01 @xmath0 0.01 ( see table 6 ) . since star b might also have been partly hidden itself , we can not say for sure based on one measurement whether star b is brighter than @xmath14 = 14.01 @xmath0 0.01 , but it certainly can not be fainter . this conclusion is consistent with the historical light curve . johnson et al . ( 2005 ) show that prior to 1990 , the system was at times as much as 1 mag brighter than it appears today . if we take the difference between the combined light ( star a + star b ) and the light of star a alone to be @xmath37 mag , then the magnitude difference between star a and b is 0.45 . the mean @xmath14 magnitude out - of - eclipse based on the vvo data ( see fig . 10 ) shows that star a has a mean @xmath14 = 14.47 @xmath0 0.03 mag . assuming a minimum @xmath14 magnitude of 14.01 @xmath0 0.01 for star b , the mean brightness of star b is 0.46 @xmath0 0.03 mag brighter than star a. to make further progress we wish to phase the light curve with the appropriate period for the system , a task to which we now turn . it is important to establish the period of this system to the highest accuracy possible so that the data can be properly phased and the evolution of the light curve with time correctly displayed . it is now known that the 48-day period comes from the orbital motion within the binary . as such , the most definitive determination of the period should come from radial - velocity measurements and the spectroscopic - binary solution . this yields a period of 48.38 days with an uncertainty of about 0.01 days ( johnson et al . the problem with this method is that only a portion of the full orbit of one star can currently be seen , and it is the part near apastron which provides the least leverage on the solution . a second complication is that the visible star may suffer from a rossiter effect ( worek 1996 ) of unknown size as it enters or emerges from eclipse , depending on the inclination angle of the rotation axis to the line of sight . therefore , the data that provide the greatest leverage on the radial - velocity solution ( i.e. , those taken during late egress or early ingress ) may also be the data which are compromised in this regard . hence , we believe it is important to ask independently of the radial velocities whether the period of the system can be determined . the problem with the photometric method is , of course , that the light curve is evolving with time due to the progressive occultation by a foreground screen . this puts an additional time - dependent behavior into the system , albeit on a much longer time scale than the orbital period . to proceed , we first did a fourier transform of all of the modern photometric data using the scargle periodogram technique ( scargle 1982 ; horne & baliunas 1986 ) . this yielded a highly significant peak in the periodogram at 48.367 days ( see fig . 11 ) , which we round to 48.37 , as discussed below . this is gratifyingly close to the spectroscopic period and to the previously reported photometric period of 48.36 days ( herbst et al . 2002 ) . we attempted to refine the photometric period by focusing on a particular feature in the light curve whose phase we believe is stable and whose location we could determine from season to season . the only such feature that exists is the peak of the central reversal ( see fig . 12 ) , which we take to be the point in the orbit when the currently unseen star is closest to the edge of the obscuring cloud . assuming that this is a stable , fixed location in the orbit , if we could phase these central peaks we would have a good estimate of the period . unfortunately , it turns out that this method lacks discrimination because the location of the central peak is not sufficiently well defined by the light curves , especially the early ones where the data density is too low . a difference of 0.01 days in period is not significant and does not affect our results , as it corresponds to only 0.75 days over the 75 cycles covered by our monitoring and the early data do not constrain the time of the central peak to better than that . therefore , we find it impossible to improve on the accuracy of the period determinations with just these data . it might be possible to do this photometrically using the historical data , but any discussion of that is postponed to a later contribution ( winn et al . 2005 , in prep . ) . here we simply adopt the result of the periodogram analysis , 48.37 days , noting that it is consistent with the spectroscopic determination ; the 0.01 day difference does not affect any of the analysis or conclusions reported here . during the 20012004 observing seasons , a great deal of effort was made to observe kh 15d out - of - eclipse as well as in eclipse . significantly more data populated the bright phase of the light curve during these seasons than ever before , and small - scale variations ( @xmath130.08 mag over the course of 814 days in 2002/2003 ) are evident ( see fig . 4 and section 3.1 ) . since the visible star in the kh 15d system is a weak - lined t tauri star ( wtts ; hamilton et al . 2001 ) , it is most likely spotted in some fashion . it has been determined that this k7 star is still actively accreting ( hamilton et al . 2003 ) , and could therefore exhibit accretion hot spots on its surface , in addition to any dark , magnetically induced star spots . if this is the case , a closer look at the out - of - eclipse data from the 20012004 seasons may provide an estimate for the rotation period of the star . the analysis above indicates that systematic errors between data sets are relatively small , being at the level of 0.02 mag or less . however , when searching for the tiny amplitudes ( less than 0.1 mag ) that may characterize rotationally induced spotted - star variations , even errors of this size may make the search more difficult . therefore , we have searched each observatory s data set independently for periodicity during the out - of - eclipse phases . it is also prudent to confine the search to a single season of observation since spots are well known to evolve on time scales of less than one year . hence , one rarely finds phase coherence of spot variations extending over more than one observing season . detecting spots on wttss that are not undergoing eclipses is difficult enough , and generally only 15% or less of the stars in a cluster field will exhibit coherent periodic variations over a season s observing . in this star we have greater difficulty because during about half the observing season the star is in eclipse . nonetheless , we have found that periodogram analyses of the available data sets reveals highly significant periods in three cases ( see fig . 13 ) , as listed in table 9 . notably , two of these periods are identical to within the errors , averaging to 9.6 days . the chance of finding two exactly equal , highly significant periods in these data is vanishingly small if the periodicity is not real ; we thus identify the detected period as the likely rotation period of the visible star . it is interesting that the third significant period found in the periodograms ( 7.95 d ) is very close to the beat period between 9.6 and 48.37 days . supporting evidence for our interpretation comes from the measured @xmath2sin@xmath3 of the star . as discussed in appendix a , we derive a value of @xmath2sin@xmath3 = 6.9 @xmath0 0.3 km s@xmath4 for the visible component based on high - resolution spectra obtained out of eclipse . since sin@xmath3 is likely to be very close to 1.0 for this star ( assuming that the spin and orbital angular momenta vectors are nearly aligned ) , one can easily calculate that for a radius of @xmath5r@xmath6 ( hamilton et al . 2001 ) , the expected rotation period is 9.6 @xmath0 0.1 days , consistent with our result . both the photometry and spectroscopy of this system out - of - eclipse concur in suggesting a rotation period of 9.6 days for the visible component . to within the errors , this is a 5:1 resonance with the orbital period ( 48.37/5 = 9.67 ) , although it is not clear whether this is physically significant or just a numerical coincidence . a rotation period of 9.6 days is somewhat long for a 0.6 m@xmath6 wtts in ngc 2264 , although not unprecedented . the ( bi)modal values for stars more massive than 0.25 m@xmath6 are around 1 and 4 days . however , about 12% of the 182 cluster stars in that mass range with known rotational periods have @xmath38 days ( lamm et al . could the slightly slower than typical rotation of the visible star in the kh 15d system be caused by a tidal interaction between the components ? the effect of tidal friction on rotation can be estimated quantitatively based on the seminal theory of zahn ( 1977 ) for stars with convective envelopes . he shows that the synchronization time for a star s rotation to become tidally locked to its revolution is approximately given by @xmath39 where @xmath40 is the mass ratio , @xmath41 is the semimajor axis , @xmath21 is the stellar radius , and @xmath42 is given in years . most of the torque in a highly eccentric system will naturally occur at periastron . with @xmath43 and @xmath44 , we find a synchronization time of @xmath135 my . hence , it is not unreasonable to suggest that tidal friction is responsible for some degree of slowing of this wtts . we further note that since the orbit is rather eccentric , one would not expect to reach synchronous rotation , but only pseudosynchronous rotation . the pseudosynchronization timescale describes a near synchronization of revolution and rotation around periastron ( hut 1981 ) . hut ( 1981 ) has shown in elegant fashion that for binaries with eccentricities exceeding @xmath130.3 , as must surely be the case for kh 15d , tidal interaction near periastron is expected to produce an equilibrium angular rotation velocity of about 0.8 times the orbital angular velocity at periastron . one may write , therefore , that the pseudosynchronous rotation period ( @xmath45 ) predicted by this theory is related to the orbital period ( @xmath46 ) by @xmath47 where @xmath48 is near 0.8 for @xmath49 and is given precisely by hut ( 1981 ) . the range of particular interest for kh 15d is @xmath50 as discussed by johnson et al . ( 2004 ) , because these solutions are consistent with the radial velocities and with the inferred masses of the components based on astrophysical constraints . over this limited range , and somewhat beyond it , @xmath48 = 0.81 @xmath0 0.01 . adopting @xmath51 and @xmath46 = 48.37 @xmath0 0.01 yields a rather precise prediction for @xmath1 of 0.65 @xmath0 0.01 . this is the required orbital eccentricity if the star is in pseudosynchronous rotation and if the theory of hut ( 1981 ) is correct . remarkably , this eccentricity is very close to those inferred on the basis of two independent methods . as already noted , the orbital solution based on radial velocities , together with astrophysical constraints on the masses of the components , leads to @xmath52 , just barely outside of the present result . model fits to the light curve and its evolution over more than fifty years , by winn et al . ( 2004 ) and winn et al . ( 2005 , in prep . ) , also suggest an eccentricity of between 0.55 and 0.7 . since the rotation rate of the visible component is notably slower than that of most stars of comparable mass and age , and since it agrees so well with the predicted pseudosynchronous period , we suggest that the star has , indeed , been tidally locked into its current configuration . that , in turn , puts a rather severe limit on the possible eccentricity of the orbit , assuming that the locking has reached equilibrium and that the theory of hut ( 1981 ) is accurate in its prediction of @xmath48 . we note that the strong sensitivity of @xmath45 to @xmath1 means that there is only a very small range of measured rotation periods which would have given consistent results with the orbital solution and light - curve fitting techniques . for comparison , a normal rotation period of around 14 days for this star would have predicted a value of @xmath1 = 0.80.95 for this binary , rather extreme solutions that are inconsistent with the other results . while most of the monitoring has been in the @xmath14 band due to the brightness of the star during eclipse , we have also obtained some data at other wavelengths which reveal interesting features of the system . here we discuss the color results obtained at ctio in 2001/2002 and at the usno during several seasons . the discovery reported by herbst et al . ( 2002 ) that the star is generally bluer when fainter is confirmed , but we now have much more detailed and intriguing information on the color variations and their phase dependence . extensive @xmath20-band and @xmath10-band observations were made during the 2001/2002 season at ctio with the yale 1 m telescope ( see sections 2.8 and 3.1 for details ) and are shown in figures 14 and 15 . the solid and dashed lines in figure 14 are provided to indicate the approximate beginning of ingress and ending of egress , respectively . while the johnson magnitudes can be transformed to cousins for comparison with other data sets , there is no need to do so in this section , so we plot the results in the johnson system . the high density of color data obtained at ctio during this season reveals an interesting new feature of the behavior of kh 15d : it becomes bluer by a small but significant amount in very steady fashion as it enters eclipse and shows an analogous reddening as it emerges from eclipse . this is quite easily seen in both the time and phased versions of the light curves ( fig . 14 and 15 , respectively ) . it is also confirmed by the less dense but more temporally extended color monitoring done at the usno as described in the next section . while the general trend of bluer when fainter had been recognized for this star previously ( herbst et al . 2002 ) and has been seen in other young stars , such as uxors ( e.g. , herbst & shevchenko 1998 ) , the smooth decrease ( increase ) in @xmath53 seen during ingress ( egress ) in figure 15 is remarkable and unprecedented . before discussing its interpretation , we complete a description of the color behavior during eclipse . as both figures 14 and 15 show , the color change that occurs as the star goes into eclipse does not continue smoothly throughout the eclipse . while the data are noisy due to the low brightness of the star at those phases , it is clear that the star does not maintain a steady color with phase during eclipse . our impression of figure 15 is that there are actually three phases when the star has its bluest colors , peaking at about @xmath54 = 1.21.3 mag , or 0.30.4 mag bluer than the out - of - eclipse value of @xmath54 = 1.6 mag . these phases occur around @xmath00.17 and close to phase 0 . it is interesting that there is also a distinctive feature in the light curve at each of these three phases . at @xmath00.17 phase ( in 2001/2002 ) there is a distinct change of slope in the decline rate during ingress and the rise rate during egress . while this distinctive change of slope can be seen in the 2001/2002 light curve , it is even more noticeable in the light curves of the last two years ( see fig . 8) , where it occurs at phases of about @xmath00.20 and @xmath00.24 , respectively . the other feature in the light curve which seems to correspond to a blue extreme in the colors is the well - known central reversal near phase 0 . before discussing the interpretation of these interesting features , we turn to a description of the other major set of color data , that obtained at the usno . @xmath20 , @xmath55 , and @xmath56-band monitoring took place at the usno 1.3 m telescope throughout the 2001/2002 season , while @xmath19 , @xmath20 , @xmath55 , and @xmath56-band monitoring took place at the 1.0 m telescope during 2002/2003 ( see section 2.1 for details ) . it was determined from the first observing season that the object was simply too faint in @xmath19 to provide any reliable photometry at those wavelengths during eclipse . over the second season of monitoring , @xmath19-band measurements were obtained out - of - eclipse , in addition to the standard @xmath20 , @xmath55 , and @xmath56 observations . during 2003/2004 , only @xmath20 , @xmath55 , and @xmath56 data were obtained during three consecutive egresses . in figure 16 , we show the behavior of all of the cousins - band colors monitored at the usno during the 2002/2003 season . the trend is clearly quite similar to what was seen in the ctio data in 2001/2002 namely , there is a distinctive bluing of the system just as it enters eclipse or corresponding reddening as it emerges from eclipse . in addition , there are clearly some significant color variations out - of - eclipse which we attribute to star - spot activity on the visible star . combining the color data obtained over many seasons at the usno , we illustrate the overall trend of color with brightness in figure 17 . this confirms our impression that , out of eclipse , the star is about as red as it gets and shows relatively little color variation beyond what may be explained through star spots . as it enters ( emerges from ) eclipse the star gets bluer ( redder ) by 0.10.2 mag . near minimum light there are substantial color variations ; the star is sometimes at its bluest extreme , which is 0.30.4 mag bluer in @xmath53 than out - of - eclipse . however , sometimes it shows very little , if any , color effect . a full explanation of the color data must await improved understanding of the details of this system , which will only come from an extensive phenomenological model currently being constructed ( winn et al . 2005 , in prep ) . however , qualitative explanations for the observed trends can be proposed based on a few simple arguments . these lead to a set of questions that need to be addressed , and they suggest further observations that may clarify the situation . basically , blue colors indicate either the importance of wavelength - dependent scattering , such as from small grains or molecules , or the presence of a hotter component in the system , or both . it is likely that there is a hotter component in the system , since the currently invisible companion to the k7 star is known to be more luminous , as discussed above . all models of pre - main - sequence contraction indicate that the more massive star in a coeval pair is also more luminous and hotter , so it is reasonable to suppose that the currently invisible component of this system is also bluer than the k7 star . however , since the system was never substantially brighter in the past , even when both stars were visible , it is also true that this invisible component can not be much more massive , luminous , or hotter than the k7 star . in fact , the lack of detectable change of spectral type during minimum means there is little other than k6 or k7 light in the system , as far as a stellar component . the original interpretation proposed for the color variation by herbst et al . ( 2002 ) is that near minimum light , we see the system primarily or only by reflected light and that some small grains are involved in the reflection . this is supported by the increased polarization detected near minimum light by agol et al . the current data show , however , that there is almost certainly not a single color to which the star moves during eclipse , but rather that there is a great deal of real variability , some or all of which is phase dependent . therefore , it can not simply be that all of the light of the system is coming from distant , scattered radiation . in particular , we are impressed by the relative peak in blueness that occurs near phase zero in the color - phase plots ( fig . 15 and 18 ) . this is the time when the hotter , currently invisible , star is closest to the edge of the occulting cloud . we propose that the blue peak at this phase may be caused by this fact . we are apparently seeing a very small amount of either transmitted light or increased importance in the reflected component due to the light of this star , or , perhaps , the extension of a hot , circumstellar nebula associated with it , above the obscuring wall . in any event , we tentatively attribute the central peak in blue color and the phase - dependent variations around it as due to the orbital motion of a slightly bluer component ( than the k7 star ) relative to the edge of the occulting disk and , perhaps , relative to the principal scatterers . the blue peaks associated with phases @xmath00.17 in 2001/2002 are an entirely new feature of the light of this system revealed by the intensive color monitoring at ctio during that season and deserve some explanation . we find it interesting , as noted above , that they seem to correlate with a distinct change in slope that occurs in the decline ( rise ) in magnitude during late ingress ( early egress ) . one problem in interpreting these data is that we do not yet know how sharp the occulting edge of the cloud is . let us suppose , for the purposes of this qualitative study , that the edge is very sharp . in that case , we can identify the change of slope in decline ( rise ) rate as the point where the photosphere has just been completely covered ( has first appeared ) . the fact that the color peaks blue at that point indicates that the blue light is closely associated with the photosphere , arising just above it . we suggest , therefore , that the light in this system contains a hotter component associated with an extended chromosphere or corona of the k7 star ( and possibly its invisible companion as well ) . we have no idea whether the physical nature of this is a scaled - up solar - like chromosphere , a magnetically channeled accretion column , or something else . but the color evidence suggests an extended hotter , dense , optically thick zone located close to the photosphere of the k7 star . the persistence of the bluing of the light curve shows that it must be extended on a length scale comparable to a stellar radius . an alternative model is that one is seeing a peculiar scattering effect just as the star enters ( emerges ) from full eclipse . one could imagine , for example , that strongly forward scattered light might glance off the top of an occulting disk providing a bluish glint " at these phases . this would require , of course , wavelength - dependent scattering and , therefore , some component other than the obscuring cloud ( which produces no reddening of the transmitted light ) . if wavelength - dependent reflection is heavily involved here , then there should be some dramatic increases in polarization during these phases . it would definitely be interesting to extend the polarization study of agol et al . ( 2004 ) to a wider variety of phases . it will also be important to continue to monitor the color during these phases in future eclipses , hopefully including the @xmath57 band with larger telescopes , which will give important information on the nature of the scattering or the temperature of the emission zone . the light curve of the remarkable system kh 15d continues to evolve , evidently as a result of an obscuring screen moving across the orbit of a binary . a periodogram analysis of our data indicates that the photometric period is in excellent agreement with the spectroscopic period . analysis of the data collected outside of eclipse suggests that the visible star has a rotation period of @xmath139.6 days . this conclusion is in accord with the predicted rotation period , which is derived from the measured @xmath2sin@xmath3 ( 6.9 km s@xmath4 ) and taking sin@xmath3 @xmath12 1 . interpretation of the color information leads us to believe that some extended blue emission is revealed in the colors as the currently visible star goes into or comes out of eclipse . the eclipse length is now greater than half the orbital period of the system and continues to evolve , expanding at a rate that is @xmath132 days year@xmath4 . continued photometric monitoring will help to further constrain the models . this object will clearly provide us with the opportunity to learn about disk evolution , the circumstellar environment of young stars , and the possibility of planet formation . thus , we strongly encourage continued monitoring before it disappears ! we would like to thank josh winn for helpful discussions regarding this system . c. m. h. thanks christopher johns - krull for supporting her during the last year of this project and providing the k7 template used in the determination of @xmath2sin@xmath3 . w. h. gratefully acknowledges support by the national aeronautics and space administration under grant nag5 - 12502 , issued through the origins of solar systems program . p. a. , m. k. , a. m. , j. b. , and s. c. are supported by the hungarian national science fund under numbers t043739 and t043504 . a. v. f. and w. l. are supported by nsf grant ast-0307894 . a. v. f. is also grateful for a miller research professorship at uc berkeley , during which part of this work was completed . v. j. s. b. would like to acknowledge prof . rafael rebolo for informing him of the international campaign to monitor kh 15d and for inviting him to participate . kait was made possible by generous donations from sun microsystems , inc . , the hewlett - packard company , autoscope corporation , lick observatory , the national science foundation , the university of california , and the sylvia & jim katzman foundation . in order to determine the @xmath2sin@xmath3 of the visible star in the kh 15d system , the uves / vlt spectra ( hamilton et al . 2003 ) were revisited . first , a k7 v model spectrum was produced using a temperature of 4000 k , a log(@xmath58 ) = 3.5 ( @xmath58 in cgs units ) , and a macroturbulence value of 0 . the template was then artificially broadened to resemble that of a higher velocity star by convolving it with a rotation profile that was generated from a theoretical model , and including a macroturbulence value of 2 km s@xmath4 . this was accomplished by using an idl code and various values of @xmath2sin@xmath3 believed to span the range within which the @xmath2sin@xmath3 of the visible k7 star is expected to be found . each broadened spectrum was then cross - correlated against the original narrow - lined template , and the fwhm of the peak of the cross - correlation function was measured . a plot of the fwhm values versus @xmath2sin@xmath3 a calibration curve was produced by fitting a second - order polynomial function to the data . the 29 nov . 2001 spectrum was then cross - correlated against the narrow - lined template and the fwhm value of the cross - correlation peak was measured . this value was determined to be 0.489 . the polynomial fit to the calibration curve was evaluated at this fwhm value to obtain a @xmath2sin@xmath3 = 6.9 @xmath0 0.3 km s@xmath4 for the k7 star . the uncertainty was determined by first broadening our template spectrum to the value of 6.9 km s@xmath4 . we then performed a monte carlo test where noise , appropriate to that which is found in the uves / vlt spectrum , was added to the broadened template . this broadened , noisy template was then cross - correlated against the narrow - lined model spectrum . the fwhm of the peak of the cross - correlation function was measured for each test , and the standard deviation of these values provided us with our estimated uncertainty . vvo & 0.6 m & 0.5k @xmath35 0.5k , 0.6@xmath28 pix@xmath4 & @xmath14 & 27 oct 1995 25 mar 1998 + vvo & 0.6 m & 1k @xmath35 1k , 0.6@xmath28 pix@xmath4 & @xmath14 & 10 dec 1998 16 mar 2004 + eso & 2.2 m & 8k @xmath35 8k , 0.238@xmath28 pix@xmath4 & @xmath14 & 30 dec 2000 1 mar 2001 + ctio & 1.0 m & 2k @xmath35 2k , 0.3@xmath28 pix@xmath4 & @xmath59 & 22 aug 2001 18 apr 2002 + mmo & 1.5 m & 2k @xmath35 0.8k , 0.26@xmath28 pix@xmath4 & @xmath14 & 12 sept 2001 18 mar 2004 + usno & 1.3 m & 2k @xmath35 4k , 0.6@xmath28 pix@xmath4 & @xmath60 & 28 nov 2001 3 apr 2002 + usno & 1.0 m & 1k @xmath35 1k , 0.68@xmath28 pix@xmath4 & @xmath61 & 11 nov 2002 30 apr 2003 + usno & 1.0 m & 1k @xmath35 1k , 0.68@xmath28 pix@xmath4 & @xmath60 & 23 dec 2003 12 mar 2004 + kpno & 0.9 m & 2k @xmath35 2k , 0.6@xmath28 pix@xmath4 & @xmath62 & 1 22 dec 2001 + wise & 1.0 m & 1k @xmath35 1k , 0.7@xmath28 pix@xmath4 & @xmath14 & 2 22 dec 2001 + tenagra & 0.81 m & 1k @xmath35 1k , 0.87@xmath28 pix@xmath4 & @xmath14 & 2 nov 2002 4 apr 2004 + kait & 0.76 m & 0.5k @xmath35 0.5k , 0.8@xmath28 pix@xmath4 & @xmath14 & 20 sept 2002 30 mar 2003 + teide & 0.82 m & 1k @xmath35 1k , 0.432@xmath28 pix@xmath4 & @xmath14 & 28 sept 2002 28 mar 2003 + konkoly & 1.0 m & 1k @xmath35 1k , 0.33@xmath28 pix@xmath4 & @xmath14 & 1 oct 2002 2 mar 2003 + vvo & 7 ( 4.2@xmath28 ) & 10 ( 6.0@xmath28 ) & 5 ( 3.0@xmath28 ) + mmo & 10 ( 2.6@xmath28 ) & 11 ( 2.86@xmath28 ) & 5 ( 1.3@xmath28 ) + usno & 4.5 ( 3.06@xmath28 ) & 6.5 ( 4.42@xmath28 ) & 2 ( 1.36@xmath28 ) + tenagra & 3 ( 2.61@xmath28 ) & 5 ( 4.35@xmath28 ) & 2 ( 1.74@xmath28 ) + kait & 3 ( 2.4@xmath28 ) & 5 ( 3.2@xmath28 ) & 2 ( 1.6@xmath28 ) + teide & 2.5 ( 1.1@xmath28 ) & 20 ( 8.6@xmath28 ) & 8 ( 3.5@xmath28 ) + konkoly & 12 ( 3.96@xmath28 ) & 15 ( 4.95@xmath28 ) & 5 ( 1.65@xmath28 ) + a & 14.401 @xmath0 0.022 & 14.115 @xmath0 0.021 & 13.368 @xmath0 0.020 & 12.952 @xmath0 0.020 & 12.548 @xmath0 0.021 + b & 13.514 @xmath0 0.022 & 13.341 @xmath0 0.021 & 12.624 @xmath0 0.020 & 12.196 @xmath0 0.020 & 11.734 @xmath0 0.021 + c & 13.563 @xmath0 0.022 & 13.566 @xmath0 0.021 & 12.969 @xmath0 0.020 & 12.617 @xmath0 0.020 & 12.240 @xmath0 0.021 + d & 15.808 @xmath0 0.028 & 15.921 @xmath0 0.022 & 15.013 @xmath0 0.020 & 14.328 @xmath0 0.021 & 13.645 @xmath0 0.021 + e & 15.803 @xmath0 0.028 & 15.004 @xmath0 0.022 & 13.885 @xmath0 0.020 & 13.211 @xmath0 0.021 & 12.614 @xmath0 0.021 + f & 15.319 @xmath0 0.024 & 14.745 @xmath0 0.021 & 13.869 @xmath0 0.020 & 13.375 @xmath0 0.021 & 12.902 @xmath0 0.021 + g & 16.842 @xmath0 0.041 & 15.642 @xmath0 0.022 & 14.230 @xmath0 0.020 & 13.350 @xmath0 0.021 & 12.479 @xmath0 0.021 + * 16d & 14.303 @xmath0 0.001 & 14.304 @xmath0 0.001 & 14.302 @xmath0 0.002 + * 17d & 15.085 @xmath0 0.001 & 15.098 @xmath0 0.002 & 15.158 @xmath0 0.003 + * 21d & 15.023 @xmath0 0.001 & 14.991 @xmath0 0.002 & 15.006 @xmath0 0.003 + * 27d & 13.783 @xmath0 0.001 & 13.785 @xmath0 0.001 & 13.794 @xmath0 0.002 + * 31d & 14.423 @xmath0 0.001 & 14.426 @xmath0 0.001 & 14.411 @xmath0 0.002 + * * * * * 2450017.8375 & 14.449 & 0.011 & vvo + 2450021.7443 & 14.521 & 0.024 & vvo + 2450026.7416 & 17.132 & 0.074 & vvo + 2450028.6983 & 14.687 & 0.014 & vvo + 2450028.8949 & 14.435 & 0.015 & vvo + 2450031.7696 & 14.014 & 0.007 & vvo + 2450039.7466 & 14.562 & 0.010 & vvo + 2450054.6473 & 14.466 & 0.009 & vvo + 2450056.7206 & 14.466 & 0.007 & vvo + 2450056.8106 & 14.457 & 0.009 & vvo + 1.619 & 0.037 & 14.475 & 0.024 & 2452237.7643 & usno + 1.582 & 0.024 & 14.525 & 0.018 & 2452241.8421 & usno + 1.416 & 0.064 & 18.061 & 0.037 & 2452261.8464 & kpno + 1.379 & 0.040 & 17.833 & 0.026 & 2452262.8831 & kpno + 1.516 & 0.047 & 17.601 & 0.031 & 2452263.8295 & kpno + 1.565 & 0.039 & 16.739 & 0.023 & 2452264.8739 & kpno + 1.584 & 0.013 & 14.695 & 0.009 & 2452266.7740 & usno + 1.621 & 0.014 & 14.519 & 0.008 & 2452269.8189 & usno + 1.680 & 0.025 & 14.435 & 0.021 & 2452279.6966 & usno + 1.716 & 0.091 & 17.427 & 0.073 & 2452306.8015 & usno + 1995/1996 & 0.200.80 + 1996/1997 & 0.200.80 + 1997/1998 & 0.200.80 + 1998/1999 & 0.250.75 + 1999/2000 & 0.250.75 + 2000/2001 & 0.250.75 + 2001/2002 & 0.300.70 + 2002/2003 & 0.300.70 + 2003/2004 & 0.350.65 +

we present the measurements of scatter broadening time - scales ( ) for 124 pulsars at 327 mhz , using the upgraded ooty radio telescope ( ort ) . these pulsars lie in the dispersion measure range of 37 503 pc and declination ( ) range of . new estimates for 58 pulsars are presented , increasing the sample of all such measurements by about 40% at 327 mhz . using all available measurements in the literature , we investigate the dependence of on dispersion measure . our measurements , together with previously reported values for , affirm that the ionized interstellar medium upto 3 kpc is consistent with kolmogorov spectrum , while it deviates significantly beyond this distance . # 1#2 # 1^#1^ # 1~#1~

1501.05401

free electrons in the interstellar medium ( ism ) affect the pulsar signal in three different ways : cold plasma dispersion , free - free absorption and scattering . the broadband pulsar signal gets dispersed in the ism resulting in delay in pulse arrival time , which is a function of wavelength , @xmath5 , and its magnitude is described by the quantity known as dispersion measure @xmath6 pc @xmath1 , where @xmath7 is the column electron density per @xmath8 and d is the distance to the pulsar in parsec . @xmath9 for a pulsar is obtained as a part of its discovery and if the distance @xmath10 to the pulsar is known ( by parallax method for example ) , then the mean electron density in the ism can be estimated . alternatively , if the mean @xmath7 is known ( either from observations or from the free electron density distribution model of the galaxy such as ne2001 by @xcite ) , the approximate distance to the pulsar can be determined . the pulsar signals also get absorbed by the free electrons in the ism . this process is known as free - free absorption which is responsible for the turnover observed in many of the pulsar spectra at lower frequencies @xcite . pulsar signals are also affected by the fluctuations of the electron density in the ism . such fluctuations give rise to random variations in the refractive index of the medium . the pulsar signal traverses through such irregularities and gets scattered in the process . this phenomenon , known as ` interstellar scattering ' of pulsar signals , was first investigated by @xcite . it essentially arises due to multipath propagation of the pulsar signal . @xcite demonstrated that a narrow pulse with a sharp rise time , while passing through a gaussian distribution of irregularities will broaden in time due to scattering , giving rise to an exponential decay of the pulse with a characteristic timescale @xmath0 , known as the scatter broadening time . the multipath propagation also leads to a diffraction pattern in the observer s plane , which decorrelates over a characteristic bandwidth @xmath11 , such that @xmath12 , where @xmath13 is of the order of unity . these scattering parameters are strongly frequency ( @xmath14 ) dependent and scale as @xmath15 , where @xmath16 is the power law index . if the same volume of electrons are responsible for dispersion and scattering , a simple scaling relation @xmath17 , can be shown to exist in the model of @xcite . in general , the scattering strength can be attributed to a power law electron density spectrum of the form @xmath18 , where @xmath19 is the wavenumber , and @xmath20 @xcite . for a kolmogorov distribution of irregularities , where @xmath21 , the scaling relation is expected to be @xmath22 @xcite . several attempts to verify these scaling relations and fathom an understanding of the ionized component of the ism exist in the literature @xcite . these investigations show that the observed values of @xmath16 is much lower ( @xmath23 4.0 ) than the theoretical value of 4.4 for several of the high @xmath9 pulsars . hence , accurate @xmath0 measurements over multiple frequencies are essential for these studies . at meter wavelengths , the @xmath0 values for @xmath24 pc @xmath1 dominate over the intrinsic pulse width and it is possible to obtain reliable @xmath0 measurements . a survey of a number of highly dispersed pulsars with a possible detectability at 327 mhz with the aim of finding @xmath0 was carried out at ooty radio telescope ( ort ) by @xcite and @xcite for 47 pulsars out of 706 pulsars discovered by that time . currently @xmath252200 pulsars are known and the ort has been upgraded for improved sensitivity . hence , a large number of accurate @xmath0 measurements at 327 mhz using the ort are now possible . with this aim , we have launched a survey of @xmath0 measurements . in this paper , we report results from the current survey , which includes several new @xmath0 measurements and significantly improved @xmath0 estimates for pulsars with previously measured values . observations were conducted for 124 pulsars from 2013 january to 2014 september using the ort , situated in southern india at a latitude of 11@xmath26 n. it has an offset parabolic cylindrical reflector , 530-m long in north - south direction and 30-m wide in east - west direction , placed on a hillside with a north - south slope of 11@xmath26 @xcite . the feed array consists of 1056 dipoles , oriented along the north - south direction . hence , the telescope is sensitive only to the north - south polarization . the telescope operates at a central frequency of 326.5 mhz . the front - end electronics downconverts the signal to an intermediate frequency ( if ) of 30 mhz with a bandpass of 16 mhz @xcite . the 1056 dipoles are grouped in 22 modules with half of them each in north and south halves of the telescope . a separate if is obtained from each half after appropriately phasing the modules and combining their output to form 12 beams , which cover @xmath27 in the sky . recently , a new pulsar ort new digital efficient receiver , ponder , has been commissioned at the ort to process the if outputs @xcite . the two if outputs of beam 7 ( central beam ) are converted to baseband and 8-bit digitized in ponder after processing through a 16-mhz low pass filter . the digitized signals from the two halves are added in ponder and a 4096-point fast fourier transform is carried out in real - time across the 16 mhz passband , providing 2048 channels across the passband with a spectral resolution of 7.8125 khz . the channelised data were then incoherently dedispersed in real - time , to the nominal @xmath9 taken from the atnf pulsar catalogue @xcite . ponder was also used to fold the dedispersed data in real - time to an appropriate number of bins , using tempo2 predictors @xcite with most of the ephemerides taken from atnf pulsar catalogue . in some cases , particularly for glitching pulsars , ephemerides from recently published rotational parameters are used . see @xcite for more details on ponder and the real - time processing capabilities used for this study . the folded profiles were then analysed using the procedure described in section 2.2 . the ort has been recently upgraded and currently has a sensitivity which is 1.6 times better than what it was in the previous similar studies @xcite . ponder is designed to digitize the 16 mhz bandwidth in contrast with the previous studies , which used a bandwidth of 10 mhz , giving a further improvement in sensitivity by a factor of 1.2 . additionally , the telescope has been refurbished which improved the gain of the telescope by a factor of 1.3 bringing it to 3.3 k / jy ( eg . , @xcite ) . ponder has been particularly useful for this project due to the following reasons . firstly , it eliminated the need for the storage of raw data with its real - time capability for channelisation , incoherent dedispersion and folding . this results in a smaller dedispersed time series and integrated profile as standard data products , which allowed a faster observations to analysis turn - around time , even for 8 hour long observations . secondly , it has a flexible spectral resolution , which enables minimization of the dispersion smear across a channel , and consequently allows for robust estimation of scatter - broadening parameters . it also has real - time coherent dedispersion capability eliminating this smear completely . for this project , a fixed spectral resolution of 7.8125 khz was used . lastly , it will allow real - time measurements of dynamic spectra , which will also be useful for measurements of ism effects on pulsed signals in future ( see @xcite for more details ) . the above mentioned capabilities are useful for future similar investigations of pulsar scatter - broadening . we selected 124 pulsars for our study using the following method . the pulsars in our sample lie in the ort observable declination , ( @xmath2 ) range of @xmath28 and have a flux density at 327 mhz greater than 1 mjy . the flux density at 327 mhz was estimated using available spectral index in the atnf catalogue . where the spectral index was not available , flux density at the frequency available in the atnf catalogue was scaled using a spectral index of -1.6 @xcite . the above criteria yielded a large number of pulsars , and the list was further pruned by removing pulsars with @xmath9s that are not sufficiently high , given that their @xmath0 values are likely to be significantly smaller than the intrinsic pulse profile width . pulsars with sufficiently large @xmath9 , where the estimated scattered pulse width ( obtained by scaling an existing @xmath0 measurement at another frequency by @xmath29 ) exceeds the pulsar period were also removed . the final list of pulsars lie in the @xmath9 range of 37@xmath3503 pc @xmath1 and is listed in table [ tab1 ] . the observing time for each pulsar was fixed in the following manner . while for every pulsar at least 2000 pulses were acquired to get a stable profile , pulsars with low signal to noise ratio were integrated until a minimum peak signal to noise ratio of 5 was obtained . most of the observations were for 30 minutes duration , whereas observations were carried out in some special cases up to 8 hours due to the low signal to noise ratio of the pulsar . [ cols= " < , > " , ] the observed pulse profile @xmath30 is a convolution of intrinsic pulse shape @xmath31 with ( 1 ) the impulse response characterizing the scatter broadening in the ism @xmath32 , ( 2 ) the dispersion smear across the narrow spectral channel d(t ) and the instrumental impulse response , @xmath33 defined as @xmath34 where @xmath35 denotes convolution . the rise times of the receivers and back end are small enough to consider the effect of @xmath33 , while @xmath36 is a rectangular function of temporal width given by the dispersion smearing in the narrow spectral channel for incoherent dedispersion . we use the ism transfer function @xmath37 @xcite . the prime difficulty in extracting @xmath0 stems from the fact that the intrinsic unscattered pulse profile @xmath38 is unknown . in several earlier studies , various assumptions were made regarding @xmath38 with no clear advantages or disadvantages . for example , @xcite and @xcite assumed the @xmath38 to have a gaussian shape and simultaneously fitted for the pulse width and @xmath0 . @xcite used a clean based algorithm , where they claim to recover the intrinsic pulse profile and simultaneously give the pulse broadened profile , while @xcite used a high frequency unscattered profile as a model for @xmath38 . in this paper , we follow the method described by @xcite , hereafter referred as lkmll01 . for every pulsar in our sample , we searched for a high frequency profile in the epn data base , where the expected @xmath0 is significantly smaller than the pulse width . in cases , where such profiles were not available , often the widths of the profile at a high frequency were found in the literature . we used this width to construct a gaussian profile as a template for @xmath38 . finally , this model @xmath38 was used to obtain a best fit model by minimizing the normalized @xmath39 value @xcite defined by @xmath40 ^ 2 , \label{eq2}\ ] ] where @xmath41 is the off - pulse root - mean - square ( rms ) , @xmath42 is the observed pulse profile , @xmath43 is the model profile and @xmath44 is the total number of bins in the profile . the model profile @xmath45 is scaled with the pulse amplitude @xmath46 , shifted by a constant offset @xmath47 in phase and fitted to a baseline @xmath48 to minimize the @xmath49 . for fitting purpose , we use the nonlinear fitting routine `` mrqmin '' given in numerical recipes @xcite , where the errors in @xmath0 is obtained from the covariance matrix . two examples of the fitting procedure used to estimate @xmath0 values are shown in figure [ fig1 ] . the top panel of the left hand figure is the fit to the profile of psr b1900 + 01 , and the bottom panel is the high frequency profile obtained at 1.4 ghz , which has a single component and is used as a template @xmath50 . the fitting process yields an excellent fit to the data ( shown in the top panel of the figure ) with @xmath51 and @xmath52 ms . it is important to note that we have ignored the effect of pulse width and component evolution with observing frequency in our fitting algorithm . it is known that pulse width increases with increasing wavelength as @xmath53 due to radius to frequency mapping @xcite , and hence @xmath0 value that we obtain is an upper limit . however , in several cases , particularly for high @xmath9 pulsars , the @xmath0 evolution is significantly higher than the width evolution and hence the measured values of @xmath0 are robust . in the example of psr b1900 + 01 , the pulse width is expected to increase from a higher frequency profile by a factor of 1.5 whereas the expected increase in @xmath0 is a factor of 40 , indicating that the @xmath0 measurement is reliable . however , the values of @xmath0 can be affected by the presence of an unidentified pulse component on the trailing part of the profile , as discussed by lkmll01 . they suggested that an increased error bar to 3 standard deviations , is usually sufficient to account for these effects . it is noteworthy that for @xmath2570% of the pulsars in our sample , the above procedure serves as an acceptable model . the remaining @xmath2530% of pulsars in our sample has multiple component profiles , both at higher and lower frequencies ( except three cases that are discussed below ) . for these pulsars , applying the method of lkmll01 is not possible . the profile evolution in pulsars is such that the location and amplitude of the high frequency profile components change significantly at a lower frequency . however , if the separation of a profile component ( identified at both higher and lower frequencies ) from the adjacent one is significantly larger than @xmath0 , then the lkmll01 method can be applied to the component . the second example of psr b0736@xmath340 shown in the right hand panel of figure [ fig1 ] pertains to such a case . the high frequency template for psr b0736@xmath340 has two components , and as shown in the figure , we have considered only the trailing component as a model and fitted this to the trailing component of the pulse profile at 327 mhz . this has resulted in significantly better fitting and hence estimation of @xmath0 . we have applied this procedure to our data wherever applicable ( @xmath2530% of the pulsars presented here ) . to gain additional confidence in this method , we have performed extensive simulations , where we created unscattered templates with two gaussian components and convolved with the exponential transfer function to get the scattered profile for an input @xmath0 value . we then used the procedure described above to extract @xmath0 from the scattered profile . multiple templates were obtained by varying the amplitude ratio and the separation of the gaussian components , and for each case the corresponding @xmath0 value was estimated . we find that for majority of the cases , where components could be resolved , we are able to recover the @xmath0 value ( within 3 standard deviation ) . when the separation between the components and/or the ratio of the amplitudes are too small , we tend to overestimate @xmath0 by about 10% , as can be seen in figure 64 of the supplementary material . there are 3 pulsars in our sample , namely b1737@xmath330 , b1834@xmath310 and b1859 + 03 , which have a slow rising edge and heavily scatter - broadened trailing edge . for these pulsars , the lkmll01 model is found inadequate , since the rising edge of the profiles are poorly modelled . if we _ assume _ that there is no unidentified emission component that changes the profile shape in the rising edge for these pulsars , then there exist a possibility that the transfer function of the ism towards these pulsars are responsible for these profile changes . we have explored whether the transfer function for a uniformly distributed ism with gaussian density fluctuation ( see @xcite ) given by @xmath54 , can model the data , since the rising edge of this transfer function increases much more slowly as seen in the data . in all the three cases , we found reasonable fits and they yielded @xmath0 values of 62@xmath553 , 203@xmath5523 and 103@xmath552 , respectively ( the fitted profiles are shown separately in the online supplementary material ) . however , before we conclude that the effect seen in these pulsars are related to a different transfer function , futher modelling of the evolution of pulse profiles based on high quality multifrequency observations need to be done , which is beyond the scope of the present study . table [ tab1 ] lists the estimated @xmath0 , the error @xmath56 and the @xmath39 for all the 124 pulsars in our sample . the left panel of figure [ fig2 ] shows the distribution of reduced @xmath39 for our sample , which lie in a range of 0.5 to 1.5 and peaks around unity . this confirms that the model profile @xmath57 is a good fit to the data for majority of pulsars in our sample . in the right panel of figure [ fig2 ] , @xmath0 values for 31 pulsars which are common between our sample ( the abscissa ) and that of @xcite and @xcite ( the ordinate ) are shown for comparison . the @xmath0 values obtained between these two methods are generally in good agreement with each other . the primary goal of this work is to provide a large database of @xmath0 measured at 327 mhz ( p - band ) . in this paper , we have estimated @xmath0 values for 124 pulsars , in which , as far as we know , 58 are new measurements at p - band , which thereby increases the @xmath0 measurements at 327 mhz by almost 40% . including our current sample and measurements available in the literature , we found 154 pulsars for which scattering measurements exist at p - band : out of which 136 have @xmath0 measurements , with 124 from this paper and 12 from @xcite and @xcite . 18 have decorrelation bandwidth @xmath11 measurements from @xcite . additionally , we found 134 pulsars having @xmath0 values and 70 pulsars with @xmath11 values measured at different frequencies . thus , currently scattering measurements exist for 358 pulsars spanning a dm range of 2.9 1073 pc @xmath1 . + @c@ + + generally , scattering properties along various lines of sight to the pulsars are examined by studying the relation between @xmath0 and @xmath9 . figure [ taudm ] shows the @xmath0 versus @xmath9 plot at 327 mhz for all the 358 pulsars . the filled circles are direct measurements at 327 mhz . the black triangles are @xmath11 measurements at 327 mhz which have been converted to @xmath0 using the relation @xmath58 , where @xmath13 is assumed to be 1.16 ( see @xcite for discussion ) . for convenience of discussion , let us define four different @xmath9 regions in the @xmath0 versus @xmath9 plot : region @xmath59 for @xmath9 range 0 to 40 pc @xmath1 , region @xmath60 for @xmath9 range 40 to 100 pc @xmath1 , region @xmath61 for @xmath9 range 100 to 500 pc @xmath1 and region @xmath62 for @xmath9 range above 500 pc @xmath1 . it may be noted that in the region @xmath60 , no reliable @xmath0 or @xmath11 measurements are currently available at p - band . in this range at p - band , the @xmath0 becomes unreliable since the expected @xmath0 values are significantly smaller than the profile widths . the @xmath11 values are also difficult to measure since they decorrelate over a significantly small bandwidth . further @xmath0 measurements at p - band in the region @xmath62 is not possible since the scattering time exceeds the pulsar period . however , the 327 mhz data reveal the basic feature of the @xmath0 versus @xmath9 relation , where at low @xmath9 i.e. , region @xmath59 , the slope is flatter compared to the high @xmath9 region @xmath61 @xcite . in order to fill in data at p - band for the missing @xmath9 ranges @xmath60 and @xmath62 in figure [ taudm ] , we rely on scattering measurements at higher frequencies . in region @xmath60 , @xmath11 measurements and in region @xmath62 , @xmath0 measurements are available at higher frequencies . these measurements therefore need to be scaled to 327 mhz using a frequency scaling law @xmath63 . the properties of the kolmogorov spectrum of electron density irregularities , which has been widely used as a reference to understand the @xmath0 versus @xmath9 curve , predicts @xmath644.4 . however , observationally there is evidence that the frequency scaling of @xmath644.4 might not be uniform across the pulsar sample . for dms below about 100 pc @xmath1 @xcite and @xcite found @xmath16 to be consistent with 4.4 , whereas for higher dm values , @xcite and @xcite reported a much smaller value of @xmath16 @xmath25 3.44 . the scaling relation for these pulsars were obtained on a case by case basis using detailed modelling and measurement of @xmath0 and @xmath11 . @xcite conducted an exhaustive study for obtaining @xmath16 treating it as a parameter ( constant across the whole dm range ) and fitted for @xmath16 using direct estimates of @xmath0 and @xmath11 at the observing frequencies . it yielded a value of [email protected] . currently more accurate measurements are required to establish the frequency scaling as the function of dm . here we use @xmath64 4.4 , which can be referenced to the kolmogorov spectrum and scaled the scattering measurements to 327 mhz using the frequency dependence of @xmath65 ( note that the @xmath11 measurements are converted to @xmath0 using @xmath66 ) . the scaled @xmath11 and @xmath0 values are shown as open triangles in figure [ taudm ] . the combined @xmath0 versus @xmath9 plot is obtained using all currently available scattering measurement data of 358 pulsars . we fitted the empirical relation @xmath67 , as was suggested by @xcite . by fixing @xmath68 , the term @xmath69 can provide a useful description for assessing the deviation of @xmath0 with @xmath9 from the kolmogorov theory . the best form of the fit ( shown as solid curve in the figure ) corresponds to @xmath70 which is consistent with earlier studies of @xcite and @xcite . note that there is a large spread in @xmath0 for a given @xmath9 and in the bottom panel of figure [ taudm ] , the fractional residual ( i.e. ( data - model)/model ) is shown , where it appears that with increasing @xmath9 , the spread in @xmath71 increases . there are a few reasons for the spread in @xmath0 measurements . as discussed earlier , the methods applied for estimating @xmath0 can lead to uncertainties due to ambiguity in the intrinsic pulse shape . this mostly affects @xmath0 values in regions @xmath59 and @xmath60 at p - band . however , for larger @xmath9s , i.e. , regions @xmath61 and @xmath62 , @xmath0 values are significantly larger than the pulse profile widths , and hence the ism properties are mostly the reason for the observed spread in @xmath0 . clearly our analysis reveals several anomalous lines of sights , for example psr b1015@xmath356 has a @xmath9 of 436 pc @xmath1 while the measured @xmath0 is 13 ms which is a factor of 10 lower than the mean @xmath0 for that @xmath9 . anomalous scattering is a commonly observed phenomenon , and is most probably related to inhomogeneous distribution of scattering material along pulsar lines of sight . the indirect ( i.e. non p - band measured ) @xmath0 values in figure [ taudm ] depends on @xmath13 and the frequency scaling . the change in @xmath13 in the lower @xmath9 range can lead to a change in the slope of the @xmath0 versus @xmath9 relation . as argued by @xcite and @xcite , there is very little allowance in the theory for changing @xmath13 values from close to unity . however the frequency scaling of @xmath65 used in figure [ taudm ] has been a subject of scrutiny in recent studies by @xcite . these careful studies of multifrequency @xmath0 measurements revealed that at high @xmath9 ( @xmath72500 pc @xmath1 ) the frequency scaling is about 3.4 , which is significantly smaller than the kolmogorov value . this can result in flattening of the slope in the region @xmath62 of the @xmath0 versus @xmath9 curve , and only a systematic multifrequency study to obtain scattering estimates for a large sample of pulsars can resolve some of these issues . however there is significant overlap of pulsars with direct p - band @xmath0 measurements ( filled circles ) and frequency scaled @xmath0 measurements in the @xmath9 region @xmath61 , where no differences in @xmath0 values can be seen . we can hence speculate that for pulsars below @xmath9 500 pc @xmath1 , the frequency scaling of @xmath65 is a good approximation ( also a result supported by @xcite ) . it has been also speculated that the breakdown of the homogeneity condition in the ism @xcite can lead to the steepening of @xmath0 values . this can be checked by evaluating the scattering strength @xmath73 along pulsar lines of sight . in the top panel of figure [ cn2 ] , the estimated @xmath74 for all the 358 pulsars as a function of pulsar distance in kpc the @xmath73 values were obtained for a homogeneous medium with kolmogorov spectrum using the expression given by @xcite as @xmath75 , where @xmath14 is the observing frequency in ghz , @xmath10 is distance to the pulsar in kpc and @xmath11 is the decorrelation bandwidth in mhz . the distance to the pulsars were obtained using the @xcite model of galactic free electron density distribution . the dashed line in the top panel of figure [ cn2 ] correspond to the case of homogeneous medium where @xmath76 @xcite . our results are consistent with all earlier studies @xcite where lines of sight to pulsars below 3 kpc has a value closer to @xmath33.5 . for the distant pulsars , the deviation is significant towards larger @xmath74 values which indicate that high level of inhomogeneity exists along these lines of sight . the bottom panel of figure [ cn2 ] shows the distribution of all 358 pulsars in the galaxy , where majority of the pulsars lie towards the inner regions of the galaxy and hence are likely to encounter several hii regions . note that the enhanced scattering seen around galactic longitude of 260@xmath77 is due to the gum nebula @xcite . in this paper , we report new observations of 124 pulsars using the ooty radio telescope at 327 mhz with the aim of estimating scatter broadening ( @xmath0 ) values . our attempt has yielded 58 new @xmath0 measurements at 327 mhz , which increases the sample with known @xmath0 measurement at 327 mhz by about 40% . the major difficulty in estimating @xmath0 lies in the uncertainty of knowing the intrinsic ( unscattered ) pulse shape . we have hence used the method suggested by @xcite in which the intrinsic pulse profile is obtained from higher frequencies where the profile is unscattered . this method significantly reduces the uncertainty in @xmath0 measurements . further we have also assimilated 234 scattering measurements from the literature and have used them to study the ism properties . our results are consistent with earlier works where , for low @xmath9 ( or below 3 kpc ) , the ism is consistent with kolmogorov spectrum , but it deviates significantly for high @xmath9 pulsars . in figure [ taudm ] , the @xmath0 values at p - band for several pulsars were obtained estimating @xmath0 from @xmath11 using a frequency scaling of @xmath65 from higher frequencies . the high sensitivity of the upgraded ort and the flexibility of ponder will be useful to enhance the sample of such observations further , particularly to weaker pulsars , which were not covered in this study . we also note that these measurements can be improved by observations of pulsars at multiple frequencies with instruments , such as the gmrt , covering the low @xmath9 sample with frequencies including and below 300 mhz and the high @xmath9 sample at 610 or 1420 mhz and such observations are planned in future . we acknowledge the kind help and support provided by the members of radio astronomy centre during these observations . the ort is operated and maintained at the radio astronomy centre by the national centre for radio astrophysics of the tata institute of fundamental research . we also thank the anonymous referee for a critical review which helped in improving the manuscript significantly . alurkar s. k. , slee o. b. & bobra a. d. , 1986 , aust . j. phys . , 39 , 433 bhat , n.d.r , cordes , j.m , camilo f , nice , d.j & lorimer , d.r , 2004 , apj , 605 , 759 bhat , n. d. r , rao , a. p. & gupta , y , 1999 , apjs , 121 , 483 blitz , leo . , fich , michel . , stark , a. antony . , 1982 , apj , 49 , 183 clifton , t. r. , lyne , a. g. , jones , a. w. , mckenna , j. & ashworth , m. , 1992 , mnras , 254 , 177 - 184 . cordes j.m . , 1986 , apj , 311 , 183 cordes j.m & lazio , t.j.w , 2002 , astro - ph , arxiv : astro - ph/0207156 cordes , j.m , weisberg j.m & boriakoff v , 1985 , apj , 288 , 221 dewey , r. j. , cordes , j. m. , wolszczan , a. & weisberg , j. m. , 1988a . interstellar scintillations of binary pulsars . radio wave scattering in the interstellar medium , aip conference proceedings no . 174 , 217 - 221 , eds cordes , j. , rickett , b. j. & backer , d. c. , american institute of physics hobbs , g. b. , edwards , r. t. & manchester , r. n. , 2006 , mnras , 369 , 655 johnston , s. , 1990 . phd thesis , the university of manchester . johnston , simon . , nicastro , luciano . , koribalski , barbel . , 1998 , mnras,297 , 108 . lang , k.r , 1971 , apj , 164 , 249 leahy , d.a . , tian , w.w . , 2007 , a&a , 461 , 1013 lewandowski w , dembska m , kijak j & kowalinska m , 2013 , mnras , 434 , 69 lhmer o , kramer m , mitra d , lorimer d.r & lyne a.g , 2001 , apj , 562 , 157 lhmer o , mitra d , gupta y , kramer m & ahuja a , 2004 , a&a , 425 , 569 lorimer , d. r. , yates , j. a. , lyne , a. g. & gould , d. m , 1995,mnras,273,411 manchester , r. n. , hobbs , g. b. , teoh , a. & hobbs , m. , 1993 ( 2005 ) , aj , 129 manoharan , p.k , 2010 , soph , 265 , 137 maiz apellaniz , j. , alfaro , e. j. & sota , a. , 2008 , astro - ph , 2008arxiv0804.2553 m mitra , d & ramachandran , r , 2001 , a&a , 370 , 586 mitra , d & rankin , j.m . , apj , 2002 , 577 , 322 naidu , a . , joshi , b.c . , manoharan , p.k . & krishnakumar , m.a , 2014 , exp . astronomy ( submitted ) press , w.h . , flannery , b.p . , teukolsky , s.a . , vetterling , w.t . , 2001 , numerical recipes in fortran 77 : the art of scientific computing ramachandran r , mitra d , deshpande a.a , mcconnel d.m & ables j.g , 1997 , mnras , 290 , 260 ramkumar , p.s . , prabhu , t. , girimaji , m. , & markandeyulu , g. , 1994 , japa , 15 , 343 rickett , b. j. , 1977 , ara&a , 15 , 479 roberts , j. a. & ables , j. g. , 1982 , mnras , 201 , 1119 - 1138 . romani , r. w. , narayan , r. , blandford , r. , 1986 , mnras , 220,19 scheuer , p. a. g. , 1968 , nature , 218 , 920 selvanayagam a.j , praveenkumar a , nandagopal d & velusamy t , 1993 , iete tech . , 10 , 333 sharpless , stewart , 1959 , apj , 4 , 257 sieber , w , 1973 , a&a , 28 , 237 sutton , j. m. , 1971 , mnras , 155 , 51 swarup g , sarma n.v.g , joshi m.n , et al , 1971 , nat . , 230 , 185 taylor j.h , manchester r.n & lyne a.g , 1993 , apjs , 88 , 529 thorsett , s. e. , 1991 , apj , 377,263 wang n. , manchester r.n . & johnston s. , 2007 , mnras , 377 , 1383 weisberg j.m , pildis r , cordes j.m , spangler s.r , & clifton t. , 1990 , baas , 22 , 1244 williamson , ian p. , 1972 , mnras , 157 , 55 williamson , ian p. , 1973 , mnras , 163 , 345 1 & b0447 - 12 @xmath78 & 211.08 & -32.63 & 37.04 & [email protected] & 33.77 & 1.0 & 464 & 1.9 & 0.02 & -1.38 + 2 & b0450 - 18 @xmath78 & 217.08 & -34.09 & 39.90 & [email protected] & 37.33 & 0.8 & 487 & 2.4 & 0.02 & -2.16 + 3 & b0458 + 46 & 160.36 & 3.08 & 42.19 & 19@xmath551 & 36.79 & 1.2 & 483 & 1.4 & 0.05 & -0.57 + 4 & b0523 + 11 @xmath78 & 192.70 & -13.25 & 79.34 & [email protected] & 19.90 & 1.0 & 478 & 3.1 & 0.21 & -2.27 + 5 & b0531 + 21 & 184.56 & -5.78 & 56.79 & [email protected] & 27.90 & 1.0 & 124 & 1.7 & 0.28 & -1.52 + 6 & b0559 - 05 @xmath78 & 212.20 & -13.48 & 80.54 & [email protected] & 26.52 & 1.0 & 415 & 3.9 & 0.43 & -2.29 + 7 & b0611 + 22 & 188.79 & 2.40 & 96.91 & [email protected] & 19.51 & 0.8 & 392 & 2.1 & 0.3 & -1.75 + 8 & b0621 - 04 @xmath78 & 213.79 & -8.04 & 70.83 & [email protected] & 67.47 & 1.1 & 382 & 2.8 & 0.2 & -1.84 + 9 & b0626 + 24 @xmath78 & 188.82 & 6.22 & 84.19 & [email protected] & 28.63 & 1.0 & 482 & 2.2 & 0.32 & -1.76 + 10 & * b0736 - 40 * @xmath78 & 254.19 & -9.19 & 160.80 & 84@xmath551 & 254.84 & 1.0 & 125 & 2.6 & 117 & -0.51 + 11 & b0740 - 28 @xmath78 & 243.77 & -2.44 & 73.78 & [email protected] & 17.28 & 1.1 & 383 & 2.1 & 0.0001 & -2.22 + 12 & j0749 - 4247 & 257.07 & -8.35 & 104.59 & [email protected] & 29.95 & 0.8 & 253 & 0.3 & 21 & 0.46 + 13 & * b0808 - 47 * & 263.30 & -7.96 & 228.30 & 78@xmath552 & 134.67 & 1.2 & 253 & 0.3 & 131 & 1.26 + 14 & b0818 - 41 @xmath78 & 258.75 & -2.74 & 113.40 & [email protected] & 234.38 & 1.1 & 124 & 0.3 & 24 & 0.55 + 15 & b0833 - 45 & 263.55 & -2.79 & 67.99 & [email protected] & 50.97 & 0.9 & 125 & 0.2 & 7.01 & 0.39 + 16 & * b0835 - 41 * & 260.90 & -0.34 & 147.29 & [email protected] & 32.30 & 0.5 & 1021 & 1.0 & 2.9 & -1.14 + 17 & b0840 - 48 & 267.18 & -4.10 & 196.85 & 5@xmath551 & 20.14 & 1.0 & 253 & 0.3 & 93 & 0.28 + 18 & * b0844 - 35 * @xmath78 & 257.19 & 4.71 & 94.16 & 5@xmath551 & 82.48 & 0.9 & 430 & 0.6 & 0.00001 & -0.35 + 19 & * b0853 - 33 * & 256.85 & 7.517 & 86.64 & [email protected] & 34.15 & 1.0 & 331 & 0.6 & 0.00001 & -0.21 + 20 & j0857 - 4424 & 265.46 & 0.82 & 184.43 & 16@xmath553 & 26.81 & 1.0 & 252 & 1.9 & 11 & -0.90 + 21 & * b0903 - 42 * & 265.08 & 2.86 & 145.80 & 11@xmath551 & 45.24 & 1.3 & 253 & 0.6 & 7 & -0.13 + 22 & j0905 - 4536 @xmath78 & 267.24 & 1.01 & 179.70 & 18@xmath552 & 216.19 & 1.1 & 253 & 2.3 & 0.02 & -0.82 + 23 & b0906 - 49 & 270.27 & -1.02 & 180.37 & 7@xmath551 & 65.13 & 1.0 & 97 & 2.6 & 0.03 & -1.44 + 24 & * b0950 - 38 * & 268.70 & 12.03 & 167.00 & [email protected] & 96.60 & 1.1 & 253 & 6.5 & 0.16 & -1.99 + 25 & * b1015 - 56 * & 282.73 & 0.34 & 439.10 & 13@xmath551 & 47.20 & 2.4 & 126 & 8.9 & 112 & -2.18 + 26 & j1036 - 4926 @xmath78 & 281.52 & 7.73 & 136.53 & 2@xmath551 & 11.96 & 1.1 & 125 & 4.0 & 1.2 & -2.19 + 27 & * b1039 - 55 * @xmath78 & 285.19 & 2.99 & 306.50 & 9@xmath551 & 45.73 & 1.1 & 254 & 6.3 & 34 & -2.04 + 28 & b1044 - 57 & 287.07 & 0.73 & 240.20 & 19@xmath551 & 37.52 & 1.2 & 125 & 4.4 & 24 & -1.47 + 29 & b1119 - 54 & 290.08 & 5.87 & 204.70 & [email protected] & 58.60 & 1.2 & 125 & 5.2 & 8.3 & -1.90 + 30 & j1123 - 4844 @xmath78 & 288.29 & 11.61 & 92.92 & [email protected] & 12.43 & 1.1 & 253 & 2.9 & 0.2 & -2.17 + 31 & j1210 - 5559 @xmath78 & 297.14 & 6.42 & 174.35 & [email protected] & 6.56 & 1.0 & 253 & 4.3 & 7 & -2.57 + 32 & b1317 - 53 & 307.30 & 8.64 & 97.60 & [email protected] & 10.93 & 0.9 & 125 & 2.3 & 1.4 & -1.92 + 33 & b1325 - 49 @xmath78 & 309.12 & 13.07 & 118.00 & [email protected] & 63.54 & 1.0 & 509 & 3.7 & 1.1 & -2.30 + 34 & b1523 - 55 & 323.64 & 0.59 & 362.70 & 59@xmath556 & 49.16 & 1.0 & 253 & 5.2 & 128 & -1.21 + 35 & * j1557 - 4258 * & 335.27 & 7.95 & 144.50 & [email protected] & 22.01 & 1.1 & 253 & 4.6 & 10 & -1.98 + 36 & * b1600 - 49 * @xmath78 & 332.15 & 2.44 & 140.80 & [email protected] & 19.18 & 0.6 & 253 & 5.1 & 15 & -2.46 + 37 & * b1609 - 47 * & 334.57 & 2.84 & 161.20 & 7@xmath551 & 22.41 & 0.9 & 253 & 3.7 & 9.4 & -1.72 + 38 & * j1614 - 3937 * @xmath78 & 340.00 & 8.21 & 152.44 & [email protected] & 28.64 & 0.9 & 254 & 3.8 & 1.5 & -2.20 + 39 & b1620 - 42 & 338.89 & 4.62 & 295.00 & 52@xmath556 & 27.06 & 1.0 & 253 & 6.6 & 13 & -1.44 + 40 & j1625 - 4048 @xmath78 & 340.61 & 5.93 & 145.00 & 4@xmath552 & 147.22 & 1.1 & 252 & 3.1 & 5 & -1.56 + 41 & b1635 - 45 & 338.50 & 0.46 & 258.91 & 26@xmath551 & 63.79 & 1.1 & 362 & 3.8 & 60 & -1.24 + 42 & j1648 - 3256 & 349.59 & 7.75 & 128.28 & [email protected] & 16.23 & 1.1 & 396 & 3.1 & 2.7 & -1.72 + 43 & * b1647 - 52 * & 335.01 & -5.17 & 179.10 & [email protected] & 29.77 & 1.1 & 253 & 3.8 & 12 & -1.73 + 44 & j1700 - 3312 @xmath78 & 351.06 & 5.49 & 166.97 & 4@xmath551 & 57.73 & 1.0 & 397 & 3.6 & 3 & -1.92 + 45 & b1658 - 37 & 347.76 & 2.83 & 303.40 & 543@xmath5555 & 1112.15 & 1.1 & 125 & 5.2 & 61 & -0.40 + 46 & b1700 - 32 @xmath78 & 351.79 & 5.39 & 110.31 & [email protected] & 71.10 & 0.9 & 406 & 2.3 & 2.5 & -1.44 + 47 & j1703 - 4851 & 338.99 & -4.51 & 150.29 & 10@xmath551 & 60.01 & 0.9 & 253 & 3.0 & 2.9 & -1.42 + 48 & * j1705 - 3423 * & 350.72 & 3.98 & 146.36 & 41@xmath553 & 68.84 & 0.9 & 252 & 2.9 & 10 & -0.86 + 49 & j1708 - 3426 & 351.08 & 3.41 & 190.70 & 120@xmath5513 & 51.37 & 1.0 & 125 & 3.6 & 11 & -0.65 + 50 & * b1718 - 32 * & 354.56 & 2.53 & 126.06 & [email protected] & 51.84 & 1.0 & 402 & 2.4 & 13 & -1.10 + 51 & b1727 - 47 & 342.57 & -7.67 & 123.33 & [email protected] & 71.31 & 0.9 & 125 & 2.8 & 3 & -1.40 + 52 & * b1729 - 41 * & 347.98 & -4.46 & 195.30 & 26@xmath554 & 24.53 & 1.2 & 252 & 3.9 & 28 & -1.27 + 53 & b1737 - 30@xmath79}$ ] @xmath35 & 358.29 & 0.24 & 152.15 & 117@xmath556 & 200.28 & 1.0 & 196 & 2.7 & 11 & -0.44 + 54 & b1738 - 08 @xmath78 & 16.96 & 11.30 & 74.90 & [email protected] & 105.52 & 1.1 & 481 & 2.2 & 0.3 & -1.67 + 55 & b1737 - 39 & 350.56 & -4.75 & 158.50 & [email protected] & 56.03 & 1.0 & 253 & 3.2 & 5.3 & -1.27 + 56 & b1740 - 13 & 12.70 & 8.21 & 116.30 & [email protected] & 19.35 & 1.0 & 374 & 2.9 & 1.5 & -2.56 + 57 & b1740 - 31 & 357.29 & -1.15 & 193.05 & 296@xmath5524 & 244.03 & 1.0 & 471 & 3.3 & 29 & -0.27 + 58 & * b1742 - 30 * @xmath78 & 358.55 & -0.963 & 88.37 & [email protected] & 35.88 & 1.1 & 253 & 1.9 & 2.7 & -1.41 + 59 & * j1750 - 3503 * & 355.31 & -4.08 & 189.35 & 51@xmath557 & 101.54 & 1.1 & 252 & 3.9 & 48 & -1.03 + 60 & * b1756 - 22 * & 7.47 & 0.81 & 177.16 & [email protected] & 19.72 & 1.0 & 441 & 3.6 & 17 & -1.59 + 61 & b1758 - 03 & 23.60 & 9.26 & 120.37 & [email protected] & 29.66 & 1.0 & 432 & 3.6 & 0.7 & -2.58 + 62 & b1804 - 12 @xmath78 & 17.14 & 4.42 & 122.41 & [email protected] & 40.83 & 1.0 & 125 & 2.8 & 2.2 & -1.93 + 63 & b1804 - 08 @xmath78 & 20.06 & 5.59 & 112.38 & [email protected] & 22.47 & 1.0 & 150 & 2.7 & 4.4 & -1.74 + 64 & * b1804 - 27 * & 3.84 & -3.26 & 312.98 & 37@xmath552 & 90.54 & 1.3 & 125 & 5.0 & 31 & -1.35 + 65 & j1808 - 0813 & 20.63 & 5.75 & 151.27 & 12@xmath551 & 43.80 & 0.8 & 396 & 3.8 & 2 & -1.54 + 66 & b1813 - 26 @xmath78 & 5.22 & -4.91 & 128.12 & [email protected] & 58.51 & 0.9 & 149 & 3.1 & 2.8 & -1.92 + 67 & j1817 - 3837 & 354.68 & -10.41 & 102.85 & 4@xmath551 & 13.52 & 1.2 & 252 & 2.5 & 1 & -1.55 + 68 & b1818 - 04 @xmath78 & 25.46 & 4.73 & 84.44 & [email protected] & 48.15 & 2.7 & 469 & 1.9 & 3.7 & -1.53 + 69 & b1819 - 22 @xmath78 & 9.35 & -4.37 & 121.20 & [email protected] & 161.05 & 1.0 & 253 & 3.0 & 2.7 & -1.27 + 70 & * j1823 - 0154 * & 28.08 & 5.26 & 135.87 & [email protected] & 20.95 & 1.2 & 396 & 3.6 & 8.2 & -1.75 + 71 & b1821 - 19 & 12.28 & -3.11 & 224.65 & 56@xmath553 & 77.01 & 1.0 & 173 & 4.7 & 25 & -1.15 + 72 & b1823 - 11 & 19.80 & 0.29 & 320.58 & 163@xmath5513 & 149.52 & 1.0 & 444 & 4.6 & 133 & -0.73 + 73 & * b1826 - 17 * & 14.60 & -3.42 & 217.11 & 92@xmath554 & 132.43 & 1.1 & 266 & 4.7 & 90 & -0.96 + 74 & b1829 - 08 @xmath78 & 23.27 & 0.29 & 300.87 & 52@xmath555 & 54.98 & 0.9 & 361 & 4.9 & 239 & -1.20 + 75 & * b1831 - 03 * & 27.66 & 2.27 & 234.54 & 50@xmath551 & 181.46 & 1.0 & 289 & 5.1 & 29 & -1.26 + 76 & j1835 - 1106 & 21.22 & -1.51 & 132.68 & [email protected] & 12.06 & 1.0 & 395 & 2.8 & 4.0 & -1.51 + 77 & b1834 - 10 @xmath35 & 22.26 & -1.42 & 316.98 & 410@xmath55184 & 124.17 & 0.7 & 297 & 4.5 & 96 & -0.38 + 78 & b1839 - 04 @xmath78 & 28.35 & 0.17 & 195.98 & 215@xmath5511 & 618.78 & 1.1 & 331 & 4.7 & 2.4 & -0.71 + 79 & b1841 - 04 & 28.10 & -0.55 & 123.16 & 43@xmath552 & 74.55 & 1.0 & 449 & 2.9 & 3.2 & -0.87 + 80 & b1844 - 04 & 28.88 & -0.94 & 141.98 & 33@xmath551 & 91.29 & 0.9 & 474 & 3.3 & 6 & -1.05 + 81 & * b1845 - 01 * & 31.34 & 0.04 & 159.53 & 115@xmath555 & 140.48 & 1.3 & 456 & 4.0 & 106 & -0.75 + 82 & * j1848 - 1414 * & 19.90 & -5.84 & 134.47 & 8@xmath551 & 8.19 & 1.2 & 397 & 3.6 & 33 & -1.84 + 83 & b1846 - 06 & 26.77 & -2.50 & 148.17 & 24@xmath551 & 78.46 & 1.3 & 478 & 3.4 & 4 & -1.20 + 84 & b1852 + 10 & 42.89 & 4.22 & 207.20 & 36@xmath552 & 44.78 & 1.3 & 252 & 6.9 & 2 & -1.50 + 85 & b1851 - 14 & 20.46 & -7.21 & 130.40 & [email protected] & 51.70 & 1.0 & 485 & 3.7 & 1.2 & -2.49 + 86 & b1859 + 01 & 35.82 & -1.37 & 105.39 & [email protected] & 8.62 & 1.0 & 298 & 2.8 & 0.45 & -2.30 + 87 & b1859 + 03 @xmath80 & 37.21 & -0.64 & 402.08 & 177@xmath556 & 390.28 & 0.8 & 170 & 7.3 & 182 & -1.08 + 88 & b1859 + 07 & 40.57 & 1.06 & 252.81 & 23@xmath555 & 27.93 & 1.0 & 434 & 5.5 & 80 & -1.58 + 89 & * b1900 + 05 * & 39.50 & 0.21 & 177.49 & 29@xmath553 & 30.43 & 1.1 & 394 & 4.7 & 29 & -1.37 + 90 & b1900 + 06 & 39.81 & 0.34 & 502.90 & 197@xmath5520 & 265.14 & 1.5 & 123 & 8.4 & 1085 & -1.16 + 91 & b1900 + 01 & 35.73 & -1.96 & 245.17 & [email protected] & 81.19 & 1.0 & 1021 & 3.3 & 43 & -1.12 + 92 & * b1900 - 06 * & 28.48 & -5.68 & 195.61 & [email protected] & 92.98 & 0.9 & 365 & 5.4 & 20 & -1.61 + 93 & * j1904 + 0004 * & 34.45 & -2.81 & 233.61 & 15@xmath552 & 25.10 & 1.2 & 112 & 5.7 & 11 & -1.78 + 94 & j1904 - 1224 & 23.29 & -8.49 & 118.23 & [email protected] & 26.28 & 0.9 & 397 & 3.4 & 1 & -1.69 + 95 & * b1902 - 01 * & 33.69 & -3.55 & 229.13 & [email protected] & 25.67 & 0.5 & 473 & 6.0 & 10 & -2.07 + 96 & j1908 + 0500 & 39.29 & -1.40 & 201.42 & 9@xmath552 & 6.82 & 1.0 & 252 & 5.2 & 25 & -1.89 + 97 & b1907 + 02 @xmath78 & 37.61 & -2.71 & 171.73 & [email protected] & 37.20 & 0.9 & 476 & 4.9 & 3.9 & -2.31 + 98 & b1907 + 10 @xmath78 & 44.83 & 0.99 & 149.98 & [email protected] & 11.36 & 0.8 & 1021 & 4.2 & 3.2 & -2.40 + 99 & b1907 - 03 @xmath78 & 32.28 & -5.68 & 205.53 & [email protected] & 24.64 & 0.9 & 1021 & 6.1 & 1.4 & -2.45 + 100 & j1910 + 0714 & 41.52 & -0.87 & 124.06 & 4@xmath551 & 42.38 & 0.7 & 509 & 4.1 & 4.2 & -2.03 + 101 & b1907 + 12 & 46.22 & 1.59 & 258.64 & 52@xmath555 & 31.60 & 1.0 & 363 & 7.4 & 22 & -1.53 + 102 & b1911 - 04 @xmath78 & 31.31 & -7.12 & 89.39 & [email protected] & 29.23 & 0.6 & 1020 & 2.8 & 1.65 & -2.11 + 103 & b1911 + 13 @xmath78 & 47.88 & 1.59 & 145.05 & [email protected] & 25.75 & 1.0 & 402 & 5.1 & 6.4 & -2.92 + 104 & b1914 + 13 & 47.58 & 0.45 & 237.01 & 6@xmath551 & 10.68 & 1.0 & 392 & 5.0 & 11 & -1.99 + 105 & b1915 + 13 & 48.26 & 0.62 & 94.54 & [email protected] & 13.33 & 1.2 & 216 & 4.0 & 3.05 & -2.84 + 106 & b1918 + 19 & 53.87 & 2.67 & 153.85 & [email protected] & 86.43 & 1.0 & 282 & 5.6 & 0.62 & -1.64 + 107 & b1920 + 21 @xmath78 & 55.28 & 2.94 & 217.09 & [email protected] & 44.48 & 1.1 & 409 & 7.6 & 4.6 & -2.69 + 108 & b1922 + 20 & 55.02 & 2.33 & 213.00 & 9@xmath554 & 3.98 & 0.8 & 235 & 7.3 & 1.3 & -2.15 + 109 & b1923 + 04 & 40.98 & -5.67 & 102.24 & [email protected] & 26.67 & 0.9 & 440 & 3.8 & 0.2 & -2.15 + 110 & b1924 + 14 @xmath78 & 49.92 & -1.04 & 211.41 & 19@xmath551 & 36.55 & 1.0 & 432 & 6.7 & 13 & -1.82 + 111 & b1924 + 16 & 51.86 & 0.06 & 176.88 & [email protected] & 18.66 & 1.0 & 370 & 5.8 & 7 & -2.22 + 112 & b1929 + 20 & 55.58 & 0.64 & 211.15 & 19@xmath551 & 28.60 & 1.1 & 419 & 6.9 & 18 & -1.84 + 113 & b1930 + 22 & 57.36 & 1.55 & 219.20 & [email protected] & 5.58 & 1.1 & 230 & 7.5 & 1.4 & -3.13 + 114 & b1933 + 16 & 52.44 & -2.09 & 158.52 & [email protected] & 28.62 & 1.0 & 373 & 5.6 & 0.9 & -2.32 + 115 & b1944 + 22 & 59.30 & -1.07 & 140.00 & 12@xmath551 & 31.28 & 1.2 & 253 & 5.4 & 1 & -1.81 + 116 & b1946 + 35 & 70.70 & 5.05 & 129.07 & [email protected] & 138.63 & 0.8 & 473 & 5.8 & 28 & -1.48 + 117 & b2000 + 32 & 69.26 & 0.88 & 142.21 & 9@xmath551 & 25.07 & 0.9 & 414 & 5.7 & 0.6 & -1.96 + 118 & b2002 + 31 & 69.01 & 0.02 & 234.82 & [email protected] & 52.67 & 1.2 & 478 & 7.5 & 10 & -2.18 + 119 & * b2011 + 38 * & 75.93 & 2.48 & 238.22 & 22@xmath552 & 34.99 & 1.0 & 325 & 8.4 & 16 & -1.94 + 120 & b2027 + 37 & 76.89 & -0.73 & 190.66 & [email protected] & 53.89 & 1.1 & 426 & 6.3 & 3 & -1.95 + 121 & b2053 + 36 & 79.13 & -5.59 & 97.31 & [email protected] & 16.56 & 0.9 & 385 & 4.6 & 5.8 & -1.97 + 122 & b2106 + 44 & 86.91 & -2.01 & 139.83 & 16@xmath551 & 33.32 & 1.1 & 494 & 5.0 & 1.4 & -1.64 + 123 & b2111 + 46 @xmath78 & 89.00 & -1.27 & 141.26 & [email protected] & 234.48 & 0.7 & 235 & 4.5 & 1 & -2.36 + 124 & b2319 + 60 @xmath78 & 112.10 & -0.57 & 94.59 & 11@xmath551 & 152.90 & 1.2 & 425 & 3.0 & 1 & -1.40 + [ tab1 ]

the x - ray afterglows ( ags ) of gamma - ray bursts ( grbs ) and x - ray flashes ( xrfs ) have , after the fast decline phase of their prompt emission , a temporal behaviour varying between two extremes . a large fraction of these ags has a ` canonical ' light curve which , after an initial shallow - decay ` plateau ' phase , ` breaks smoothly ' into a fast power - law decline . very energetic grbs , contrariwise , appear not to have a ` break ' , their ag declines like a power law from the start of the observations . breaks and ` missing breaks ' are intimately related to the geometry and deceleration of the jets responsible for grbs . in the frame of the ` cannonball ' ( cb ) model of grbs and xrfs , we analyze the cited extreme behaviours ( canonical and pure power law ) and intermediate cases spanning the observed range of x - ray ag shapes . we show that the entire panoply of x - ray light - curve shapes measured with swift and other satellites are as anticipated in the cb model . we test the expected correlations between the ag s shape and the peak- and isotropic energies of the prompt radiation , strengthening a simple conclusion of the analysis of ag shapes : in energetic grbs the break is not truly ` missing ' , it is hidden under the tail of the prompt emission , or it occurs too early to be recorded . we also verify that the spectral index of the unabsorbed ags and the temporal index of their late power - law decline differ by half a unit , as predicted . -.5 cm

0712.1527

the isotropic distribution of gamma ray bursts ( grbs ) in the sky and their number distribution as function of intensity , measured with the batse instrument aboard the compton gamma ray observatory , provided the first observational evidence that gamma ray bursts ( grbs ) originate at large cosmological distances ( meegan et al . moreover , the rapid variation of their light curves ( bahat et al . 1992 ) indicated that their huge energy is emitted from a very small volume . in the original fireball ( fb ) model of grbs ( e.g. paczynski 1986 ; goodman 1986 ; rees & mszros 1992 ) the emission was spherically symmetric . the implied isotropic energy release of grbs in @xmath0-rays often exceeded @xmath1 , creating an ` energy crisis ' . indeed , such a mighty , abrupt , compact , and @xmath0-ray - efficient source was unforeseen . a simple solution to this puzzle was suggested by shaviv and dar ( 1995 ) : the @xmath0-ray emission is narrowly collimated by the relativistic motion of their _ jetted _ source , which is seen when it points closely enough to the observer . in this view , grbs are not produced by fireballs , but by inverse compton scattering of light by highly relativistic jets of ordinary matter , ejected in violent stellar processes such as supernova explosions , mergers of neutron stars ( paczynski 1986 ; goodman , dar & nussinov 1987 ; dar et al . 1992 ) , or the direct collapse of massive stars to black holes without a supernova ( woosley 1993 ) . the sky localization of grbs by bepposax ( costa et al . 1997 ) led to the discovery ( groot et al . 1997 ; van paradijs et al . 1997 ) of their optical afterglows ( ags ) and their host galaxies ( sahu et al . 1997 ) which were used to extract their cosmological redshifts ( metzger et al . the ags seemed to follow an achromatic power - law decline , as expected from a highly relativistic expanding fireball that drives a blast wave into the circumburst environment ( e.g. , mszros & rees 1997 ) . this prediction of the spherical fireball model ( see e.g. , piran 1999 ) being independent of the assumption of spherical symmetry , it was also argued that the ags , like the grbs themselves , are produced by narrowly collimated jets ( dar 1997 , 1998 ) . the concept of jets was incorporated into the fb model by the substitution of its spherical shells by conical sections thereof , the mechanism for the @xmath0-ray emission still being synchrotron radiation from shock - accelerated @xmath2 pairs in a baryon - poor material ( see , e.g. piran 1999 , 2000 ; mszros 2002 and references therein ) , in spite of the difficulties that such a radiation mechanism encounters ( ghisellini et al . 2000 ) . an elegant and simple way to distinguish between a conical jet and a spherical fireball was suggested by rhoads ( 1997 ) : the ag of a decelerating conical jet will show an achromatic steepening a _ jet break_ in its power - law decline when the relativistic beaming angle of its radiation becomes larger than the opening angle of the jet . soon afterwards , better sampled data on the optical afterglow of grbs showed the existence of what appeared to be such achromatic jet breaks ( harrison et al . 1999 ; stanek et al . 1999 ) , and the spherical fb model was modified into a _ collimated fireball _ model ( e.g. piran , sari & halpern 1999 ) . in this model grb pulses are produced by synchrotron radiation from the collision between conical sections of shells . the collision of the ensemble of shells with the interstellar matter ( ism ) generates the ag by synchrotron radiation from the forward shock propagating in the ism , and/or from the backwards shock within the merged shells . rhoads ( 1999 ) and sari , piran and halpern ( 1999 ) derived a relation between the opening angle of the conical jet and the time of the jet break . this relation has been applied extensively to the pre - swift data to infer the opening angle of the conical jet and to determine the ` true ' energy of grbs , posited to be an approximate standard candle ( frail et al . 2001 ) . since the launch of swift the above generally - accepted ` standard ' paradigm has been challenged , due to the absence of breaks in the ags of many grbs ( panaitescu et al . 2006 ; burrows and racusin 2006 ) , to the chromatic behaviour of the ag of other grbs having the alleged ` jet break ' ( stanek et al . 1999 , harrison et al . 1999 ) , and to the failure of the frail relation ( frail et al . 2001 ) in many swift grbs ( kocevski & butler 2007 ) . in the fireball model , the jet breaks need not be sharp ; they are often parametrized with a varying smoothness ( stanek et al . 1999 ) . allowing for such breaks , covino et al . ( 2006 ) could not identify a swift grb with a fully achromatic break . liang et al . ( 2007 ) have extended this study , and analyzed the swift x - ray data for the 179 grbs detected between january 2005 and january 2007 and the optical ags of 57 pre- and post - swift grbs . they found that not a single burst satisfies all the criteria of a jet break . this brings us fully into the question of the nature and properties of the jets responsible for grbs and their ags and , more specifically in this paper , to the understanding of ` breaks ' and ` missing breaks ' . an alternative to the fireball scenario is offered by the ` cannonball ' ( cb ) model of grbs [ dar & de rjula 2000a , 2004 , hereafter dd2000a , 2004 ; dado , dar & de rjula ( hereafter ddd ) 2002 ; for a recent review see de rjula 2007 ] . in this model _ long - duration _ grbs and their ags are produced by bipolar jets of cbs ( shaviv & dar 1995 ; dar & plaga 1999 ) , ejected in _ ordinary core - collapse _ supernova ( sn ) explosions as matter is accreted onto the newly - formed compact object ( de rjula 1987 ) . the ` cannon - balls ' are made of _ ordinary - matter plasma_. the @xmath0-rays of a single pulse of a grb are produced as a cb coasts through the sn _ glory _ the initial sn light , scattered away from the radial direction by the ` wind ' : the ejecta puffed by the progenitor star in a succession of pre - sn flares . the electrons enclosed in the cb raise the glory s photons to grb energies by _ inverse compton scattering ( ics)_. as a cb coasts through the glory , the distribution of the glory s light becomes increasingly radial and its density decreases rapidly . consequently , the energy of the up - scattered photons is continuously shifted to lower energies and their number decreases swiftly , resulting in a fast softening and decline of the prompt emission ( dd2004 , 2007a , b ) . in the cb model , the ag of a grb is due to _ synchrotron radiation ( sr ) _ from swept - in ism electrons spiraling in the cb s enclosed turbulent magnetic field , generated by the intercepted ism nuclei and electrons ( ddd2002 ) . at x - ray energies , the sr afterglow begins to dominate the ics prompt emission only during the fast - decline phase of the latter ( ddd2006 ) . in the cb model , the _ beau rle _ in the understanding of grbs is played by the doppler factor , @xmath3 , relating times , energies and fluxes in a cb s rest system to those in the observer s system . its form in terms of the observer s angle @xmath4 ( relative to the cb s direction of motion ) and the time - dependent lorentz factor , @xmath5 , of a cb , is : @xmath6}\approx { 2\ , \gamma(t)\over 1+[\gamma(t)\ , \theta]^2}\ ; , \label{delta}\ ] ] where the approximation is excellent for @xmath7 and @xmath8 . the decrease of @xmath5 with time , as a cb encounters the particles of the ism , is calculable on grounds of energy - momentum conservation ( ddd2002 , dar & de rjula 2006 , thereafter dd2006 ) . the energy - integrated energy flux of the ag of a grb , is @xmath9 . let @xmath10 . consider a cb that is observed almost on axis , so that @xmath11 : the observer is _ ab initio _ within the opening cone of the relativistically beamed radiation . as @xmath5 decreases , @xmath3 monotonically decreases and so does the observed ag . consider the same cb , viewed by an observer at a much larger angle , so that @xmath12 is ` a few ' . as @xmath5 decreases , @xmath3 in eq . ( [ delta ] ) _ increases _ , reflecting the fact that the characteristic opening angle of the radiation , @xmath13 , is reaching the observer s direction . past the point @xmath14 , the decrease of @xmath3 is monotonic , as in the first case we considered . the ag radiation parallels again the behaviour of @xmath3 . for observers of the same grb from different angles , as @xmath4 increases at fixed @xmath5 , the ag s flux decreases . all these simple facts , supported by the corresponding explicit derivations , are reflected in fig . [ f1]a , which we have copied from ddd2002 , as it foretells the progressive variety of ag shapes to be studied here . there is more to fig . [ f1]a than what we said . the lorentz factor @xmath5 of a cb only begins to change significantly , in a calculable manner , when the increase in its mass induced by the energy influx of the swept - in ism particles becomes comparable to the cb s initial mass . this happens , as we shall review , at a time @xmath15/\gamma_0 ^ 3 $ ] . at fixed @xmath5 , as reflected in fig . [ f1]a , a larger @xmath4 entails a larger @xmath16 . this achromatic ` deceleration bend ' at @xmath17 , we believe , was often interpreted in fb models as a putative jet break . naturally , the values of @xmath18 and @xmath19 of a given cb also affect the properties of its prompt ics - dominated radiation ( we are presenting this introductory discussion as if there was a single cb generating the prompt and ag radiations , a simplification to be undone when needed ) . in the cb model the ics - dictated @xmath20 dependences of a cb s isotropic energy , peak energy and peak luminosity are @xmath21 , @xmath22 and @xmath23 ( dd2000b ) . the conditions for these quantities to be relatively large ( a relatively small @xmath4 or a large @xmath18 ) are the ones leading to a luminous ag with a small @xmath16 . the basis for one of these expected correlations , studied before in detail in ddd2007c , is illustrated in fig . [ f1]b . if the deceleration bend at time @xmath16 takes place _ after _ the fast - decline phase of the prompt emission , it is observable , and the unabsorbed x - ray light curve is canonical ( ddd2002 ) . in these cases , there is a ` break ' . if @xmath16 takes place earlier , it is hidden under the prompt emission , and only the tail of the canonical behaviour , namely the ` late ' power - law decline of the unabsorbed synchrotron afterglow , is observable . in these cases , the break is missing . the transition from long - plateau , clearly ` broken ' ags , to power - law like ` unbroken ' ags should be anticorrelated with the trend from under-`energetic ' to over - energetic grbs . in the cb model the late - time spectral energy density @xmath24 of the x - ray and optical ag tends to a time and energy - dependence @xmath25 , with @xmath26 the spectral index of the electrons accelerated within a cb and cooled by the emission of the very sr seen as the ag . a prediction that we have not emphasized before is that the temporal power decline should be , grb by grb , half a unit steeper than the spectral decline . in ddd2006 , 2007a we have demonstrated that the most common light curves of the x - ray ag of grbs are well described by the cb model . we have also explained there the various origins of the chromatic behaviours of ags . in ddd2007b we have focused on the fast decline phase of the prompt emission and we have demonstrated that the rapid spectral evolution observed during this phase is also as expected in the cb model . in ddd2007c we have shown , for large ensembles of grbs , how the observed correlations between @xmath27 , @xmath28 [ dar & de rjula 2000b ( dd2000b ) , amati et al . 2002 ] , @xmath29 , and other prompt observables ( pulse rise - time , lag - time and variability ) follow mainly from the same simple geometrical considerations that we have reviewed above on the case - by - case variability of the doppler factor . in the cb model , xrfs are simply grbs seen at relatively large @xmath4 ( ddd2004a ) , even the particularly interesting xrf 060218 is in no way exceptional ( ddd2007a ) . in this paper we focus on the shape of the light curves of the x - ray afterglow of grbs , with and without breaks , measured with the x - ray telescope ( xrt ) aboard swift . we show that the shapes of the x - ray light curves of grbs and xrfs predicted in fig . [ f1]a , and the correlation between @xmath16 and @xmath27 illustrated fig . [ f1]b ( and the consequent apparent presence or absence of breaks in the ag ) agree with the cb - model s expectations . we also analize the @xmath30 correlation on the same light . finally , we investigate the relation between the temporal power - law index of the post - break decline and the photon spectral index , reaching satisfactory results . to do all this , we investigate 16 grbs chosen to reflect the full span of the question of the presence or absence of breaks . the selected grbs range from the faintest known grb ( 980425 , of supernova - association fame ) , which also has the most pronounced plateau and the latest break time , to the brightest swift grb ( 061007 ) , with the most luminous and longest - observed unbroken power - law x - ray ag . in the cb model , the mechanism for the emission of the prompt radiation of grbs and xrfs is inverse compton scattering . the temporal and spectral properties of the prompt phase , including its fast decline , are summarized in a ` master formula ' ( dd2004 ) that we have already contrasted with swift data ( ddd2006 , 2007a , b , c , d ) . we shall not repeat it here as our emphasis in the current study is on breaks in the x - ray light curves of grb afterglows though we shall use it to describe the fading of the prompt emission until the take - over by the synchrotron - ag emission , and the occasional late x - ray flares . neither do we discuss here the optical ags ( ddd2007a ) . the extinction in the optical- and , more so , in the radio- domain ( within the cbs , in the circumburst environment , in the ism of the host galaxy and ours , and in the intergalactic medium ) are difficult to model as reliably as the x - ray extinction . we shall see once again that the x - ray light curves ( corrected for extinction ) carry clear and direct information on the radiation mechanisms that dominate the prompt emission and the ag phase ( ics and sr , respectively , in the cb model ) . during the initial phase of @xmath0-ray emission in a grb , the lorentz factor @xmath0 of a cb stays put at its initial value @xmath31 , for the deceleration induced by the interactions with the ism has not yet had a significant effect . the doppler factor by which the light emitted by a cb is boosted in energy is given by eq . ( [ delta ] ) . since the emitted light is forward - collimated into a cone of characteristic opening angle @xmath32 , the boosted energetic radiation is easiest to detect for @xmath33 . thus , typically , @xmath34 . as a cb ploughs through the ism , fully ionized by the preceding @xmath0 radiation , it gathers and scatters the ism ions , mainly protons . these encounters are ` collisionless ' since , at about the time it becomes transparent to radiation , a cb also becomes ` transparent ' to hadronic interactions . as a consequence of momentum conservation , the scattered and re - emitted protons inevitably exert an inwards ` pressure ' on the cb . we have assumed that the main effect of this pressure is to slow the cb s expansion , posited to be relativistic at the emission time . in the approximation of isotropic re - emission in the cb s rest frame and a constant ism density @xmath35 , one then finds that , typically within minutes of observer s time @xmath36 , a cb reaches a roughly ` coasting ' radius , @xmath37 , which increases slowly until the cb finally stops and blows up ( dd2006 ) . up to the end of the coasting phase , and in a constant density ism , @xmath5 obeys ( ddd2002 ) : @xmath38 where @xmath39 if the ism particles re - emitted fast by the cb are a small fraction of the flux of the intercepted ones . in the opposite limit , @xmath40 . in the cb model of cosmic rays ( dd2006 ) , the observed spectrum strongly favours @xmath39 , used here in our fits . we have also concluded from previous analysis of swift x - ray data that @xmath41 is the right choice . as indicated by first - principle calculations of the relativistic merger of two plasmas ( frederiksen et al . 2004 ) , the ism ions continuously impinging on a cb generate within it turbulent magnetic fields , which we assume to be in approximate energy equipartition with the energy of the intercepted ism , @xmath42 . in this field , the intercepted electrons emit synchrotron radiation . the sr , isotropic in the cb s rest frame , has a characteristic frequency , @xmath43 , the typical frequency radiated by the electrons that enter a cb at time @xmath36 with a relative lorentz factor @xmath5 . in the observer s frame : @xmath44 ^ 3\ , \delta(t)\over 10^{12}}\ , \left[{n\over 10^{-1}\;\rm cm^3}\right]^{1/2 } { \rm hz}. \label{nub}\ ] ] where @xmath45 . the spectral energy density of the sr from a single cb at a luminosity distance @xmath46 is given by ( ddd2003a ) : @xmath47^{-1/2}\ , \left[1 + { \nu\over\nu_b(t)}\right]^{-(p-1)/2}\ , , \label{fnu}\ ] ] where @xmath48 is the typical spectral index ) is only correct for @xmath49 , for otherwise the norm diverges . the cutoffs for the @xmath50 distribution are time - dependent , dictated by the acceleration and sr times of electrons and their ` larmor ' limit . the discussion of these processes being complex ( ddd2003a , dd2006 ) , we shall satisfy ourselves here with the statement that for @xmath51 the ag s normalization is not predicted . ] of the fermi accelerated electrons , @xmath52 is the fraction of the impinging ism electron energy that is synchrotron re - radiated by the cb , and @xmath53 is the attenuation of photons of observed frequency @xmath50 along the line of sight through the cb , the host galaxy ( hg ) , the intergalactic medium ( igm ) and the milky way ( mw ) : @xmath54 . } \label{attenuation}\ ] ] the opacity @xmath55 at very early times , during the fast - expansion phase of the cb , may strongly depend on time and frequency . the opacity of the circumburst medium [ @xmath56 at early times ] is affected by the grb and could also be @xmath36- and @xmath50-dependent . the opacities @xmath56 and @xmath57 should be functions of @xmath36 and @xmath50 , for the line of sight to the cbs varies during the ag observations , due to the hyperluminal motion of cbs . these facts , the different @xmath58 dependences of the ics and sr emissions , and the dependence of the synchrotron ag on @xmath43 , are responsible for the complex observed chromatic behaviour of the ags . to a fair approximation , though , the deceleration bend , if occurring late enough , is achromatic from x - ray energies to the optical domain ( ddd2002 ) but not as far as radio ( ddd2003a ) . the swift x - ray bands are above the characteristic frequency @xmath59 in eq . ( [ nub ] ) at all times . it then follows from eq . ( [ fnu ] ) that the _ unabsorbed _ x - ray spectral energy density has the form : @xmath60 where we have used the customary notation @xmath61 . the functions @xmath62 and @xmath63 of eqs . ( [ delta],[deceleration ] ) evolve slowly , up until a time : @xmath64\;t_0\nonumber\\ \!\!\!\!\!\!\!\!&&\approx ( 130\,{\rm s})\ , [ 1 + 2\,\gamma_0 ^ 2\ , \theta^2]\,(1+z ) \left[{\gamma_0\over 10 ^ 3}\right]^{-3}\ , \left[{n\over 10^{-1}\ , { \rm cm}^{-3}}\right]^{-1 } \left[{r\over 10^{14}\,{\rm cm}}\right]^{-2 } \left[{n_{_{\rm b}}\over 10^{50}}\right ] \ ! , \label{tbreaks}\end{aligned}\ ] ] where we scaled the result to typical cb - model values of @xmath65 and a cb s baryon number , @xmath66 . the combination of the parameters @xmath35 , @xmath65 and @xmath66 appearing in eq . ( [ tbreaks ] ) is best constrained by the excellent x - ray observations discussed here . our previous results on optical and radio ags ( for fixed @xmath65 and @xmath66 ) favoured 10 times a smaller @xmath35 at the much larger sampled times , not an inconsistency , since a cb travels for @xmath67 light - days in one day of grb data . we have chosen to normalize @xmath35 as in eq . ( [ tbreaks ] ) , rather than to reproduce long discussions on the distributions of cb - model parameters ( e.g. de rjula 2007 ) . the quantity @xmath16 in eq . ( [ tbreaks ] ) characterizes the _ deceleration bend - time _ of the cb model ; eq . ( [ deceleration ] ) for @xmath68 describes the gradual character of this ` break ' . at later times ( [ deceleration ] ) implies that @xmath69 , and eq . ( [ delta ] ) that @xmath70 . thus , at @xmath71 , eq . ( [ fnux ] ) yields : @xmath72 with , as announced , a power decay in time half a unit steeper than in frequency . in the cb model , the peak energy of grbs satisfies : @xmath73 where @xmath74 ev is the typical energy of the glory s photons , that of the associated - supernova early light just prior to the ejection of cbs . the isotropic ( or spherical equivalent ) energy of a grb is ( dd2004 ) : @xmath75 where @xmath76 is the mean supernova early optical luminosity , @xmath77 is the number of cbs in the jet , @xmath78 is the comoving early expansion velocity of a cb ( in units of @xmath79 ) , and @xmath80 is the thomson cross section . for @xmath81 ( schaefer 2007 ) , the early sn luminosity required to produce the mean isotropic energy , @xmath82 erg , of ordinary long grbs , is @xmath83 , the estimated early luminosity of sn1998bw . all quantities in eq . ( [ eiso ] ) are normalized to their typical cb - model values . we have normalized to @xmath84 , an adequate mean number of prominent x - ray pulses in the subset of grbs analized here . the results in eqs . ( [ ep],[eiso ] ) are based on the assumption that ics is the mechanism generating the prompt radiation . they depend on @xmath18 and @xmath19 , two parameters also appearing in the description of the sr afterglow . that is why we shall be able to test the implied correlations , grb by grb , between the shape of the ag and the energetics of the prompt radiation , the very strong dependence of the @xmath85 on @xmath4 playing once more the major role . according to eqs . ( [ delta],[ep],[eiso ] ) , cbs with large @xmath18 , and , more so , small @xmath4 , produce the largest values of @xmath28 and @xmath27 : they generate the brightest grbs . according to eq . ( [ tbreaks ] ) , such @xmath86 values entail a small @xmath16 , an expectation that our analysis will validate . in such cases the deceleration bend or ` break ' of the synchrotron ag may take place before the beginning of the xrt observations and/or be hidden under the prompt compton emission . according to eq . ( [ fnux ] ) , these ags must be very luminous at early times , and according to eq . ( [ asymptotic ] ) , they must be well approximated from starters by the asymptotic power law behaviour given by eq . ( [ asymptotic ] ) . our analysis will verify all these predictions . to date , swift has detected and localized nearly 300 long grbs , and for most of them it followed their x - ray emission until it faded into the background . incapable of discussing all of them , we analyze the light curves of the x - ray afterglow of a set of grbs with and ` without ' jet breaks , which represent fairly well the entire spectrum of canonical and non - canonical x - ray afterglows of grbs . they include the most extreme cases of canonical and non canonical behaviour ( grb 0980425 and grb 061126 , respectively ) , the longest - measured canonical and non canonical x - ray light curves ( grb 060729 and grb 061007 ) and a variety of light curves with and without breaks , with and without superimposed x - ray flares . since many cb - model fits to canonical light curves of x - ray afterglows with ` breaks ' were included in previous publications ( ddd2002 , 2006 , 2007a , b , d ) we shall discuss in this paper more cases of grbs with an approximate power - law ag than of grbs with a canonical ag . we start the fits to the x - ray light curves during the transition between the rapid decline phase of the ics - dominated prompt emission to the sr - dominated ag phase . it suffices to include the ics contribution of the last prompt - emission pulse ( or the last two ) , because of an exponential factor in the pulse shape that suppresses very fast the relative contribution of the earlier pulses by the time the data sample the later ones ( ddd2007a , b ) . for the synchrotron contribution , it usually suffices to consider a common emission angle @xmath4 and an average initial lorentz factor @xmath18 for the ensemble of cbs . the ism density along the cbs trajectories is approximated by a constant . we then fit the entire observations of the x - ray ag of the selected grbs by using the ` master formula ' [ dd2004 , eq . ( 11 ) of ddd2007a ] , for the tail of the ics prompt emission contribution , and eq . ( [ fnux ] ) for the sr . many grbs have late x - ray flares , which we interpret as dying pangs of the engine , that is , the emission of cbs in late episodes of accretion into the recently collapsed central compact object . these cbs , whose ics - generated flares can only be seen on the weak background of a decaying sr ` after'-glow ( quote - unquote , since the ` ag ' is observable _ before _ the late ` prompt ' flares ) , are also modeled with the same master formula . the calculated shape of the energy - integrated x - ray ag , eq . ( [ fnux ] ) , depends only on three fit parameters . two of them are the product @xmath87 and the deceleration - bend time , @xmath16 , for an on - axis observer , as given in the first line of eq . ( [ deceleration ] ) . they determine the deceleration - bend time , @xmath16 , observed at a viewing angle @xmath4 , see eq . ( [ tbreaks ] ) . the third fit parameter is the index @xmath26 in the @xmath0 and @xmath85 time - dependent factors of eq . ( [ fnux ] ) . unlike in previous analysis , we let @xmath26 be a free parameter , unrelated to the spectral index @xmath88 , independently extracted by the observers from the shape of the x - ray spectrum . this way we shall be able to test explicitly the cb - model prediction implied by eq . ( [ fnux ] ) , @xmath89 , or by its more readable asymptotic form , eq . ( [ asymptotic ] ) . in all the cases we study , but two ( grbs 071020 and 050416a ) , a single cb or an ` average ' cb suffice to describe the ag . the occasional need for two cbs in the ag light - curve description is not a novelty . the most notable instance is that of grb 030329 ( ddd2003b ) . a comparison between the observed and predicted light curves of the 16 selected grbs is shown in figs . [ f2 ] to [ f5 ] . when well - measured , the ` break ' time , @xmath16 , is indicated in the figure by an arrow . the best fit values , of @xmath26 , @xmath87 , and @xmath16 are listed in table [ t1 ] , along with additional observational information on these grbs ( redshift , peak energy , equivalent isotropic energy , the start - time of the swift xrt observations , the spectral index of the unabsorbed ag , and the @xmath90 per degree of freedom of the fits ) . afterglows which exhibit nearly a pure power - law decline , have a @xmath16 smaller than @xmath91 , the time after trigger when the xrt started its observations of the ag , or a @xmath16 smaller than the time when the afterglow became brighter than the tail of the prompt emission . such ags have a nearly power - law shape , @xmath92 . their fits , however , return upper limits for @xmath93 and @xmath16 , above which the shape of the ag deviates from the data . these limiting values are also reported in table [ t1 ] , but the corresponding limit-@xmath16 location will not be shown in the figures in the case - by - case analysis , as it generally falls off - limits . in most cases ( including many swift ags studied in ddd2006 , 2007a , b , but not shown here ) , the cb model produces good fits with reduced @xmath90 values close to unity . even if the @xmath90 figures are good , we generally have refrained in the past from reporting them . one reason is that it is easy to obtain an excellent @xmath90 for a fit that has many data points , but misses some that clearly reflect a significant structure ( such as a supernova , see ddd2002 ) , or is , even within errors , systematically above the data in one region and below it in another . for that reason , and the occasional local scatter of the data , we consider the eye to be a better judge than any statistical measure . we comment on the @xmath90 values when they are ` bad ' . the values of @xmath12 , @xmath16 ( or @xmath94 ) , and @xmath26 returned by our fits and reported in table [ t1 ] have formal errors of a few percent . the error - correlation matrix has relatively small off - diagonal elements . the reduced-@xmath90 values are very close to unity , once the occasional flares are taken into account , to reveal the presence of a smoother sr background . one reason for all this is that @xmath94 sets the overall time scale , @xmath12 determines the shape of the bend , and @xmath26 is sensitive to the whole sr light curve , playing a major role in its power - law tail . this means that when a light curve is well sampled ( over orders of magnitude in flux and time ) , the fit is very sensitive to its parameters . naturally , the results depend also on the deceleration law , eq . ( [ deceleration ] ) , meant to be an approximation . therefore the extracted parameters have ` systematic ' errors reflecting the approximate nature of eq . ( [ deceleration ] ) . we can argue explicitly why the approximation should be better than it looks at first sight , even case by case ( on average , and independently , it leads to the correct spectrum of cosmic rays from non - relativistic energies up to the ` knee ' at some @xmath95 gev , dd2006 ) . the continuation of this rather formal argument on errors would take us well beyond the scope of this paper . in this section we comment one by one on the 16 grbs or xrfs whose x - ray light curves we discuss . the results of the cb - model fits are shown in figs . [ f2 ] to [ f5 ] , and the parameters relevant to our discussion are listed in table [ t1 ] . the first eight grbs are shown in the order of decreasing @xmath16 . for the next four , only an upper limit on @xmath16 can be extracted from the fits . the last four have very complex ags . the presence or absence of visible breaks in x - ray light curves and their different ` look ' the panoply of possibilities that we illustrate with our grb choices depend not only on @xmath16 , but on its value relative to @xmath91 ( the start - time of xrt observations ) and relative to the duration of the initial period of prompt - radiation dominance over the synchrotron ag . for this reason , it is easier to compare the plethora of looks of our grbs in an order slightly different from that of a decreasing @xmath16 . this we do ( only ) in the next paragraph . grbs 980425 and 060729 have light curves with a complete and simple canonical shape : one or two very clear prompt x - ray flares , a pronounced fast decay , a long plateau , a very visible ` break ' smoothly bending at @xmath16 to become a power - law decay . in grbs 050401 , 060105 , 060418 , 061007 and 050717 , the plateau is becoming less and less pronounced , so that the ag s @xmath16 is hiding better and better under the prompt signal , to the point that the last two are close to a pure power - law tail . in grbs 060813 , 070508 and 050505 , the prompt radiation ended early enough not to be caught by swift s xrt ( in the last case the follow - up started very late ) , but this trio displays very canonical ags , with their neat plateaus softly bending into a late power law . grbs 071025 , 061126 and 070125 are again approximate power - law tails , in which neither the early x - ray flares nor the putative bend are seen . grbs 071020 , 050416a and certainly 060607a , are very complex . the first two require contributions to the ag from two distinct cbs , 050416a having also a late flare . the unsightly x - ray light curve of grb 060607a can be described by the cb model without any new ingredients , but not much is learned from fitting it . * grb 980425 . * the light curve of this memorable single - peak grb , as observed by bepposax ( pian et al . 2000 ) , is shown in fig . the dotted line is the fit in ddd2002 , showing what we called a pronounced ` plateau ' . we have added to it the last ( predicted ) data point , measured with chandra by kouveliotou et al . ( 2004 ) , some 1285 days after burst ! to be consistent with the analysis here , we have re - fit the ensemble of data in the same manner as for all the other grbs to be discussed . the result is the continuous curve in the figure . this grb has , so far , the record large values @xmath96 and @xmath97 s , resulting in a light curve that rises before it falls , as explained in the introduction and illustrated in fig . this is the behaviour expected for far - off - axis grbs ( dd2000a ) . this one barely missed the official classification as an xrf : it s @xmath28 is @xmath98 kev , as opposed to @xmath99 kev ( see dado & dar 2005 for further comment on this point ) . * grb 060729 * and its x - ray light curve were studied in detail by grupe et al . it has a canonical shape , the longest follow - up observations with swift xrt , and the record - high @xmath100s , among the swift grbs . in fig . [ f2]b we show its cb - model description with , superimposed on its prompt decline phase , four ics x - ray flares included in the fit , as discussed in detail in ddd2007a . this grb being ` canonical ' and having a very clear ` break ' as several others also discussed in ddd2007a is included here to illustrate the start of the transition from ` breaks ' to ` missing breaks ' . although the best fit to the x - ray ag appears to be excellent , it yields a large @xmath101 , mostly due to many far - flung isolated data points in the swift data . more accurate data from xmm newton ( grupe et al . 2007 ) do not show such outliers . eliminating their contribution yields the @xmath90 value reported in table 1 . * grb 050505 , * whose x - ray light curve was studied in detail by hurkett et al . ( 2006 ) . at @xmath102 , this grb is amongst the most distant with a known redshift . due to an earth - limb constraint , swift was unable to slew to it until 47 minutes after the grb s trigger , and started measuring its x - ray light curve only 2883s after burst , during the transition of the ag from its shallow - decline phase to a power - law decline . as can be seen from fig . [ f2]c , the cb model describes very well the xrt light curve , except when the counting rate becomes comparable to the background . * grb 050401 , * whose x - ray light curve , studied in detail by de pasquale et al . ( 2006 ) , evolves smoothly from the tail of the prompt emission at around 200s to a short decaying plateau , which suavely breaks into a power - law decline at @xmath103s . its cb - model description is shown in fig . this grb had a very bright x - ray ag , even though it originated at a fairly large redshift , @xmath104 , and had a very large extragalactic absorbing column density along its sight - line , inferred from its x - ray spectrum to be @xmath105 ( de pasquale et al . 2006 ) , or @xmath106 ( watson et al . such a column density implies a very strong extinction of the optical ag and , consequently , an extreme chromaticity : more than 10-magnitude extinction in the v band ( zombeck 1990 ) and more than 30 magnitudes at 1800 @xmath107 . indeed , the optical ag was very dim ( a fitted spectral index , @xmath108 , between the x - band and the optical band , compared to @xmath109 to @xmath110 in ` normal ' grbs . in fact , according to jakobson et al . ( 2004 ) , grb 050401 qualified as a ` dark burst ' . it would not be a good case to discuss ( unattenuated ) chromaticity , or the lack thereof . * grb 070508 . * swift s xrt started to measure the x - ray light curve of this grb 82s after the grb trigger . even at this early time , it already displays the shallow - decay plateau phase of a canonical ag , which later bends into a power - law decline , as shown in fig . [ f3]a . * grb 060813 * , shown in fig . [ f3]b , is a case in which the prompt radiation is not seen by the xrt , and the ag has no obvious flares . in spite of some evidence for local variability , the smoothly bending ag is well described by the cb model ( @xmath111_dof _ @xmath112 for 254 _ dof _ ) . had the break happened a bit earlier , as in other cases , the x - ray ag would look like a power - law . the last data point lies below the fit , it could be due to an overestimated background . * grb 060418 , * whose achromatic ag was studied in detail by molinari et al . its x - ray ag evolves fast into a power - law decline , see fig . the cb - model fit returns an early break at @xmath113s , well hidden under the flaring activity during the fast - decline phase of the prompt emission . the transition from an ics - dominated regime to one in which sr is prevalent is corroborated by the fast spectral softening of the tail of the flare from around @xmath114s ( evans et al . 2007 ) , which suddenly turns , at @xmath115s , into the much harder time - independent power - law spectrum characteristic of the synchrotron ag ( ddd2007b ) . we have checked that the reasonable @xmath111_dof _ @xmath116 for 295 _ dof _ of the fit shown in the figure can be reduced to @xmath111_dof _ @xmath117 by including x - ray flares between 5 to 10 ks , or by replacing the fluctuating data points by average values . * grb 050717 , * studied in detail by krimm et al . it had the largest inferred peak energy of all swift grbs , @xmath118 kev , despite its estimated large redshift , @xmath119 at this @xmath120-limit , @xmath121 erg , and the local peak energy is @xmath122 kev . it also had an initially very bright x - ray ag , after the fast declining prompt emission , with a power - law decline from @xmath123s onwards . the fit in fig . [ f3]d returns an early break - time limit , @xmath124s , well hidden under the prompt - emission tail . the cb - model interpretation of the transition from a prompt ics radiation to a synchrotron ag is supported by the observed rapid spectral softening of the tail of the prompt emission and its sudden change at @xmath123s into the harder time - independent power - law spectrum of the synchrotron ag ( ddd2007b ) . in the case of this grb the best - fit value , @xmath125 , does not satisfy eq . ( [ asymptotic ] ) , with @xmath126 , as inferred from the x - ray spectrum with a fixed column density limited to the galactic one ( krimm et al . however , @xmath125 is consistent with @xmath127 of the ag for @xmath128s , the spectral index reported by zhang et al . ( 2007 ) , after inclusion of host - galaxy and igm absorption . * grb 061126 , * studied in detail by perley et al . ( 2008 ) had two major prompt pulses . due to an earth - limb constraint , swift slewed to the burst s direction only 23 minutes after its localization by its burst alert telescope ( bat ) . its light curve , measured by the xrt between 1.6 ks and 1.88 ms , is shown in fig . the x - ray light curve was reported to be well fit by a power - law in time with index @xmath129 ( sbarufatti et al . 2006 ) . a cb - model fit , with @xmath130 and @xmath131s , is shown in fig . there is a possible indication in the data of a steeper decay between 1.6 ks and 3.6 ks , which might belong to the tail of another cb with a smaller @xmath16 . cases of ags clearly requiring two cbs will be discussed anon . * grb 071025 . * swift s xrt started observations of the x - ray light curve 146s after the bat trigger . the initial relatively hard spectrum ( @xmath132 ) softened beyond 300s and the light curve declined like a single power - law , consistent with the cb - model s asymptotic power - law decline with a power - law index @xmath133 , as shown in fig . the data suggest a flaring activity between 4 ks and 40 ks . the effect of such flares on the cb - model x - ray light curve is illustrated in the figure by adding an ics flare around 40 ks with parameters ( peak time , width and normalization ) chosen , as in all other cases with clear flares , to best fit the data . * grb 070125 , * studied in detail by bellm et al . it was detected by mars odyssey , suzaku , integral , and rhessi . it is one of the swift - era grbs with the largest measured values of @xmath134 erg , @xmath135 , and source - frame @xmath136 kev . the initial detection of this grb occurred while it was not in the bat field of view during the beginning of the prompt emission , and its xrt light curve starts at 46 ks after the burst . as shown in fig . [ f4]c its power - law decline is well described by the cb model . the feature at @xmath137 ks can be interpreted as an x - ray flare , as in the figure . * grb 061007 , * whose ag was studied in detail by schady et al . ( 2006 ) and mundell et al . ( 2007 ) , was the brightest grb detected by swift and was accompanied by an exceptionally luminous x - ray and uv / optical afterglow , which decayed as a power law with an index @xmath138 . it had the largest values of @xmath134 erg , @xmath139 and an emission - point peak energy , @xmath140 kev ( golenetskii et al . this grb is the best example to date of a bright x - ray ag , well - sampled from the start of the xrt observations ( 86s after the bat trigger ) to @xmath141 s. the ag , shown in fig . [ f4]d , is power - law behaved right after the tail of the prompt emission . the cb - model fit returns @xmath142s , below which the @xmath111_dof _ ( a reasonable 1.13 for 1030 _ dof ) _ stays put . * grb 071020 , * measured by swift s xrt between 68s and 1.7 ms after trigger , and shown in fig . holland et al . ( 2007 ) fitted the data with a broken power - law with an initial decay index of @xmath143 , a break at @xmath144s , and a late - time decay index of @xmath145 . the fit is poor between 1.5 ks and 1.5 ms . a cb - model fit with a single cb is also unsatisfactory . the addition of a second cb to the ag s description , as in the fit shown in fig . [ f5]a , greatly improves the fit to @xmath111_dof _ @xmath146 for 174 _ dof _ , acceptable in view of what appears to be evidence for flaring activity , from 1.5 to 15 ks , which we have not endeavoured to describe , given the scarcity of data . * xrf 050416a . * the complex x - ray light curve of this xrf was monitored up to 74 days after the burst ( mangano et al . the late decline rate of the light curve is significantly slower than expected in the cb model from the observed photon spectral index @xmath88 , namely @xmath147 . the prompt signal of xrf 050416a had two clear pulses which , in the cb model , correspond to two separate cbs . the x - ray light curve , modeled with two cbs and shown in fig . [ f5]b , has a sr - component late - power decay that although it is not readable ` by eye ' due to the late - occurring ics flare is compatible with the predicted one . * grb 060105 , * whose x - ray light curve was studied in detail by tashiro et al . ( 2007 ) . following the prompt emission , which ended with a very steep decay , the light curve is canonical , it has a shallow decay after 180s and steepens at around 500s to a fast power - law decline , with a weak flaring activity superposed on it . the deviations from a smooth x - ray light curve may be caused by the flaring activity , not included in this particular fit , whose @xmath111_dof _ @xmath148 for 854 _ dof , _ is not inadequate . * grb 060607a , * was studied by molinari et al . its complex x - ray light curve , like that of quite a few other grbs , is dominated by strong flaring activity , as can be seen in fig . [ f5]d , with its many flares superimposed on the ag of a fitted , single , dominant cb . this fit , which can be improved by splitting the last flare into two , is a very rough description ( @xmath111_dof _ @xmath149 for 440 _ dof _ ) , not a proof of the quality of a prediction . moreover , in cases with such a prominent flaring activity , the photon spectral index of the ag data is an average between the typical index of flares , @xmath150 , and that of synchrotron ag , @xmath151 , i.e. , an average significantly smaller than that of the synchrotron ag . thus , we do not expect such a labyrinthine ag to satisfy the cb - model spectral - index relations , eqs . ( [ fnux],[asymptotic ] ) . we have summarized in eq . ( [ fnu ] ) the predicted form of the spectral energy density of the ag of a grb , in which the time - dependence and the energy - dependence are explicitly concatenated . in the large - frequency limit of the x - ray domain , the expression simplifies to that of eq . ( [ fnux ] ) , implying a predicted relation between the temporal index @xmath26 ( which we fit to the xrt light curve of the x - ray ag ) and the spectral index @xmath88 , independently fitted by the swift team to the x - ray ag spectrum after correcting for attenuation , and reported by zhang et al . the prediction is particularly simple , and is most transparently readable in the late-@xmath36 limit for the ags dependence on @xmath36 and @xmath50 , eq . ( [ asymptotic ] ) , in which both the time and the frequency functional forms are separate power laws . the values of @xmath26 and @xmath88 are listed in table [ t1 ] . notice that @xmath88 varies over a significant range of central values , 1.61 to 2.25 , and that the measurements are not compatible within errors with a common value . to illustrate the prediction in eqs . ( [ fnux],[asymptotic ] ) , we have plotted in fig . [ f6 ] the ratio @xmath152 ( predicted to be unity ) for the various grbs analized in this paper , and added a few other analized in the same fashion . the results are quite satisfactory . the mean value of @xmath153 , for instance , is @xmath154 for the grbs analized here , @xmath155 for the ensemble plotted in fig . [ f6 ] . in the cb model , the functional dependence on @xmath4 and @xmath18 of the deceleration - bend time of the synchrotron ag , @xmath16 , as well as its normalization , are specified by eq . ( [ tbreaks ] ) . this is also the case for the parameters , @xmath28 and @xmath27 of eqs . ( [ ep ] ) and ( [ eiso ] ) , of the prompt ics signal . as we saw in the introduction , this implies explicit correlations between @xmath16 and the prompt observables . the @xmath156 correlation is illustrated in fig . [ f1]b for various choices of @xmath4 and @xmath18 , with the rest of the parameters in @xmath16 and @xmath27 fixed to reference values in eqs . ( [ tbreaks],[eiso ] ) . in fig . [ f7 ] we plot , in the @xmath157 $ ] plane , the values returned by our analysis of the grbs we have discussed , see table [ t1 ] . the grbs represented by arrows reflect the fact that some data are just upper limits . the large shaded contour plot in the figure is the boundary of the domain covered by letting @xmath18 vary from 500 to 1500 , @xmath4 from 0 to 8 mrad , typical ranges encountered in the cb - model analysis of grbs . moreover , the normalization of @xmath16 in eq . ( [ tbreaks ] ) was varied from its central value in eq . ( [ tbreaks ] ) to 1/2 order of magnitude above it , and the normalization of @xmath27 in eq . ( [ eiso ] ) from its central value to 1/2 order of magnitude below and above it . the variability in these normalizations is best ascertained by the current analysis , it has been chosen to make fig . [ f7 ] ` look good ' . we have added to the figure the results for a few grbs which we have previously analyzed along the same lines in ddd2007a , b . there is no reason to expect the data to populate uniformly the region bounded by the contour in fig . [ f7 ] . on the contrary . the relativistically beamed radiation from a point in a cb initially subtends an angle @xmath158 . observers at an angle @xmath4 from the axial direction have a chance @xmath159 of being illuminated . at @xmath160 this chance decreases abruptly , given the fast fall of the doppler factor . all in all , @xmath161 is the optimal observation angle , for _ any _ @xmath18 . most grbs , then , should be seen at @xmath162 . the thick straight line in fig . [ f7 ] is @xmath163 at fixed @xmath12 , for which @xmath164 and @xmath165 . thus : @xmath166 the data follow this trend well , but at the high-@xmath27 end , at which they bend as in fig . [ f1]b . in fig . [ f8 ] we plot , in the @xmath167 $ ] plane , the corresponding results of our analysis . the shaded domain is obtained with the same ranges in @xmath18 , @xmath4 and in the normalization of @xmath16 as in the previous paragraph . the normalization of @xmath28 has been allowed to vary from 1/3 to 1/6 of its value in eq . ( [ ep ] ) of eq . ( [ ep ] ) is the peak energy at the start of a pulse ; set it to @xmath168 . the energy of the radiation is predicted to decrease during the pulse s duration : @xmath169 $ ] , with @xmath170 the width parameter ( the full width at half - maximum , fwhm , is @xmath171 ) . observers usually report @xmath28 at the peak s maximum , expected to be @xmath172 , or its pulse - averaged value : @xmath173 over the fwhm . we have not corrected for these facts , which may explain the choice of the ` best ' domain . ] . the points plotted as ellipses have an unknown @xmath120 , which we have let vary from 0 to 2.75 , the average for swift - era gbrs ( greiner , @xmath174 ) . at fixed @xmath12 , @xmath175 , so that : @xmath176^{-3/2}. \label{naivetbep}\ ] ] the rest of the comments are as in the discussion of the @xmath156 correlation . another direct way to ascertain the variability of the parameters governing the normalizations of @xmath28 and @xmath27 is to study their scatter plot ( dd2000b , amati et al . 2002 , dd2004 , ddd2007c , amati 2006 ) for a large collection of grbs and xrfs . this is done in fig . [ f9 ] , where the varying - power correlation predicted by the cb model ( ddd2007c ) is shown , and to which the grbs with known @xmath120 , @xmath28 and @xmath27 , among those studied here , are added . the figure shows that a total uncertainty of a factor of 2 in the norm of @xmath28 and of one an order of magnitude in the norm of @xmath27 ( as we have adopted ) is adequate to bracket the data . we have also tested elsewhere ( ddd2007d ) the correlation , apparent in fig . [ f1]a , between @xmath16 and the normalization of the ag . willingale et al . ( 2007 ) and nava et al . ( 2006 ) had collected and analyzed a large set of grbs , and made a scatter plot of @xmath16 versus the total ag energy in the 15 - 150 kev x - ray band up to time @xmath16 . to use this available information , we studied this correlation in its willingale - nava form . like the ones in figs . [ f7 ] and [ f8 ] , it turns out to follow the pattern expected in the cb model . the correlations between @xmath16 and @xmath27 or @xmath28 demonstrate that ` sub - energetic ' grbs ( or xrfs ) have large ` break ' times and , consequently , easily observable ` breaks ' . as grbs become more ` energetic ' , @xmath16 decreases and the chances increase to ` miss the break ' , which may be hidden under the prompt radiation , or may precede the swift slew - time minimum , or the start of the observations . a virtue of astrophysical x - ray data is that , in many instances and relative to lower - frequency bands , the corrections for attenuation are simpler and more reliable . the strength of swift in dealing with transient phenomena is , as the satellite s name reflects , the prompt start of its data - taking after an alert . this has made the swift results an excellent testing ground for theories of grbs and xrfs . in particular , the ability to monitor the x - ray flux over a very wide range of times has provided decisive tests of the theoretical predictions . filling the pre - swift gap in the data between ` prompt ' and ` afterglow ' radiations has led to a better understanding of the mechanisms responsible for them . in the cb model they are different : inverse compton scattering and synchrotron radiation , respectively . we have previously argued that the strong case for an ics origin of the ` prompt ' radiation ( dd2004 ) has been reconfirmed by the analysis of the swift x - ray flares , and the fast decay of the prompt signals ( ddd2007b ) . in this respect the study of x - ray and optical data is also particularly meaningful ( ddd2007a ) . the observed correlations between prompt observables @xmath27 , @xmath28 , @xmath29 and pulse rise - time , lag - time and variability also agree with the cb - model ( see ddd2007c and references therein ) . these correlations follow from the same simple considerations , that we have emphasized in this paper , on the dependence of the cited prompt observables on the lorentz and doppler factors of the radiation emitted by a quasi - point - like relativistically moving source . the cb - model s expectations for the interplay between ics and sr were confirmed by the analysis of the ` canonical ' shape of many swift x - ray light curves ( ddd2007a ) . the extreme canonical case is still grb 980425 , shown in fig . ( [ f2]a ) . the trend of the ` hardness ratios ' reflecting the spectral behaviour , and the spectral index itself , also corroborate the expected transitional behaviour ( ddd2007b ) , as the dominant mechanism evolves from ics to sr . in this paper we have shown in detail how the variety of x - ray ag shapes , with and without ` breaks ' , is also to be expected from a decelerating jet of effectively pointlike cannonballs , as in fig . ( [ f1]a ) . that the ag emission mechanism is sr from cbs slowing down in the way approximated by eq . ( [ deceleration ] ) is confirmed by the detailed frequency and time - dependences of eq . ( [ fnu ] ) for the spectral energy density . we have presented a study of the correlation between the synchrotron ags @xmath36- and @xmath50- dependences , specified in the x - ray domain by eq . ( [ fnux ] ) . this results in a relation between the ags spectral index @xmath88 , and the index @xmath26 appearing in their time dependence , a very simple relation at the late times at which the time dependence is also a power - law , see eq . ( [ asymptotic ] ) . the prediction is tested in fig . [ f6 ] . in the cb model , the understanding of ags with breaks or no breaks turns out to be clear : the ` missing ' sr breaks are hiding under the prompt ics radiation , or occur too early to be seen . this sounds like a trivial and model - independent excuse . it is not . it is supported by our case - by - case analysis of ag shapes . moreover , a crucial ingredient the angle of observation of the jet , compared to the beaming angle of its doppler - boosted radiation is validated by the correlations , e.g. the luminous ags are the ones with early or even undetectable breaks , as in fig . [ f1]a , and as in many of the examples we discussed here ( the correlation between @xmath16 and the energy in the x - ray ag was studied in ddd2007d ) . our conclusions are also supported by the correlations between the cbs deceleration - bend ` break - times ' , @xmath16 ( in the synchrotron ags ) , and the values of @xmath27 and @xmath28 ( in the prompt compton signal ) . these correlations , shown in figs . [ f7 ] and [ f8 ] , reconfirm the consistency of the overall picture . we have given no comment in the conclusions to our fits to the grbs and xrfs that we have studied . this is because the point we would like to make is _ not _ that the cb model can be used to fit the data very well . the main issue , in our view , is _ how _ a model , preferably in a predictive manner and in terms of very few concrete concepts like its radiation mechanisms , the aperture of its jets and the angle from which they are viewed can be used to understand the ensemble of long - duration grbs , and xrfs . after all , phenomena that require ever - increasingly complex explanations are of limited scientific interest . 980425 & 0.0085 & 56 & 6.9 e47 & 2.1 @xmath177 ? & 2.20 & 9.17 & 145000 & 36000 & 31/0 + 060729 & 0.54 & & @xmath178 7 e51 & @xmath179 & 2.20 & 2.51 & 8300 & 130 & 1966/207 + 050505 & 4.27 & 214 & & @xmath180 & 2.22 & 1.57 & 1980 & 2833 & 114/95 + 050401 & 2.90 & 132 & 3.5 e53 & @xmath181 & 2.20 & 0.80 & 1660 & 133 & 353/299 + 070508 & 0.82 ? & 188 & 7.0 e52 & @xmath182 & 2.12 & 1.22 & 260 & 82 & 610/469 + 060813 & & 214 & & @xmath183 & 1.70 & 1.13 & 190 & 85 & 256/239 + 060418 & 1.49 & 230 & 9 e52 & @xmath184 & 2.20 & 1.73 & 123 & 84 & 339/280 + 050717 & @xmath185 2.7 ? & 2401 & @xmath185 1 e54 & @xmath186 & 1.67 & ( 0.08 ) & @xmath178 55 & 91 & 114/78 + 061126 & 1.159 & 620 & 1.1 e53 & @xmath187 & 1.89 & ( 1.87 ) & @xmath178 104 & 1604 & 506/261 + 071025 & & & & & 2.20 & ( 0.90 ) & @xmath178 68 & 150 & 330/243 + 070125 & 1.547 & 440 & 9.4 e53 & @xmath188 & 2.38 & ( 1.19 ) & @xmath178 8060 & 47000 & 28/28 + 061007 & 1.261 & 498 & 1.0 e54 & @xmath189 & 2.26 & ( 0.05 ) & @xmath178 89 & 86 & 1147/1015 + 071020 & 2.145 & 322 & 8.0 e52 & @xmath190 & 1.86 & 0.67 & 90 & 68 & 234/154 + @xmath191 `` & & & & & 1.86 & 1.43 & 15100 & 68 & + 050416a & 0.6535 & 15 & 1.2 e51 & @xmath192 & 2.00 & 1.05 & 944 & 85 & 101/92 + @xmath191 '' & & & & & & 2.00 & 14800 & 85 & + 060105 & & 424 & & @xmath193 & 2.33 & 0.53 & 510 & 96 & 1879/839 + 060607a & 3.082&&& & 2.20 & 0.97 & 164 & 73 & 1119/404 + + the values of the peak energy , @xmath28 ( in kev ) and @xmath27 ( in erg ) of the grbs are from gcn reports of data of konus - wind , rhessi and suzaku . the grb redshifts are from gcn reports from ground - based optical telescopes . the start times @xmath91 , of the xrt data after the bat trigger , are from the swift repository ( evans et al . the unabsorbed spectral indices @xmath88 are from swift gcn reports and zhang , liang & zhang ( 2007 ) . the cb - model fits return @xmath26 , @xmath93 and @xmath16 . the parenthesized @xmath93 are for @xmath16 at its upper limit .